position calibration of acoustic sensors and actuators

Position Calibration of Acoustic Sensors and Actuators on Distributed General Purpose Computing Platforms

Vikas Chandrakant Raykar | University of Maryland, CollegePark

Motivation

Many multimedia applications are emerging which use multiple audio/video sensors and actuators.

Microphones

Cameras

Speakers

Displays

Dis

trib

ute

d

Cap

ture

Dis

trib

ute

d

Ren

der

ing

Other Applications

Number Crunching

Cur

rent

The

sis

X

What can you do with multiple microphones…

Speaker localization and tracking.

Beamforming or Spatial filtering.

Some Applications…

Audio/Video Surveillance

Smart ConferenceRoomsAudio/Image Based

Rendering

Meeting RecordingSource separation and

Dereverberation

Speech Recognition

Hands free voice communication

Speaker Localizationand tracking

Multichannel speech Enhancement

MultiChannel echoCancellation

Novel Interactive audio Visual Interfaces

More Motivation…

Current work has focused on setting up all the sensors and actuators on a single dedicated computing platform.

Dedicated infrastructure required in terms of the sensors, multi-channel interface cards and computing power.

On the other hand

Computing devices such as laptops, PDAs, tablets, cellular phones,and camcorders have become pervasive.

Audio/video sensors on different laptops can be used to form a distributed network of sensors.

Common TIME and SPACE

Put all the distributed audio/visual input/output capabilities of all the laptops into a common TIME and SPACE.

This thesis deals with common SPACE i.e estimate the 3D positions of the sensors and actuators.

Why common SPACE

Most array processing algorithms require that precise positions of microphones be known.

Painful, tedious and imprecise to do a manual measurement.

This thesis is about..

X

YZ

If we know the positions of speakers….

If distances are not exact

If we have more speakers

X

Y

?

Solve in the least squaresense

If positions of speakers unknown…

Consider M Microphones and S speakers.

What can we measure?Distance between each speaker and all microphones.

Or Time Of Flight (TOF)

MxS TOF matrix

Assume TOF corrupted by Gaussian noise.

Can derive the ML estimate.

Calibration signal

Nonlinear Least Squares..More formally can

derive the ML estimateusing a Gaussian

Noise model

Find the coordinates which minimizes this

Maximum Likelihood (ML) Estimate..

we can define a noise modeland derive the ML estimate i.e. maximize the likelihood ratio

Gaussian noise

If noise is Gaussianand independentML is same asLeast squares

Reference Coordinate SystemReference Coordinate system | Multiple Global minima

X axis

Positive Y axis

OriginSimilarly in 3D

1.Fix origin (0,0,0)

2.Fix X axis

(x1,0,0)

3.Fix Y axis

(x2,y2,0)

4.Fix positive Z axis

x1,x2,y2>0

Which to choose? Later…

On a synchronized platform all is well..

However On a Distributed system..

The journey of an audio sample..

NetworkThis laptop wants to play a calibration signal on the other laptop.

Play comand in software.

When will the sound be actually played out fromThe loudspeaker.

Operating system

Multimedia/multistream applications

Audio/video I/O devices

I/O bus

t

t

jtsSignal Emitted by source j

Signal Received by microphone i

ijFOT ˆ

itmijTOF

Capture Started

Playback Started

Time Origin

On a Distributed system..

Joint Estimation..

Speaker Emission Start Times

S

Microphone Capture Start Times

M -1Assume tm_1=0

Microphone and speakerCoordinates

3(M+S)-6

MS TOF Measurements

Totally

4M+4S-7 parameters to estimates

MS observations

Can reduce the number of parameters

Use Time Difference of Arrival (TDOA)..

Formulation same as above but less number of parameters.

Assuming M=S=K Minimum K required..

Nonlinear least squares..

Levenberg Marquadrat method

Function of a large number of parameters

Unless we have a good initial guess may not convergeto the minima.

Approximate initial guess required.

Closed form Solution..

Say if we are given all pairwise distances between N points can we get the coordinates.

1 2 3 4

1 X X X X

2 X X X X

3 X X X X

4 X X X X

Classical Metric Multi Dimensional Scaling

dot product matrixSymmetric positive definiterank 3

Given B can you get X ?....Singular Value Decomposition

Same as Principal component Analysis

But we can measureOnly the pairwise distance matrix

How to get dot product from the pairwise distance matrix…

k

ijd

kjd

kid

i

j

Centroid as the origin…

Later shift it to our

orignal reference

Slightly perturb each location of GPCinto two to get the initial guess for the microphone and speaker coordinates

Example of MDS…

• Instead of pairwise distances we can use pairwise “dissimilarities”.

• When the distances are Euclidean MDS is equivalent to PCA.

• Eg. Face recognition, wine tasting

• Can get the significant cognitive dimensions.

MDS is more general..

