Reply by Shaw, David G David November 24, 20032003-11-24
The network generally has ECs for voice. These are
very seldom co-located with the central office equipment. Part of
the reason why the tail lengths are so long in modern ECs
is because of the ambiguity as to where the echo producing
device is relative to the EC. If you think about the contributors
to the significant impulse response that forms the echo, it
is dominated by the filtering in the ulaw or alaw codec.
These filters most often produce an impulse response that is
down sufficiently with 10-15 msec.
That codec is generally in the central office, or maybe in a
pedestal in a neighborhood. If the EC were in the central office,
then the tail of the EC would only need to cover 10-15 msec instead
of the 96-128 msec supplied by modern cancellers.

In the context of modems, years ago it was determined that
the double talk detection process implemented in network ECs -
typically Geigel's algorithm - is not reliable for modem signals.
This means that the modem signal causes the algorithm to falsely
determined that energy is arriving from only one end, thus the
canceller tries to converge in the presence of double talking
data signals.
Two major problems occur.
1) Time varying echo path seen by the modem.
2) Totally incorrect convergence.
For these reasons the answer tone from the modem (2100 Hz) includes
phase reversals every 450 msec which are intended to tell
the network echo cancellers to disable themselves. Now with
echo cancellation in the network disabled, the modem has to
do the work to kill the remote echo. Thus, it needs to know
the round trip delay to correctly "place" the far echo canceller.

I hope this answers your questions.

Regards,
Dave Shaw

-----Original Message-----
From: Huo Jiaquan [mailto:]
Sent: Monday, November 24, 2003 12:54 AM
To:
Subject: RE: [echocancel] Re: LEC and NEC The papers i read gave me the impression that the
hybrids at the central offices are equipped with ECs.
So they are not? Or those ECs do not provide
sufficient suppression?
--- "Shaw, David G (David)" <> wrote:
> I will discuss V.32, since it is the first case of
> this
> type and all others are at least similar.
> The calling modem sees answer tone (2100 Hz) and
> sends a tone at 1800 Hz.
> The answer modem then switches from answer tone to a
> pair of tones
> at 600 and 3000 Hz. After a short delay, the answer
> modem
> reverses the phase of the 2 tones and starts a
> timer. When
> the calling modems sees the phase change, it waits a
> predetermined
> 64 symbols at 2400 symbols per second and reverses
> the phase
> of the 1800, while starting a timer. When the
> answer modem sees
> the phase change at 1800 it now has an estimate of
> the round trip
> delay by subtracting the 64 symbols from the total
> on the timer.
> The answer modem now waits that same 64 symbols and
> reverses the
> phase of the two tones again. The calling modem
> sees the phase
> changes again and also has an estimate of the round
> trip delay, again by subtracting the 64 symbols from
> its timer.
> The assumption here is that the echo producing
> mechanism is the
> hybrid in the central office closest to the remote
> modem, so that
> the measured delay end to end is just a little
> longer than the round trip
> delay I would expect to see before the remote echo
> comes back.
> All of this does not take into account that
> possibility
> of another/other echo(es) somewhere else in the
> network.
>
> Regards,
> Dave Shaw
>
> -----Original Message-----
> From: Huo Jiaquan [mailto:]
> Sent: Friday, November 21, 2003 2:10 AM
> To:
> Subject: RE: [echocancel] Re: LEC and NEC > Sorry, i have been working with acoustic echoes, not
> very familiar with modem designs. Could u explain
> 'determined through an exchange between the
> end-points
> during startup' in a bit more details? Or sugguest
> some good ref.?
>
> thx
>
> --- "Shaw, David G (David)" <>
> wrote:
> > For modem, the bulk delay is determined through an
> > exchange between the end-points during the startup
> > of the modem.
> > For voice calls, you will have to come up with a
> > clever algorithm to determine the delay.
> >
> > Dave Shaw
> >
> > -----Original Message-----
> > From: Huo Jiaquan [mailto:]
> > Sent: Monday, November 17, 2003 8:50 PM
> > To:
> > Subject: Re: [echocancel] Re: LEC and NEC
> >
> >
> > How do u determine the bulk delay for the far-end
> > echo?
> > _____________________________________
> 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
> group.
>
> _____________________________________
> About this discussion group:
>
> To Join:
>
> To Post:
>
> To Leave:
>
> Archives:
> http://www.yahoogroups.com/group/echocancel
>
> Other DSP-Related Groups: http://www.dsprelated.com > ">http://docs.yahoo.com/info/terms/ >

http://personals.yahoo.com.au - Yahoo! Personals
New people, new possibilities. FREE for a limited time.
_____________________________________
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 group.

