Disclaimer:
This material is presented to ensure timely dissemination of
scholarly and technical work. Copyright and all rights therein are
retained by authors or by other copyright holders. All person copying
this information are expected to adhere to the terms and constraints
invoked by each author's copyright. In most cases, these works may not
be reposted without the explicit permission of the copyright holder.
Submitted Journal Publications:
 R.
G. Baraniuk, V. Cevher, M. F. Duarte, and C. Hegde, “Modelbased
compressive sensing,” submitted to IEEE Transactions on
Information Theory.
Abstract
BiBTeX
Paper:
PDF
Compressive sensing (CS) is an alternative to
Shannon/Nyquist sampling for acquisition of sparse
or compressible signals that can be well
approximated by just K << N elements from an
Ndimensional basis. Instead of taking periodic
samples, we measure inner products with M < N random
vectors and then recover the signal via a sparsityseeking
optimization or greedy algorithm. The standard CS
theory dictates that robust signal recovery is
possible from M = O(K log(N/K)) measurements. The
goal of this paper is to demonstrate that it is
possible to substantially decrease M without
sacrificing robustness by leveraging more realistic
signal models that go beyond simple sparsity and
compressibility by including dependencies between
values and locations of the signal coefficients. We
introduce a model based CS theory that parallels the
conventional theory and provides concrete guidelines
on how to create modelbased recovery algorithms
with provable performance guarantees. A highlight is
the introduction of a new class of
modelcompressible signals along with a new
sufficient condition for robust model compressible
signal recovery that we dub the restricted
amplification property (RAmP). The RAmP is the
natural counterpart to the restricted isometry
property (RIP) of conventional CS. To take practical
advantage of the new theory, we integrate two
relevant signal models — wavelet trees and block
sparsity — into two stateoftheart CS recovery
algorithms and prove that they offer robust recovery
from just M = O(K) measurements. Extensive numerical
simulations demonstrate the validity and
applicability of our new theory and algorithms.
@Unpublished{IEEE_TSP_Cevher_UNP2,
author = "R. G. Baraniuk and V. Cevher and M. F.
Duarte and C. Hegde", title = "Modelbased
compressive sensing", note = "preprint, under revision at IEEE Transactions on
Information Theory",}
Journal:

R. G. Baraniuk, V.
Cevher, and M. Wakin, “Lowdimensional models for
dimensionality reduction and signal recovery: A geometric
perspective,” accepted to Proceedings of the IEEE.
Abstract
BiBTeX
Paper:
PDF
We compare and contrast from a geometric perspective a
number of lowdimensional signal models that support stable
informationpreserving dimensionality reduction. We consider
sparse and compressible signal models for deterministic and
random signals, structured sparse and compressible signal
models, point clouds, and manifold signal models. Each model
has a particular geometrical structure that enables signal
information in to be stably preserved via a simple linear
and nonadaptive projection to a much lower dimensional space
whose dimension either is independent of the ambient
dimension at best or grows logarithmically with it at worst.
As a bonus, we point out a common misconception related to
probabilistic compressible signal models, that is, that the
generalized Gaussian and Laplacian random models do not
support stable linear dimensionality reduction.
@Unpublished{PROC_IEEE_BCW,
author = "R. G. Baraniuk and V. Cevher and M. Wakin", title = "Lowdimensional
models for dimensionality reduction and signal recovery: A
geometric perspective", note = "accepted to
Proceedings of the IEEE",}

