A 2D LIDAR SYSTEM FOR SPACE APPLICATIONS

Dipl.-Ing. Sebastian Ehrenreich

von Hoerner & Sulger GmbH, Schlossplatz 8, 68723 Schwetzingen, Germany, ehrenreich@vh-s.de

ABSTRACT

A prototype of a low power, low mass fully autonomous2D Light Detection and Ranging (LIDAR) system is pre-sented. A sample application is scanning of the sur-rounding area for the navigation of autonomous rovers.The system features a distance measurement range of0 . . . 74.5 m and an angular coverage of 336. It usesa quasi continuous wave (qcw) square signal modulatedlaser diode for high efficiency and a software defined re-ceiver concept for the distance measurement based onsignal phase shift. Maximum supported sample rate is39.06 kHz. Measurement speed can be traded against ac-curacy and vice versa.

Key words: LIDAR, phase shift, CORDIC, DSP, rover.

1. INTRODUCTION

Many space applications require a fast and precise rangefinder device, which should be of low mass and volume.Sample applications are:

• Rendezvous and docking of two spacecraft in space

• Landing of a spacecraft on planetary/asteroid sur-face

• Rover navigation on Mars/Moon

• Optical metrology for spacecraft formation flying

The maximum supported distance range, as well as theaccuracy requirements vary for these applications and arealso mission phase dependent. For instance, early ren-dezvous maneuvers require just a coarse resolution (<1 m) at distances up to 5km, whereas in the approach anddocking phase ranges < 50 m with better accuracies arerequired. The presented 2D Light Detection and Rang-ing (LIDAR) system has been optimized for autonomousrover navigation applications for detection of the posi-tion of (non-cooperative) obstacles in the periphery 40maround the rover with accuracies of ±1 cm. However, themeasurement principle described in this paper allows totrade accuracy vs. maximum distance range, basically

supporting a wide set of applications.The document is organized as follows: Part 2 discussesthe electrical and system level design aspects, whereassection 3 deals with the optical and mechanical design.Part 4 lists some performance metrics of the prototype,followed by a conclusion in section 5.

2. ELECTRICAL SYSTEM ARCHITECTURE

The presented instrument measures the phase delay be-tween a quasi continuous wave (qcw) modulated lasersignal and the back scattered received signal. For bestefficiency, a square wave modulation signal is used todrive a low power solid-state laser diode. For the phasedelay measurement, a direct-conversion software definedreceiver approach is used. A full block diagram is shownin Fig. 1. Key component is an Actel ProAsic3 field pro-grammable gate array (FPGA), which can be easily sub-stituted by a radiation hardened RTAX device for a flightmodel. The FPGA generates the laser diode square wavemodulation signal, which is user adjustable in steps be-tween f0 = 2 and 5 MHz. In addition, the laser outputpower is user controllable.The reflected laser light is received by an avalanche photodiode (APD) and amplified by several programmablegain stages. The APD bias high voltage is fully con-trolled by the micro-controller built into the FPGA. Thefirmware contains a temperature compensation curve forthe APD bias to maintain maximum sensitivity over thecomplete operating range. The programmable gain stageshave a bandpass transfer function, limiting the total noisebandwidth and acting as an anti-aliasing filter for the16 bit, 40MS/s analog-to-digital converter (ADC) at thesame time. The prototype uses a commercial of theshelf (COTS) device from Analog Devices, however rad-hard replacements are available. Preamplifier gain isprogrammable to 0, 20 and 40dB. All remaining signal-processing is done in the digital domain inside the FPGA.The digital signal processing (DSP) logic part consists oftwo I/Q mixers followed by a second order infinite im-pulse response (IIR) low-pass filter (fc = 50 kHz). TheIIR filter reduces the noise bandwidth and does there-fore improve overall signal-to-noise ratio (SNR) of thereceiver. The IIR filter output is connected to a deci-mating cascaded-integrator-comb (CIC) filter. Besidesof the down sampling functionality, the CIC filter does

DSP

Q

I

8051

Core

Ethernet

Laser Driver

fs=40MHz

ADCTIA

Bandpass

fc=5MHz

0/20dB 0/20dB

sin(phi)

cos(phi)

programmable rate

IIR

IIR

CIC

CIC

Decimation filter

2nd

2nd

50kHz LP Filter

cos(f0)

sin(f0)

