Blind Signal Separation in MIMO OFDM Systems Using ICA and Fractional Sampling

This paper addresses the problem of Blind Signal Separation (BSS) as it pertains to Multiple Input Multiple Output (MIMO) systems utilizing Orthogonal Frequency Division Multiplexing (OFDM). In systems with N subcarriers affected by frequency selective channels, when sampling the received signals at the Nyquist rate, the original BSS is transformed into a set of N standard Independent Component Analysis (ICA) problems. In this paper, fractional sampling is employed to increase the number of received signals and improve diversity at the receiver. The up-sampling of the OFDM frames is analyzed in the frequency domain with an up-sampling factor of 2. This doubles the number of ICA problems which provide N new solutions to aid in the recovery of the original data symbols. The additional solutions using equal gain combining improve signal-to-noise ratio (SNR) and therefore the data recovery. To achieve convergence of the ICA algorithm for the over-sampled data symbols, a specialized rotation of constellations on adjacent subcarriers is introduced. The simulation results demonstrate effectiveness of the overall system in its resiliency to inter-carrier interference and the AWGN channel.