Robust Detection with Stable Throughput over Ill-Conditioned Channels for High-Order MIMO Systems

The conventional depth-first sphere decoder (DF-SD) and breadth-first K-Best algorithm for MIMO detection are subject to unstable output throughput and error-rate degradation, respectively, over ill- conditioned channels. These problems become more serious in high-order MIMO systems with more antennas and larger constellation size. In this paper, we propose a condition-number driven decoder for high-order MIMO systems to overcome these problems. Specifically, we first apply K-Best algorithm with Winner-Path Enumeration (WPE) method on high layers of the detection tree to guarantee the throughput stability and apply DF-SD on remaining layers to lower the bit-error-rate (BER). As the throughput fluctuation or, equivalently, complexity fluctuation of DF-SD often happens with a large condition number, we adaptively adjust the number of layers applying K-Best WPE by comparing the condition number with a predefined threshold, such that the complexity can be further decreased. Through theoretical analyses and simulation verifications, we derive a criterion of determining the number of layers adopting K-Best algorithm. Performance analyses and simulation evaluations show that our proposed scheme has much lower average complexity than K-Best WPE, while achieving near-optimal BER performance and stable complexity / throughput.