V. Cevher and L.
Kaplan, “Acoustic sensor network design for position estimation,”
ACM
Transactions on Sensor Networks, vol. 4, no. 3, May
2009.
Abstract
BiBTeX
Paper:
PDFIn this paper, we develop tractable mathematical
models and approximate solution algorithms for a
class of integer optimization problems with
probabilistic and deterministic constraints, with
applications to the design of distributed sensor
networks that have limited connectivity. For a given
deployment region size, we calculate the Pareto
frontier of the sensor network utility at the
desired probabilities for dconnectivity and
kcoverage. As a result of our analysis, we
determine (i) the number of sensors of different
types to deploy from a sensor pool, which offers a
cost vs. performance tradeoff for each type of
sensor, (ii) the minimum required radio transmission
ranges of the sensors to ensure connectivity, and
(iii) the lifetime of the sensor network. For
generality, we consider randomly deployed sensor
networks and formulate constrained optimization
techniques in a Bayesian experimental design
framework to obtain the best point estimates of a
given stateofnature, represented by a finite
number of parameters. The approach is guided and
validated using an unattended acoustic sensor
network design. Finally, approximations of the
complete statistical characterization of the
acoustic sensor networks are given, which enable
average network performance predictions of any
combination of acoustic sensors.
@article{ACM_TOSNCevher,
author = "V. Cevher and L. Kaplan", title = "Acoustic
sensor network design for position estimation", journal = "ACM
Transactions on Sensor Networks", volume = "5", number = "3", year = "May
2009",}
 V. Cevher,
R. Chellappa, and J. H. McClellan, “Vehicle speed estimation using
acoustic wave patterns,” to be published at IEEE Transactions on Signal
Processing.
Abstract
BiBTeX
Paper:
PDFWe estimate a vehicle’s speed, wheelbase length, and its
tire track length by jointly estimating its acoustic
wavepattern using a single passive acoustic sensor that
records the vehicle’s driveby noise. The acoustic
wavepattern is determined using three envelope shape (ES)
components, which approximate the shape variations of the
received signal’s power envelope. We incorporate the
parameters of the ES components along with estimates of the
vehicle engine RPM and the number of cylinders, and the
vehicle’s loudness and speed to form a vehicle profile
vector. This vector provides a compressed statistics that
can be used for vehicle identification and classification.
We also provide possible reasons for why some of the
existing methods are unable to provide unbiased vehicle
speed estimates using the same framework. The approach is
illustrated using vehicle speed estimation and
classification results obtained with field data.
@Unpublished{IEEE_TSP_Cevher_UNP2, author = "V. Cevher and R. Chellappa and J. H. McClellan", title = "Vehicle speed estimation using acoustic wave
patterns", note = "accepted at IEEE Transactions on Signal
Processing",}
 V. Cevher,
R. Velmurugan, and J. H. McClellan, “Acoustic multi target tracking
using directionofarrival batches,”
IEEE Transactions on Signal Processing,
vol. 55, no. 6, pp. 28102825, June 2007.
Abstract
BiBTeX Paper:
PDF
In
this paper, we propose a particle filter acoustic
directionofarrival (DOA) tracker to track multiple
maneuvering targets using a state space approach. The
particle filter determines its state vector using a batch of
DOA estimates. The filter likelihood treats the observations
as an image, using template models derived from the state
update equation, and also incorporates the possibility of
missing data as well as spurious DOA observations. The
particle filter handles multiple targets, using a
partitioned statevector approach. The particle filter
solution is compared with three other methods: the extended
Kalman filter, Laplacian filter, and another particle filter
that uses the acoustic microphone outputs directly. We
discuss the advantages and disadvantages of these methods
for our problem. In addition, we also demonstrate an
autonomous system for multiple target DOA tracking with
automatic target initialization and deletion. The
initialization system uses a trackbeforedetect approach
and employs the matching pursuit idea to initialize multiple
targets. Computer simulations are presented to show the
performances of the algorithms.
@article{IEEE_TSP_Cevher_MMT, author = "V.
Cevher and R. Velmurugan and J. H. McClellan", title = "Acoustic
multi target tracking using directionofarrival batches", journal = "IEEE
Transactions on Signal Processing", volume = "55", number = "6", pages = "28102825", year = "June 2007",}
 V. Cevher,
A. Sankaranarayanan, J. H. McClellan, and R. Chellappa, “Target
tracking using a joint acoustic video system,”
IEEE
Transactions on Multimedia,
vol. 9, no. 4, pp. 715727, June 2007.
Abstract
BiBTeX Paper:
PDF
In this paper, we present a particle filter that exploits
multi modal information for robust target tracking. We
demonstrate a Bayesian framework for combining acoustic and
video information using a state space approach. A proposal
strategy for joint acoustic and video statespace tracking
using particle filters is given by carefully placing the
random support of the joint filter where the final posterior
is likely to lie. By using the KullbackLeibler divergence
measure, it is shown that the joint filter posterior
estimate decreases the worst case divergence of the
individual modalities. Hence, the joint tracking filter is
robust against video and acoustic occlusions. We also
introduce a timedelay variable to the joint state space to
handle the acousticvideo data synchronization issue, caused
by acoustic propagation delay. Computer simulations are
presented with field and synthetic data to demonstrate the
filter’s performance.
@article{IEEE_MM05_Cevher, author = "V.
Cevher and A. Sankaranarayanan and J. H. McClellan and R.
Chellappa", title = "Target tracking using a
joint acoustic video system", journal = "IEEE
Transactions on Multimedia", volume = "9", number = "4", pages = "715727", year = "June 2007",}
 M. Alam, V.
Cevher, J. H. McClellan, G. D. Larson, and W. R. Scott Jr. “Optimal
maneuvering of seismic sensors for localization of subsurface
targets,”
IEEE
Transactions on Geoscience and Remote Sensing,
vol. 45, no. 5, pp. 12471257, May 2007.
Abstract
BiBTeX Paper:
PDF
We consider the problem of detecting and locating buried
land mines and subsurface objects by using a maneuvering
array that receives scattered seismic surface waves. We
demonstrate an adaptive system that moves an array of
receivers according to an optimal positioning algorithm
based on the theory of optimal experiments. The goal is to
minimize the number of distinct measurements (array
movements) needed to localize mines. The adaptive
localization algorithm has been tested using experimental
data collected in a laboratory facility at Georgia Tech. The
performance of algorithm is exhibited for cases with one or
two targets and in the presence of common types of clutter
like rocks found in the soil. It has also been tested for
the case where the propagation properties of the medium vary
spatially. In almost all test cases the mines were located
exactly using three or four array movements. It is
envisioned that future systems could incorporate this new
method into a portable mobile minelocation system.
@article{IEEE_TGRS05_ALAM, author = "M. Alam
and V. Cevher and J. H. McClellan and G. D. Larson and W. R.
Scott Jr.", title = "Optimal maneuvering of
seismic sensors for localization of subsurface targets", journal = "IEEE
Transactions on Geosciences and Remote Sensing", volume = "45", number = "5", pages = "12471257", year = "May 2007",}
 M. Borkar,
V. Cevher, and J. H. McClellan, “A MonteCarlo method for
initializing distributed tracking algorithms with acoustic
propagation delay compensation,” Journal of VLSI Signal Processing
Systems, vol. 48, no. 12, pp. 109125, August 2007.
Abstract
BiBTeX Paper:
PDF
Decentralized processing algorithms are attractive
alternatives to centralized algorithms for target tracking
applications in smart sensor networks since they provide the
ability to scale, reduce vulnerability, reduce communication
and share processing responsibilities among individual
nodes. Sharing the processing responsibilities allows
parallel processing of raw data at the individual nodes.
However, this introduces other difficulties in multimodal
smart sensor networks, such as non observability of the
target state at any individual node and various delays such
as varying processing delays, communication delays and
signal propagation delays for the different modalities. In
this paper, we provide a novel algorithm to determine the
initial probability distribution of multiple target states
in a decentralized manner. The targets state vector consists
of the target positions and velocities on the 2D plane. Our
approach can determine the state vector distribution even if
the individual sensors alone are not capable of observing
it. Our approach can also compensate for varying delays
among the assorted modalities. The resulting distribution
can be used to initialize various tracking algorithms. Our
approach is based on MonteCarlo methods, where the state
distributions are represented as a weighted set of discrete
state realizations. A robust weighting strategy is
formulated to account for missed detections, clutter and
estimation delays. To demonstrate the effectiveness of the
algorithm, we simulate a network with directionofarrival
nodes and rangedoppler nodes.
@article{JVLSI05_Borkar, author = "M. Borkar and V. Cevher and J. H. McClellan", title = "A
MonteCarlo method for initializing distributed tracking
algorithms with acoustic propagation delay compensation", journal = "Journal
of VLSI Signal Processing Systems", volume = "48", number = "12", pages = "109125", year = "August 2007",}
 M. Borkar,
V. Cevher, and J. H. McClellan, “Low computation and low latency
algorithms for distributed sensor network initialization,” Signal, Image and Video Processing (Springer),
vol. 1, no.2, pp 133148, June 2007.
Abstract
BiBTeX Paper:
PDF
In this paper, we show how an underlying system’s state
vector distribution can be determined in a distributed
heterogeneous sensor network with reduced subspace
observability at the individual nodes. The presented
algorithm can generate the initial state vector distribution
for networks with a variety of sensor types as long as the
collective set of measurements from all the sensors provides
full state observability. Hence the network, as a whole, can
be capable of observing the target state vector even if the
individual nodes are not capable of observing it locally.
Initialization is accomplished through a novel distributed
implementation of the particle filter that involves serial
particle proposal and weighting strategies that can be
accomplished without sharing raw data between individual
nodes. If multiple events of interest occur, their
individual states can be initialized simultaneously without
requiring explicit data association across nodes. The
resulting distributions can be used to initialize a variety
of distributed joint tracking algorithms. We present two
variants of our initialization algorithm: a low complexity
implementation and a low latency implementation. To
demonstrate the effectiveness of our algorithms we provide
simulation results for initializing the states of multiple
maneuvering targets in smart sensor networks consisting of
acoustic and radar sensors.
@article{SIVP07_Borkar, author = "M. Borkar
and V. Cevher and J. H. McClellan", title = "Low
computation and low latency algorithms for distributed
sensor network initialization", journal = "Signal,
Image and Video Processing (Springer)", volume = "1", number = "2", pages = "133148", year = "June 2007",}
 V. Cevher
and J. H. McClellan, “Acoustic node calibration using moving sources,”
IEEE Transactions on Aerospace and Electronic Systems, vol.
42, no. 2, pp 585600, April 2006.
Abstract
BiBTeX Paper:
PDF
Acoustic nodes, each
containing an array of microphones, can track targets in xy
space from their received acoustic signals, if the node
positions and orientations are known exactly. However, it is
not always possible to deploy the nodes precisely, so a
calibration phase is needed to estimate the position and the
orientation of each node before doing any tracking or
localization. An acoustic node can be calibrated from
sources of opportunity such as beacons or a moving source.
In this paper, we derive and compare several calibration
methods for the case where the node can hear a moving source
whose position can be reported back to the node. Since
calibration from a moving source is, in effect, the dual of
a tracking problem, methods derived for acoustic target
trackers are used to obtain robust and high resolution
acoustic calibration processes. For example, two
directionofarrivalbased calibration methods can be
formulated based on combining angle estimates, geometry, and
the motion dynamics of the moving source. In addition, a
maximumlikelihood (ML) solution is presented using a
narrowband acoustic observation model, along with a
Newtonbased search algorithm that speeds up the calculation
the likelihood surface. The CramerRao lower bound on the
node position estimates is also derived to show that the
effect of position errors for the moving source on the
estimated node position is much less severe than the
variance in angle estimates from the microphone array. The
performance of the calibration algorithms is demonstrated on
synthetic and field data.
@article{IEEE_AES_CEVHER06, author = "V. Cevher and J. H. McClellan", title = "Acoustic node calibration using moving sources", journal = "IEEE Transactions on Aerospace and Electronic
Systems", volume = "42", number = "2", pages = "585600", year = "April 2006",}
 V. Cevher
and J. H. McClellan, “General directionofarrival tracking with acoustic
nodes,” IEEE Transactions on Signal Processing, vol. 53, no.
1, pp. 112, January 2005.
Abstract
BiBTeX Paper:
PDF
In this paper, we propose a particle filter acoustic
directionofarrival (DOA) tracker to track multiple
maneuvering targets using a state space approach. The
particle filter determines its state vector using a batch of
DOA estimates. The filter likelihood treats the observations
as an image, using template models derived from the state
update equation, and also incorporates the possibility of
missing data as well as spurious DOA observations. The
particle filter handles multiple targets, using a
partitioned statevector approach. The particle filter
solution is compared with three other methods: the extended
Kalman filter, Laplacian filter, and another particle filter
that uses the acoustic microphone outputs directly. We
discuss the advantages and disadvantages of these methods
for our problem. In addition, we also demonstrate an
autonomous system for multiple target DOA tracking with
automatic target initialization and deletion. The
initialization system uses a trackbeforedetect approach and employs the matching pursuit idea to initialize
multiple targets. Computer simulations are presented to show
the performances of the algorithms.
@article{IEEE_TSP_CEVHER05, author = "V. Cevher and J. H. McClellan", title = "General directionofarrival tracking with acoustic
nodes", journal = "IEEE Transactions on Signal Processing", volume = "53", number = "1", pages = "112", year = "January 2005",}
Conference:

