Adaptive Signal Detection in Subspace Interference with Uncertain Prior Knowledge

Abstract

We present a new Bayesian learning algorithm, referred to as the mSKL-GAMP, for signal detection in subspace interference with uncertain/partial prior knowledge of the subspace. It is an extension of the recently introduced SKL algorithm [1] that employs a fixed dictionary for subspace recovery, which causes the grid-mismatch problem. mSKL-GAMP overcomes the problem via a subspace refining procedure. In addition, it integrates the generalized approximate message passing (GAMP) for posterior approximation, which bypasses iterative matrix inversions required by SKL, and thus is computationally much simpler. Numerical results show mSKL-GAMP yields improved detection performance over SKL and other benchmark schemes.