Q

I

Estimator

Magnitude

16CORDIC

I

Q

32

32

Scalerϕ

32

mag

AGC

−> dϕ

DSP Core

Target

Reference LED Driver

VapdAPD Bias Control

f0=4MHz

Reference/Calibration signal

ϑ

Angular Sensor

Motor

Motor CtlPID

24

Power Control

f0=4MHz

IR LED

Housekeeping

FPGA

Overdrive

32:1 CIC Filter

32:1 CIC Filter

Figure 1. Block diagram of the proposed 2D-LIDAR system.

also act as a high order low-pass filter for the mixeroutput signal, which significantly improves the attenu-ation of the carrier and other unwanted mixer productsin the output signal. Actually, only the DC componentis of interest. The CIC filter decimation rate is pro-grammable by the user, supporting final sample frequen-cies of fs = 0.6 Hz . . . 39.06 kHz. The phase shift be-tween sent and received signal can be calculated usingEq. 1:

φ = arctanQ

I(1)

This operation is performed by a 32 bit wide, 31 iterationsteps serial CORDIC [1, 2] operating in vectoring modewhich is implemented into the FPGA [3]. The 32 bitwide data-bus supports distance resolutions of 1/100mm.The CORDIC does not only determine the phase ofthe complex I/Q vector, but also the length

√I2 +Q2

(neglecting the CORDIC gain here) is computed at thesame time. The vector length is an indicator for thereceived signal strength (RSSI), which is used by theautomatic gain control (AGC) logic to adjust the frontendgain. In addition, the signal strength value is sent to theuser along with the distance measurement, because itcontains valuable information regarding the reflectivityof the target.

Due to the large dynamic range of the input signal, AGCis a key building block. As already mentioned, the RSSIvalue is one input to the AGC, however, the value is fore-most valid at the end of a measurement cycle. Therefore,another input to the AGC is a signal strength estimateoriginating from a fast but low accuracy vector length es-timator block, directly using the raw output of the I/Qmixers. The estimator uses the approximation [4]:

RSSI ≈

I + 1

2Q for I ≥ QQ+ 1

2I for I < Q(2)

Using this simplification, the estimator can be imple-mented with low hardware overhead. A 32:1 decima-tor located before the input of the magnitude estimatorgives a good compromise between speed and accuracy ofthe estimation. The third input to the AGC block is theADC overdrive signal, which instantly triggers a frontendgain change. The data processing pipeline is automati-cally flushed in case a gain change has occurred, which isespecially important for the digital filters to get rid of badvalues due to over- or underdrive conditions.One challenge with the AGC is the change in intrinsicpreamplifier phase delay for different gain settings. Toaccommodate for this effect, a novel self calibration tech-nique has been deployed. During the periodic instrumentself-calibration, the laser signal is switched off and a ref-erence signal emitted by a light emitting diode (LED) is

activated instead. The power of the LED is adjusted bythe algorithm to a value, which is in the overlapping re-gion of two adjacent gain steps. A series of phase delaymeasurements using both gain settings is performed, re-sulting in a calibrated value for the intrinsic phase delay.This algorithm is repeated for all gain settings.Data frame transfer and telecommand processing is per-formed by a 8051 micro-controller implemented as softcore in the FPGA. The prototype uses an Ethernet in-terface for the actual communication, however other in-terfaces, e.g. SpaceWire can be easily implemented in-stead. To support the micro-controller, conversion of theCORDIC phase measurement into a distance value basedon Eq. 3 is done by a dedicated arithmetic block, signifi-cantly reducing processor load.

d =c

4πf0· φ (3)

In this equation, c is the speed of light, f0 is the signalmodulation frequency. The micro-controller does alsocare for the housekeeping data, e.g. temperatures of theCPU and APD.A DC motor is used to drive the scan head. The H-bridgedriver electronics is controlled by a fully digital PID reg-ulator built into the FPGA. Besides of a constant speedmode, the system does also support a ”pointing mode”,which allows commanding of the scan head to a user de-fined position. A 4096 steps incremental encoder directlyattached to the DC motor shaft is used for angular mea-surement. Together with the gear ratio of the drive anangular resolution of 0.0107 is achieved.