V.
Cevher, “Learning with compressible priors,” NIPS,
Vancouver, B.C., Canada, 712 December 2009.
Abstract
BiBTeX
Paper:
PDF
Software:
randcs.m
We
describe a set of probability distributions, dubbed
compressible priors, whose independent and
identically distributed (iid) realizations result in
pcompressible signals. A signal x in R^N is called
pcompressible with magnitude R if its sorted
coefficients exhibit a powerlaw decay as x_(i) <=
R i^d, where the decay rate d is equal to 1/p.
pcompressible signals live close to Ksparse
signals (K<<N) in the ell_rnorm (r > p) since their
best Ksparse approximation error decreases with O(R
K^{1/r1/p}). We show that the membership of
generalized Pareto, Student’s t, lognormal, Frechet,
and loglogistic distributions to the set of
compressible priors depends only on the distribution
parameters and is independent of N. In contrast, we
demonstrate that the membership of the generalized
Gaussian distribution (GGD) depends both on the
signal dimension and the GGD parameters: the
expected decay rate of Nsample iid realizations
from the GGD with the shape parameter q is given by
1/[q log (N/q)]. As stylized examples, we show via
experiments that the wavelet coefficients of natural
images are 1.67compressible whereas their pixel
gradients are 0.95 log (N/0.95)compressible, on the
average. We also leverage the connections between
compressible priors and sparse signals to develop
new iterative reweighted sparse signal recovery
algorithms that outperform the standard ell_1norm
minimization. Finally, we describe how to learn the
hyperparameters of compressible priors in
underdetermined regression problems.
@inproceedings{IEEE_NIPS09_Cevher,
author = "V. Cevher", title = "Learning
with compressible priors", booktitle = "NIPS", address= "Vancouver,
B.C., Canada", year = "712 December 2008",}

M. F.
Duarte, V. Cevher, and R. G. Baraniuk, “Modelbased
compressive sensing for signal ensembles,” Proceedings of
the 47rd Allerton Conference on Communication, Control, and
Computing, Monticello, IL, 30 September2 October 2009.
Abstract
BiBTeX
Paper:
PDF
Compressive sensing (CS) is an alternative to
Shannon/Nyquist sampling for acquiring sparse or
compressible signals. Instead of taking N periodic
samples, we measure M << N inner products with
random vectors and then recover the signal via a
sparsityseeking optimization or greedy algorithm. A
new framework for CS based on unions of subspaces
can improve signal recovery by including
dependencies between values and locations of the
signal's significant coefficients. In this paper, we
extend this framework to the acquisition of signal
ensembles under a common sparse supports model. The
new framework provides recovery algorithms with
theoretical performance guarantees. Additionally,
the framework scales naturally to large sensor
networks: the number of measurements needed for each
signal does not increase as the network becomes
larger. Furthermore, the complexity of the recovery
algorithm is only linear in the size of the network.
We provide experimental results using synthetic and
realworld signals that confirm these benefits.
@inproceedings{Allerton09_DCB,
author = "M. F. Duarte and V. Cevher and R. G.
Baraniuk", title = "Modelbased
compressive sensing for signal ensembles", booktitle = "Allerton", address= "Monticello,
IL", year = "30 September2 October 2009",}

V.
Cevher, P. Indyk, C. Hegde, R. G. Baraniuk, “Recovery of
clustered sparse signals from compressive measurements,”
SAMPTA 2009, Marseille, France, 1822 May
2009.
Abstract
BiBTeX
Paper:
PDF
We introduce a new signal model, called (K,C)sparse,
to capture Ksparse signals in N dimensions whose
nonzero coefficients are contained within at most C
clusters, with C < K
<<
N. In contrast to the existing work in the sparse
approximation and compressive sensing literature on
block sparsity, no prior knowledge of the locations
and sizes of the clusters is assumed. We prove that
O (K + C log(N/C))) random projections are
sufficient for (K,C)model sparse signal recovery
based on subspace enumeration. We also provide a
robust polynomialtime recovery algorithm for (K,C)model
sparse signals with provable estimation guarantees.
@inproceedings{IEEE_SAMPTA09_Cevher, author = "V. Cevher and P. Indyk and C. Hegde
and R. G. Baraniuk", title = "Recovery of
clustered sparse signals from compressive
measurements", booktitle = "SAMPTA", address= "Marseille,
France", year = "1822 May 2009",}

V.
Cevher, P. Boufounos, R. G. Baraniuk, A. C. Gilbert, and M.
J. Strauss, “Nearoptimal Bayesian localization via
incoherence and sparsity,” IPSN 2009, San Francisco, CA, 1316 April
2009.
Abstract
BiBTeX
Paper:
PDF
Source localization using a network of sensors is a
classical problem with applications in tracking,
habitat monitoring, etc. A solution to this
estimation problem must satisfy a number of
competing resource constraints, such as estimation
accuracy, communication and energy costs, signal
sampling requirements and computational complexity.
This paper exploits recent developments in sparse
approximation and compressive sensing to efficiently
perform localization in a sensor network. We
introduce a Bayesian framework for the localization
problem and provide sparse approximations to its
optimal solution. By exploiting the spatial sparsity
of the posterior density, we demonstrate that the
optimal solution can be computed using fast sparse
approximation algorithms. We show that exploiting
the signal sparsity can reduce the sensing and
computational cost on the sensors, as well as the
communication bandwidth. We further illustrate that
the sparsity of the source locations can be
exploited to decentralize the computation of the
source locations and reduce the sensor
communications even further. We also discuss how
recent results in 1bit compressive sensing can
impact the sensor communications by transmitting
only the timing information relevant to the problem.
Finally, we develop a computationally efficient
algorithm for bearing estimation using a network of
sensors with provable guarantees.
@inproceedings{IEEE_IPSN09_Cevher, author = "V. Cevher and P. Boufounos and R. G.
Baraniuk and A. C. Gilbert and M. J. Strauss", title =
"Nearoptimal Bayesian localization via
incoherence and sparsity", booktitle = "IPSN", address= "San
Francisco, CA", year = "1316 April 2009",}

C. Hegde,
M. F. Duarte, and V. Cevher, “Compressive sensing recovery
of spike trains using a structured sparsity model,”
SPARS'09, SaintMalo, France, 69 April 2009.
Abstract
BiBTeX
Paper:
PDF
The
theory of Compressive Sensing (CS) exploits a
wellknown concept used in signal compression –
sparsity – to design new, efficient techniques for
signal acquisition. CS theory states that for a
lengthN signal x with sparsity level K, M = O(K
log(N/K)) random linear projections of x are
sufficient to robustly recover x in polynomial time.
However, richer models are often applicable in
realworld settings that impose additional structure
on the sparse nonzero coefficients of x. Many such
models can be succinctly described as a union of
Kdimensional subspaces. In recent work, we have
developed a general approach for the design and
analysis of robust, efficient CS recovery algorithms
that exploit such signal models with structured
sparsity.
We apply our framework to a new signal model
which is motivated by neuronal spike trains. We
model the firing process of a single Poisson neuron
with absolute refractoriness using a union of
subspaces. We then derive a bound on the number of
random projections M needed for stable embedding of
this signal model, and develop a algorithm that
provably recovers any neuronal spike train from M
measurements. Numerical experimental results
demonstrate the benefits of our modelbased approach
compared to conventional CS recovery techniques.
@inproceedings{SPARS09_Hegde, author = "C. Hegde, M. F. Duarte, and V. Cevher", title = "Compressive sensing recovery of spike
trains using a structured sparsity model", booktitle = "SPARS'09", address= "SaintMalo,
France", year = "1316 April 2009",}

M. F.
Duarte, C. Hegde, V. Cevher, and Richard G. Baraniuk
“Recovery of Compressible Signals in Unions of Subspaces,”
CISS, Baltimore, MD, 1820 March 2009.
Abstract
BiBTeX
Paper:
PDF
Compressive sensing (CS) is an alternative to
Shannon/Nyquist sampling for acquisition of sparse
or compressible signals; instead of taking periodic
samples, we measure inner products with M < N random
vectors and then recover the signal via a sparsityseeking
optimization or greedy algorithm. Initial research
has shown that by leveraging stronger signal models
than standard sparsity, the number of measurements
required for recovery of an structured sparse signal
can be much lower than that of standard recovery. In
this paper, we introduce a new framework for
structured compressible signals based on the unions
of subspaces signal model, along with a new
sufficient condition for their recovery that we dub
the restricted amplification property (RAmP). The
RAmP is the natural counterpart to the restricted
isometry property (RIP) of conventional CS.
Numerical simulations demonstrate the validity and
applicability of our new framework using
wavelettree compressible signals as an example.
@inproceedings{CISS09_Duarte, author = "M. F.
Duarte, C. Hegde, V. Cevher, and Richard G. Baraniuk", title = "Recovery of Compressible Signals in Unions
of Subspaces", booktitle = "CISS", address= "Baltimore,
MD", year = "1820 March 2009",}

