ADEPT: Task-specific adaptive beamforming.
R. DeLap · Deep Blue (University of Michigan) · 1994
In this dissertation we develop an approach to adaptive beamforming which we call Adaptive Detection/Estimation for Particular Tasks (ADEPT). The ADEPT approach seeks to optimize the best achievable signal detection or parameter estimation performance at the output of a beamsummer array. The philosophy behind our approach is that the adaptation criterion for adaptive beamforming weights should be designed to optimize achievable performance for the primary task of interest. The methodology behind our approach is the use of weight-dependent detection criteria and estimation theoretic lower bounds to specify adaptation criteria appropriate to the specific task of interest. We focus on designing unconstrained adaptive beamsummers for signal detection, estimation of parameters of a spatially-invariant constant-modulus signal amplitude, and for signal direction-of-arrival (DOA) estimation. The adaptive beamsummer is formulated for sensor arrays operating in a broadband environment which is characteristic of slow Rayleigh fading signals. In this environment signal components in successive snapshots are uncorrelated but signal amplitudes remain coherent over the array. For the task of signal detection, we introduce a beamsummer weight adaptation rule which asymptotically maximizes a beamsummer "deflection index" in the limit of a large number of snapshots. The deflection index is closely related to the maximum detection probability achievable at the beamsummer output. We establish that the weight adaptation rule reduces to the Applebaum and Frost beamsummers in the limits of narrowband and broadband array operation, respectively. For the tasks of constant modulus parameter estimation, and DOA estimation, we introduce beamsummer weight adaptation rules which asymptotically minimize the Cramer-Rao lower bound on estimator variance at the beamsummer output. We present simulations comparing simple beampattern-based detection and estimation algorithms for our optimal beamsummer weights and the Applebaum and Frost weights. We also compare our performance to an ideal generalized likelihood ratio signal detector implemented with Akaike's signal selection criterion, and an ideal maximum-likelihood signal DOA estimator, both of which assume that raw multiple-sensor data is available, and the signal and noise powers are known. Even with such unfair disadvantages, our beamsummer-based adaptive algorithms perform remarkably well in comparison.