Digital Measurement What Whereby How

Embedded Systems Engineering WS10Armin Wasicek

Digital Measurement 2

Overviewbull What

ndash Measurement Unitsndash Signals

(Classification Sampling Theorem Noise)bull Whereby

ndash Sensors hellipbull How

ndash Measurement Errorndash Measurement with microcontroller

(analog comparator ADC DAC Codes and value representation)ndash Data processing and filtering

Digital Measurement 3

Measurement Units

A measurement consists of the product of the measurand value and its unit

x = v ux the measurementv a number representing the measurand value (the physical or chemical

quantity property or condition that is measured)u the respective unit

bull Standardized by the international system of units(Systegraveme International dlsquoUniteacutes) SI-System

Digital Measurement 4

SI-System - Unitsbull 7 Base units

kg m s A K mol cd

bull Derived Units (examples)Hz [1s] N [(kgmiddotm)ssup2)W [(kgmiddotmsup2)s] Pa [kg(mmiddotssup2)]degC [T_kelvin ndash 27315] Joule [(kgmiddotmsup2)ssup2]

Digital Measurement 5

SI-System - Unit-Prefixkilo k 103 milli m 10-3

mega M 106 micro micro u 10-6

giga G 109 nano n 10-9

tera T 1012 pico p 10-12

peta P 1015 femto f 10-15

exa E 1018 atto a 10-18

zetta Z 1021 zepto z 10-21

yotta Y 1024 yocto y 10-24

Digital Measurement 6

Signals - Classification

Digital Measurement 7

Signals ndash Sampling Theorem

bull Sampling is the process of converting an signal into a numeric sequence (analog value to a discrete value)

bull A band-limited (0-fmax) time dependent function f(t) can be reconstructed by sample-points if sampling frequency is more than the twice of fmax

fsample gt 2middotfmax

Digital Measurement 8

Signals - Noise

bull Noise is a stochastic changing of currentvoltage of a signal which is caused by several affectsndash Thermal noisendash Atmospheric and GalacticCosmic noise

bull Impact of noise on the measurement should be minimized by appropriate design measures egndash Shielding to reduce the impact of external noisendash Appropriate hardware layout to eliminate avoidable sources of

internal noise(more details are given in the lecture bdquoHardware Design for Embedded Systemsldquo)

Digital Measurement 9

Signals ndash Noise Types|X(f)| [dB]

10 100 1000 10000f log0

White Noise

|X(f)| [dB]

10 100 1000 10000f log0

Pink Noise

|X(f)| [dB]

10 100 1000 10000f log0

BrownRed Noise-40-40

|X(f)| [dB]

10 100 1000 10000f log0

Blue Noise

|X(f)| [dB]

10 100 10000f log0

Purple Noise

1 fsup2-6dBOctave

+3dBOctave

-40-40

|X(f)| [dB]

10 100 1000 10000f log0

Gray Noise

1000

+6dBOctave

1 f-3dBOctave

White Noise (constant)Gray Noise (approxto const

psychoacoustic loudness)Pink Noise (-3dBOctave)BrownRed N (-6dBOctave)Blue Noise (+3dBOctave)Purple Noise (+6dBOctave)

10

Sensors ndash Some Definitions

bull A device that responds to a physical or chemical stimulus (such as heat pressure flow acceleration etc) and affects or generates an electrical signal

bull Facilitates to quantitatively or qualitatively acquire the physical or chemical properties of an object

bull A sensor is a transducer that converts the measurand into a signal carrying information

11

Classification of Sensors (1)

bull Passive Sensorsndash A sensor whose physical measurement variable

controls or affects the energy of somethingsomeone else

ndash eg strain gauge (ldquoDMSrdquo) capacitive sensors

bull Active Sensorsndash A sensor that generates by itself some form of energy

as its measurement signalndash eg photo transistor piezo-electric sensors

12

Classification of Sensors (2)

bull Type of measurandndash Mechanical quantities (eg pressure position force)ndash Thermal quantities (eg temperature heat flow)ndash Electrostatic and magnetic fieldsndash Radiation intensity (eg electromagnetic)ndash Chemical quantities (eg humidity gas)ndash Biological quantities (eg antigens antibodies)

13

Classification of Sensors (3)

bull Nature of Output Signalndash analog output continuous signal in its magnitude andor temporal

(eg temperature)ndash digital output output signal in the form of discrete steps or states

(eg switch)

bull Physical Measurement Variablendash resistancendash inductancendash capacitancendash etc

selecting the most appropriatesensor is not a trivial task

14

Selecting a Sensor (1)bull What should be measured

ndash distance brightness hellipndash sometimes it is easier to measure a related value (voltage

instead of current)

bull Can we access the physicalchemical property directlyndash eg temperature within a melting pot

bull How should it be measuredndash measuring wheel propagation delay triangulationndash required precision

15

Selecting a Sensor (2)bull Interface

ndash digital analog restrictions dynamics hellip

bull Requirements in the field of applicationndash mechanical stress (eg heat pressure etc)ndash costs (mass production vs prototype setup)

bull Effects biasing the measurementndash Never forget the physics behind the sensorndash Example

bull two infrared sensors (triangulation)bull almost all measurements are rubbishbull beam from sensor A was detected by senor B

16

Some Common Sensing MethodsMeasurand Method

Displacement Position resistive capacitive opto-electronic Hall effect variable reluctance

Distancetriangulation measuring wheel radar echelon capacitiveinductiveproximity

Temperature Thermistor (NTC PTC) infrared radiation thermocouple

Pressure piezoresisitve capacitive piezoelectric strain gauge

Velocity Hall effect opto-electronic variable reluctance

Luminance photo-resistor photo-diode photo-transistor

17

Resistive Sensors

bull Common passive sensors variation of resistance R

bull Potentiometer ndash changing L due to mechanical displacement (slide rotation)ndash liquid level sensor rotation and angle sensor

bull Thermistor ndash change of resistance due to change of NTC (negative k) PTC (positive k)ndash heat sensor heat flow sensor

bull Piezoresistive Sensors ndash resistor diffused in silicon compression decreases resistance tension increases resistancendash mechanical stress

R k T

18

Resistive Sensors

bull Potentiometer ndash eg angle sensor (105deg) 5k Ohm linear

bull Thermistor ndash eg PTC with range 0 hellip 55deg C

(figures taken from RS componentsonline catalogue)

19

Capacitive Sensorsbull Common passive sensors variation of capacitance C

bull Pressure sensorndash typically capacitor diffused into silicon chipndash distance between capacitor plates is varied due to mechanical stress

bull Liquid level sensorndash typically two or three electrodes (measured object itself forms one

electrode)

bull Capacitive proximity switchndash contact-free detection of liquids or solid materials

20

Revolution Sensor

bull Speed indicator signal provided by DC fan (ESE-LU board)ndash special kind of onoff switchndash two pulses per revolution

(figure taken from ebmpapst DCfan data sheet)

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 2

Overviewbull What

ndash Measurement Unitsndash Signals

(Classification Sampling Theorem Noise)bull Whereby

ndash Sensors hellipbull How

ndash Measurement Errorndash Measurement with microcontroller

(analog comparator ADC DAC Codes and value representation)ndash Data processing and filtering

Digital Measurement 3

Measurement Units

A measurement consists of the product of the measurand value and its unit

x = v ux the measurementv a number representing the measurand value (the physical or chemical

quantity property or condition that is measured)u the respective unit

bull Standardized by the international system of units(Systegraveme International dlsquoUniteacutes) SI-System

Digital Measurement 4

SI-System - Unitsbull 7 Base units

kg m s A K mol cd

bull Derived Units (examples)Hz [1s] N [(kgmiddotm)ssup2)W [(kgmiddotmsup2)s] Pa [kg(mmiddotssup2)]degC [T_kelvin ndash 27315] Joule [(kgmiddotmsup2)ssup2]

Digital Measurement 5

SI-System - Unit-Prefixkilo k 103 milli m 10-3

mega M 106 micro micro u 10-6

giga G 109 nano n 10-9

tera T 1012 pico p 10-12

peta P 1015 femto f 10-15

exa E 1018 atto a 10-18

zetta Z 1021 zepto z 10-21

yotta Y 1024 yocto y 10-24

Digital Measurement 6

Signals - Classification

Digital Measurement 7

Signals ndash Sampling Theorem

bull Sampling is the process of converting an signal into a numeric sequence (analog value to a discrete value)

bull A band-limited (0-fmax) time dependent function f(t) can be reconstructed by sample-points if sampling frequency is more than the twice of fmax

fsample gt 2middotfmax

Digital Measurement 8

Signals - Noise

bull Noise is a stochastic changing of currentvoltage of a signal which is caused by several affectsndash Thermal noisendash Atmospheric and GalacticCosmic noise

bull Impact of noise on the measurement should be minimized by appropriate design measures egndash Shielding to reduce the impact of external noisendash Appropriate hardware layout to eliminate avoidable sources of

internal noise(more details are given in the lecture bdquoHardware Design for Embedded Systemsldquo)

Digital Measurement 9

Signals ndash Noise Types|X(f)| [dB]

10 100 1000 10000f log0

White Noise

|X(f)| [dB]

10 100 1000 10000f log0

Pink Noise

|X(f)| [dB]

10 100 1000 10000f log0

BrownRed Noise-40-40

|X(f)| [dB]

10 100 1000 10000f log0

Blue Noise

|X(f)| [dB]

10 100 10000f log0

Purple Noise

1 fsup2-6dBOctave

+3dBOctave

-40-40

|X(f)| [dB]

10 100 1000 10000f log0

Gray Noise

1000

+6dBOctave

1 f-3dBOctave

White Noise (constant)Gray Noise (approxto const

psychoacoustic loudness)Pink Noise (-3dBOctave)BrownRed N (-6dBOctave)Blue Noise (+3dBOctave)Purple Noise (+6dBOctave)

10

Sensors ndash Some Definitions

bull A device that responds to a physical or chemical stimulus (such as heat pressure flow acceleration etc) and affects or generates an electrical signal

bull Facilitates to quantitatively or qualitatively acquire the physical or chemical properties of an object

bull A sensor is a transducer that converts the measurand into a signal carrying information

