channel equalier doubt?

Started by Unknown August 15, 2001
Hi friends,

HAve any one worked on Adaptive equalization using Kalman filters.
In that case, I would like to know about the channel matrix for indoor

HAri -------------
This mail sent through : http://mail.sify.com

May I ask what does 'near-end signal' means? The signal observed at
the output of the microphone (array) consists of 'near-end speech
plus signal' and 'echo'. If 'near-end signal' refers to the speech
signal from the near-end talker plus any background noise originated
at the near-end, step 4 appears to be pretty odd to me, coz there is
no way on can observe this signal.

There are two ways (maybe I should say three ways) to handle double
talk, as I know. One is use a double talk detector and inhibit the
adaptation whenever double talk is detected. The second one is to
estimate the optimal stepsize (for NLMS) and use that for the
adaptation. The third is quite rough, just adapt on an auxiliary
filter and copy the filter weights to the main filter (used for
cancelling the echo). 'Acoustic echo control: an application of very-
high-order adaptive filters' by Christina Breining et. al. (IEEE SP
Mag. July 1999) contains some good discussion of this topic. And also
chapter 5 of the book 'acoustic signal processing for
telecommunication' by Gay and Benesty.

--- In echocancel@y..., Tapas Guha Neogi _98007017_ <taps8ue@e...>
> hi all..
> I had worked on a project on acoustic echo cancellation and
would like
> to share my knowledge in the field with each of you.I had used an
> algorithm to detect double talk while using the NLMS
algorithm.Though it
> worked fine with the NLMS algorithm it didn't perform well with the
> algorithms I had worked on.
> The algorithm I had followed was as follows.
> 1.determine presence of farend and near end speech by using
> appropriate values of SNRs on each side.While chosing the SNR the
> signal and corresponding noise energy on each side was taken itno
> consideration.
> 2.set up a counter say cnt.
> 3.adapt the filter coefficients only if far end signal
> 4.if far end signal and near end signal present than stop
> adaptation.cnt=cnt+1.
> 5.if cnt > num then increase cnt by a greater factor say 50.dont
> 6.if cnt > highnum then flush all the filter coefficients.restart.
> this algorithm worked for me with the NLMS algorithms under
certain test
> conditions.I dont have a good exposure in this area, so it would be
> adisable to seek advice from more knowledgeable people
around.However i
> thought I would bring forth my algorithm so that i get feedabcks
> drawbacks that exist in the model.Waiting for your kind feedbacks.
> Regards,
> Tapas Guha Neogi
> Senior Undergraduate,
> Electrical Engineering Department,
> Indian Institute of Technology,Bombay,
> India > On Sat, 11 Aug 2001, Fuliang yin wrote:
> > Hi,
> >
> > An echo canceller must include the double-talk module. But in
> > echo canceller, the energy of the acoustic echo often is larger
than that of
> > far-end speech because of effect of power amplifier in TV. This
> > often occurs in the videoconference system. Can anybody tell me
how to
> > efficiently detect the double-talk ? Any information (Article,
> > software, resource, etc) will be appreciated.
> >
> > Thanks
> >
> > Fuliang Yin
> >
> >
> >
> >
> >
> > _____________________________________
> > Note: If you do a simple "reply" with your email client, only the
author of this message will receive your answer. You need to do
a "reply all" if you want your answer to be distributed to the entire
> >
> > _____________________________________
> > About this discussion group:
> >
> > To Join: echocancel-subscribe@y...
> >
> > To Post: echocancel@y...
> >
> > To Leave: echocancel-unsubscribe@y...
> >
> > Archives: http://www.yahoogroups.com/group/echocancel
> >
> > Other DSP-Related Groups: http://www.dsprelated.com
> >
> >
> > ">http://docs.yahoo.com/info/terms/
> >
> >