Forums

a question about map equalizer

Started by zqchen July 4, 2007
A question puzzled me for days, have to look for help here:

>From the derivation of MLSE, a matched filter is obviously needed
before the equalizer. Starting from continuous signal, the operation of matched filter and sampling is so natural. But so far as I read about MAP equalizer, I doubt the matched filter is needed no more. Or maybe I've missed something, at least it's not so clearly stated as that of MLSE. Somebody please give me an answer or some references on this issue.
On Jul 4, 12:30 am, zqchen <zhiqun.c...@gmail.com> wrote:
> A question puzzled me for days, have to look for help here: > > >From the derivation of MLSE, a matched filter is obviously needed > > before the equalizer. > Starting from continuous signal, the operation of matched filter and > sampling is so > natural. But so far as I read about MAP equalizer, I doubt the matched > filter is needed > no more. Or maybe I've missed something, at least it's not so clearly > stated as that of > MLSE. > > Somebody please give me an answer or some references on this issue.
It depends on what transmit pulse shape you use. If that pulse already has zero-ISI (inter-symbol interference), then you don't need a matched filter beyond your anti-aliasing filter. Depending on the realization you can get away without it. However, if your system requires that you have a matched filter so that the *effective* pulse shape has zero-ISI, then if you omit the matched filter, your pulse itself has ISI. And on top of that, you have your ISI channel. Sure, you can compensate for all of that, but then your equalizer has to be more complex/longer/ has more states. Does that make sense? Julius

zqchen wrote:

> A question puzzled me for days, have to look for help here: > >>From the derivation of MLSE, a matched filter is obviously needed > before the equalizer. > Starting from continuous signal, the operation of matched filter and > sampling is so > natural. But so far as I read about MAP equalizer, I doubt the matched > filter is needed > no more. Or maybe I've missed something, at least it's not so clearly > stated as that of > MLSE. > Somebody please give me an answer or some references on this issue.
MLSE is trying to find the information sequence which produces the signal closest to the output of your channel. That's the whole idea. Sampling and matched filters are just the details of the particular implementation; they may or may not be used. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com
On Jul 5, 4:28 am, julius <juli...@gmail.com> wrote:
> On Jul 4, 12:30 am, zqchen <zhiqun.c...@gmail.com> wrote: > > It depends on what transmit pulse shape you use. If that pulse > already has zero-ISI (inter-symbol interference), then you don't > need a matched filter beyond your anti-aliasing filter. Depending > on the realization you can get away without it. > > However, if your system requires that you have a matched filter > so that the *effective* pulse shape has zero-ISI, then if you omit > the matched filter, your pulse itself has ISI. And on top of that, > you have your ISI channel. Sure, you can compensate for all > of that, but then your equalizer has to be more complex/longer/ > has more states. > > Does that make sense? > Julius
Sorry, I think maybe the matched filter you mentioned is not the same thing. In MLSE, the matched filter is matched to the channel impulse response. Its target is not zero ISI. ISI is compensated by the following sequence estimator. zqchen
On Jul 5, 5:28 am, Vladimir Vassilevsky <antispam_bo...@hotmail.com>
wrote:
> > MLSE is trying to find the information sequence which produces the > signal closest to the output of your channel. That's the whole idea. > Sampling and matched filters are just the details of the particular > implementation; they may or may not be used. > > Vladimir Vassilevsky > > DSP and Mixed Signal Design Consultant > > http://www.abvolt.com
Thanks. To be more specific, you also think matched filter is not needed for a MAP equalizer, right? MLSE is a more beautiful theory to put all these together. Then the sampling in MAP is governed by the sampling theorem? Is there possibly a fractionally spaced MAP? zqchen