11

Classification of Sensors (1)

bull Passive Sensorsndash A sensor whose physical measurement variable

controls or affects the energy of somethingsomeone else

ndash eg strain gauge (ldquoDMSrdquo) capacitive sensors

bull Active Sensorsndash A sensor that generates by itself some form of energy

as its measurement signalndash eg photo transistor piezo-electric sensors

12

Classification of Sensors (2)

bull Type of measurandndash Mechanical quantities (eg pressure position force)ndash Thermal quantities (eg temperature heat flow)ndash Electrostatic and magnetic fieldsndash Radiation intensity (eg electromagnetic)ndash Chemical quantities (eg humidity gas)ndash Biological quantities (eg antigens antibodies)

13

Classification of Sensors (3)

bull Nature of Output Signalndash analog output continuous signal in its magnitude andor temporal

(eg temperature)ndash digital output output signal in the form of discrete steps or states

(eg switch)

bull Physical Measurement Variablendash resistancendash inductancendash capacitancendash etc

selecting the most appropriatesensor is not a trivial task

14

Selecting a Sensor (1)bull What should be measured

ndash distance brightness hellipndash sometimes it is easier to measure a related value (voltage

instead of current)

bull Can we access the physicalchemical property directlyndash eg temperature within a melting pot

bull How should it be measuredndash measuring wheel propagation delay triangulationndash required precision

15

Selecting a Sensor (2)bull Interface

ndash digital analog restrictions dynamics hellip

bull Requirements in the field of applicationndash mechanical stress (eg heat pressure etc)ndash costs (mass production vs prototype setup)

bull Effects biasing the measurementndash Never forget the physics behind the sensorndash Example

bull two infrared sensors (triangulation)bull almost all measurements are rubbishbull beam from sensor A was detected by senor B

16

Some Common Sensing MethodsMeasurand Method

Displacement Position resistive capacitive opto-electronic Hall effect variable reluctance

Distancetriangulation measuring wheel radar echelon capacitiveinductiveproximity

Temperature Thermistor (NTC PTC) infrared radiation thermocouple

Pressure piezoresisitve capacitive piezoelectric strain gauge

Velocity Hall effect opto-electronic variable reluctance

Luminance photo-resistor photo-diode photo-transistor

17

Resistive Sensors

bull Common passive sensors variation of resistance R

bull Potentiometer ndash changing L due to mechanical displacement (slide rotation)ndash liquid level sensor rotation and angle sensor

bull Thermistor ndash change of resistance due to change of NTC (negative k) PTC (positive k)ndash heat sensor heat flow sensor

bull Piezoresistive Sensors ndash resistor diffused in silicon compression decreases resistance tension increases resistancendash mechanical stress

R k T

18

Resistive Sensors

bull Potentiometer ndash eg angle sensor (105deg) 5k Ohm linear

bull Thermistor ndash eg PTC with range 0 hellip 55deg C

(figures taken from RS componentsonline catalogue)

19

Capacitive Sensorsbull Common passive sensors variation of capacitance C

bull Pressure sensorndash typically capacitor diffused into silicon chipndash distance between capacitor plates is varied due to mechanical stress

bull Liquid level sensorndash typically two or three electrodes (measured object itself forms one

electrode)

bull Capacitive proximity switchndash contact-free detection of liquids or solid materials

20

Revolution Sensor

bull Speed indicator signal provided by DC fan (ESE-LU board)ndash special kind of onoff switchndash two pulses per revolution

(figure taken from ebmpapst DCfan data sheet)

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 3

Measurement Units

A measurement consists of the product of the measurand value and its unit

x = v ux the measurementv a number representing the measurand value (the physical or chemical

quantity property or condition that is measured)u the respective unit

bull Standardized by the international system of units(Systegraveme International dlsquoUniteacutes) SI-System

Digital Measurement 4

SI-System - Unitsbull 7 Base units

kg m s A K mol cd

bull Derived Units (examples)Hz [1s] N [(kgmiddotm)ssup2)W [(kgmiddotmsup2)s] Pa [kg(mmiddotssup2)]degC [T_kelvin ndash 27315] Joule [(kgmiddotmsup2)ssup2]

Digital Measurement 5

SI-System - Unit-Prefixkilo k 103 milli m 10-3

mega M 106 micro micro u 10-6

giga G 109 nano n 10-9

tera T 1012 pico p 10-12

peta P 1015 femto f 10-15

exa E 1018 atto a 10-18

zetta Z 1021 zepto z 10-21

yotta Y 1024 yocto y 10-24

Digital Measurement 6

Signals - Classification

Digital Measurement 7

Signals ndash Sampling Theorem

bull Sampling is the process of converting an signal into a numeric sequence (analog value to a discrete value)

bull A band-limited (0-fmax) time dependent function f(t) can be reconstructed by sample-points if sampling frequency is more than the twice of fmax

fsample gt 2middotfmax

Digital Measurement 8

Signals - Noise

bull Noise is a stochastic changing of currentvoltage of a signal which is caused by several affectsndash Thermal noisendash Atmospheric and GalacticCosmic noise

bull Impact of noise on the measurement should be minimized by appropriate design measures egndash Shielding to reduce the impact of external noisendash Appropriate hardware layout to eliminate avoidable sources of

internal noise(more details are given in the lecture bdquoHardware Design for Embedded Systemsldquo)

Digital Measurement 9

Signals ndash Noise Types|X(f)| [dB]

10 100 1000 10000f log0

White Noise

|X(f)| [dB]

10 100 1000 10000f log0

Pink Noise

|X(f)| [dB]

10 100 1000 10000f log0

BrownRed Noise-40-40

|X(f)| [dB]

10 100 1000 10000f log0

Blue Noise

|X(f)| [dB]

10 100 10000f log0

Purple Noise

1 fsup2-6dBOctave

+3dBOctave

-40-40

|X(f)| [dB]

10 100 1000 10000f log0

Gray Noise

1000

+6dBOctave

1 f-3dBOctave

White Noise (constant)Gray Noise (approxto const

psychoacoustic loudness)Pink Noise (-3dBOctave)BrownRed N (-6dBOctave)Blue Noise (+3dBOctave)Purple Noise (+6dBOctave)

10

Sensors ndash Some Definitions

bull A device that responds to a physical or chemical stimulus (such as heat pressure flow acceleration etc) and affects or generates an electrical signal

bull Facilitates to quantitatively or qualitatively acquire the physical or chemical properties of an object

bull A sensor is a transducer that converts the measurand into a signal carrying information

11

Classification of Sensors (1)

bull Passive Sensorsndash A sensor whose physical measurement variable

controls or affects the energy of somethingsomeone else

ndash eg strain gauge (ldquoDMSrdquo) capacitive sensors

bull Active Sensorsndash A sensor that generates by itself some form of energy

as its measurement signalndash eg photo transistor piezo-electric sensors

12

Classification of Sensors (2)

bull Type of measurandndash Mechanical quantities (eg pressure position force)ndash Thermal quantities (eg temperature heat flow)ndash Electrostatic and magnetic fieldsndash Radiation intensity (eg electromagnetic)ndash Chemical quantities (eg humidity gas)ndash Biological quantities (eg antigens antibodies)

13

Classification of Sensors (3)

bull Nature of Output Signalndash analog output continuous signal in its magnitude andor temporal

(eg temperature)ndash digital output output signal in the form of discrete steps or states

(eg switch)

bull Physical Measurement Variablendash resistancendash inductancendash capacitancendash etc

selecting the most appropriatesensor is not a trivial task

14

Selecting a Sensor (1)bull What should be measured

ndash distance brightness hellipndash sometimes it is easier to measure a related value (voltage

instead of current)

bull Can we access the physicalchemical property directlyndash eg temperature within a melting pot

bull How should it be measuredndash measuring wheel propagation delay triangulationndash required precision

15

Selecting a Sensor (2)bull Interface

ndash digital analog restrictions dynamics hellip

bull Requirements in the field of applicationndash mechanical stress (eg heat pressure etc)ndash costs (mass production vs prototype setup)

bull Effects biasing the measurementndash Never forget the physics behind the sensorndash Example

bull two infrared sensors (triangulation)bull almost all measurements are rubbishbull beam from sensor A was detected by senor B

16

Some Common Sensing MethodsMeasurand Method

Displacement Position resistive capacitive opto-electronic Hall effect variable reluctance

Distancetriangulation measuring wheel radar echelon capacitiveinductiveproximity

Temperature Thermistor (NTC PTC) infrared radiation thermocouple

Pressure piezoresisitve capacitive piezoelectric strain gauge

Velocity Hall effect opto-electronic variable reluctance

Luminance photo-resistor photo-diode photo-transistor

17

Resistive Sensors

bull Common passive sensors variation of resistance R

bull Potentiometer ndash changing L due to mechanical displacement (slide rotation)ndash liquid level sensor rotation and angle sensor

bull Thermistor ndash change of resistance due to change of NTC (negative k) PTC (positive k)ndash heat sensor heat flow sensor

bull Piezoresistive Sensors ndash resistor diffused in silicon compression decreases resistance tension increases resistancendash mechanical stress

R k T

18

Resistive Sensors

bull Potentiometer ndash eg angle sensor (105deg) 5k Ohm linear

bull Thermistor ndash eg PTC with range 0 hellip 55deg C

(figures taken from RS componentsonline catalogue)

19

Capacitive Sensorsbull Common passive sensors variation of capacitance C

bull Pressure sensorndash typically capacitor diffused into silicon chipndash distance between capacitor plates is varied due to mechanical stress

bull Liquid level sensorndash typically two or three electrodes (measured object itself forms one

electrode)

bull Capacitive proximity switchndash contact-free detection of liquids or solid materials

20

Revolution Sensor

bull Speed indicator signal provided by DC fan (ESE-LU board)ndash special kind of onoff switchndash two pulses per revolution

(figure taken from ebmpapst DCfan data sheet)

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 4

SI-System - Unitsbull 7 Base units

kg m s A K mol cd

bull Derived Units (examples)Hz [1s] N [(kgmiddotm)ssup2)W [(kgmiddotmsup2)s] Pa [kg(mmiddotssup2)]degC [T_kelvin ndash 27315] Joule [(kgmiddotmsup2)ssup2]

