Robust Adaptive Beamforming Based on Steering Vector Estimation With as Little as Possible Prior Information

Arash Khabbazibasmenj, Sergiy A. Vorobyov, Aboulnasr Hassanien

IEEE Transactions on Signal Processing · 2012 · 280 citations · 33 references

Concepts

TL;DR

Differences between MVDR robust adaptive beamforming techniques arise mainly from the varying prior information used for steering‑vector estimation. The study proposes a new MVDR RAB that relies on only minimal, easily obtainable imprecise prior information. It formulates the steering‑vector estimate as a non‑convex quadratically constrained quadratic program that maximizes beamformer output power under normalization and interference‑avoidance constraints, using only antenna geometry and angular‑sector knowledge, and provides an efficient rank‑one solution. Simulations demonstrate that this method outperforms previously developed RAB techniques.

Abstract

A general notion of robustness for robust adaptive beamforming (RAB) problem and a unified principle for minimum variance distortionless response (MVDR) RAB techniques design are formulated. This principle is to use standard MVDR beamformer in tandem with an estimate of the desired signal steering vector found based on some imprecise prior information. Differences between various MVDR RAB techniques occur only because of the differences in the assumed prior information and the corresponding signal steering vector estimation techniques. A new MVDR RAB technique, which uses as little as possible and easy to obtain imprecise prior information, is developed. The objective for estimating the steering vector is the maximization of the beamformer output power, while the constraints are the normalization condition and the requirement that the estimate does not converge to any of the interference steering vectors and their linear combinations. The prior information used is only the imprecise knowledge of the antenna array geometry and angular sector in which the actual steering vector lies. Mathematically, the proposed MVDR RAB is expressed as the well known non-convex quadratically constrained quadratic programming problem with two constraints, which can be efficiently and exactly solved. Some new results for the corresponding optimization problem such as a new algebraic way of finding the rank-one solution from the general-rank solution of the relaxed problem and the condition under which the solution of the relaxed problem is guaranteed to be rank-one are derived. Our simulation results demonstrate the superiority of the proposed method over other previously developed RAB techniques.

References

33