ViPer MIMO: Increasing Large MIMO Efficiency via Practical Vector-Perturbation

摘要

Large multi-user MIMO systems with spatial multiplexing are among the most promising approaches for increasing wireless throughput while serving many clients. Yet, the achievable spectral efficiency of current large MIMO systems is limited by the adoption of simple, but sub-optimal, linear precoding techniques (e.g, minimum-mean-square-error (MMSE)). Nonlinear precoding methods, like Vector Perturbation (VP), claim to be able to provide improved network throughput. However, such methods are still purely theoretical and they do not account for the practical aspects of actual wireless systems, as the corresponding complexity and latency requirements, or the need for feasible rate adaptation. This paper presents ViPer, the first practical VP-based MIMO system design. ViPer substantially reduces the latency requirements of VP by employing massively parallel processing and realizes a practical rate adaptation method that efficiently translates VP's signal-to-noise-ratio (SNR) gains into actual throughput gains. In our first systematic experimental evaluation of VP-based precoders, we show that ViPer can deliver in practice up to 30% higher throughput than MMSE precoding with comparable latency requirements. In addition, ViPer can match the performance of state-of-the-art parallel VP precoding schemes, by utilizing less than one tenth of the processing elements.