loading page

Approximate Noise-Whitening in MIMO Detection via Banded-Inverse Extension
  • Sha Hu ,
  • Hao Wang

Abstract

In this paper, we propose a novel approximate noise-whitening for multi-input multi-output (MIMO) detection with banded-inverse extension (BIE). If the noise covariance matrix origins from a Gauss-Markov process (GMP), the inverse is banded and the proposed method is exact yielding no accuracy-losses. For general matrices to be inverted, the approximation-errors introduce a mismatched MIMO detection-model, which can cause performance-degradation. Hence, we further develop an information-theoretic tool using generalized mutual information (GMI) to assess the impacts from approximation-errors on the final achievable rate. We show that the approximate noise-whitening method based on BIE not only minimizes the noise distortion measured from Kullback-Leibler (KL) divergence, but also maximizes the GMI of the MIMO system asymptotically as signal-to-noise ratio  (SNR) increases. Besides, the BIE also provides a unified framework for approximate matrix-inverse with an adjustable band-size.