Digital Measurement 5

SI-System - Unit-Prefixkilo k 103 milli m 10-3

mega M 106 micro micro u 10-6

giga G 109 nano n 10-9

tera T 1012 pico p 10-12

peta P 1015 femto f 10-15

exa E 1018 atto a 10-18

zetta Z 1021 zepto z 10-21

yotta Y 1024 yocto y 10-24

Digital Measurement 6

Signals - Classification

Digital Measurement 7

Signals ndash Sampling Theorem

bull Sampling is the process of converting an signal into a numeric sequence (analog value to a discrete value)

bull A band-limited (0-fmax) time dependent function f(t) can be reconstructed by sample-points if sampling frequency is more than the twice of fmax

fsample gt 2middotfmax

Digital Measurement 8

Signals - Noise

bull Noise is a stochastic changing of currentvoltage of a signal which is caused by several affectsndash Thermal noisendash Atmospheric and GalacticCosmic noise

bull Impact of noise on the measurement should be minimized by appropriate design measures egndash Shielding to reduce the impact of external noisendash Appropriate hardware layout to eliminate avoidable sources of

internal noise(more details are given in the lecture bdquoHardware Design for Embedded Systemsldquo)

Digital Measurement 9

Signals ndash Noise Types|X(f)| [dB]

10 100 1000 10000f log0

White Noise

|X(f)| [dB]

10 100 1000 10000f log0

Pink Noise

|X(f)| [dB]

10 100 1000 10000f log0

BrownRed Noise-40-40

|X(f)| [dB]

10 100 1000 10000f log0

Blue Noise

|X(f)| [dB]

10 100 10000f log0

Purple Noise

1 fsup2-6dBOctave

+3dBOctave

-40-40

|X(f)| [dB]

10 100 1000 10000f log0

Gray Noise

1000

+6dBOctave

1 f-3dBOctave

White Noise (constant)Gray Noise (approxto const

psychoacoustic loudness)Pink Noise (-3dBOctave)BrownRed N (-6dBOctave)Blue Noise (+3dBOctave)Purple Noise (+6dBOctave)

10

Sensors ndash Some Definitions

bull A device that responds to a physical or chemical stimulus (such as heat pressure flow acceleration etc) and affects or generates an electrical signal

bull Facilitates to quantitatively or qualitatively acquire the physical or chemical properties of an object

bull A sensor is a transducer that converts the measurand into a signal carrying information

11

Classification of Sensors (1)

bull Passive Sensorsndash A sensor whose physical measurement variable

controls or affects the energy of somethingsomeone else

ndash eg strain gauge (ldquoDMSrdquo) capacitive sensors

bull Active Sensorsndash A sensor that generates by itself some form of energy

as its measurement signalndash eg photo transistor piezo-electric sensors

12

Classification of Sensors (2)

bull Type of measurandndash Mechanical quantities (eg pressure position force)ndash Thermal quantities (eg temperature heat flow)ndash Electrostatic and magnetic fieldsndash Radiation intensity (eg electromagnetic)ndash Chemical quantities (eg humidity gas)ndash Biological quantities (eg antigens antibodies)

13

Classification of Sensors (3)

bull Nature of Output Signalndash analog output continuous signal in its magnitude andor temporal

(eg temperature)ndash digital output output signal in the form of discrete steps or states

(eg switch)

bull Physical Measurement Variablendash resistancendash inductancendash capacitancendash etc

selecting the most appropriatesensor is not a trivial task

14

Selecting a Sensor (1)bull What should be measured

ndash distance brightness hellipndash sometimes it is easier to measure a related value (voltage

instead of current)

bull Can we access the physicalchemical property directlyndash eg temperature within a melting pot

bull How should it be measuredndash measuring wheel propagation delay triangulationndash required precision

15

Selecting a Sensor (2)bull Interface

ndash digital analog restrictions dynamics hellip

bull Requirements in the field of applicationndash mechanical stress (eg heat pressure etc)ndash costs (mass production vs prototype setup)

bull Effects biasing the measurementndash Never forget the physics behind the sensorndash Example

bull two infrared sensors (triangulation)bull almost all measurements are rubbishbull beam from sensor A was detected by senor B

16

Some Common Sensing MethodsMeasurand Method

Displacement Position resistive capacitive opto-electronic Hall effect variable reluctance

Distancetriangulation measuring wheel radar echelon capacitiveinductiveproximity

Temperature Thermistor (NTC PTC) infrared radiation thermocouple

Pressure piezoresisitve capacitive piezoelectric strain gauge

Velocity Hall effect opto-electronic variable reluctance

Luminance photo-resistor photo-diode photo-transistor

17

Resistive Sensors

bull Common passive sensors variation of resistance R

bull Potentiometer ndash changing L due to mechanical displacement (slide rotation)ndash liquid level sensor rotation and angle sensor

bull Thermistor ndash change of resistance due to change of NTC (negative k) PTC (positive k)ndash heat sensor heat flow sensor

bull Piezoresistive Sensors ndash resistor diffused in silicon compression decreases resistance tension increases resistancendash mechanical stress

R k T

18

Resistive Sensors

bull Potentiometer ndash eg angle sensor (105deg) 5k Ohm linear

bull Thermistor ndash eg PTC with range 0 hellip 55deg C

(figures taken from RS componentsonline catalogue)

19

Capacitive Sensorsbull Common passive sensors variation of capacitance C

bull Pressure sensorndash typically capacitor diffused into silicon chipndash distance between capacitor plates is varied due to mechanical stress

bull Liquid level sensorndash typically two or three electrodes (measured object itself forms one

electrode)

bull Capacitive proximity switchndash contact-free detection of liquids or solid materials

20

Revolution Sensor

bull Speed indicator signal provided by DC fan (ESE-LU board)ndash special kind of onoff switchndash two pulses per revolution

(figure taken from ebmpapst DCfan data sheet)

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 5

SI-System - Unit-Prefixkilo k 103 milli m 10-3

mega M 106 micro micro u 10-6

giga G 109 nano n 10-9

tera T 1012 pico p 10-12

peta P 1015 femto f 10-15

exa E 1018 atto a 10-18

zetta Z 1021 zepto z 10-21

yotta Y 1024 yocto y 10-24

Digital Measurement 6

Signals - Classification

Digital Measurement 7

Signals ndash Sampling Theorem

bull Sampling is the process of converting an signal into a numeric sequence (analog value to a discrete value)

bull A band-limited (0-fmax) time dependent function f(t) can be reconstructed by sample-points if sampling frequency is more than the twice of fmax

fsample gt 2middotfmax

Digital Measurement 8

Signals - Noise

bull Noise is a stochastic changing of currentvoltage of a signal which is caused by several affectsndash Thermal noisendash Atmospheric and GalacticCosmic noise

bull Impact of noise on the measurement should be minimized by appropriate design measures egndash Shielding to reduce the impact of external noisendash Appropriate hardware layout to eliminate avoidable sources of

internal noise(more details are given in the lecture bdquoHardware Design for Embedded Systemsldquo)

Digital Measurement 9

Signals ndash Noise Types|X(f)| [dB]

10 100 1000 10000f log0

White Noise

|X(f)| [dB]

10 100 1000 10000f log0

Pink Noise

|X(f)| [dB]

10 100 1000 10000f log0

BrownRed Noise-40-40

|X(f)| [dB]

10 100 1000 10000f log0

Blue Noise

|X(f)| [dB]

10 100 10000f log0

Purple Noise

1 fsup2-6dBOctave

+3dBOctave

-40-40

|X(f)| [dB]

10 100 1000 10000f log0

Gray Noise

1000

+6dBOctave

1 f-3dBOctave

White Noise (constant)Gray Noise (approxto const

psychoacoustic loudness)Pink Noise (-3dBOctave)BrownRed N (-6dBOctave)Blue Noise (+3dBOctave)Purple Noise (+6dBOctave)

10

Sensors ndash Some Definitions

bull A device that responds to a physical or chemical stimulus (such as heat pressure flow acceleration etc) and affects or generates an electrical signal

bull Facilitates to quantitatively or qualitatively acquire the physical or chemical properties of an object

bull A sensor is a transducer that converts the measurand into a signal carrying information

11

Classification of Sensors (1)

bull Passive Sensorsndash A sensor whose physical measurement variable

controls or affects the energy of somethingsomeone else

ndash eg strain gauge (ldquoDMSrdquo) capacitive sensors

bull Active Sensorsndash A sensor that generates by itself some form of energy

as its measurement signalndash eg photo transistor piezo-electric sensors

12

Classification of Sensors (2)

bull Type of measurandndash Mechanical quantities (eg pressure position force)ndash Thermal quantities (eg temperature heat flow)ndash Electrostatic and magnetic fieldsndash Radiation intensity (eg electromagnetic)ndash Chemical quantities (eg humidity gas)ndash Biological quantities (eg antigens antibodies)

13

Classification of Sensors (3)

bull Nature of Output Signalndash analog output continuous signal in its magnitude andor temporal

(eg temperature)ndash digital output output signal in the form of discrete steps or states

(eg switch)

bull Physical Measurement Variablendash resistancendash inductancendash capacitancendash etc

selecting the most appropriatesensor is not a trivial task

14

Selecting a Sensor (1)bull What should be measured

ndash distance brightness hellipndash sometimes it is easier to measure a related value (voltage

instead of current)

bull Can we access the physicalchemical property directlyndash eg temperature within a melting pot

bull How should it be measuredndash measuring wheel propagation delay triangulationndash required precision

15

Selecting a Sensor (2)bull Interface

ndash digital analog restrictions dynamics hellip

bull Requirements in the field of applicationndash mechanical stress (eg heat pressure etc)ndash costs (mass production vs prototype setup)

bull Effects biasing the measurementndash Never forget the physics behind the sensorndash Example

bull two infrared sensors (triangulation)bull almost all measurements are rubbishbull beam from sensor A was detected by senor B

16

Some Common Sensing MethodsMeasurand Method

Displacement Position resistive capacitive opto-electronic Hall effect variable reluctance

