拘束条件つきスライド窓形RLSのシストリックアレイ

概要

論文の詳細を見る
Recently, there have been many works concerning systolic array implementations of recursive least squares(RLS). It has been shown that based on the QR decomposition (QRD), i.e., X=QR with X, the data matrix, Q, an orthogonal matrix, R, an upper triangular matrix, the weight extraction can be done by updating R^<-1>(inverse updating) on a lower triangular array and by combining this array with an upper triangular array for updating R, the overall rhombic array is obtained (McWhirter). Also, Sakai has derived a similar array by using the modified Gram-Schmidt (MGS) method. Recently, a QRD algorithm of square-root-free MGS pseudo orthogonalization (MGSPO) for sliding window block RLS has been derived by Hsieh, Liu, and Yao where in this method both updating for new data coming in and downdating for old data discarded out must be done. In based on the inverse updating recursion has been derived. In beamforming applications, the constrained RLS(CRLS), that is, RLS with multiple linear constriants on the weight coefficients has been considered and a systolic array is derived for block CRLS with a forgetting factor in. In this paper the corresponding result for sliding window CRLS is presented.
社団法人電子情報通信学会の論文
1995-03-27