1

I have a question regarding the implementation of MMSE-DFE. Here is the conventional block diagram:

DFE

The signal y shown in the pic is, for clarification, y(kT) i.e. the signal y(t) coming out from the matched filter (not shown in the pic) sampled at times t=kT with T the symbol period.

Now, suppose I have a vector of N samples [y(0), y(T), ... , y((N-1)T)] as the input of the feedforward filter (FFF), and let the FFF have length M, then the output of the FFF will be a signal of length N+M-1 (convolution of 2 signals of lengths N, M).

So I'm expecting a signal of length N+M-1 as the input of the data detector block, but since I must only detect N symbols... it looks like I must discard some of the samples (actually M-1 samples) that go into the detector. Which ones must be discarded, and why?

Thanks in advance!

Javi
  • 121
  • 4

1 Answers1

1

Let's assume y to be a sequence of individually detected bits and the system in your image tries to catch a multibit pattern (or a class of patterns) from the stream regardless some errors in bit by bit detections. The information about the existence of the interesting pattern has been spred over the whole response of your FFF so you should not discard anything assuming your system is properly designed and does not have fully useless redundant delay.

You will not get the indication of the existence of the interesting pattern as soon as the last bit has been inputted to FFF, but some amount of time later.

Think about a radar which sends continuous signal. Echos from separate targets overlap. Nothing really should be discarded. If you know there's no overlaps, you can obviously discard everything between a detection and the first possible arrival of the first bit of the next interesting pattern.

  • I understand. Actually, to be more precise in my case y is a bandbase representation of the received symbols from a 16-QAM. This means each component of the sequence y is a complex number which could already be used for detection, except for the fact that there is intersymbol interference present, and so the DFE appears to remove ISI. The problem is I don't know what to do with the output of the DFE to decide which were the symbols transmitted. – Javi Nov 27 '17 at 15:10