Distancetriangulation measuring wheel radar echelon capacitiveinductiveproximity

Temperature Thermistor (NTC PTC) infrared radiation thermocouple

Pressure piezoresisitve capacitive piezoelectric strain gauge

Velocity Hall effect opto-electronic variable reluctance

Luminance photo-resistor photo-diode photo-transistor

17

Resistive Sensors

bull Common passive sensors variation of resistance R

bull Potentiometer ndash changing L due to mechanical displacement (slide rotation)ndash liquid level sensor rotation and angle sensor

bull Thermistor ndash change of resistance due to change of NTC (negative k) PTC (positive k)ndash heat sensor heat flow sensor

bull Piezoresistive Sensors ndash resistor diffused in silicon compression decreases resistance tension increases resistancendash mechanical stress

R k T

18

Resistive Sensors

bull Potentiometer ndash eg angle sensor (105deg) 5k Ohm linear

bull Thermistor ndash eg PTC with range 0 hellip 55deg C

(figures taken from RS componentsonline catalogue)

19

Capacitive Sensorsbull Common passive sensors variation of capacitance C

bull Pressure sensorndash typically capacitor diffused into silicon chipndash distance between capacitor plates is varied due to mechanical stress

bull Liquid level sensorndash typically two or three electrodes (measured object itself forms one

electrode)

bull Capacitive proximity switchndash contact-free detection of liquids or solid materials

20

Revolution Sensor

bull Speed indicator signal provided by DC fan (ESE-LU board)ndash special kind of onoff switchndash two pulses per revolution

(figure taken from ebmpapst DCfan data sheet)

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 6

Signals - Classification

Digital Measurement 7

Signals ndash Sampling Theorem

bull Sampling is the process of converting an signal into a numeric sequence (analog value to a discrete value)

bull A band-limited (0-fmax) time dependent function f(t) can be reconstructed by sample-points if sampling frequency is more than the twice of fmax

fsample gt 2middotfmax

Digital Measurement 8

Signals - Noise

bull Noise is a stochastic changing of currentvoltage of a signal which is caused by several affectsndash Thermal noisendash Atmospheric and GalacticCosmic noise

bull Impact of noise on the measurement should be minimized by appropriate design measures egndash Shielding to reduce the impact of external noisendash Appropriate hardware layout to eliminate avoidable sources of

internal noise(more details are given in the lecture bdquoHardware Design for Embedded Systemsldquo)

Digital Measurement 9

Signals ndash Noise Types|X(f)| [dB]

10 100 1000 10000f log0

White Noise

|X(f)| [dB]

10 100 1000 10000f log0

Pink Noise

|X(f)| [dB]

10 100 1000 10000f log0

BrownRed Noise-40-40

|X(f)| [dB]

10 100 1000 10000f log0

Blue Noise

|X(f)| [dB]

10 100 10000f log0

Purple Noise

1 fsup2-6dBOctave

+3dBOctave

-40-40

|X(f)| [dB]

10 100 1000 10000f log0

Gray Noise

1000

+6dBOctave

1 f-3dBOctave

White Noise (constant)Gray Noise (approxto const

psychoacoustic loudness)Pink Noise (-3dBOctave)BrownRed N (-6dBOctave)Blue Noise (+3dBOctave)Purple Noise (+6dBOctave)

10

Sensors ndash Some Definitions

bull A device that responds to a physical or chemical stimulus (such as heat pressure flow acceleration etc) and affects or generates an electrical signal

bull Facilitates to quantitatively or qualitatively acquire the physical or chemical properties of an object

bull A sensor is a transducer that converts the measurand into a signal carrying information

11

Classification of Sensors (1)

bull Passive Sensorsndash A sensor whose physical measurement variable

controls or affects the energy of somethingsomeone else

ndash eg strain gauge (ldquoDMSrdquo) capacitive sensors

bull Active Sensorsndash A sensor that generates by itself some form of energy

as its measurement signalndash eg photo transistor piezo-electric sensors

12

Classification of Sensors (2)

bull Type of measurandndash Mechanical quantities (eg pressure position force)ndash Thermal quantities (eg temperature heat flow)ndash Electrostatic and magnetic fieldsndash Radiation intensity (eg electromagnetic)ndash Chemical quantities (eg humidity gas)ndash Biological quantities (eg antigens antibodies)

13

Classification of Sensors (3)

bull Nature of Output Signalndash analog output continuous signal in its magnitude andor temporal

(eg temperature)ndash digital output output signal in the form of discrete steps or states

(eg switch)

bull Physical Measurement Variablendash resistancendash inductancendash capacitancendash etc

selecting the most appropriatesensor is not a trivial task

14

Selecting a Sensor (1)bull What should be measured

ndash distance brightness hellipndash sometimes it is easier to measure a related value (voltage

instead of current)

bull Can we access the physicalchemical property directlyndash eg temperature within a melting pot

bull How should it be measuredndash measuring wheel propagation delay triangulationndash required precision

15

Selecting a Sensor (2)bull Interface

ndash digital analog restrictions dynamics hellip

bull Requirements in the field of applicationndash mechanical stress (eg heat pressure etc)ndash costs (mass production vs prototype setup)

bull Effects biasing the measurementndash Never forget the physics behind the sensorndash Example

bull two infrared sensors (triangulation)bull almost all measurements are rubbishbull beam from sensor A was detected by senor B

16

Some Common Sensing MethodsMeasurand Method

Displacement Position resistive capacitive opto-electronic Hall effect variable reluctance

Distancetriangulation measuring wheel radar echelon capacitiveinductiveproximity

Temperature Thermistor (NTC PTC) infrared radiation thermocouple

Pressure piezoresisitve capacitive piezoelectric strain gauge

Velocity Hall effect opto-electronic variable reluctance

Luminance photo-resistor photo-diode photo-transistor

17

Resistive Sensors

bull Common passive sensors variation of resistance R

bull Potentiometer ndash changing L due to mechanical displacement (slide rotation)ndash liquid level sensor rotation and angle sensor

bull Thermistor ndash change of resistance due to change of NTC (negative k) PTC (positive k)ndash heat sensor heat flow sensor

bull Piezoresistive Sensors ndash resistor diffused in silicon compression decreases resistance tension increases resistancendash mechanical stress

R k T

18

Resistive Sensors

bull Potentiometer ndash eg angle sensor (105deg) 5k Ohm linear

bull Thermistor ndash eg PTC with range 0 hellip 55deg C

(figures taken from RS componentsonline catalogue)

19

Capacitive Sensorsbull Common passive sensors variation of capacitance C

bull Pressure sensorndash typically capacitor diffused into silicon chipndash distance between capacitor plates is varied due to mechanical stress

bull Liquid level sensorndash typically two or three electrodes (measured object itself forms one

electrode)

bull Capacitive proximity switchndash contact-free detection of liquids or solid materials

20

Revolution Sensor

bull Speed indicator signal provided by DC fan (ESE-LU board)ndash special kind of onoff switchndash two pulses per revolution

(figure taken from ebmpapst DCfan data sheet)

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 7

Signals ndash Sampling Theorem

bull Sampling is the process of converting an signal into a numeric sequence (analog value to a discrete value)

bull A band-limited (0-fmax) time dependent function f(t) can be reconstructed by sample-points if sampling frequency is more than the twice of fmax

fsample gt 2middotfmax

Digital Measurement 8

Signals - Noise

bull Noise is a stochastic changing of currentvoltage of a signal which is caused by several affectsndash Thermal noisendash Atmospheric and GalacticCosmic noise

bull Impact of noise on the measurement should be minimized by appropriate design measures egndash Shielding to reduce the impact of external noisendash Appropriate hardware layout to eliminate avoidable sources of

internal noise(more details are given in the lecture bdquoHardware Design for Embedded Systemsldquo)

Digital Measurement 9

Signals ndash Noise Types|X(f)| [dB]

10 100 1000 10000f log0

White Noise

|X(f)| [dB]

10 100 1000 10000f log0

Pink Noise

|X(f)| [dB]

10 100 1000 10000f log0

BrownRed Noise-40-40

|X(f)| [dB]

10 100 1000 10000f log0

Blue Noise

|X(f)| [dB]

10 100 10000f log0

Purple Noise

1 fsup2-6dBOctave

+3dBOctave

-40-40

|X(f)| [dB]

10 100 1000 10000f log0

Gray Noise

1000

+6dBOctave

1 f-3dBOctave

White Noise (constant)Gray Noise (approxto const

psychoacoustic loudness)Pink Noise (-3dBOctave)BrownRed N (-6dBOctave)Blue Noise (+3dBOctave)Purple Noise (+6dBOctave)

10

Sensors ndash Some Definitions

bull A device that responds to a physical or chemical stimulus (such as heat pressure flow acceleration etc) and affects or generates an electrical signal

bull Facilitates to quantitatively or qualitatively acquire the physical or chemical properties of an object

bull A sensor is a transducer that converts the measurand into a signal carrying information

11

Classification of Sensors (1)

bull Passive Sensorsndash A sensor whose physical measurement variable

controls or affects the energy of somethingsomeone else

ndash eg strain gauge (ldquoDMSrdquo) capacitive sensors

bull Active Sensorsndash A sensor that generates by itself some form of energy

as its measurement signalndash eg photo transistor piezo-electric sensors

12

Classification of Sensors (2)

bull Type of measurandndash Mechanical quantities (eg pressure position force)ndash Thermal quantities (eg temperature heat flow)ndash Electrostatic and magnetic fieldsndash Radiation intensity (eg electromagnetic)ndash Chemical quantities (eg humidity gas)ndash Biological quantities (eg antigens antibodies)

13

Classification of Sensors (3)

bull Nature of Output Signalndash analog output continuous signal in its magnitude andor temporal

(eg temperature)ndash digital output output signal in the form of discrete steps or states

(eg switch)

bull Physical Measurement Variablendash resistancendash inductancendash capacitancendash etc

selecting the most appropriatesensor is not a trivial task

14

Selecting a Sensor (1)bull What should be measured

ndash distance brightness hellipndash sometimes it is easier to measure a related value (voltage

instead of current)

bull Can we access the physicalchemical property directlyndash eg temperature within a melting pot

bull How should it be measuredndash measuring wheel propagation delay triangulationndash required precision