V.
Cevher, M. F. Duarte, C. Hegde, and R. G. Baraniuk, “Sparse
signal recovery using Markov random fields,” NIPS,
Vancouver, B.C., Canada, 811 December 2008.
Abstract
BiBTeX
Paper:
PDF
Compressive Sensing (CS) combines sampling and
compression into a single subNyquist linear
measurement process for sparse and compressible
signals. In this paper, we extend the theory of CS
to include signals that are concisely represented in
terms of a graphical model. In particular, we use
Markov Random Fields (MRFs) to represent sparse
signals whose nonzero coefficients are clustered.
Our new modelbased reconstruction algorithm, dubbed
Lattice Matching Pursuit (LaMP), stably recovers MRFmodeled
signals using many fewer measurements and
computations than the current stateoftheart
algorithms.
@inproceedings{IEEE_NIPS08_Cevher, author = "V. Cevher and and
M. F. Duarte and C. Hegde and R. G. Baraniuk", title = "Sparse
signal recovery using Markov random fields", booktitle = "NIPS", address= "Vancouver,
B.C., Canada", year = "811 December 2008",}

V.
Cevher, A. C. Sankaranarayanan, M. F. Duarte, D. Reddy, R.
G. Baraniuk, and R. Chellappa, “Compressive sensing for
background subtraction,” ECCV, Marseille, France, 1218
October 2008.
Abstract
BiBTeX
Paper:
PDF
Compressive sensing (CS) is an emerging field that
provides a framework for image recovery using subNyquist
sampling rates. The CS theory shows that a signal
can be reconstructed from a small set of random
projections, provided that the signal is sparse in
some basis, e.g., wavelets. In this paper, we
describe a method to directly recover background
subtracted images using CS and discuss its
applications in some communication constrained
multicamera computer vision problems. We show how
to apply the CS theory to recover object silhouettes
(binary background subtracted images) when the
objects of interest occupy a small portion of the
camera view, i.e., when they are sparse in the
spatial domain. We cast the background subtraction
as a sparse approximation problem and provide
different solutions based on convex optimization and
total variation. In our method, as opposed to
learning the background, we learn and adapt a low
dimensional compressed representation of it, which
is sufficient to determine spatial innovations;
object silhouettes are then estimated directly using
the compressive samples without any auxiliary image
reconstruction. We also discuss simultaneous
appearance recovery of the objects using compressive
measurements. In this case, we show that it may be
necessary to reconstruct one auxiliary image. To
demonstrate the performance of the proposed
algorithm, we provide results on data captured using
a compressive singlepixel camera. We also
illustrate that our approach is suitable for image
coding in communication constrained problems by
using data captured by multiple conventional cameras
to provide 2D tracking and 3D shape reconstruction
results with compressive measurements.
@inproceedings{IEEE_ECCV08_Cevher, author = "V. Cevher and A. C. Sankaranarayanan and
M. F. Duarte and D. Reddy and R. G. Baraniuk and R.
Chellappa", title = "Compressive Sensing for
Background Subtraction", booktitle = "ECCV", address= "Marseille,
France", year = "1218 October 2008",}

V.
Cevher, M. F. Duarte, and R. G. Baraniuk, “Distributed
target localization via spatial sparsity,” EUSIPCO 2008,
Lausanne, Switzerland, 2529 August 2008.
Abstract
BiBTeX
Paper:
PDF
We
propose an approximation framework for distributed
target localization in sensor networks. We represent
the unknown target positions on a location grid as a
sparse vector, whose support encodes the multiple
target locations. The location vector is linearly
related to multiple sensor measurements through a
sensing matrix, which can be locally estimated at
each sensor. We show that we can successfully
determine multiple target locations by using linear
dimensionalityreducing projections of sensor
measurements. The overall communication bandwidth
requirement per sensor is logarithmic in the number
of grid points and linear in the number of targets,
ameliorating the communication requirements.
Simulations results demonstrate the performance of
the proposed framework.
@inproceedings{IEEE_EUSIPCO08_Cevher, author = "V. Cevher and M. F. Duarte and R. G.
Baraniuk", title = "Distributed target
localization via spatial sparsity", booktitle = "EUSIPCO", address= "Lausanne,
Switzerland", year = "2529 August
2008",}

V.
Cevher, and L. Kaplan, “Pareto frontiers of sensor networks
for localization,” IPSN 2008, St. Louis, MO, 2224 April
2008.
Abstract
BiBTeX
Paper:
PDF
We
develop a theory to predict the localization
performance of randomly distributed sensor networks
consisting of various sensor modalities when only a
constant active subset of sensors that minimize
localization error is used for estimation. The
characteristics of the modalities include
measurement type (bearing or range) and error,
sensor reliability, FOV, sensing range, and
mobility. We show that the localization performance
of a sensor network is a function of a weighted sum
of the total number of each sensor modality. We also
show that optimization of this weighted sum is
independent of how the sensor management strategy
chooses the active sensors. We combine the utility
objective with other objectives, such as lifetime,
coverage and reliability to determine the best mix
of sensors for an optimal sensor network design. The
Pareto efficient frontier of the multi objectives
are obtained with a dynamic program, which also
accommodates additional convex constraints.
@inproceedings{IEEE_IPSN08_Cevher, author = "V. Cevher and L. Kaplan", title =
"Pareto frontiers of sensor networks for
localization", booktitle = "IPSN", address= "St.
Louis, MO", year = "2224 April 2008",}

V.
Cevher, and R. G. Baraniuk, “Compressive sensing for sensor
calibration,” SAM 2008 Workshop, Darmstadt, Germany, 2123
July 2008.
Abstract
BiBTeX
Paper:
PDF
We
consider a calibration problem, where we determine
an unknown sensor location using the known track of
a calibration target and a known reference sensor
location. We cast the calibration problem as a
sparse approximation problem where the unknown
sensor location is determined over a discrete
spatial grid with respect to the reference sensor.
To achieve the calibration objective, low
dimensional random projections of the sensor data
are passed to the reference sensor, which
significantly reduces the intersensor communication
bandwidth. The unknown sensor location is then
determined by solving an ℓ1norm minimization
problem (linear program). Field data results are
provided to demonstrate the effectiveness of the
approach.
@inproceedings{IEEE_SAM08_Cevher, author = "V. Cevher and R. G. Baraniuk", title = "Compressive
sensing for sensor calibration", booktitle = "SAM
2008 Workshop", address= "Darmstadt,
Germany", year = "2123 July 2008",}

D.
Reddy, A. C. Sankaranarayanan, V. Cevher, and R. Chellappa, “Compressed
sensing for multiview tracking and 3D voxel reconstruction,”
ICIP 2008, San Diego, CA, 1215 October 2008.
Abstract
BiBTeX
Paper:
PDF
Compressed sensing(CS) suggests that a signal,
sparse in some basis, can be recovered from a small
number of random projections. In this paper, we
apply the CS theory on sparse backgroundsubtracted
silhouettes and show the usefulness of such an
approach in various multiview estimation problems.
The sparsity of the silhouette images corresponds to
sparsity of object parameters (location, volume
etc.) in the scene. We use random projections
(compressed measurements) of the silhouette images
for directly recovering object parameters in the
scene coordinates. To keep the computational
requirements of this recovery procedure reasonable,
we tessellate the scene into a bunch of
nonoverlapping lines and perform estimation on each
of these lines. Our method is scalable in the number
of cameras and utilizes very few measurements for
transmission among cameras. We illustrate the
usefulness of our approach for multiview tracking
and 3D voxel reconstruction problems.
@inproceedings{IEEE_ICIP08Reddy, author = "D. Reddy and A. C. Sankaranarayanan and V.
Cevher and R. Chellappa", title = "Compressed
sensing for multiview tracking and 3D voxel
reconstruction", booktitle = "ICIP 2008", address= "San
Diego, CA", year = "1215 October
2008",}

V.
Cevher, A. C. Sankaranarayanan, and R. Chellappa, “Factorized
variational approximations for acoustic multi source
localization,” ICASSP 2008, Las Vegas, NV,
30 March4 April 2008.
Abstract
BiBTeX
Paper:
PDF
Estimation based on received signal strength (RSS)
is crucial in sensor networks for sensor
localization, target tracking, etc. In this paper,
we present a Gaussian approximation of the Chi
distribution that is applicable to general RSS
source localization problems in sensor networks.
Using our Gaussian approximation, we provide a
factorized variational Bayes (VB) approximation to
the location and power posterior of multiple sources
using a sensor network. When the source signal and
the sensor noise have uncorrelated Gaussian
distributions, we demonstrate that the envelope of
the sensor output can be accurately modeled with a
multiplicative Gaussian noise model. In turn, our
factorized VB approximations decrease the
computational complexity and provide computational
robustness as the number of targets increases.
Simulations are provided to demonstrate the
effectiveness of the proposed approximations.
@inproceedings{IEEE_ICASSP08Cevher2, author = "V. Cevher and A. C. Sankaranarayanan and
R. Chellappa", title = "Factorized
variational approximations for acoustic multi source
localization", booktitle = "ICASSP 2008", address= "Las
Vegas, NV", year = "30 March4 April
2008",}

