Reply by Richard_K February 22, 20082008-02-22
Please can anyone help me on this topic? Thanks.

> >Hi, > >I wrote a MATLAB program for single carrier cyclic prefix (SC-CP) >frequency domain equalizer (FDE)that is adapted by LMS algorithm. Based
on
>the simulation results, it converges well. However, when the LMS
algorithm
>is modified to NLMS algorithm, it couldn't converge. > >If given that x[n] is the column input vector to the FDE for the kth >block, how should I normalize the step size? Is the normalization
process
>of the step size in frequency domain still the same as it is in the time >domain? > >step_size = normalized_step_size/{Hermitian_of _x[n]*x[n]+ alpha) > >where: >0 < normalized_step_size < 2 >alpha is a small positive constant. >Hermitian_of_x[n] = Hermitian of x[n] > >Please help. Many thanks. >
Reply by Richard_K February 20, 20082008-02-20
Hi,

I wrote a MATLAB program for single carrier cyclic prefix (SC-CP)
frequency domain equalizer (FDE)that is adapted by LMS algorithm. Based on
the simulation results, it converges well. However, when the LMS algorithm
is modified to NLMS algorithm, it couldn't converge.

If given that x[n] is the column input vector to the FDE for the kth
block, how should I normalize the step size? Is the normalization process
of the step size in frequency domain still the same as it is in the time
domain?

step_size = normalized_step_size/{Hermitian_of _x[n]*x[n]+ alpha)

where:
0 < normalized_step_size < 2
alpha is a small positive constant.
Hermitian_of_x[n] = Hermitian of x[n]

Please help. Many thanks.