Low complexity turbo equalization over unknown frequency selective Rayleigh channels

For turbo equalization, the use of Minimum Mean Square Error or trellis-based algorithms give remarkable performances over static channels, particularly when efficient error correcting codes are used. For the selective time varying channels, when the fading rate is sufficiently slow, a separate trained Least Mean Square (LMS) or Recursive Least Square (RLS) channel estimator may be used to inform the equalizer module. However, when the fading rate is high, it is very difficult to estimate the channel with great precision which can significantly degrade the bit error rate (BER). Channel estimators based on the use of the Kalman filter, maximum likelihood and the Wiener filter are efficient for fast fading channels. However, they require knowledge of channel statistics such as Doppler shift and noise variance. In this paper, we focus on turbo equalizers in realistic scenarios (i.e. statistics are unknown). We propose and evaluate a low complexity iterative receiver, integrating the channel statistics estimation. Simulation results show that the proposed receiver can achieve performances near those obtained with known statistics. It is also proved that if the Doppler frequency is set to a value above or below the true value, the BER will significantly degrade.