Adaptive system identification based on adaptation with data discarding

At high data rates, the adaptation processor of the adaptive filter may be unable to perform the multiplications required by the adaptation algorithm each baud interval. A direct, but expensive, solution of this problem is to use parallel processing. The present paper presents a simple alternative solution that is suitable for the case of adaptive system identification. The idea of the proposed method is to make one iteration for each several baud intervals. The idea is implemented as follows. With N being the number of adaptive filter coefficients, N successive input samples and the corresponding sample of the system output are injected to the adaptation processor to perform one iteration. Input and output samples that come before the termination of this iteration are discarded by the adaptation processor. After termination of the iteration, new N input samples and the corresponding output sample are injected to the adaptation processor and the procedure is repeated. The above idea is applied to both the LMS algorithm and the sign algorithm. The effect of data discarding on the transient performance of the identification procedure is studied by computer simulations for both white and correlated inputs of the system.