IEEE - Aerospace and Electronic Systems - November 2022 - 6

Detection, Mode Selection, and Parameter Estimation in Distributed Radar Networks
converge, but does not necessarily converge to reach a
consensus and obtain the global GLRT value.
Despite this progress in distributed target detection,
nonconvex optimization problems have to be solved
in [15] with endured computational complexity and no
closed-form expressions. Moreover, existing methods do
not evaluate the impact on the global GLRT metric when
using local information from nodes that have obscure
vision of targets, i.e., the nodes that observe low signal-tonoise
ratio (SNR) and/or signal-to-noise-and-interference
(SINR) values. In this article, we leverage clutter suppression
techniques [18], [19] to obtain an analytical expression
for the global GLRT metric that possesses constant
false alarm rate (CFAR) property. We show how this
global metric can be achieved with the distributed algorithm
that requires sharing of only the scalar GLRT values
among the neighboring nodes. We also demonstrate that
poor performing nodes can significantly improve their
detection performance by cooperating with good performing
nodes, i.e., the nodes that observe high SNR values.
However, in some cases, in order to avoid degradation in
global detection performance, it may be necessary to suspend
one or more poor performing nodes from their active
mode of operation. These nodes can be made momentarily
passive until such time when their corresponding SINRs
sufficiently rise. This approach, referred to as mode selection,
shifts resources allocated to poor performing nodes
to good performing nodes so the latter can pursue active
mode of operation with increased resources. This strategy
is deemed to significantly improve resource utilization of
the distributed radar network. The nodes that become passive
sensors can angularly localize RF emitters and interferers
in the environment and share this information with
the network active nodes to improve range, Doppler, and
angle estimation of radar targets.
Direction of arrival (DoA) estimation of multiple signal
sources is a classical and standard problem in array
signal processing and radar systems [20]. One of the key
objectives is to resolve as many sources as possible with a
given number of array elements and aperture size. Sparse
arrays (minimum redundancy array, nested and coprime
arrays [21], [22], [23]) can detect more sources than the
available number of physical antennas (sensors). For an
example, a uniform linear array (ULA) with Ne sensors
can resolve up to Ne 1 noncoherent sources, whereas
sparse arrays can identify up to N2
e uncorrelated sources.
Moreover, it is known from the literature that the DoA
estimation methods based on compressive sensing (CS)
and sparse signal recovery [24], [25], [26] provide superior
performance to subspace-based method, such as multiple
signal classification (MUSIC), in their abilities to deal
with coherent and highly correlated sources. In this regard,
greedy sparse recovery algorithms, such as orthogonal
matching pursuit (OMP), have been extended to distributed
OMP (DOMP) [27], which has found applications in
6
distributed sensor networks. The distributed versions of
subspace pursuit [28], such as simultaneous subspace pursuit
and decentralized and collaborative subspace pursuit
[29], have been proposed for distributed networks.
These methods have been modified and extended to
through the wall imaging applications in [30]. A key feature
of these methods is that all nodes require communications
with each other. For example, the modified DOMP
of [30] requires averaging the observation vectors or their
iterative versions at all distributed nodes. This requirement
can be at odds with poor wireless links associated with
distant nodes. In typical networks with various geometries,
all nodes are not necessarily near each other. In this case,
large delays and signal attenuation impede data sharing
for far-off distant nodes, rendering the above approaches
less effective. This calls for the need of developing DoA
estimation approaches for distributed systems in which
only neighboring nodes are allowed to communicate.
In this article, we propose distributed DoA estimation
of targets in a network of radar nodes, in which only the
neighboring nodes communicate and share limited TRI.
The sparsity of the target scene is exploited and distributed
DoA estimation is formulated as the estimation of
sparse vectors with which the communicating nodes aim
to achieve a consensus. The estimation problem is solved
iteratively with the alternating direction method of multipliers
(ADMM) method [31], [32], which can be shown
to converge to the global solution due to the convexity
of the underlying problem. The proposed approach is
general and can be applied to any node antenna arrangement
or array configuration. However, we consider one
well-known sparse array structure, namely coprime
arrays, at each node to leverage more degrees of freedom,
permitting direction finding of more sources than
the number of the node available antennas. It is worth
noting that the distributed DoA estimation can also be
developed with an objective of increasing the total number
of degrees of freedom of a system in which each
node has fewer antennas than the number of sources/targets.
The distributed nodes and their antennas can be
located in such a way that limited signal observations,
for example, quantized signals from the nodes, can be
collected at a central coordinator to form a virtual coarray
corresponding to a certain sparse array configuration
and to enhance the overall degrees of freedom of the distributed
radar network. While such design approach
opens up new research directions, in this article, we
focus on the objective of enhancing DoA estimation of
all nodes, wherein the nodes observing low SNR values
would benefit by cooperating with the ones observing
high SNR values.
Delay and Doppler estimation problems in radar systems
are typically solved as follows. The received sampled
signal is collected for P number of radar pulses to form a
data matrix of size P L, where L is the number of data
IEEE A&E SYSTEMS MAGAZINE
NOVEMBER 2022
IEEE - Aerospace and Electronic Systems - November 2022

Table of Contents for the Digital Edition of IEEE - Aerospace and Electronic Systems - November 2022
