Efficient Sphere Detector Algorithm for Massive MIMO using GPU Hardware Accelerator

To further enhance the capacity of next generation wireless communication systems, massive MIMO has recently appeared as a necessary enabling technology to achieve high performance signal processing for large-scale multiple antennas. However, massive MIMO systems inevitably generate signal processing overheads, which translate into ever-increasing rate of complexity, and therefore, such system may not maintain the inherent real-time requirement of wireless systems. We redesign the non-linear sphere decoder method to increase the performance of the system, cast most memory-bound computations into compute-bound operations to reduce the overall complexity, and maintain the real-time processing thanks to the GPU computational power. We show a comprehensive complexity and performance analysis on an unprecedented MIMO system scale, which can ease the design phase toward simulating future massive MIMO wireless systems.