Reply by mobi July 24, 20062006-07-24
Hi all,
I am having problems regarding weight extraction in QRD-LSL algorithm
i.e., I want to know the unknown filter coefficients of the QRD-LSL
adaptive filter. I have implemented QRD-LSL algorithm, given on page
666 of symon hakin book on adaptive filtering. On page 686 he mentions
weight extraction, which states that in order to compute the weight
vector we need backward prediction coefficients from order 0 to N-1
(where N is filter length) and the regression coefficient, denoted as
kappa in his book (eq 15.191). I know that using levinson durbin
recursion backward prediction coefficients from order 0 to N-1 can be
obtained using eqns 15.44 and 15.45. Now the problem is that I compute
the weight coefficients and compare them with that of RLS algorithm
(both are initialized exactly and the square difference of the errors
is -300 dBs down ) and i see that the first and the last coefficients
match quiet close (-300 dB down in square difference), but rest of the
coefficients donot go below -100 dBs. Any suggestions?