College of Engineering and Applied Science
UW STAT Information
The University of Wyoming Source Tracking Array Testbed, UW STAT [1], was specifically
developed to perform experimental evaluation of high-resolution source tracking algorithms.
The array is a six element uniformly spaced linear array. UW STAT is a compact testbed
and allows for precision source motion. The motion is controlled by a motor and the
angular position is recorded by an encoder, thus allowing for a direct comparison
of the tracking algorithm's performance with the true position recorded by the encoder.
In the past decade, many advanced algorithms have been developed to track moving sources
using an array of sensors. Most of these algorithms rely on an updating of the signal
or noise subspaces and the use of a high-resolution direction finding algorithm. Although
a significant amount of research has been performed in this area, very few experimental
tracking results have been reported in the open literature for actual sensor array
systems.
The UW STAT allows for a variety of experimental scenarios with precise control of
source motion. The tracking array is a compact ultrasonic array. The compactness of
the testbed and the precisely controlled environment allows for detailed studies of
cause and effect. Source motion is controlled using a precision motor to turn the
shaft of a radial arm. An important aspect of the system is the true source position
being precisely measured using an encoder located on the shaft. Tracking algorithm
performance is therefore compared to the actual position recorded by the encoder.
Many of the new tracking algorithms use high-resolution source localization methods,
which estimate the direction-of-arrival of a source at one instant in time. These
high-resolution algorithms are very computationally intensive because they involve
the separation of the received signal into a signal subspace and a noise subspace.
Much of the recent research therefore has concentrated on updating these subspaces
after each snapshot of array data is received. These algorithms can usually be classified
into one of two groups --
adaptive or
recursive. For
adaptive tracking algorithms the update of the subspace will asymptotically approach the true
subspace if the data is wide sense stationary. For
recursive tracking algorithms the update of the exact subspace, using a new data vector, results
in the exact subspace for the given modified data matrix. An excellent discussion
and bibliography of many subspace tracking algorithms are given in [2]. These algorithms
have been investigated with some theoretical analysis and simulation studies, but
little has been reported on the experimental capabilities and limitations of the methods
using actual array data.
A front view of the mechanical system for UW STAT is shown in Figure 1. The mechanical
system, where signal transmission and reception take place, is contained in an echoless
anechoic chamber and consists of the following: two transmitting sources capable of
independent movement in the same arc about the linear receiving sensor array, a DC
motor for moving one of the sources, and an encoder for precise position knowledge
of the moving source. The remainder of the testbed is placed outside the anechoic
chamber and consists of the following: transmission hardware, reception hardware,
carrier signal generation hardware, message signal generation devices (white noise
generators and arbitrary function generators), and a PC/data acquisition system.
Figure 1
A doublexi-sideband suppressed-carrier (DSBSC) modulated signal is generated by the
transmission hardware. Message signals with an acoustic frequency range of 0-2 kHz
from the message signal devices can be used for modulation with a carrier signal.
The carrier signal is generated at a frequency of 40 kHz so that piezoelectric transducers
with natural frequencies of 40 kHz can be utilized for signal transmission and reception.
This ultrasonic carrier frequency allows for the physical dimensions to be practical
for a testbed. The approximate narrowband plane wave generated by the transmitting
transducer has a wavelength of l = 8.275 mm. Due to the physical size of the receiving
transducers, the element spacing of the linear array is 2.11. This sensor spacing
limits the field of view for a spatially non-aliased sector from -13.5 to 13.5 degrees.
This non-aliased range for the angle-of-arrival of a transmitting source corresponds
well to the limited range over which the transducers can be considered omnidirectional.
Movement of one of the transmitting sources is controlled with a DC motor geared to
the sources radial arm shaft as shown in Figure 1. This allows for motion control
and experimental repeatability of movement for one of the sources. In addition to
the DC motor control, this source also has an incremental pulse encoder coupled to
its shaft as illustrated in Figure 1. The encoder allows for the position of the source
to be known at all times as it moves through the sensor array field-of-view. Source
tracking algorithm performance can be evaluated against the precisely known position
of this source. The second transmitting source is able to move independently from
this source at approximately the same arc length, but it does not have motor control
or encoder position knowledge. This second source can be used as an interfering source
to see how well source tracking algorithms perform when two sources occupying the
same temporal frequency band are near each other, or it could be used to create multipath
signal effects
The system is approximately modeled as far field. Gain and phase characteristics of
the transmitted DSBSC signal are affected by differences in the characteristics of
the six receiving transducers. The reception hardware performs the in-phase and quadrature
demodulation of the received DSBSC signal. This gives both the real and imaginary
components of the message signal so that the recovered message signal information
is complete.
After quadrature demodulation of the signal received at the sensor array, a 12 bit
data acquisition card is utilized to transform the recovered analog data into digital
data for storage in a PC. The data acquisition system has 16 channels. Since it is
desirable to sample the recovered demodulated signals from the six sensors in the
array at the same time, simultaneous sample and hold circuitry is utilized in conjunction
with the data acquisition card. The data acquisition card also acquires data from
the encoder to allow for the precise position knowledge of one of the transmitting
sources. One of the transmitted message signals may also be sampled. The digital data
is then stored in the PC for processing at a later time. The array is calibrated for
gain, phase, and mutual coupling errors using the method described in [3]. Thirty
one measured direction vectors were used to estimate the gain, phase, and mutual coupling
calibration matrix for the system.
[1]J.W. Pierre, E.D. Scott, and M.P. Hays, "A Sensor Array Testbed for Source Tracking
Algorithms," Proc. of ICASSP '97, Munich, Germany, April 1997.
[2]E.M. Dowling, L.P. Ammann, and R.D. DeGroat, "ATQR-iteration based adaptive SVD
for real time angle and frequency tracking," IEEE Trans. on Signal Processing, vol.
42, April 1994.
[3]J.W. Pierre and M. Kaveh, "Experimental evaluation of high resolution direction-finding
algorithms using a calibrated sensor array testbed," Digital Signal Processing: A
Review Journal, October 1995.
