Adaptive transmission for radar arrays using Weiss-Weinstein bounds

TL;DR

This paper introduces an adaptive radar array transmission algorithm that optimizes measurement parameters in real-time using Weiss-Weinstein bounds, Bayesian filtering, and efficient computation methods to improve Doppler and DoA estimation accuracy.

Contribution

It proposes a novel adaptive measurement selection method based on Weiss-Weinstein bounds, combining Bayesian filtering and computationally efficient approximations for real-time radar parameter estimation.

Findings

Effective real-time adaptive measurement selection demonstrated in simulations.

Algorithm outperforms non-adaptive methods in estimation accuracy.

Feasible implementation using look-up tables or neural networks.

Abstract

We present an algorithm for adaptive selection of pulse repetition frequency or antenna activations for Doppler and DoA estimation. The adaptation is performed sequentially using a Bayesian filter, responsible for updating the belief on parameters, and a controller, responsible for selecting transmission variables for the next measurement by optimizing a prediction of the estimation error. This selection optimizes the Weiss-Weinstein bound for a multi-dimensional frequency estimation model based on array measurements of a narrow-band far-field source. A particle filter implements the update of the posterior distribution after each new measurement is taken, and this posterior is further approximated by a Gaussian or a uniform distribution for which computationally fast expressions of the Weiss-Weinstein bound are analytically derived. We characterize the controller's optimal choices in terms of SNR and variance of the current belief, discussing their properties in terms of the ambiguity function and comparing them with optimal choices of other Weiss-Weinstein bound constructions in the literature. The resulting algorithms are analyzed in simulations where we showcase a practically feasible real-time evaluation based on look-up tables or small neural networks trained off-line.