_____________________________________
About this discussion group:

To Join:

To Post:

To Leave:

Archives: http://www.yahoogroups.com/group/echocancel

Other DSP-Related Groups: http://www.dsprelated.com ">http://docs.yahoo.com/info/terms/



Reply by Shaw, David G David November 21, 20032003-11-21
I will discuss V.32, since it is the first case of this
type and all others are at least similar.
The calling modem sees answer tone (2100 Hz) and sends a tone at 1800 Hz.
The answer modem then switches from answer tone to a pair of tones
at 600 and 3000 Hz. After a short delay, the answer modem
reverses the phase of the 2 tones and starts a timer. When
the calling modems sees the phase change, it waits a predetermined
64 symbols at 2400 symbols per second and reverses the phase
of the 1800, while starting a timer. When the answer modem sees
the phase change at 1800 it now has an estimate of the round trip
delay by subtracting the 64 symbols from the total on the timer.
The answer modem now waits that same 64 symbols and reverses the
phase of the two tones again. The calling modem sees the phase
changes again and also has an estimate of the round
trip delay, again by subtracting the 64 symbols from its timer.
The assumption here is that the echo producing mechanism is the
hybrid in the central office closest to the remote modem, so that
the measured delay end to end is just a little longer than the round trip
delay I would expect to see before the remote echo comes back.
All of this does not take into account that possibility
of another/other echo(es) somewhere else in the network.

Regards,
Dave Shaw

-----Original Message-----
From: Huo Jiaquan [mailto:]
Sent: Friday, November 21, 2003 2:10 AM
To:
Subject: RE: [echocancel] Re: LEC and NEC Sorry, i have been working with acoustic echoes, not
very familiar with modem designs. Could u explain
'determined through an exchange between the end-points
during startup' in a bit more details? Or sugguest
some good ref.?

thx

--- "Shaw, David G (David)" <> wrote:
> For modem, the bulk delay is determined through an
> exchange between the end-points during the startup
> of the modem.
> For voice calls, you will have to come up with a
> clever algorithm to determine the delay.
>
> Dave Shaw
>
> -----Original Message-----
> From: Huo Jiaquan [mailto:]
> Sent: Monday, November 17, 2003 8:50 PM
> To:
> Subject: Re: [echocancel] Re: LEC and NEC > How do u determine the bulk delay for the far-end
> echo?
>


_____________________________________
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 group.

_____________________________________
About this discussion group:

To Join:

To Post:

To Leave:

Archives: http://www.yahoogroups.com/group/echocancel

Other DSP-Related Groups: http://www.dsprelated.com ">http://docs.yahoo.com/info/terms/



Reply by Shaw, David G David November 18, 20032003-11-18
For modem, the bulk delay is determined through an
exchange between the end-points during the startup
of the modem.
For voice calls, you will have to come up with a
clever algorithm to determine the delay.

Dave Shaw

-----Original Message-----
From: Huo Jiaquan [mailto:]
Sent: Monday, November 17, 2003 8:50 PM
To:
Subject: Re: [echocancel] Re: LEC and NEC How do u determine the bulk delay for the far-end
echo?

--- anjuman2 <> wrote: > --- In
, "lee"
> <hao_li@y...> wrote:
> > Can someone tell me what the difference is between
> a Line Echo
> > Canceller and a Network Echo Canceller? To me, it
> seems that the
> only
> > difference is the echo path delay. Or say, they
> are just different
> > echo paths.
> >
> > If that's right, then a echo canceller with, say,
> 128ms delay
> should
> > be able to cancel near end and far end echos. Or
> is there a better
> > approach?
> >
> > Please advice. Thanks.
>
> Dont know the terminology difference between line
> and network EC's.
> But there's a big difference between how to
> implement a near LEC or a
> far LEC. Having one long EC of (say) 128msec is not
> optimal if in
> fact, there are two distinct sources--a near and a
> far one. The best
> way there is to have two short EC's, one that spans
> the near-echo
> (typically <16msec) and another that spans the
> far-end echo but with
> a large (also called bulk) delay in front. But the
> second EC could
> also be spanning only 16msec of echo. This
> arrangment is used in
> modems a lot and is computationally very efficient. >

http://personals.yahoo.com.au - Yahoo! Personals
New people, new possibilities. FREE for a limited time.
_____________________________________
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 group.

