Abstract:

For MIMO systems, we introduce two novel architectures that significantly reduce the required number of expensive RF chains and still perform as well as a conventional full complexity solution for many cases of interest. The architectures we propose combine RF pre-processing - based soley on channel statistics and not instananeous channel state - with selection and corresponding baseband signal processing. For a system with L RF chains, Nt transmit and Nr receive antennas, the first architecture uses an L x Nr RF pre-processing matrix that outputs only L streams followed by baseband signal processing. The second mone uses an Nr x Nr RF pre-processing matrix followed by selection of L streams and baseband signal processing. We derive the optimal RF pre-processing matrices and the corresponding baseband combining vectors that maximize the average output signal to noise ratio. A beamforming-based geometric intuition is also developed. With the contraints of tody's IC fabrication technology in mind, an approximation that requires only variable phase-shifters and adders, is also proposed and shown to incur a negligible loss in performance. In the process, a fair benchmark for comparing with the previously proposed FFT Butler preprocessing is provided.