V.
Cevher, A. C. Gurbuz, J. H. McClellan, and R. Chellappa, “Compressive
wireless arrays for bearing estimation of sparse sources in
angle domain,” ICASSP 2008, Las Vegas, NV,
30 March4 April 2008.
Abstract
BiBTeX
Paper:
PDF
Joint processing of sensor array outputs improves
the performance of parameter estimation and
hypothesis testing problems beyond the sum of the
individual sensor processing results. When the
sensors have high data sampling rates, arrays are
tethered, creating a disadvantage for their
deployment and also limiting their aperture size. In
this paper, we develop the signal processing
algorithms for randomly deployable wireless sensor
arrays that are severely constrained in
communication bandwidth. We focus on the acoustic
bearing estimation problem and show that when the
target bearings are modeled as a sparse vector in
the angle space, functions of the low dimensional
random projections of the microphone signals can be
used to determine multiple source bearings as a
solution of an ℓ1norm minimization problem. Field
data results are shown where only 10bits of
information is passed from each microphone to
estimate multiple target bearings.
@inproceedings{IEEE_ICASSP08Cevher1, author = "V. Cevher and
A.
C. Gurbuz and J. H. McClellan and R.
Chellappa", title = "Compressive
wireless arrays for bearing estimation of sparse
sources in angle domain", booktitle = "ICASSP
2008", address= "Las Vegas, NV", year = "30
March4 April 2008",}

A. C.
Gurbuz, V. Cevher, and J. H. McClellan, “A compressive
beamforming method,” ICASSP 2008, Las Vegas, NV,
30 March4 April 2008.
Abstract
BiBTeX
Paper:
PDF
Compressive Sensing (CS) is an emerging area which
uses a relatively small number of nontraditional
samples in the form of randomized projections to
reconstruct sparse or compressible signals. This
paper considers the directionofarrival (DOA)
estimation problem with an array of sensors using
CS. We show that by using random projections of the
sensor data, along with a full waveform recording on
one reference sensor, a sparse angle space scenario
can be reconstructed, giving the number of sources
and their DOA’s. The number of projections can be
very small, proportional to the number sources. We
provide simulations to demonstrate the performance
and the advantages of our compressive beamformer
algorithm.
@inproceedings{IEEE_ICASSP08Gurbuz, author = "A. C. Gurbuz and V. Cevher and J. H.
McClellan", title = "A compressive
beamforming method", booktitle = "ICASSP
2008", address= "Las Vegas, NV", year = "30
March4 April 2008",}

V. Cevher,
R. Chellappa, and J. H. McClellan, “Gaussian approximations for
energybased detection and localization in sensor networks,” IEEE
Statistical Signal Processing Workshop, Madison, WI, 2629 August
2007.
Abstract
BiBTeX
Paper:
PDF
Energybased detection and estimation are crucial in sensor
networks for sensor localization, target tracking, etc. In
this paper, we present novel Gaussian approximations that
are applicable to general energybased source detection and
localization problems in sensor networks. Using our
approximations, (i) we derive receiver operating
characteristics curves and CramerRao bounds, and (ii) we
provide a factorized variational Bayes approximation to the
location and source energy posterior for centralized or
decentralized estimation. When the source signal and the
sensor noise have uncorrelated Gaussian distributions, we
demonstrate that the envelope of the sensor output can be
accurately modeled with a multiplicative Gaussian noise
model, which results in smaller estimation biases than the
other Gaussian models typically used in the literature. We
also prove that additive Gaussian noise models result in
negatively biased speed estimates under the same signal
assumptions, which can be circumvented by the proposed
approximations.
@inproceedings{IEEE_SSP07Cevher, author = "V. Cevher
and R. Chellappa and J. H.
McClellan", title = "Gaussian
approximations for energybased detection and localization
in sensor networks", booktitle = "IEEE
Statistical Signal Processing Workshop", address= "Madison,
WI", year = "2629 August 2007",}

R.
Velmurugan, V. Cevher, and J. H. McClellan, “Implementation of
batchbased particle filters for multisensor tracking,” IEEE CAMSAP
2007, U.S. Virgin Islands, 1214 December 2007.
Abstract
BiBTeX
Paper:
PDF
Energybased detection and estimation are crucial in sensor
networks for sensor localization, target tracking, etc. In
this paper, we present novel Gaussian approximations that
are applicable to general energybased source detection and
localization problems in sensor networks. Using our
approximations, (i) we derive receiver operating
characteristics curves and CramerRao bounds, and (ii) we
provide a factorized variational Bayes approximation to the
location and source energy posterior for centralized or
decentralized estimation. When the source signal and the
sensor noise have uncorrelated Gaussian distributions, we
demonstrate that the envelope of the sensor output can be
accurately modeled with a multiplicative Gaussian noise
model, which results in smaller estimation biases than the
other Gaussian models typically used in the literature. We
also prove that additive Gaussian noise models result in
negatively biased speed estimates under the same signal
assumptions, which can be circumvented by the proposed
approximations.
@inproceedings{IEEE_CAMSAP07Velmurugan, author = "R. Velmurugan and V. Cevher and J. H. McClellan", title = "Implementation
of batchbased particle filters for multisensor tracking", booktitle = "IEEE
CAMSAP", address= "U.S. Virgin Islands", year = "1214
December 2007",}

R.
Velmurugan, S. Subramanian, V. Cevher, J. H. McClellan, and
D. V. Anderson “Mixedmode implementation of particle
filters,” IEEE PACRIM 2007, Victoria, B.C., CA, 2224 August
2007.
Abstract
BiBTeX Paper:
PDF
In this paper, we develop new mixedmode
implementations for particle filters and compare
them to a digital implementation. The motivation for
the mixedmode implementation is to achieve
lowpower implementation of particle filters. The
specific application considered is a bearingsonly,
singletarget tracking algorithm. Specifically, we
develop mixedmode implementations that use analog
components to realize nonlinear functions in the
particle filter algorithm. The analog implementation
of nonlinear functions use multipleinput
translinear element (MITE) networks. MITEs operate
in the subthreshold region and hence dissipate
lowpower. Simulation results for onmixedmode
implementation of the bearingsonly tracker are
presented. We show that of the two mixedmode
implementations, one approach dissipates almost same
power as that of a digital implementation. The
second approach has nearly twenty times less power
dissipation, but requires careful analog design.
@inproceedings{IEEE_PACRIM07Velmurugan, author = "R. Velmurugan and S.
Subramanian and V. Cevher and J. H. McClellan and D.
V. Anderson", title = "Mixedmode
implementation of particle filters", booktitle = "IEEE PACRIM", address= "Victoria,
B.C., Canada", year = "2224 August
2007",}

V. Cevher,
R. Chellappa, and J. H. McClellan, “Joint acousticvideo
fingerprinting of vehicles, part I,” ICASSP 2007, Honolulu, Hawaii,
1620 April 2007.
Abstract
BiBTeX
Paper:
PDF
We address
vehicle classification and mensuration problems using
acoustic and video sensors. In this paper, we show how to
estimate a vehicle’s speed, width, and length by jointly
estimating its acoustic wavepattern using a single passive
acoustic sensor that records the vehicle’s driveby noise.
The acoustic wavepattern is approximated using three
envelope shape (ES) components, which approximate the shape
of the received signal’s power envelope. We incorporate the
parameters of the ES components along with estimates of the
vehicle engine RPM and number of cylinders to create a
vehicle profile vector that forms an intuitive
discriminatory feature space. In the companion paper, we
discuss vehicle classification and mensuration based on
silhouette extraction and wheel detection, using a video
sensor. Vehicle speed estimation and classification results
are provided using field data.
@inproceedings{IEEE_ICASSP07Cevher1, author = "V. Cevher
and R. Chellappa and J. H.
McClellan", title = "Joint
acousticvideo fingerprinting of vehicles, Part I", booktitle = "ICASSP 2007", address= "Honolulu,
Hawaii", year = "1520 April 2007",}

V. Cevher,
F. Guo, A. C. Sankaranarayanan, and R. Chellappa, “Joint
acousticvideo fingerprinting of vehicles, part II,” ICASSP 2007,
Honolulu, Hawaii, 1620 April 2007.
Abstract
BiBTeX Paper:
PDF
In this second paper, we first
show how to estimate the wheelbase length of a vehicle using
line metrology in video. We then address the vehicle
fingerprinting problem using vehicle silhouettes and color
invariants. We combine the acoustic metrology and
classification results discussed in Part I with the video
results to improve estimation performance and robustness.
The acoustic video fusion is achieved in a Bayesian
framework by assuming conditional independence of the
observations of each modality. For the metrology density
functions, Laplacian approximations are used for
computational efficiency. Experimental results are given
using field data.
@inproceedings{IEEE_ICASSP07Cevher2, author = "V. Cevher and F. Guo and A. C.
Sankaranarayanan and R. Chellappa and J. H.
McClellan", title = "Joint acousticvideo fingerprinting of
vehicles, Part II", booktitle = "ICASSP 2007", address= "Honolulu, Hawaii", year = "1520 April 2007",}

