DFT: Tk,m=1/sqrt(M)*exp((-j*2*pi*m*k)/M)
DCT:  T*k,m =β(k)*cos(k*(2*m+1)*pi/M)
β(0)=1/sqrt(M) et β(k)=sqrt(2/M)
Algo
X’i= [x(i)……….x(i-M+1)]*T
ldk(i)=b*lbk(i-1)+(1-b)*norm(x’i(k) )^2
Di=diag(ldk)
e(i)=d(i)-x’i*H’i-1
Hi=Hi-1+nu*D-1i*x’*i*e(i)
I use this algorithm to build an dftlms and dctlms, something I don't
understand, that all my tests with a speech signal show that the Nlms is
the better. So my question is it normal? If it's the case, why use dftlms
and dctlms who are more complex and take more time to compute?
Thanks
Moc