High throughput MIMO-OFDM detection with graphics processing units

Graphics Processing Units (GPUs) have been one of the most important General Purpose Processors (GPPs) for Software Defined Radio (SDR), since they are specialized for vector-based data-intensive highly parallel computation. Among the main modules of SDR platform, multiple-input multiple-output (MIMO) detection is most time-consuming and computational complexity. A novel strategy is proposed to implement the minimum mean square error (MMSE)-based detector for multiple-input multiple-output (MIMO) wireless communication systems with orthogonal frequency-division multiplexing (OFDM). The key component of the strategy is a massively parallel implementation of the scalable matrix inversion, which is the performance bottleneck in the MMSE detector, on graphics processing units (GPUs). A series of optimization methods such as multi-threaded matrix inversion with multiple data frames, maximizing the utilization of the fast on-chip memories, and overlap kernel execution with data transfer between CPU and GPU, are proposed to construct a reconfigurable and flexible MMSE detector. Experimental results demonstrate that the throughput for 4×4 MIMO-OFDM system can achieve over 18×10⁶ symbol/s, which is equivalent to 100 Mb/s for 64QAM and can satisfy the requirement of the 4th generation wireless communication standards like LTE/LTE-Advanced.