V. Cevher, F. Shah, R. Velmurugan, and J. H. McClellan, “An
acoustic multitarget tracking system using random sampling
consensus,” 2007 IEEE Aerospace Conference, Big Sky, Montana, 310
March 2007.
Abstract
BiBTeX
Paper:
PDF
In this paper, we present an
acoustic directionofarrival (DOA) tracking system to track
multiple maneuvering targets using a state space approach.
The system consists of three blocks: beamformer, random
sampling, and particle filter. The beamformer block
processes the received acoustic data to output bearing
batches as point statistics. The random sampling block
determines temporal clustering of the bearings in a batch to
determine regionofinterests (ROIs). Based on the trackbeforedetect approach, each ROI indicates the
presence of a possible target. We describe three random
sampling algorithms called RANSAC, MSAC, and NAPSAC to use
in the random sampling block. The particle filter then
tracks the targets via its interactions with the beamformer
and the random sampling blocks. We present a computational
analysis of the random sampling blocks and show tracking
results with field data.
@Inproceedings{IEEE_AESCONF07_Cevher1, author = "V. Cevher and R. Chellappa and
F. Shah and R. Velmurugan and J. H. McClellan", title = "An
acoustic multitarget tracking system using random sampling
consensus", booktitle = "2007 IEEE Aerospace Conference", address= "Big Sky, Montana", year = "310 March 2007",}

M. Borkar,
V. Cevher, and J. H. McClellan, “A MonteCarlo approach for tracking
mobile personnel,” 2007 IEEE Aerospace Conference, Big Sky, Montana,
310 March 2007.
Abstract
BiBTeX
Paper:
PDF
In this paper, we propose a MonteCarlo method based on the
particle filter framework to track footfall locations
generated by mobile personnel using seismic arrays. While
the particle proposal function follows a simple bootstrap
approach, the novelty in our algorithm comes from a unique
weighting strategy that takes into account the sparse nature
of the seismic footfall signal and is robust against missed
detections and clutter which could appear in the form of
other impulsive sources or other walkers. Our weighting
strategy automatically makes use of the wavefront shape,
either planar or circular, and assigns weights in xy space.
Data association is built into the system, eliminating the
need to explicitly associate the received footfall impulses
with different walkers. Hence our algorithm is ideal for
tracking multiple mobile personnel. We also demonstrate the
fusion of our system with range information available by
means of radar. Fusion with radar improves xy tracking when
range resolution is lost due to a large distance between the
target and the seismic array leading to planar wavefronts.
@Inproceedings{IEEE_AESCONF07_Borkar1, author = "M.
Borkar and V. Cevher and J. H. McClellan", title = "A
Monte Carlo approach for tracking mobile personnel", booktitle = "2007 IEEE Aerospace Conference", address= "Big Sky, Montana", year = "310 March 2007",}

L.
Kaplan and V. Cevher, “Design considerations for
heterogeneous network of bearingsonly sensors using sensor
management,” 2007 IEEE Aerospace Conference, Big Sky,
Montana, 310 March 2007.
Abstract
BiBTeX
Paper:
PDFThis paper presents the design characterization of a
heterogeneous sensor network with the goal of geolocation
accuracy. It is assumed that the network exploits sensor
management to conserve node usage. Each available node
modality is a bearingsonly sensor of varying capability.
The optimal mixture of modalities is discussed under the
constraint of the overall network cost. Finally, simulations
verify the theory and demonstrate design choices for a
network consisting of two modes analogous to acoustic arrays
and cameras.
@Inproceedings{IEEE_AESCONF07_Kaplan1, author = "L. Kaplan and V. Cevher", title = "Design
considerations for heterogeneous network of bearingsonly
sensors using sensor management", booktitle = "2007 IEEE Aerospace Conference", address= "Big Sky, Montana", year = "310 March 2007",}

S. Ozgur, V.
Cevher, D. B. Williams, and J. H. McClellan, “Convergence Analysis
for Sequential Monte Carlo Receivers in Communications
Applications,” IEEE DSP Workshop 2006, Grand Teton National Park,
Wyoming, 2427 September 2006.
Abstract
BiBTeX
Paper:
PDFRecently, sequential Monte Carlo methods have been applied
in the telecommunications field finding application in
receiver design. These receivers do not require channel
state information or training, making them bandwidth
efficient as no communication bandwidth needs to be spent on
training. The receivers are optimal in the sense that they
achieve minimum symbol error rate regardless of the noise
distribution, nonlinearities in the system, and distribution
of the transmitted symbols. Moreover, these receivers are
capable of producing softinformation outputs, which enables
the designer to utilize iterative receiver architectures for
nearoptimal performance. In this work we investigate the
convergence properties of these algorithms when utilized in
various types of receivers and we quantify the convergence
rate. We describe how various parameters (e.g., noise power,
number of particles) and factors (e.g., dimensionality of
the problem) affect the convergence rate and point out
factors that should be improved first to gain speed and
accuracy in the convergence.
@inproceedings{IEEE_DSP06_Ozgur, author = "S. Ozgur and V. Cevher and D. B. Williams and J.
H. McClellan", title = "Convergence Analysis
for Sequential Monte Carlo Receivers in Communications
Applications", booktitle = "IEEE DSP Workshop
2006", address= "Grand Teton National Park,
Wyoming", year = "2427 September 2006",}

V.
Cevher, M. Borkar, and J. H. McClellan, “A joint
radaracoustic particle filter tracker with acoustic
propagation delay compensation,” EUSIPCO 2006, Florence,
Italy, 48 September 2006.
Abstract
BiBTeX Paper:
PDF
In this paper, a novel particle filter tracker is
presented for target tracking using collocated radar
and acoustic sensors. Realtime tracking of the
target's position and velocity in Cartesian
coordinates is performed using batches of range and
directionofarrival estimates. For robustness, the
filter aligns the radar and acoustic data streams to
account for acoustic propagation delays. The filter
proposal function uses a Gaussian approximation to
the full tracking posterior for improved efficiency.
To incorporate the aligned acoustic data into the
tracker, a twostage weighting strategy is proposed.
Computer simulations are provided to demonstrate the
effectiveness of the algorithm.
@inproceedings{EUSIPCO06Cevher1, author = "V. Cevher and M. Borkar and J. H.
McClellan", title = "A joint radaracoustic particle filter
tracker with acoustic propagation delay
compensation", booktitle = "EUSIPCO 2006", address= "Florence, Italy", year = "48 September 2006",}

R.
Velmurugan, S. Subramanian, V. Cevher, D. Abramson, K. M.
Odame, J. D. Gray, H.J. Lo, J. H. McClellan, and D. V.
Anderson “On lowpower analog implementations of particle
filters for target tracking,” EUSIPCO 2006, Florence, Italy,
48 September 2006.
Abstract
BiBTeX Paper:
PDF
We propose a lowpower, analog and mixedmode,
implementation of particle filters. Lowpower analog
implementation of nonlinear functions such as
exponential and arctangent functions is done using
multipleinput translinear element (MITE) networks.
These nonlinear functions are used to calculate the
probability densities in the particle filter. A
bearingsonly tracking problem is simulated to
present the proposed lowpower implementation of the
particle filter algorithm.
@inproceedings{EUSIPCO06Velmurugan1, author = "R. Velmurugan and S. Subramanian and V.
Cevher and D. Abramson and K. M. Odame and J. D.
Gray and H.J. Lo and J. H. McClellan and D. V.
Anderson", title = "On lowpower analog implementations of
particle filters for target tracking", booktitle = "EUSIPCO 2006", address= "Florence, Italy", year = "48 September 2006",}

V.Cevher, R. Velmurugan, and J. H. McClellan, “A particle
filter range tracker,” ICASSP 2006, Toulouse, France, 1519 May 2006.
Abstract
BiBTeX Paper:
PDF
We propose a particle filter tracker to track
multiple maneuvering targets using a batch of range
measurements. The state update is formulated through
a locally linear motion model and the observability
of the state vector is proved using geometrical
arguments. The data likelihood treats the range
observations as an image using template models
derived from the state update equation, and
incorporates the possibility of missing data as well
as spurious range observations. The particle filter
handles multiple targets, using a partitioned
statevector approach. The filter proposal function
uses a Gaussian approximation of the fullposterior
to cope with target maneuvers for improved
efficiency. By treating the range measurements as
images and using smoothness constraints, the
particle filter is able to avoid the data
association problems. Computer simulations
demonstrate the performance of the tracking
algorithm.
@inproceedings{IEEE_ICASSP06Cevher1, author = "V. Cevher and R. Velmurugan and J. H.
McClellan", title = "A particle filter range tracker", booktitle = "ICASSP 2006", address= "Toulouse, France", year = "1519 May 2006",}