Can we use MDS..Two problems

s1 s2 s3 s4 m1 m2 m3 m4

s1 ? ? ? ? X X X X

s2 ? ? ? ? X X X X

s3 ? ? ? ? X X X X

s4 ? ? ? ? X X X X

m1 X X X X ? ? ? ?

m2 X X X X ? ? ? ?

m3 X X X X ? ? ? ?

m4 X X X X ? ? ? ?

1. We do not have the complete pairwise distances

2. Measured distances Include the effect of lack of synchronization

UNKNOW

N

UNKNOW

N

Clustering approximation…

Clustering approximation…

j i

j j

i j

i i

Finally the complete algorithm…

ApproxDistance matrixbetween GPCs

Approxts

Approx tm

Clustering

Approximation

Dot product matrix

Dimension and coordinate system

MDS to get approx GPC locations

perturb

TOF matrix

Approx. microphone and speaker

locations

TDOA basedNonlinear

minimization

Microphone and speakerlocations tm

Sample result in 2D…

Algorithm Performance… •The performance of our algorithm depends on

•Noise Variance in the estimated distances.•Number of microphones and speakers.•Microphone and speaker geometry

•One way to study the dependence is to do a lot of monte carlo simulations.

•Else can derive the covariance matrix and bias of the estimator.

•The ML estimate is implicitly defined as the minimum of a certain error function.

•Cannot get an exact analytical expression for the mean and variance.

• Or given a noise model can derive bounds on how worst can our algortihm perform.

•The Cramer Rao bound.

Can use implicit function theorem and Taylors series expansion to get approximate expressions for bias and variance.

•J A Fessler. Mean and variance of implicitly defined biased estimators (such as penalized maximum likelihood): Applications to tomography. IEEE Tr. Im. Proc., 5(3):493-506, 1996. •Amit Roy Chowdhury and Rama Chellappa, "Statistical Bias and the Accuracy of 3D Reconstruction from Video", Submitted to International Journal of Computer Vision

Using first order taylors series expansion

Jacobian

Rank Deficit..remove theKnown parameters

Estimator Variance…

Gives the lower bound on the variance of any unbiased estimator.

Does not depends on the estimator. Just the data and the noise model.

Basically tells us to what extent the noise limits our performance i.e. you cannot get a variance lesser than the CR bound.

Jacobian

Rank Deficit..remove theKnown parameters

Different Estimators..

Number of sensors matter…

Number of sensors matter…

Geometry also matters…

Geometry also matters…

Calibration Signal…

• Compute the cross-correlation between the signals received at the two microphones.

• The location of the peak in the cross correlation gives an estimate of the delay.

• Task complicated due to two reasons 1.Background noise. 2.Channel multi-path due to room reverberations.• Use Generalized Cross Correlation(GCC).

• W(w) is the weighting function. • PHAT(Phase Transform) Weighting

Time Delay Estimation…

Time Delay Estimation…

Synchronized setup | bias 0.08 cm sigma 3.8 cm

Mic 3

Mic 1

Mic 2

Mic 4

Speaker 1

Sp

eake

r 4S

pea

ker

2

Speaker 3

X

Z

Roo

m L

engt

h =

4.2

2 m

Room Width = 2.55 m

Room Height = 2.03 m

Distributed Setup…

Initialization phase Scan the network and find the number of GPC’s and the UPnP services available

MasterGPC 1 GPC 2 GPC M

•GPC 1 (Speaker) GPC 2 (Mic)•Calibration signal parameters

TOA Computation

TOATOA matrix

Position estimation

Play Calibration Signal

Play ML Sequence

Experimental results using real data

Related Previous work…

J. M. Sachar, H. F. Silverman, and W. R. Patterson III. Position calibration of

large-aperture microphone arrays. ICASSP 2002

Y. Rockah and P. M. Schultheiss. Array shape calibration using sources in unknown

locations Part II: Near-field sources and estimator implementation. IEEE Trans. Acoust.,

Speech, Signal Processing, ASSP-35(6):724-735, June 1987.

J. Weiss and B. Friedlander. Array shape calibration using sources in unknow locations a maximum-likelihood approach. IEEE Trans. Acoust., Speech, Signal Processing , 37(12):1958-1966, December 1989.

R. Moses, D. Krishnamurthy, and R. Patterson. A self-localization method for wireless

sensor networks. Eurasip Journal on Applied Signal Processing Special Issue on Sensor

Networks, 2003(4):348{358, March 2003.

index.htm

Our Contributions…

•Novel setup for array processing.

•Position calibration in a distributed scenario.

•Closed form solution for the non-linear minimization routine.

•Expression for the mean and variance of the esimators.

•Study the effect of sensor geometry.

Acknowledgements…

• Dr. Ramani Duraiswami and Prof. Rama Chellappa

• Prof. Yegnanarayana

• Dr. Igor Kozintsev and Dr. Rainer Lienhart, Intel Research

• Prof. Min Wu and Prof. Shihab Shamma

• Prof. Larry Davis

Thank You ! | Questions ?