15

Selecting a Sensor (2)bull Interface

ndash digital analog restrictions dynamics hellip

bull Requirements in the field of applicationndash mechanical stress (eg heat pressure etc)ndash costs (mass production vs prototype setup)

bull Effects biasing the measurementndash Never forget the physics behind the sensorndash Example

bull two infrared sensors (triangulation)bull almost all measurements are rubbishbull beam from sensor A was detected by senor B

16

Some Common Sensing MethodsMeasurand Method

Displacement Position resistive capacitive opto-electronic Hall effect variable reluctance

Distancetriangulation measuring wheel radar echelon capacitiveinductiveproximity

Temperature Thermistor (NTC PTC) infrared radiation thermocouple

Pressure piezoresisitve capacitive piezoelectric strain gauge

Velocity Hall effect opto-electronic variable reluctance

Luminance photo-resistor photo-diode photo-transistor

17

Resistive Sensors

bull Common passive sensors variation of resistance R

bull Potentiometer ndash changing L due to mechanical displacement (slide rotation)ndash liquid level sensor rotation and angle sensor

bull Thermistor ndash change of resistance due to change of NTC (negative k) PTC (positive k)ndash heat sensor heat flow sensor

bull Piezoresistive Sensors ndash resistor diffused in silicon compression decreases resistance tension increases resistancendash mechanical stress

R k T

18

Resistive Sensors

bull Potentiometer ndash eg angle sensor (105deg) 5k Ohm linear

bull Thermistor ndash eg PTC with range 0 hellip 55deg C

(figures taken from RS componentsonline catalogue)

19

Capacitive Sensorsbull Common passive sensors variation of capacitance C

bull Pressure sensorndash typically capacitor diffused into silicon chipndash distance between capacitor plates is varied due to mechanical stress

bull Liquid level sensorndash typically two or three electrodes (measured object itself forms one

electrode)

bull Capacitive proximity switchndash contact-free detection of liquids or solid materials

20

Revolution Sensor

bull Speed indicator signal provided by DC fan (ESE-LU board)ndash special kind of onoff switchndash two pulses per revolution

(figure taken from ebmpapst DCfan data sheet)

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 8

Signals - Noise

bull Noise is a stochastic changing of currentvoltage of a signal which is caused by several affectsndash Thermal noisendash Atmospheric and GalacticCosmic noise

bull Impact of noise on the measurement should be minimized by appropriate design measures egndash Shielding to reduce the impact of external noisendash Appropriate hardware layout to eliminate avoidable sources of

internal noise(more details are given in the lecture bdquoHardware Design for Embedded Systemsldquo)

Digital Measurement 9

Signals ndash Noise Types|X(f)| [dB]

10 100 1000 10000f log0

White Noise

|X(f)| [dB]

10 100 1000 10000f log0

Pink Noise

|X(f)| [dB]

10 100 1000 10000f log0

BrownRed Noise-40-40

|X(f)| [dB]

10 100 1000 10000f log0

Blue Noise

|X(f)| [dB]

10 100 10000f log0

Purple Noise

1 fsup2-6dBOctave

+3dBOctave

-40-40

|X(f)| [dB]

10 100 1000 10000f log0

Gray Noise

1000

+6dBOctave

1 f-3dBOctave

White Noise (constant)Gray Noise (approxto const

psychoacoustic loudness)Pink Noise (-3dBOctave)BrownRed N (-6dBOctave)Blue Noise (+3dBOctave)Purple Noise (+6dBOctave)

10

Sensors ndash Some Definitions

bull A device that responds to a physical or chemical stimulus (such as heat pressure flow acceleration etc) and affects or generates an electrical signal

bull Facilitates to quantitatively or qualitatively acquire the physical or chemical properties of an object

bull A sensor is a transducer that converts the measurand into a signal carrying information

11

Classification of Sensors (1)

bull Passive Sensorsndash A sensor whose physical measurement variable

controls or affects the energy of somethingsomeone else

ndash eg strain gauge (ldquoDMSrdquo) capacitive sensors

bull Active Sensorsndash A sensor that generates by itself some form of energy

as its measurement signalndash eg photo transistor piezo-electric sensors

12

Classification of Sensors (2)

bull Type of measurandndash Mechanical quantities (eg pressure position force)ndash Thermal quantities (eg temperature heat flow)ndash Electrostatic and magnetic fieldsndash Radiation intensity (eg electromagnetic)ndash Chemical quantities (eg humidity gas)ndash Biological quantities (eg antigens antibodies)

13

Classification of Sensors (3)

bull Nature of Output Signalndash analog output continuous signal in its magnitude andor temporal

(eg temperature)ndash digital output output signal in the form of discrete steps or states

(eg switch)

bull Physical Measurement Variablendash resistancendash inductancendash capacitancendash etc

selecting the most appropriatesensor is not a trivial task

14

Selecting a Sensor (1)bull What should be measured

ndash distance brightness hellipndash sometimes it is easier to measure a related value (voltage

instead of current)

bull Can we access the physicalchemical property directlyndash eg temperature within a melting pot

bull How should it be measuredndash measuring wheel propagation delay triangulationndash required precision

15

Selecting a Sensor (2)bull Interface

ndash digital analog restrictions dynamics hellip

bull Requirements in the field of applicationndash mechanical stress (eg heat pressure etc)ndash costs (mass production vs prototype setup)

bull Effects biasing the measurementndash Never forget the physics behind the sensorndash Example

bull two infrared sensors (triangulation)bull almost all measurements are rubbishbull beam from sensor A was detected by senor B

16

Some Common Sensing MethodsMeasurand Method

Displacement Position resistive capacitive opto-electronic Hall effect variable reluctance

Distancetriangulation measuring wheel radar echelon capacitiveinductiveproximity

Temperature Thermistor (NTC PTC) infrared radiation thermocouple

Pressure piezoresisitve capacitive piezoelectric strain gauge

Velocity Hall effect opto-electronic variable reluctance

Luminance photo-resistor photo-diode photo-transistor

17

Resistive Sensors

bull Common passive sensors variation of resistance R

bull Potentiometer ndash changing L due to mechanical displacement (slide rotation)ndash liquid level sensor rotation and angle sensor

bull Thermistor ndash change of resistance due to change of NTC (negative k) PTC (positive k)ndash heat sensor heat flow sensor

bull Piezoresistive Sensors ndash resistor diffused in silicon compression decreases resistance tension increases resistancendash mechanical stress

R k T

18

Resistive Sensors

bull Potentiometer ndash eg angle sensor (105deg) 5k Ohm linear

bull Thermistor ndash eg PTC with range 0 hellip 55deg C

(figures taken from RS componentsonline catalogue)

19

Capacitive Sensorsbull Common passive sensors variation of capacitance C

bull Pressure sensorndash typically capacitor diffused into silicon chipndash distance between capacitor plates is varied due to mechanical stress

bull Liquid level sensorndash typically two or three electrodes (measured object itself forms one

electrode)

bull Capacitive proximity switchndash contact-free detection of liquids or solid materials

20

Revolution Sensor

bull Speed indicator signal provided by DC fan (ESE-LU board)ndash special kind of onoff switchndash two pulses per revolution

(figure taken from ebmpapst DCfan data sheet)

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 9

Signals ndash Noise Types|X(f)| [dB]

10 100 1000 10000f log0

White Noise

|X(f)| [dB]

10 100 1000 10000f log0

Pink Noise

|X(f)| [dB]

10 100 1000 10000f log0

BrownRed Noise-40-40

|X(f)| [dB]

10 100 1000 10000f log0

Blue Noise

|X(f)| [dB]

10 100 10000f log0

Purple Noise

1 fsup2-6dBOctave

+3dBOctave

-40-40

|X(f)| [dB]

10 100 1000 10000f log0

Gray Noise

1000

+6dBOctave

1 f-3dBOctave

White Noise (constant)Gray Noise (approxto const

psychoacoustic loudness)Pink Noise (-3dBOctave)BrownRed N (-6dBOctave)Blue Noise (+3dBOctave)Purple Noise (+6dBOctave)

10

Sensors ndash Some Definitions

bull A device that responds to a physical or chemical stimulus (such as heat pressure flow acceleration etc) and affects or generates an electrical signal

bull Facilitates to quantitatively or qualitatively acquire the physical or chemical properties of an object

bull A sensor is a transducer that converts the measurand into a signal carrying information

11

Classification of Sensors (1)

bull Passive Sensorsndash A sensor whose physical measurement variable

controls or affects the energy of somethingsomeone else

ndash eg strain gauge (ldquoDMSrdquo) capacitive sensors

bull Active Sensorsndash A sensor that generates by itself some form of energy

as its measurement signalndash eg photo transistor piezo-electric sensors

12

Classification of Sensors (2)

bull Type of measurandndash Mechanical quantities (eg pressure position force)ndash Thermal quantities (eg temperature heat flow)ndash Electrostatic and magnetic fieldsndash Radiation intensity (eg electromagnetic)ndash Chemical quantities (eg humidity gas)ndash Biological quantities (eg antigens antibodies)

13

Classification of Sensors (3)

bull Nature of Output Signalndash analog output continuous signal in its magnitude andor temporal

(eg temperature)ndash digital output output signal in the form of discrete steps or states

(eg switch)

bull Physical Measurement Variablendash resistancendash inductancendash capacitancendash etc

selecting the most appropriatesensor is not a trivial task

14

Selecting a Sensor (1)bull What should be measured

ndash distance brightness hellipndash sometimes it is easier to measure a related value (voltage

instead of current)

bull Can we access the physicalchemical property directlyndash eg temperature within a melting pot

bull How should it be measuredndash measuring wheel propagation delay triangulationndash required precision

15

Selecting a Sensor (2)bull Interface

ndash digital analog restrictions dynamics hellip

bull Requirements in the field of applicationndash mechanical stress (eg heat pressure etc)ndash costs (mass production vs prototype setup)

bull Effects biasing the measurementndash Never forget the physics behind the sensorndash Example

bull two infrared sensors (triangulation)bull almost all measurements are rubbishbull beam from sensor A was detected by senor B