2.1. Maximum Supported Distance and Accuracy

There are several effects limiting the maximum supporteddistance and the measurement accuracy. One hard limit isthe signal ambiguity due to the cw modulation of the lasersignal once the signal is delayed by more than one modu-lated signal period. The default modulation frequency off0 = 4 MHz allows for a maximum range of d ≈ 37.5 m.Using Eq. 1, the phase measurement under noisy condi-tions can be modelled as follows:

φ+ ∆φ = arctanε+ 1

2A sinφ

ε+ 12A cosφ

(4)

In this expression, ∆φ is the phase measurement errorcaused by the amplitude error/noise ε. A is the amplitudeof the received carrier signal. The expression reflects thesignal after the mixer and associated low-pass filters. Theatan function makes it difficult to find a general closedexpression for ε, however, for φ = 0, the expression canbe simplified1 to:

For φ = 0:

∆φ = arctanε

ε+ 12A

(5)

1Function plots show, that φ = 0 does not result in the maximumexpected error, however the absolute maximum seems to be within 10%of this value.

And further:

For ε 1

2A:

∆φ ≈ arctan2ε

A≈ 2ε

A(6)

∆d =c

2πf0· ε

A︸︷︷︸SNR−1

(7)

Eq. 7 can be used to calculate the maximum allowednoise level to achieve a certain distance measurement ac-curacy, e.g. for an amplitude of A = 1 V, ∆d = 1 cmand f0 = 5 MHz, the resulting noise level is ε =1.05 · 10−3 Vrms. This equals a SNR of 60dB. Un-der the assumption of a total frontend noise bandwidthof 20 kHz, the resulting allowed noise density would bee = 7.41µV/

√Hz.

Fig. 2 shows the result of a model to hardware corre-lation experiment. A signal generator has been directlyconnected to the frontend input, allowing to adjust thesignal amplitude without affecting the signal phase. Sig-nal generator and FPGA share the same reference clockand are therefore phase locked. The frontend amplifierhas been switched to its maximum gain of 40dB, result-ing in an expected noise level of ε = 496.4µV, which isbased on a circuit noise analysis not discussed here. TheRSSI readout as generated by the CORDIC can be di-rectly translated into the signal amplitude, which allowsplotting the accuracy curve together with the actual hard-ware measurement samples. As one can see, there is a

3.6

3.8

4

4.2

4.4

4.6

4.8

5

0 10000 20000 30000 40000 50000

d/m

RSSI

Length Measurement, 40dB gain, Signal generator input

MeasurementAccuracy expression

Figure 2. Accuracy prediction vs. hardware measurement

good match of the distance sample distribution and theexpected accuracy envelope as defined by Eq. 7. Forhigh RSSI values, prediction is a bit optimistic, due tothe simplifications done for solving Eq. 4 and the signalgenerator phase noise that has been neglected in the anal-ysis.Accuracy of the system can be improved by two means,either by increasing the modulation frequency f0 or byimproving the SNR. The SNR is influenced by many pa-rameters, e.g. the preamplifier and APD noise, opticsaperture and also the ADC resolution. These are mainlydefined by the system design. Other parameters affecting

the SNR are the laser output power, target distance andtarget reflectivity which depend on the environment anddevice setup. It is possible to improve the SNR by usingmore samples, resulting in a reduction of the RMS noisebased on the 1/

√N law. This is one reason for the high

oversampling rate of N = 1024 at an output data rateof 39.06 kHz, which improves the SNR by 30dB. Fig. 3shows a comparison of measured and calculated 1σ rangemeasurement accuracies using a static target. The curve

0

0.5

1

1.5

2

2.5

3

3.5

0.01 0.1 1 10 100

σ/m

m

Sample rate/kHz

Standard deviation using static target (d = 4m), 1000 samples

MeasurementTheoretical curve

Figure 3. 1σ distance measurement noise vs. data rate.0dB gain setting, static target (white wall) in 4 m dis-tance, 25 mW Laser power, f0 = 4 MHz modulation fre-quency

saturates for very low data rates, due to long term vari-ations of the laser modulation characteristics which arenot compensated in this setup. Nevertheless, an accuracybetter than 0.5mm is achieved.In addition to Eq. 7, Eq. 8 can be used for estimationof the maximum achievable distance dmax based on anaccuracy target ∆d. It is based on the well known radarequation [5]:

dmax =

√√√√√∆d2πf0c︸ ︷︷ ︸

SNR−1

·Pτηa2 · S ·R ·G

8eG ·√fs

(8)

With parameters:

P Optical Laser output power [W]τ Transmission of the transmitter and re-

ceiver optics (= τR · τT )η Reflectivity of the targeteG Preamp gain dependent frontend RMS

noise density [V/√

Hz]fs Decimated sample rate [Hz]a Aperture diameter of the receiver optics

[m]∆d Range measurement error limit [m]f0 Modulation frequency [Hz]S APD sensitivity [A/W]R Transimpedance amplifier transconduc-

tance [Ω]G Preamplifier gain

02468

1012141618202224262830

0.01 0.1 1 10 100

dmax

/m

Sample Rate/kHz

Max. Distance vs. Sample Rate, 25mW Laser, f0 = 4MHz

±5mm acc.±1cm acc.

±5cm acc.±0.5mm acc.

Figure 4. Calculated max. distance vs. sample rate usingEq. 8, 40dB gain setting, target with 20% reflectivity,5cm diameter receiver optics, 25mW Laser power, f0 =4 MHz modulation frequency

The presented 2D-LIDAR system allows the user to se-lect the modulation frequency, the net sample rate andthe laser power online. This enables a wide range of ap-plications, like fast raw scans for situational awarenessfollowed by detailed low speed scans for exact values inthe millimeter range. Laser output power is mainly a con-cern for applications on earth, where eye-safety is crit-ical. For space applications, increasing the laser powersignificantly extends the measurement range, especiallyfor non-cooperative targets.

3. OPTICAL AND MECHANICAL DESIGN

The objective of the optical and mechanical design is tofind a small and low mass solution with an as large aspossible aperture of the receiving optics and a large an-gular coverage. Fig. 5 shows a schematic diagram ofthe concept used. The laser diode is mounted on a small”gibbet” above the center axis of the rotating scan head.The mount is designed as narrow as possible to keep theangular coverage of the instrument high. As a positiveside effect, the gibbet itself is used as a reference targetfor the automatic offset calibration taking place on eachspinner rotation. The gain stage phase mismatch calibra-tion sequence is also performed during this ”dead time”,when the laser and receiving optics are partly masked bythe laser mount. Exiting the laser diode, the light beam isdeflected to the target using a 90 plane mirror. The lightscattered back by the object is received by a 90 off-axisparabolic mirror with a free aperture of d = 50 mm. TheAPD is located at the focal point of the mirror, whichis placed along the rotational axis. With this approach,the light emitter and the receiver are both mounted stati-cally, which simplifies the electrical connections, becauseno sliding contacts are necessary. The initial optics con-

Laser Diode

off−axisparabolic mirror

parallax

rotating mounting plate

compensationmirror

fixed APD mount

APD

to target

fixed laser mount

IR mirror

from near target

from far target

Figure 5. Optical system schematic view.

cept also planned for a parallax compensation mirror, de-flecting the light received by near targets, which other-wise would fall outside the APD active area. However,prototype evaluation results show, that even for near tar-gets there is enough scattered light falling on the APD togive a stable distance readout even without the compensa-tion mirror. This effect does actually reduce the dynamicrange requirements for the frontend electronics, becausethe high reflected light levels of near targets are attenu-ated by the optical misalignment.All optical surfaces can be adjusted. The alignment ofthe laser is adjustable to align it with the rotational axis.The laser deflection mirror can be tilted around all axes,giving full control to the laser beam pointing direction.The tilt of the receiving parabolic mirror can be adjustedas well. Finally, the distance of the APD assembly fromthe parabolic mirror can be changed, too. This makes itpossible to put the APD in the focal plane of the receivingoptics.The spinner is driven by a cogwheel gear. A large cog-wheel with sufficient inner diameter to not shadow thereceiver optics is mounted to the spinner. This gear isdriven by a small cogwheel directly hooked up to the DCmotor, ending up with a gear ratio of 8.18. Fig. 6 showsa sectional view of the prototype mechanics. The picturealso shows the gear drive and the location of the printedcircuit boards within the chassis. Basic outer shape of theinstrument is a tube, which is optimal for a rotating sys-tem. The APD receiver assembly is located right belowthe spinner. This component does also carry the referencesignal LED used by the gain step phase offset calibrationprocess. The LED light is reflected back to the APD byan infrared (IR) mirror located in front of the APD, pri-mary used to reduce the influence of ambient light on thevery sensitive receiver. This setup ensures a constant andstable coupling of the reference LED signal even with thespinner rotating. Fig. 7 shows the prototype without theprotective cover made of translucent plastics. The coverdoes also act as optical bandpass filter, complemented bythe IR mirror in front of the APD. One can see the laser

Reference LEDAPD

Laser diode

Figure 6. Instrument sectional view.

mount (”gibbet”) and the optical surfaces as described be-fore.