M.
Alam, V. Cevher, and J. H. McClellan, “Optimal
experiments with seismic sensors,” ICASSP 2006, Toulouse,
France, 1519 May 2006.
Abstract
BiBTeX Paper:
PDF
In this paper, we consider the problem of detecting
and locating buried land mines and subsurface
objects by using seismic waves. We demonstrate an
adaptive seismic system that maneuvers an array of
receivers, according to an optimal positioning
algorithm based on the theory of optimal
experiments, to minimize the number of distinct
measurements to localize the mine. The adaptive
localization algorithm is tested using numerical
model data as well as laboratory measurements
performed in a facility at Georgia Tech. It is
envisioned that the future systems should be able to
incorporate this new method into portable mobile
mine location systems.
@inproceedings{IEEE_ICASSP06Alam1, author = "M. Alam and V. Cevher and J. H.
McClellan", title = "Optimal experiments with seismic sensors", booktitle = "ICASSP 2006", address= "Toulouse, France", year = "1519 May 2006",}

M.
Borkar, V. Cevher, and J. H. McClellan, “A MonteCarlo method
for initializing distributed tracking algorithms,” ICASSP
2006, Toulouse, France, 1519 May 2006.
Abstract
BiBTeX Paper:
PDF
Distributed processing algorithms are attractive
alternatives to centralized algorithms for target
tracking applications in sensor networks. In this
paper, we determine an initial probability
distribution of multiple target states in a
distributed manner to initialize distributed
trackers. Our approach is based on MonteCarlo
methods, where the state distributions are
represented as a weighted set of discrete state
realizations. The filter state vector is the target
positions and velocities on the 2D plane. Our
approach can determine the state vector distribution
even if the individual sensors are not capable of
observing it. The only condition is that the network
as a whole can observe the state vector. A robust
weighting strategy is formulated to account for
missed detections and clutter. To demonstrate the
effectiveness of the algorithm, we simulate a
network with directionofarrival nodes and
rangeDoppler nodes.
@inproceedings{IEEE_ICASSP06Borkar1, author = "M. Borkar and V. Cevher and J. H.
McClellan", title = "A MonteCarlo method for initializing
distributed tracking algorithms", booktitle = "ICASSP 2006", address= "Toulouse, France", year = "1519 May 2006",}

V.
Cevher, R. Velmurugan, and J. H. McClellan, “Multi target
directionofarrival tracking using road priors,” 2006 IEEE
Aerospace Conference, Big Sky, Montana, 411 March 2006.
Abstract
BiBTeX Paper:
PDF
In this paper, we present a multi target particle
filter DOA tracker that can incorporate road prior
information at a single array node. The filter uses
a batch of DOA’s to determine the state vector,
based on an image template matching idea. The filter
likelihood is derived with the joint probability
density association principles so that no DOA
measurement is associated to more than one target.
The filter state update has the target DOA, the
target velocity over range ratio, and the target
heading parameters. We present two approaches for
incorporating the road information. In the first
approach, the road prior is injected at the
weighting stage of the tracker, where a raised
mixture Gaussian distribution, derived from the road
headings at the target DOA, constraints the
particles. The second approach is based on
modifying the state update function with a compound
model, where a mixture of the constant velocity
model and the road information is used. In this
case, the filter uses an online EM algorithm to
update the state vector along with the mixture
components. Computer simulations demonstrate the
performance of the approaches.
@inproceedings{IEEE_AESCONF06_Cevher, author = "V. Cevher and R. Velmurugan and J. H.
McClellan", title = "Multi target directionofarrival tracking
using road priors", booktitle = "2006 IEEE Aerospace Conference", address= "Big Sky, Montana", year = "411 March 2006",}

V.
Cevher and J. H. McClellan, “An acoustic multiple target
tracker,” 2005 IEEE Statistical Signal Processing
Conference, Bordeaux, France, 1720 July 2005.
Abstract
BiBTeX Paper:
PDF
We propose a particle filter acoustic tracker to
track multiple maneuvering targets using a state
space formulation. The state update is formulated
through a locally linear motion model. The
observations are a batch of directionofarrival
(DOA) estimates at various frequencies. The data
likelihood incorporates the possibility of missing
data as well as spurious DOA observations. By
imposing smoothness constraints on the target
motion, the particle filter is able to avoid the
data association problems. To make the filter
computationally efficient, a proposal strategy based
on approximating the full posterior is employed.
Computer simulations are presented to show the
performance of the algorithm.
@inproceedings{IEEE_SSP05Cevher, author = "V. Cevher and J. H. McClellan", title = "An acoustic multiple target tracker", booktitle = "IEEE Statistical Signal Processing
Conference", address= "Bordeaux, France", year = "1720 July 2005",}

M.
Borkar, V. Cevher, and J. H. McClellan, “Estimating target state
distributions in a distributed sensor network using a
MonteCarlo approach,” 2005 IEEE Workshop on Machine
Learning for Signal Processing, Mystic, Connecticut, 2830
September 2005.
Abstract
BiBTeX Paper:
PDF
Distributed processing algorithms are attractive
alternatives to centralized algorithms for target
tracking applications in sensor net works. In this
paper, we address the issue of determining an
initial probability distribution of multiple target
states in a distributed manner to initialize
distributed trackers. Our approach is based on
MonteCarlo methods, where the state distributions
are represented as a discrete set of weighted
particles. The target state vector is the target
positions and velocities in the 2D plane. Our
approach can determine the state vector distribution
even if the individual sensors are not capable of
observing it. The only condition is that the network
as a whole can observe the state vector. A robust
weighting strategy is formulated to account for misdetections
and clutter. To demonstrate the effectiveness of the
algorithm, we use directionofarrival nodes and
rangeDoppler nodes.
@inproceedings{IEEE_MLSP05Borkar, author = "M. Borkar and V. Cevher and J. H.
McClellan", title = "Estimating target state distributions in a
distributed sensor network using a MonteCarlo
approach", booktitle = "IEEE MLSP 2005", address= "Mystic, Connecticut", year = "2830 September 2005",}

G.
Qian, V. Cevher, A. Sankaranarayanan, J. H. McClellan, and
R. Chellappa, “Vehicle tracking using acoustic and video
sensors,” in Army Science Conference 2004, Orlando, 29
November2 December 2004.
Abstract
BiBTeX Paper:
PDF
In target tracking, fusing multimodal sensor data
under a powerperformance tradeoff is becoming
increasingly important. Proper fusion of multiple
modalities can help in achieving better tracking
performance while decreasing the total power
consumption. In this paper, we present a framework
for tracking a target given joint acoustic and video
observations from a colocated acoustic array and a
video camera. We demonstrate on field data that
tracking of the directionofarrival of a target
improves significantly when the video information is
incorporated at time instants when the acoustic
signaltonoise ratio is low.
@inproceedings{ASC2004Qian, author = "G. Qian and V. Cevher
and A. Sankaranarayanan and J. H. McClellan and R.
Chellappa", title = "On lowpower analog implementations of
particle filters for target tracking", booktitle = "EUSIPCO 2006", address= "Florence, Italy", year = "48 September 2006",}

V.
Cevher and J. H. McClellan, “Proposal strategies for joint
statespace tracking with particle filters,” ICASSP 2005,
Philadelphia, PA, 1823 March 2005.
Abstract
BiBTeX Paper:
PDF
A proposal function determines the random particle
support of a particle filter. When this support is
distributed close to the true target density,
filter’s estimation performance increases for a
given number of particles. In this paper, a proposal
strategy for joint statespace tracking using
particle filters is given. The statespaces are
assumed Markovian and notexact; however, each
statespace is assumed to sufficiently describe the
underlying phenomenon. The joint tracking is
achieved by carefully placing the random support of
the joint filter to where the final posterior is
likely to lie. Computer simulations demonstrate
improved performance and robustness of the joint
statespace through the proposed strategy.
@inproceedings{IEEE_ICASSP05Cevher, author = "V. Cevher and J. H. McClellan", title = "Proposal strategies for joint statespace
tracking with particle filters", booktitle = "ICASSP 2005", address= "Philadelphia, PA", year = "1823 March 2005",}

V.
Cevher and J. H. McClellan, “Acoustic node calibration using
helicopter sounds and Monte Carlo Markov chain methods,”
IEEE DSP Workshop, Taos Ski Valley, NM, 14 August 2004.
Abstract
BiBTeX Paper:
PDF
A MonteCarlo method is used to calibrate a randomly
placed sensor node using helicopter sounds. The
calibration is based on using the GPS information
from the helicopter and the estimated DOA’s at the
node. The related CramerRao lower bound is derived
and the effects of the GPS errors on the position
estimates are derived. Issues related to the
processing of the field data, e.g., time
synchronization and data nonstationarity are
discussed. The effects of the GPS errors are shown
to be negligible under certain conditions. Finally,
the results of the calibration on field data are
given.
@inproceedings{IEEE_DSP04_Cevher, author = "V. Cevher and J. H. McClellan", title = "Acoustic node calibration using
helicopter sounds and Monte Carlo Markov chain
methods", booktitle = "IEEE DSP Workshop", address= "Taos Ski Valley, NM", year = "14 August 2004",}