16

Some Common Sensing MethodsMeasurand Method

Displacement Position resistive capacitive opto-electronic Hall effect variable reluctance

Distancetriangulation measuring wheel radar echelon capacitiveinductiveproximity

Temperature Thermistor (NTC PTC) infrared radiation thermocouple

Pressure piezoresisitve capacitive piezoelectric strain gauge

Velocity Hall effect opto-electronic variable reluctance

Luminance photo-resistor photo-diode photo-transistor

17

Resistive Sensors

bull Common passive sensors variation of resistance R

bull Potentiometer ndash changing L due to mechanical displacement (slide rotation)ndash liquid level sensor rotation and angle sensor

bull Thermistor ndash change of resistance due to change of NTC (negative k) PTC (positive k)ndash heat sensor heat flow sensor

bull Piezoresistive Sensors ndash resistor diffused in silicon compression decreases resistance tension increases resistancendash mechanical stress

R k T

18

Resistive Sensors

bull Potentiometer ndash eg angle sensor (105deg) 5k Ohm linear

bull Thermistor ndash eg PTC with range 0 hellip 55deg C

(figures taken from RS componentsonline catalogue)

19

Capacitive Sensorsbull Common passive sensors variation of capacitance C

bull Pressure sensorndash typically capacitor diffused into silicon chipndash distance between capacitor plates is varied due to mechanical stress

bull Liquid level sensorndash typically two or three electrodes (measured object itself forms one

electrode)

bull Capacitive proximity switchndash contact-free detection of liquids or solid materials

20

Revolution Sensor

bull Speed indicator signal provided by DC fan (ESE-LU board)ndash special kind of onoff switchndash two pulses per revolution

(figure taken from ebmpapst DCfan data sheet)

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

10

Sensors ndash Some Definitions

bull A device that responds to a physical or chemical stimulus (such as heat pressure flow acceleration etc) and affects or generates an electrical signal

bull Facilitates to quantitatively or qualitatively acquire the physical or chemical properties of an object

bull A sensor is a transducer that converts the measurand into a signal carrying information

11

Classification of Sensors (1)

bull Passive Sensorsndash A sensor whose physical measurement variable

controls or affects the energy of somethingsomeone else

ndash eg strain gauge (ldquoDMSrdquo) capacitive sensors

bull Active Sensorsndash A sensor that generates by itself some form of energy

as its measurement signalndash eg photo transistor piezo-electric sensors

12

Classification of Sensors (2)

bull Type of measurandndash Mechanical quantities (eg pressure position force)ndash Thermal quantities (eg temperature heat flow)ndash Electrostatic and magnetic fieldsndash Radiation intensity (eg electromagnetic)ndash Chemical quantities (eg humidity gas)ndash Biological quantities (eg antigens antibodies)

13

Classification of Sensors (3)

bull Nature of Output Signalndash analog output continuous signal in its magnitude andor temporal

(eg temperature)ndash digital output output signal in the form of discrete steps or states

(eg switch)

bull Physical Measurement Variablendash resistancendash inductancendash capacitancendash etc

selecting the most appropriatesensor is not a trivial task

14

Selecting a Sensor (1)bull What should be measured

ndash distance brightness hellipndash sometimes it is easier to measure a related value (voltage

instead of current)

bull Can we access the physicalchemical property directlyndash eg temperature within a melting pot

bull How should it be measuredndash measuring wheel propagation delay triangulationndash required precision

15

Selecting a Sensor (2)bull Interface

ndash digital analog restrictions dynamics hellip

bull Requirements in the field of applicationndash mechanical stress (eg heat pressure etc)ndash costs (mass production vs prototype setup)

bull Effects biasing the measurementndash Never forget the physics behind the sensorndash Example

bull two infrared sensors (triangulation)bull almost all measurements are rubbishbull beam from sensor A was detected by senor B

16

Some Common Sensing MethodsMeasurand Method

Displacement Position resistive capacitive opto-electronic Hall effect variable reluctance

Distancetriangulation measuring wheel radar echelon capacitiveinductiveproximity

Temperature Thermistor (NTC PTC) infrared radiation thermocouple

Pressure piezoresisitve capacitive piezoelectric strain gauge

Velocity Hall effect opto-electronic variable reluctance

Luminance photo-resistor photo-diode photo-transistor

17

Resistive Sensors

bull Common passive sensors variation of resistance R

bull Potentiometer ndash changing L due to mechanical displacement (slide rotation)ndash liquid level sensor rotation and angle sensor

bull Thermistor ndash change of resistance due to change of NTC (negative k) PTC (positive k)ndash heat sensor heat flow sensor

bull Piezoresistive Sensors ndash resistor diffused in silicon compression decreases resistance tension increases resistancendash mechanical stress

R k T

18

Resistive Sensors

bull Potentiometer ndash eg angle sensor (105deg) 5k Ohm linear

bull Thermistor ndash eg PTC with range 0 hellip 55deg C

(figures taken from RS componentsonline catalogue)

19

Capacitive Sensorsbull Common passive sensors variation of capacitance C

bull Pressure sensorndash typically capacitor diffused into silicon chipndash distance between capacitor plates is varied due to mechanical stress

bull Liquid level sensorndash typically two or three electrodes (measured object itself forms one

electrode)

bull Capacitive proximity switchndash contact-free detection of liquids or solid materials

20

Revolution Sensor

bull Speed indicator signal provided by DC fan (ESE-LU board)ndash special kind of onoff switchndash two pulses per revolution

(figure taken from ebmpapst DCfan data sheet)

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

11

Classification of Sensors (1)

bull Passive Sensorsndash A sensor whose physical measurement variable

controls or affects the energy of somethingsomeone else

ndash eg strain gauge (ldquoDMSrdquo) capacitive sensors

bull Active Sensorsndash A sensor that generates by itself some form of energy

as its measurement signalndash eg photo transistor piezo-electric sensors

12

Classification of Sensors (2)

bull Type of measurandndash Mechanical quantities (eg pressure position force)ndash Thermal quantities (eg temperature heat flow)ndash Electrostatic and magnetic fieldsndash Radiation intensity (eg electromagnetic)ndash Chemical quantities (eg humidity gas)ndash Biological quantities (eg antigens antibodies)

13

Classification of Sensors (3)

bull Nature of Output Signalndash analog output continuous signal in its magnitude andor temporal

(eg temperature)ndash digital output output signal in the form of discrete steps or states

(eg switch)

bull Physical Measurement Variablendash resistancendash inductancendash capacitancendash etc

selecting the most appropriatesensor is not a trivial task

14

Selecting a Sensor (1)bull What should be measured

ndash distance brightness hellipndash sometimes it is easier to measure a related value (voltage

instead of current)

bull Can we access the physicalchemical property directlyndash eg temperature within a melting pot

bull How should it be measuredndash measuring wheel propagation delay triangulationndash required precision

15

Selecting a Sensor (2)bull Interface

ndash digital analog restrictions dynamics hellip

bull Requirements in the field of applicationndash mechanical stress (eg heat pressure etc)ndash costs (mass production vs prototype setup)

bull Effects biasing the measurementndash Never forget the physics behind the sensorndash Example

bull two infrared sensors (triangulation)bull almost all measurements are rubbishbull beam from sensor A was detected by senor B

16

Some Common Sensing MethodsMeasurand Method

Displacement Position resistive capacitive opto-electronic Hall effect variable reluctance

Distancetriangulation measuring wheel radar echelon capacitiveinductiveproximity

Temperature Thermistor (NTC PTC) infrared radiation thermocouple

Pressure piezoresisitve capacitive piezoelectric strain gauge

Velocity Hall effect opto-electronic variable reluctance

Luminance photo-resistor photo-diode photo-transistor

17

Resistive Sensors

bull Common passive sensors variation of resistance R

bull Potentiometer ndash changing L due to mechanical displacement (slide rotation)ndash liquid level sensor rotation and angle sensor

bull Thermistor ndash change of resistance due to change of NTC (negative k) PTC (positive k)ndash heat sensor heat flow sensor

bull Piezoresistive Sensors ndash resistor diffused in silicon compression decreases resistance tension increases resistancendash mechanical stress

R k T

18

Resistive Sensors

bull Potentiometer ndash eg angle sensor (105deg) 5k Ohm linear

bull Thermistor ndash eg PTC with range 0 hellip 55deg C

(figures taken from RS componentsonline catalogue)

19

Capacitive Sensorsbull Common passive sensors variation of capacitance C

bull Pressure sensorndash typically capacitor diffused into silicon chipndash distance between capacitor plates is varied due to mechanical stress

bull Liquid level sensorndash typically two or three electrodes (measured object itself forms one

electrode)

bull Capacitive proximity switchndash contact-free detection of liquids or solid materials

20

Revolution Sensor

bull Speed indicator signal provided by DC fan (ESE-LU board)ndash special kind of onoff switchndash two pulses per revolution

(figure taken from ebmpapst DCfan data sheet)

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

12

Classification of Sensors (2)

bull Type of measurandndash Mechanical quantities (eg pressure position force)ndash Thermal quantities (eg temperature heat flow)ndash Electrostatic and magnetic fieldsndash Radiation intensity (eg electromagnetic)ndash Chemical quantities (eg humidity gas)ndash Biological quantities (eg antigens antibodies)

13

Classification of Sensors (3)

bull Nature of Output Signalndash analog output continuous signal in its magnitude andor temporal

(eg temperature)ndash digital output output signal in the form of discrete steps or states

(eg switch)

bull Physical Measurement Variablendash resistancendash inductancendash capacitancendash etc

selecting the most appropriatesensor is not a trivial task

14

Selecting a Sensor (1)bull What should be measured

ndash distance brightness hellipndash sometimes it is easier to measure a related value (voltage

instead of current)

bull Can we access the physicalchemical property directlyndash eg temperature within a melting pot

bull How should it be measuredndash measuring wheel propagation delay triangulationndash required precision