4. EVALUATION RESULTS

Evaluation of the built prototype is still in progress. Tab.1 gives an overview of the performance figures. Please

Table 1. Prototype system key metrics.Item Demonstrated value

Mass m = 2.44 kg w/o coverVolume 186×140 mm3

Angular coverage 336

Angular resolution 0.0107

Scan head speed 0 . . . 400 rpmNet sample rate 0.6 Hz . . . 39.06 kHz

Theo. max. distance 74.5 m (@f0 = 2 MHz)Range resolution 1/100 mm

Input voltage range 8..32 VPower consumption Activea: 6.96 W

Idleb: 3.00 WLaser Class 3R

λ = 658 nmP0 = 25 mW

Modulation freq. 2..5 MHzInterface 10/100MBit Ethernet

aSpinner running at 91 rpm, Ethernet on, Laser power ≈ 20mW.bJust FPGA powered up, no Ethernet transfer.

note that the prototype as shown in Fig. 7 is a feasibilitystudy. The design has been trimmed for easy and cost ef-ficient manufacturing. Hence it is possible to reduce the

Figure 7. Picture of the LIDAR prototype system.

mass significantly by milling out unnecessary materialand replacing the bearings used against low mass ones.A target mass for the second design pass of < 1 kg is an-ticipated.A control software application has been developed fordemonstration and test purposes. It shows a real timeradar screen like picture based on the LIDAR measure-ments. A sample screenshot is shown in Fig. 8. It showsthe scan of an office room surrounded by shelves.

5. CONCLUSION

A medium range, fully autonomous prototypical 2D LI-DAR system design for space applications has been pre-sented. The system uses a software defined I/Q demod-ulator based approach to measure the phase delay of thereceived vs. the sent, qcw modulated laser signal. Themodulation frequency and sample rate can be changedby telecommands, influencing maximum range and ac-curacy of the instrument. This allows selecting the bestsetup matching the application requirements and missionphase, e.g. high speed but low accuracy data acquisitionat the approach phase and high accuracy but limited speedmode in the final docking phase. The system supports netrange sampling frequencies as high as 39.06 kHz. As an-other feature, the received signal strength for each mea-surement point is output as well, enabling the offline cal-culation of the target reflectivity. Usage of a solid-state

Figure 8. LIDAR demonstration software screenshotshowing an office room scan with shelves.

laser diode significantly reduces the mass and power con-sumption of the device. For easy interfacing with a space-craft, the Ethernet communication interface used by theprototype can be easily exchanged against a SpaceWireconnection.The evaluation results of the prototype are promisingand demonstrate the maturity of the design. Utiliza-tion of the Actel ProAsic3E1500 prototyping device is at69.9%, giving enough room for further algorithmic im-provements.

ACKNOWLEDGMENTS

This work was supported by the Deutsche Zentrumfur Luft- und Raumfahrt, DLR, Bonn. Contract No.FZ50RA0912.

REFERENCES

[1] Kishore Kota, Joseph R. Cavallaro, 1993, NumericalAccuracy and Hardware Tradeoffs for CORDIC Arith-metic for Special-Purpose Processors, IEEE Transac-tions on Computers, Vol. 42, No 7, July 1993

[2] Sang Yoon Park, Nam Ik Cho, 2002, Fixed Point Er-ror Analysis Of CORDIC Processor Based On TheVariance Propagation, School of Electrical Engineer-ing Seoul, Seoul National University, Seoul 151-744,Korea

[3] Actel, 2010, CoreCordic V3.0 Handbook, Actel Cor-poration

[4] Allie, M. and Lyons, R., 2005, A root of less evil[digital signal processing], IEEE Signal ProcessingMagazine, Vol. 22, No 2, pp. 93-96.

[5] Sami Kurtti, Juha Kostamovaara, 2009, Laser RadarReceiver Channel With Timing Detector Based onFront End Unipolar-to-Bipolar Pulse Shaping, IEEEJournal of Solid-State Circuits, Vol. 44, No 3, pp. 835-847.