_____________________________________
About this discussion group:

To Join:

To Post:

To Leave:

Archives: http://www.yahoogroups.com/group/echocancel

Other DSP-Related Groups: http://www.dsprelated.com ">http://docs.yahoo.com/info/terms/



Reply by Huo Jiaquan November 18, 20032003-11-18
How do u determine the bulk delay for the far-end
echo?

--- anjuman2 <> wrote: > --- In
, "lee"
> <hao_li@y...> wrote:
> > Can someone tell me what the difference is between
> a Line Echo
> > Canceller and a Network Echo Canceller? To me, it
> seems that the
> only
> > difference is the echo path delay. Or say, they
> are just different
> > echo paths.
> >
> > If that's right, then a echo canceller with, say,
> 128ms delay
> should
> > be able to cancel near end and far end echos. Or
> is there a better
> > approach?
> >
> > Please advice. Thanks.
>
> Dont know the terminology difference between line
> and network EC's.
> But there's a big difference between how to
> implement a near LEC or a
> far LEC. Having one long EC of (say) 128msec is not
> optimal if in
> fact, there are two distinct sources--a near and a
> far one. The best
> way there is to have two short EC's, one that spans
> the near-echo
> (typically <16msec) and another that spans the
> far-end echo but with
> a large (also called bulk) delay in front. But the
> second EC could
> also be spanning only 16msec of echo. This
> arrangment is used in
> modems a lot and is computationally very efficient. >

http://personals.yahoo.com.au - Yahoo! Personals
New people, new possibilities. FREE for a limited time.


Reply by Steve Underwood October 15, 20032003-10-15
anjuman2 wrote:

>Dont know the terminology difference between line and network EC's.
>But there's a big difference between how to implement a near LEC or a
>far LEC. Having one long EC of (say) 128msec is not optimal if in
>fact, there are two distinct sources--a near and a far one. The best
>way there is to have two short EC's, one that spans the near-echo
>(typically <16msec) and another that spans the far-end echo but with
>a large (also called bulk) delay in front. But the second EC could
>also be spanning only 16msec of echo. This arrangment is used in
>modems a lot and is computationally very efficient.
>
There are additional reasons for doing this with modems, although they
are becoming less important. The far echo may be frequency shifted, if
the channel includes any FDM links (many countries have completely
eliminated these, but not all). The near echo will never be shifted like
this. The carrier frequency can be expected to be very accurate, so by
locking to this, and doing a frequency shift back to the correct carrier
frequency, the far echo can be more accurately adapted to (assuming the
rx and tx paths have similar shifts). The near echo needs to be treated
separately, without the frequency correction.

Regards,
Steve



Reply by anjuman2 October 14, 20032003-10-14
--- In , "lee" <hao_li@y...> wrote:
> Can someone tell me what the difference is between a Line Echo
> Canceller and a Network Echo Canceller? To me, it seems that the
only
> difference is the echo path delay. Or say, they are just different
> echo paths.
>
> If that's right, then a echo canceller with, say, 128ms delay
should
> be able to cancel near end and far end echos. Or is there a better
> approach?
>
> Please advice. Thanks.

Dont know the terminology difference between line and network EC's.
But there's a big difference between how to implement a near LEC or a
far LEC. Having one long EC of (say) 128msec is not optimal if in
fact, there are two distinct sources--a near and a far one. The best
way there is to have two short EC's, one that spans the near-echo
(typically <16msec) and another that spans the far-end echo but with
a large (also called bulk) delay in front. But the second EC could
also be spanning only 16msec of echo. This arrangment is used in
modems a lot and is computationally very efficient.



Reply by lee August 28, 20032003-08-28
Can someone tell me what the difference is between a Line Echo
Canceller and a Network Echo Canceller? To me, it seems that the only
difference is the echo path delay. Or say, they are just different
echo paths.

If that's right, then a echo canceller with, say, 128ms delay should
be able to cancel near end and far end echos. Or is there a better
approach?

Please advice. Thanks.