15

Selecting a Sensor (2)bull Interface

ndash digital analog restrictions dynamics hellip

bull Requirements in the field of applicationndash mechanical stress (eg heat pressure etc)ndash costs (mass production vs prototype setup)

bull Effects biasing the measurementndash Never forget the physics behind the sensorndash Example

bull two infrared sensors (triangulation)bull almost all measurements are rubbishbull beam from sensor A was detected by senor B

16

Some Common Sensing MethodsMeasurand Method

Displacement Position resistive capacitive opto-electronic Hall effect variable reluctance

Distancetriangulation measuring wheel radar echelon capacitiveinductiveproximity

Temperature Thermistor (NTC PTC) infrared radiation thermocouple

Pressure piezoresisitve capacitive piezoelectric strain gauge

Velocity Hall effect opto-electronic variable reluctance

Luminance photo-resistor photo-diode photo-transistor

17

Resistive Sensors

bull Common passive sensors variation of resistance R

bull Potentiometer ndash changing L due to mechanical displacement (slide rotation)ndash liquid level sensor rotation and angle sensor

bull Thermistor ndash change of resistance due to change of NTC (negative k) PTC (positive k)ndash heat sensor heat flow sensor

bull Piezoresistive Sensors ndash resistor diffused in silicon compression decreases resistance tension increases resistancendash mechanical stress

R k T

18

Resistive Sensors

bull Potentiometer ndash eg angle sensor (105deg) 5k Ohm linear

bull Thermistor ndash eg PTC with range 0 hellip 55deg C

(figures taken from RS componentsonline catalogue)

19

Capacitive Sensorsbull Common passive sensors variation of capacitance C

bull Pressure sensorndash typically capacitor diffused into silicon chipndash distance between capacitor plates is varied due to mechanical stress

bull Liquid level sensorndash typically two or three electrodes (measured object itself forms one

electrode)

bull Capacitive proximity switchndash contact-free detection of liquids or solid materials

20

Revolution Sensor

bull Speed indicator signal provided by DC fan (ESE-LU board)ndash special kind of onoff switchndash two pulses per revolution

(figure taken from ebmpapst DCfan data sheet)

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

13

Classification of Sensors (3)

bull Nature of Output Signalndash analog output continuous signal in its magnitude andor temporal

(eg temperature)ndash digital output output signal in the form of discrete steps or states

(eg switch)

bull Physical Measurement Variablendash resistancendash inductancendash capacitancendash etc

selecting the most appropriatesensor is not a trivial task

14

Selecting a Sensor (1)bull What should be measured

ndash distance brightness hellipndash sometimes it is easier to measure a related value (voltage

instead of current)

bull Can we access the physicalchemical property directlyndash eg temperature within a melting pot

bull How should it be measuredndash measuring wheel propagation delay triangulationndash required precision

15

Selecting a Sensor (2)bull Interface

ndash digital analog restrictions dynamics hellip

bull Requirements in the field of applicationndash mechanical stress (eg heat pressure etc)ndash costs (mass production vs prototype setup)

bull Effects biasing the measurementndash Never forget the physics behind the sensorndash Example

bull two infrared sensors (triangulation)bull almost all measurements are rubbishbull beam from sensor A was detected by senor B

16

Some Common Sensing MethodsMeasurand Method

Displacement Position resistive capacitive opto-electronic Hall effect variable reluctance

Distancetriangulation measuring wheel radar echelon capacitiveinductiveproximity

Temperature Thermistor (NTC PTC) infrared radiation thermocouple

Pressure piezoresisitve capacitive piezoelectric strain gauge

Velocity Hall effect opto-electronic variable reluctance

Luminance photo-resistor photo-diode photo-transistor

17

Resistive Sensors

bull Common passive sensors variation of resistance R

bull Potentiometer ndash changing L due to mechanical displacement (slide rotation)ndash liquid level sensor rotation and angle sensor

bull Thermistor ndash change of resistance due to change of NTC (negative k) PTC (positive k)ndash heat sensor heat flow sensor

bull Piezoresistive Sensors ndash resistor diffused in silicon compression decreases resistance tension increases resistancendash mechanical stress

R k T

18

Resistive Sensors

bull Potentiometer ndash eg angle sensor (105deg) 5k Ohm linear

bull Thermistor ndash eg PTC with range 0 hellip 55deg C

(figures taken from RS componentsonline catalogue)

19

Capacitive Sensorsbull Common passive sensors variation of capacitance C

bull Pressure sensorndash typically capacitor diffused into silicon chipndash distance between capacitor plates is varied due to mechanical stress

bull Liquid level sensorndash typically two or three electrodes (measured object itself forms one

electrode)

bull Capacitive proximity switchndash contact-free detection of liquids or solid materials

20

Revolution Sensor

bull Speed indicator signal provided by DC fan (ESE-LU board)ndash special kind of onoff switchndash two pulses per revolution

(figure taken from ebmpapst DCfan data sheet)

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

14

Selecting a Sensor (1)bull What should be measured

ndash distance brightness hellipndash sometimes it is easier to measure a related value (voltage

instead of current)

bull Can we access the physicalchemical property directlyndash eg temperature within a melting pot

bull How should it be measuredndash measuring wheel propagation delay triangulationndash required precision

15

Selecting a Sensor (2)bull Interface

ndash digital analog restrictions dynamics hellip

bull Requirements in the field of applicationndash mechanical stress (eg heat pressure etc)ndash costs (mass production vs prototype setup)

bull Effects biasing the measurementndash Never forget the physics behind the sensorndash Example

bull two infrared sensors (triangulation)bull almost all measurements are rubbishbull beam from sensor A was detected by senor B

16

Some Common Sensing MethodsMeasurand Method

Displacement Position resistive capacitive opto-electronic Hall effect variable reluctance

Distancetriangulation measuring wheel radar echelon capacitiveinductiveproximity

Temperature Thermistor (NTC PTC) infrared radiation thermocouple

Pressure piezoresisitve capacitive piezoelectric strain gauge

Velocity Hall effect opto-electronic variable reluctance

Luminance photo-resistor photo-diode photo-transistor

17

Resistive Sensors

bull Common passive sensors variation of resistance R

bull Potentiometer ndash changing L due to mechanical displacement (slide rotation)ndash liquid level sensor rotation and angle sensor

bull Thermistor ndash change of resistance due to change of NTC (negative k) PTC (positive k)ndash heat sensor heat flow sensor

bull Piezoresistive Sensors ndash resistor diffused in silicon compression decreases resistance tension increases resistancendash mechanical stress

R k T

18

Resistive Sensors

bull Potentiometer ndash eg angle sensor (105deg) 5k Ohm linear

bull Thermistor ndash eg PTC with range 0 hellip 55deg C

(figures taken from RS componentsonline catalogue)

19

Capacitive Sensorsbull Common passive sensors variation of capacitance C

bull Pressure sensorndash typically capacitor diffused into silicon chipndash distance between capacitor plates is varied due to mechanical stress

bull Liquid level sensorndash typically two or three electrodes (measured object itself forms one

electrode)

bull Capacitive proximity switchndash contact-free detection of liquids or solid materials

20

Revolution Sensor

bull Speed indicator signal provided by DC fan (ESE-LU board)ndash special kind of onoff switchndash two pulses per revolution

(figure taken from ebmpapst DCfan data sheet)

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

15

Selecting a Sensor (2)bull Interface

ndash digital analog restrictions dynamics hellip

bull Requirements in the field of applicationndash mechanical stress (eg heat pressure etc)ndash costs (mass production vs prototype setup)

bull Effects biasing the measurementndash Never forget the physics behind the sensorndash Example

bull two infrared sensors (triangulation)bull almost all measurements are rubbishbull beam from sensor A was detected by senor B

16

Some Common Sensing MethodsMeasurand Method

Displacement Position resistive capacitive opto-electronic Hall effect variable reluctance

Distancetriangulation measuring wheel radar echelon capacitiveinductiveproximity

Temperature Thermistor (NTC PTC) infrared radiation thermocouple

Pressure piezoresisitve capacitive piezoelectric strain gauge

Velocity Hall effect opto-electronic variable reluctance

Luminance photo-resistor photo-diode photo-transistor

17

Resistive Sensors

bull Common passive sensors variation of resistance R

bull Potentiometer ndash changing L due to mechanical displacement (slide rotation)ndash liquid level sensor rotation and angle sensor

bull Thermistor ndash change of resistance due to change of NTC (negative k) PTC (positive k)ndash heat sensor heat flow sensor

bull Piezoresistive Sensors ndash resistor diffused in silicon compression decreases resistance tension increases resistancendash mechanical stress

R k T

18

Resistive Sensors

bull Potentiometer ndash eg angle sensor (105deg) 5k Ohm linear

bull Thermistor ndash eg PTC with range 0 hellip 55deg C

(figures taken from RS componentsonline catalogue)

19

Capacitive Sensorsbull Common passive sensors variation of capacitance C

bull Pressure sensorndash typically capacitor diffused into silicon chipndash distance between capacitor plates is varied due to mechanical stress

bull Liquid level sensorndash typically two or three electrodes (measured object itself forms one

electrode)

bull Capacitive proximity switchndash contact-free detection of liquids or solid materials

20

Revolution Sensor

bull Speed indicator signal provided by DC fan (ESE-LU board)ndash special kind of onoff switchndash two pulses per revolution

(figure taken from ebmpapst DCfan data sheet)

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

16

Some Common Sensing MethodsMeasurand Method

Displacement Position resistive capacitive opto-electronic Hall effect variable reluctance

Distancetriangulation measuring wheel radar echelon capacitiveinductiveproximity

Temperature Thermistor (NTC PTC) infrared radiation thermocouple

Pressure piezoresisitve capacitive piezoelectric strain gauge

Velocity Hall effect opto-electronic variable reluctance

Luminance photo-resistor photo-diode photo-transistor

17

Resistive Sensors

bull Common passive sensors variation of resistance R

bull Potentiometer ndash changing L due to mechanical displacement (slide rotation)ndash liquid level sensor rotation and angle sensor

bull Thermistor ndash change of resistance due to change of NTC (negative k) PTC (positive k)ndash heat sensor heat flow sensor

bull Piezoresistive Sensors ndash resistor diffused in silicon compression decreases resistance tension increases resistancendash mechanical stress

