A new stabilization technique for the fixed point prewindowed RLS algorithm

In this correspondence, a stable finite precision Recursive Least Squares (RLS) algorithm is derived for the prewindowed growing memory case (forgetting factor, , == 1). Previously, it has been shown that the prewindowed growing memory RLS algorithm diverges under fixed-point implementation [2,1]. The random walk phenomenon due to roundoff errors in the weight update causes the divergence of the algorithm. To overcome this effect, these roundoff errors are modeled such that their effect is incorporated into the algorithm. The steady-state behaviour of this new algorithm is analyzed, and it is shown that the divergence phenomenon is actually eliminated, and the new algorithm converges.

[1]  Sasan H. Ardalan,et al.  Convergence and error analysis of the fixed point RLS algorithm with correlated inputs , 1990, International Conference on Acoustics, Speech, and Signal Processing.

[2]  S. H. Ardalan,et al.  Fixed-point roundoff error analysis of the exponentially windowed RLS algorithm for time-varying systems , 1987, IEEE Trans. Acoust. Speech Signal Process..

[3]  Dirk T. M. Slock,et al.  Numerically stable fast recursive least-squares transversal filters , 1988, ICASSP-88., International Conference on Acoustics, Speech, and Signal Processing.

[4]  S. T. Alexander,et al.  A theoretical basis for divergence of conventional recursive least squares filters , 1989, International Conference on Acoustics, Speech, and Signal Processing,.

[5]  C. Caraiscos,et al.  A roundoff error analysis of the LMS adaptive algorithm , 1984 .

[6]  Shu Hung Leung,et al.  On the statistics of fixed-point roundoff error , 1985, IEEE Trans. Acoust. Speech Signal Process..