Effect of varying filter length on the performance of LMS turbo equalizers

This paper investigates the effect of varying the length of filters on the average bit error rate (BER) of the turbo equalizers. This study is carried out for two typical radio mobile channels for a coded transmission system that uses a recursive systematic convolutional (RSC) code as the outer code. The outer code in tandem with the discrete time version of the modulation channel forms a serial concatenation scheme (SCS) through an interleaver. The receiver corresponding to this SC transmitter uses an equalizer and decoder that exchange soft information following the turbo principle. The equalizer is configured as a soft interference canceller (SIC) that consists of two linear filters. Both filters use the stochastic gradient least mean square (LMS) algorithm to update their taps. Our objective is to change the length of the filters and evaluate the performance in terms of average (BER). This study is important as in a time varying mobile channel, equalizers giving acceptable performance with fast convergence properties are required and the length of the filters play a role in determining the convergence.