R k T

18

Resistive Sensors

bull Potentiometer ndash eg angle sensor (105deg) 5k Ohm linear

bull Thermistor ndash eg PTC with range 0 hellip 55deg C

(figures taken from RS componentsonline catalogue)

19

Capacitive Sensorsbull Common passive sensors variation of capacitance C

bull Pressure sensorndash typically capacitor diffused into silicon chipndash distance between capacitor plates is varied due to mechanical stress

bull Liquid level sensorndash typically two or three electrodes (measured object itself forms one

electrode)

bull Capacitive proximity switchndash contact-free detection of liquids or solid materials

20

Revolution Sensor

bull Speed indicator signal provided by DC fan (ESE-LU board)ndash special kind of onoff switchndash two pulses per revolution

(figure taken from ebmpapst DCfan data sheet)

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

17

Resistive Sensors

bull Common passive sensors variation of resistance R

bull Potentiometer ndash changing L due to mechanical displacement (slide rotation)ndash liquid level sensor rotation and angle sensor

bull Thermistor ndash change of resistance due to change of NTC (negative k) PTC (positive k)ndash heat sensor heat flow sensor

bull Piezoresistive Sensors ndash resistor diffused in silicon compression decreases resistance tension increases resistancendash mechanical stress

R k T

18

Resistive Sensors

bull Potentiometer ndash eg angle sensor (105deg) 5k Ohm linear

bull Thermistor ndash eg PTC with range 0 hellip 55deg C

(figures taken from RS componentsonline catalogue)

19

Capacitive Sensorsbull Common passive sensors variation of capacitance C

bull Pressure sensorndash typically capacitor diffused into silicon chipndash distance between capacitor plates is varied due to mechanical stress

bull Liquid level sensorndash typically two or three electrodes (measured object itself forms one

electrode)

bull Capacitive proximity switchndash contact-free detection of liquids or solid materials

20

Revolution Sensor

bull Speed indicator signal provided by DC fan (ESE-LU board)ndash special kind of onoff switchndash two pulses per revolution

(figure taken from ebmpapst DCfan data sheet)

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

18

Resistive Sensors

bull Potentiometer ndash eg angle sensor (105deg) 5k Ohm linear

bull Thermistor ndash eg PTC with range 0 hellip 55deg C

(figures taken from RS componentsonline catalogue)

19

Capacitive Sensorsbull Common passive sensors variation of capacitance C

bull Pressure sensorndash typically capacitor diffused into silicon chipndash distance between capacitor plates is varied due to mechanical stress

bull Liquid level sensorndash typically two or three electrodes (measured object itself forms one

electrode)

bull Capacitive proximity switchndash contact-free detection of liquids or solid materials

20

Revolution Sensor

bull Speed indicator signal provided by DC fan (ESE-LU board)ndash special kind of onoff switchndash two pulses per revolution

(figure taken from ebmpapst DCfan data sheet)

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

19

Capacitive Sensorsbull Common passive sensors variation of capacitance C

bull Pressure sensorndash typically capacitor diffused into silicon chipndash distance between capacitor plates is varied due to mechanical stress

bull Liquid level sensorndash typically two or three electrodes (measured object itself forms one

electrode)

bull Capacitive proximity switchndash contact-free detection of liquids or solid materials

20

Revolution Sensor

bull Speed indicator signal provided by DC fan (ESE-LU board)ndash special kind of onoff switchndash two pulses per revolution

(figure taken from ebmpapst DCfan data sheet)

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

20

Revolution Sensor

bull Speed indicator signal provided by DC fan (ESE-LU board)ndash special kind of onoff switchndash two pulses per revolution

(figure taken from ebmpapst DCfan data sheet)

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

21

Optical Sensors (1)

bull Position Encoderndash incremental position encoder

(PC mouse)bull two square waves 90deg phase-

delayedndash absolute position encoder

(parallel)bull n wiresbull Gray encoded position

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

22

Optical Sensors (2)

bull Incremental position encoder

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

23

Optical Sensors (3)

bull Absolute position encoder (5 bit gray code)

(figure taken from wwwinformatikuni-hamburgde)

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

24

Optical Sensors (4)

bull Luminance Sensorndash photo electric effect (photo transistor photo diode photo resistor)ndash output signal mostly analogndash characteristic linear amp non-linearndash some luminance sensors are programmable (gain alarm limit hellip)ndash eg NSL-19M51 photo resistor (ESE-LU board)

bull photo conductive cellbull resistance from 20-100K Ohm (light) to 20M Ohm (dark)

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

25

Piezoelectric Sensor

bull Piezoelectric Effect (P and J Curie 1880)ndash ability of crystals to generate voltage as a response to

mechanical stressndash materials quartz (SiO2) cane sugar topazndash in static operation the behavior is similar to a capacitor

bull Active sensor for measuring pressure force or acceleration (eg piezoelectric microphone)

bull Above 846degK this piezoelectric effect is lost (Curie-Temperature)

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

26

Hall Effect Sensor

bull Hall Effect (Edwin Hall 1879)ndash generation of potential difference

(Hall voltage) in a conductive material located in a stationary magnetic field through which electrical current is flowing

bull Active sensor for measuringndash displacementndash velocityndash inductive proximity switch

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

27

Ideal Sensor vs Real World

bull Ideal Sensorndash transforms highly linearly a single physical or chemical measurand

into an electrical signal while being resistant to environmental influences

bull Real Worldndash gain error clippingndash offset (bias) driftndash nonlinearityndash hysteresisndash digitization error

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 28

Measurement Error

Measurement error is the difference between the measured and the actual value

e = x ndash a erel = (x-a)a

e measurement error (absolute)erel measurement error (relative) x measured valuea actual value

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 29

Measurement Error - Types

bull Systematic errorsndash reproducible (calculation measurement) measurement deviation

(eg bias drift) which is in principle correctable (eg calibration)ndash eg zero pointoffset scaling integral linearity differential linearity

history dependent

bull Conditional errorsndash caused by external influences eg Electromagnetic

interferencepulse (EMI EMP)

bull Stochastic errorsndash measurement error that is due to random causes (eg noise)

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 30

Systematic measurement errors 1

(a) Zero pointoffset error(b) Scaling error

x(t)measure

x(t)entity

ezero

x(t)measure

x(t)entity

dx(t)entity = 1

dx(t)measure

1

1

(a) (b)

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 31

Systematic measurement errors 2

x(t)measure

x(t)entity

History dependent error(eg hysteresis error)

bull Caused by the effects of static and dynamic signal history

bull Demands an advanced strategy for measurement error calibration

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 32

Calibration

Calibration is the correction of sensor reading and physical outputs so they match a standard [JBerge]

bull Adjusting the measurement to agree with value of the applied standard within a specified accuracy

Measurement Triple- estimated value- error bound- value probability

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 33

Digital calibration model

Example The RT-Image is afflicted with a measurement error from the ADC (quantization linearization scalingoffset )

fraw(n)

Offset ErrorCorrection

Scaling ErrorCorrection Linearization RT-Image

(calibrated)

fcal(n)

RT-Image(raw)

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 34

Analog to Digital Converter - ADC

bull Changes a signallsquos time and value into discrete domain

bull Requires a sample and hold stage (SH)

bull Conversion techniquesmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 35

Digital to Analog Converter - DAC

bull Transforms a digital value with discrete time domain into analog value and time

bull Conversion techniquesndash gtmore details in bdquoTutorial Peripherals and IOldquo

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 36

Processing of Measured Databull Classification of measured data

ndash Several measurements from the same instant from different sensors (sample)

ndash Several measurements from different instants from the same sensor (series)

bull Next Sensor Fusion and Digital Filtering

Sensors

Time

sampleseries

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 37

Motivation for Sensor Fusionbull Sensor Deprivation

ndash eg loss of perception on object due to sensor break down

bull Limited spatial coverage ndash eg single measurement of water temperature returns temperature

estimation near the thermometer

bull Limited temporal coverage ndash eg set-up time of sensor limits achievable measurement frequency

bull Imprecision ndash eg inherent imprecision of deployed sensing element

bull Uncertaintyndash eg ambiguous observation due to missing features (occlusions)

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 38

Competitive vs Complementary vs Cooperative Fusion

EnvironmentA

S1 S3 S4 S5S2

B C

CompetitiveFusione g Voting

CooperativeFusion

e g Triangulation

ComplementaryFusion

Sensors

Fusion

Object AResulting data Object A + B Object C

Achievements ReliabilityAccuracy Completeness Emerging Views

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 39

Marzullorsquos Algorithm

value range

bull Each sensor measurement is represented as an interval

bull Maximum t sensors (out of n) are expected to be faulty

bull Result is a single interval M that covers all intersections shared by at least n ndash f intervals

n=4 sensors

f =1 expected to be

faulty

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 40

Confidence-weighted averagingbull Each measurement is assigned a

confidence marker indicating the uncertainty of the measurement value

bull Use statistical variance as measure for uncertainty

bull Fusion operations use both properties and yield a result of value and confidence marker

bull Confidence marker corresponds to variances

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 41

Filtering a Series of Measurements

bull A filter is a layer designed to block certain things whilst letting others through

bull Primary filtering question what is the intended information and what should be filtered out

bull Typically filter out noisebull Careful design not to remove the intended informationbull Digital filter input and output are treated as time discrete

signals

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 42

Moving Average Filter

bull Very simple implementationbull Filter kernel h[n] = 1Mbull Well-suited for signal restoration (eg noise reduction) bull But does not cut off unwanted frequencies very wellbull Tradeoff noise reduction versus rise time

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 43

Moving Average Filter (2)

M=11 M=33

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

Digital Measurement 44

bull Measurements have (systematic stochastic) errorsbull Chosen sensor type has to be aligned to (physical)

characteristics of the measurandbull Calibration (works on some of the systematic errors)bull Signal and noise typesbull Fusing samples with Marzullolsquos algorithm Confidence-

Weighted Averaginghellipbull Filtering series with Moving Average hellip

Summary

THE END

Thanks for your attention

45

THE END

Thanks for your attention

45