Hybrid ZF-ML detection for N × N spatial multiplexing systems

Both zero-forcing (ZF) and maximum-likelihood (ML) criteria are widely used for signal detection in multiple-input multiple-output (MIMO) spatial multiplexing systems. In this paper, we propose a special formulation to combine these two detection algorithms for the N × N scenario, which can provide a good compromise between the error-rate performance and computational complexity. Specifically, the detection is decomposed into some ZF stages, and at each of these stages a fraction of the data streams is nulled to zero. After that, an ML stage is performed utilizing the idea of ML with reduced dimension. We show that the resultant bit-error rate (BER) performance of this hybrid ZF-ML detection can be much better than that of the ZF detection, and the total number of floating-point operations (flops) can be comparable with that of the conventional ZF calculation.