V.
Cevher and J. H. McClellan, “Fast initialization of particle
filters using a modified MetropolisHastings algorithm:
ModeHungry approach,” ICASSP 2004, Montreal, CA, 1721 May
2004.
Abstract
BiBTeX Paper:
PDF
As a recursive algorithm, the particle filter
requires initial samples to track a state vector.
These initial samples must be generated from the
received data and usually obey a complicated
distribution. The MetropolisHastings (MH)
algorithm is used for sampling from intractable
multivariate target distributions and is well suited
for the initialization problem. Asymptotically, the
MH scheme creates samples drawn from the exact
distribution. For the particle filter to track the
state, the initial samples need to cover only the
region around its current state. This region is
marked by the presence of modes. Since the particle
filter only needs samples around the mode, we modify
the MH algorithm to generate samples distributed
around the modes of the target posterior. By
simulations, we show that this ”mode hungry”
algorithm converges an order of magnitude faster
than the original MH scheme for both unimodal and
multimodal distributions.
@inproceedings{IEEE_ICASSP04Cevher, author = "V. Cevher and J. H. McClellan", title = "Fast initialization of particle
filters using a modified MetropolisHastings
algorithm: ModeHungry approach", booktitle = "ICASSP 2004", address= "Montreal, CA", year = "1721 May 2004",}

V.
Cevher and J. H. McClellan, “Tracking of multiple wideband
targets using passive sensor arrays and particle filters,”
IEEE DSP Workshop, Callaway Gardens, GA, 1316 October 2002.
Abstract
BiBTeX Paper:
PDF
In this paper, we present a way to track multiple
maneuvering targets with varying timefrequency
signatures. A particle filter is used to track
targets that have constant speeds with changing
heading directions. The target motion dynamics help
the particle filter achieve an angular resolution
otherwise not possible by the conventional
beamforming techniques. Moreover, the particle
filter has a builtin target association that
eliminates the need for heuristic techniques
commonly used in the multiple target tracking
problems. Reference priors are used to derive the
probability distribution function of the acoustic
array outputs given the state of the multiple target
states (MTS’s). Local linearization is used to
approximate the importance function used in the
particle filter by a Gaussian pdf. Finally, computer
simulations are used to demonstrate the performance
of the algorithm with synthetic data.
@inproceedings{IEEE_DSP02_Cevher, author = "V. Cevher and J. H. McClellan", title = "Tracking of multiple wideband targets
using passive sensor arrays and particle filters", booktitle = "IEEE DSP Workshop", address= "Callaway Gardens, GA", year = "1316 October 2002",}

V.
Cevher and J. H. McClellan, “2D sensor perturbation analysis:
equivalence to AWGN on array outputs,” SAM 2002 Workshop,
WDC, 46 August 2002.
Abstract
BiBTeX Paper:
PDF
In this paper, the performance of a subspace
beamformer, namely the multiple signal
classification algorithm (MUSIC), is scrutinized in
the presence of sensor position errors. Based on a
perturbation model, a relationship between the array
autocorrelation matrix and the source
autocorrelation matrix is established. It is shown
that under certain assumptions on the source
signals, the Gaussian sensor perturbation errors can
be modeled as additive white Gaussian noise (AWGN)
for an array where sensor positions are known
perfectly. This correspondence can be used to equate
position errors to an equivalent signaltonoise
ratio (SNR) for AWGN in performance evaluation.
Finally, CramerRao bound for the position
perturbations that can be computed using the CramerRao
bound relations for the additive Gaussian noise case
at high SNR’s.
@inproceedings{SAM02Cevher, author = "V. Cevher and J. H.
McClellan", title = "2D sensor perturbation analysis:
equivalence to AWGN on array outputs", booktitle = "SAM 2002", address= "Washington, DC", year = "46 August 2002",}

V.
Cevher and J. H. McClellan, “Sensor array calibration via
tracking with the extended Kalman filter,” ICASSP 2001, vol.
5, pp. 28172820, Salt Lake City, UT, May 2001.
Abstract
BiBTeX Paper:
PDF
Starting with a randomly distributed sensor array
with unknown sensor orientations, array calibration
is needed before target localization and tracking
can be performed using classical triangulation
methods. In this paper, we assume that the sensors
are only capable of accurate direction of arrival
(DOA) estimation. The calibration problem cannot be
completely solved given the DOA estimates alone,
since the problem is not only rotationally symmetric
but also includes a range ambiguity. Our approach to
calibration is based on tracking a single target
moving at a constant velocity. In this case, the
sensor array can be calibrated from target tracks
generated by an extended Kalman filter (EKF) at each
sensor. A simple algorithm based on geometrical
matching of similar triangles will align the
separate tracks and determine the sensor positions
and orientations relative to a reference sensor.
Computer simulations show that the algorithm
performs well even with noisy DOA estimates at the
sensors.
@inproceedings{IEEE_ICASSP01Cevher, author = "V. Cevher and J. H. McClellan", title = "Sensor array calibration via tracking
with the extended Kalman filter", booktitle = "ICASSP 2001, address= "Salt Lake City, UT", year = "May 2001",}

R.
M. Dansereau, W. Kinsner, and V. Cevher, “Wavelet packet
best basis search using generalized Renyi entropy,” IEEE
CCECE 2002, vol. 2, pp. 10051008.
Abstract
BiBTeX
This paper introduces an approach to wavelet packet
best basis searches using the generalized Renyi
entropy. The approach extends work by R.R. Coifman
and M.V. Wickerhauser who showed how Shannon entropy
can be used as an additive cost function in the
wavelet packet best basis selection (see IEEE Trans.
on Inform. Theory, vol.38, no.2, p.71318, 1992).
This paper also extends the idea of an additive cost
function to an arithmetic mean. These extensions
allow for a redefinition of additive cost functions
as arithmetic means in a way consistent with the
approach of Coifman and Wickerhauser. The approach
using an arithmetic mean is then extended to include
the geometric mean. This extension to geometric
means allows us to introduce the Renyi generalized
entropy as a cost function in the best basis search.
These two extensions also allow the use of
incomplete probability distributions, whereas
Coifman and Wickerhauser's entropy based cost
function is limited to complete probability
distributions.
@inproceedings{IEEE_CCECE02_Dansereau, author = "R. M. Dansereau and W. Kinsner
and V. Cevher", title = "Wavelet packet best basis search using
generalized Renyi entropy", booktitle = "IEEE CCECE 2002",}
Thesis:
V. Cevher
A Bayesian
framework for target tracking using acoustic and image measurements,
Ph.D. thesis, Georgia
Institute of Technology, 2005.
Abstract
BiBTeX Thesis:
PDF
Target tracking is a broad subject area
extensively studied in many engineering disciplines. In
this thesis, target tracking implies the temporal
estimation of target features such as the target's
directionofarrival (DOA), the target's boundary pixels
in a sequence of images, and/or the target's position in
space. For multiple target tracking, we have introduced
a new motion model that incorporates an acceleration
component along the heading direction of the target. We
have also shown that the target motion parameters can be
considered part of a more general feature set for target
tracking, e.g., target frequencies, which may be
unrelated to the target motion, can be used to improve
the tracking performance. We have introduced an acoustic
multipletarget tracker using a flexible observation
model based on an image tracking approach by assuming
that the DOA observations might be spurious and that
some of the DOAs might be missing in the observation
set. We have also addressed the acoustic calibration
problem from sources of opportunity such as beacons or a
moving source. We have derived and compared several
calibration methods for the case where the node can hear
a moving source whose position can be reported back to
the node.
The particle filter, as a recursive algorithm, requires
an initialization phase prior to tracking a state
vector. The MetropolisHastings (MH) algorithm has been
used for sampling from intractable multivariate target
distributions and is well suited for the initialization
problem. Since the particle filter only needs samples
around the mode, we have modified the MH algorithm to
generate samples distributed around the modes of the
target posterior. By simulations, we show that this
“mode hungry” algorithm converges an order of magnitude
faster than the original MH scheme. Finally, we have
developed a general framework for the joint statespace
tracking problem. A proposal strategy for joint
statespace tracking using the particle filters is
defined by carefully placing the random support of the
joint filter in the region where the final posterior is
likely to lie. Computer simulations demonstrate improved
performance and robustness of the joint statespace when
using the new particle proposal strategy.
@phdthesis{GATECH_cevher_thesis, author = "V. Cevher", title = "A {Bayesian} framework for target tracking
using acoustic and image measurements", school = "Georgia Institute of Technology", address = "Atlanta, GA", year = "2005", }


