Parallel maximum likelihood decoding of the golden code on low-power SIMD coprocessor

The Golden Code is an optimum full-rate full-diversity STBC for 2×2 MIMO communication systems. But its optimal decoding suffers high complexity. Due to the benefits it may provide if practically implemented, reducing its decoding complexity has been of interest in recent research. One of the decoders with reduced complexity that have been proposed lately has essentially ML performance and achieves a worst-case complexity of O(N²), where N is the underlying QAM constellation size. In this paper, a parallel implementation approach for this fast essentially ML decoder is proposed. This approach is mapped on low-power processors of the SIMD type, such as the ClearSpeed CSX700. The CSX700 decoding runs up to 30 times faster than sequential decoding on a CPU. This result reveals the potential to achieve real-time low-power ML decoding of the Golden Code with the use of the suitable number of processors.