A Universal Systolic Array for Linear MIMO Detections

The detection of MIMO signals presents a formidable computation challenge due to the inherent complexity incurred by matrix-type wireless fading channels. To reduce the hardware cost, linear MIMO detection algorithms can be used instead of non-linear space-time processing. Conventionally, adaptive linear algorithms such as RLS and LMS are treated separately in implementation from matrix inversion based algorithms such as ZF and MMSE. In this paper, we identify the similarities between the QR-decomposition based forms of RLS and MMSE MIMO detection and present a unified systolic array implementation that can accommodate both. In the meanwhile, some subtle but important differences also exist. As a result, the unified array must have a size of N_r x N_r (assuming N_t les N_r) and both array inputs and array operations need to be controlled accordingly. As special cases of RLS and MMSE, LMS and ZF can also be realized in the same array, making the array flexible enough for all linear MIMO detection algorithms.