Matlab Decision Feedback Equalizer Implementation

[blade88]

Hello,

I want to simulate in Matlab a Time domain Decision Feedback Equalizer.

Let's call:
Nff the number of feedforward coefficients, W = [w₁ w₂ ... w_Nff]
Nbb the number of feedback coefficients, B = [b₁ b₂ .. b_Nbb]
k the time instant I sent N samples
d the delay of the equalizer.

I assume no oversampling and I haven't normalized the tx signal.
Also, suppose we know the W and B coefficients and that we have generated them
based either on ZF or MMSE criterion.

The steps I follow are these:

1. Generate a Nx1 vector of M-QAM symbols (M=4/16/64). This is the Tx signal.
   
   
2. Convolve this vector with a random frequency selective channel
   and then add white noise.
   
   
3. Convolve the received signal with the feedforward filter of the equalizer.
   There, I discard the first d symbols, since if I send the samples at time k
   [y_(k) y_(k+1) .... y_(k+N)] , I will receive the samples k+d time later, meaning
   [y_(k+d) y_(k+d+1).....y_(k+d+N)].
   
   
4. Convolve the Tx signal with the feedback filter, since we assume no errors
   in the decisions of the detector device. The result is the influence of the past symbols.
   
   
5. Substract the influence from the current symbols. Here I have the problem.

My question is how we decide from which symbols we substract?


Thank you in advance.

 

[Mukul Ratwani]

Hello,

Have you solved this problem. I am facing the same issue.

Instead of M-QAM I am generating BPSK data( +1 or -1) and pass it through channel ([1 0.7 0.5 0.4 0.3]) and at the receiver I am trying to equalize it using DFE. Write now I am using only training sequences to adapt the channel. But even for training sequences my weights are not converging and it is diverges.

Can you please highlight this issue. May be it is same problem as you are facing.