Antithetic Dithered 1-Bit Massive MIMO Architecture: Efficient Channel Estimation via Parameter Expansion and PML

Abstract

Drawing from a recent work on negative noise correlation in quantization and statistics, we propose a novel antithetic dithered 1-bit massive MIMO receiver architecture and develop efficient channel estimation algorithms that exploit the natural and induced negative correlated noise in the system. We illustrate that both linear and nonlinear estimators can benefit from negative correlation. We provide a rigorous analysis of a low-complexity nonlinear estimator for channel estimation. In the process, we developed a generalized statistical framework to analyze correlated quantized output arising from this generalized linear model. We formalized the approximation technique used in this work as a special case of the more general pseudo maximum likelihood method. A parameter expanded expectation maximization (PX-EM) algorithm applied to such a system is shown to exhibit fast convergence, possessing an upper bounded convergence guarantee and a graceful monotonic estimation performance over a large SNR range. Stochastic Gibbs sampling algorithms are constructed to evaluate truncated multivariate normal distributions and to implement an asymptotically exact data augmentation algorithm for comparison.