AEC | Double Talk | ERLE | G.168 | LEC | NLMS

Hi All,
       I'm doin a project on NETWORK ECHO CANCELLATION and need help to choose the best
ADAPTIVE ALGORITHM.

Everywhere its said, RLS is the best adaptive algo as well the most complex and unstable.With
the now available workholic DSP/FPGA can't we bet on RLS for the increased perfomance?

or is still there an unknown force which prevents us from doin so?

(You need to be a member of echocancel -- send a blank email to echocancel-subscribe@yahoogroups.com )

Then how abt the SPARSE ECHO CANCELLERS like PNLMS etc?How do they score
against RLS(FRLS).
-GPG

(You need to be a member of echocancel -- send a blank email to echocancel-subscribe@yahoogroups.com )

Sorry, I haven't played with the proportionating
algorithms. According to the literature, their
convergence speeds are quit fast. They can be good
choices especially when your echo path is short (that
is, you don't get much gain in complexity with
frequency domain algorithms). One extra advantage, as
I see it, of the PNLMS' is that by identifying the
active part of the echo path, you may have the
opportunity of aligning your signals to make double
talk detection easier.

--- ganesha <g...@gmail.com>дµÀ:

> Then how abt the SPARSE ECHO CANCELLERS like PNLMS
> etc?How do they score
> against RLS(FRLS).
> -GPG
>

(You need to be a member of echocancel -- send a blank email to echocancel-subscribe@yahoogroups.com )

I think you'd better bet on frequency domain
algorithms instead of RLS if you think the LMS and its
variance are not good enough. They provide similar
performance (in terms of convergence speed) whilst
maintains a low complexity.Certainly there are fast
and stablized versions of RLS such as Fast transversal
filter.
--- g...@gmail.comдµÀ:

> Hi All,
>        I'm doin a project on NETWORK ECHO
> CANCELLATION and need help to choose the best
> ADAPTIVE ALGORITHM.
> 
> Everywhere its said, RLS is the best adaptive algo
> as well the most complex and unstable.With the now
> available workholic DSP/FPGA can't we bet on RLS for
> the increased perfomance?
> 
> or is still there an unknown force which prevents us
> from doin so?
>

(You need to be a member of echocancel -- send a blank email to echocancel-subscribe@yahoogroups.com )

In my view Trade-off between  fast convergence , stability and complexity
point of  QRD - LSL/MQRD-LSL algorithms looks better than LMS and RLS Algos.

Sparse EC is useful to increase the  span of  echo tail with limited window
of echo spread.

regards
sai.

On 8/13/07, Huo Jiaquan <j...@yahoo.com> wrote:

>   Sorry, I haven't played with the proportionating
> algorithms. According to the literature, their
> convergence speeds are quit fast. They can be good
> choices especially when your echo path is short (that
> is, you don't get much gain in complexity with
> frequency domain algorithms). One extra advantage, as
> I see it, of the PNLMS' is that by identifying the
> active part of the echo path, you may have the
> opportunity of aligning your signals to make double
> talk detection easier.
>
> --- ganesha <g...@gmail.com <ganesh.nitk%40gmail.com>>$B<LF;(B:
>
> > Then how abt the SPARSE ECHO CANCELLERS like PNLMS
> > etc?How do they score
> > against RLS(FRLS).
> >
> >
> > -GPG
> >  
>

-- 
K Sai Prasad
Door No : 8-3-976/112
FLAT NO : 102
SHALIVAHANA NAGAR
YELLAREDDY GUDA
HYDERABAD
Mobile : 9849773892.

(You need to be a member of echocancel -- send a blank email to echocancel-subscribe@yahoogroups.com )

Are QRD - LSL/MQRD-LSL adaptive algos? And =BFare the available for=20
Matlab?

I'm doing an echo canceller for network communication aswell, and I'm=20
using matlab.

=BFCould anyone give me some piece of advice on any algorithm?

Thanks

--- In e...@yahoogroups.com, "Sai Prasad Kalinga"=20
<saiprasadkalinga@...> wrote:
>
> In my view Trade-off between  fast convergence , stability and=20
complexity
> point of  QRD - LSL/MQRD-LSL algorithms looks better than LMS and=20
RLS Algos.
>=20
> Sparse EC is useful to increase the  span of  echo tail with=20
limited window
> of echo spread.
>=20
> regards
> sai.
>=20
> On 8/13/07, Huo Jiaquan <jiaquan2001@...> wrote:
>=20
> >   Sorry, I haven't played with the proportionating
> > algorithms. According to the literature, their
> > convergence speeds are quit fast. They can be good
> > choices especially when your echo path is short (that
> > is, you don't get much gain in complexity with
> > frequency domain algorithms). One extra advantage, as
> > I see it, of the PNLMS' is that by identifying the
> > active part of the echo path, you may have the
> > opportunity of aligning your signals to make double
> > talk detection easier.
> >
> > --- ganesha <ganesh.nitk@... <ganesh.nitk%40gmail.com>>=BC=CC=C6=BB:
> >
> > > Then how abt the SPARSE ECHO CANCELLERS like PNLMS
> > > etc?How do they score
> > > against RLS(FRLS).
> > >
> > >
> > > -GPG
> > >
> >
> >=20=20
> >
>=20
>=20
>=20
> --=20
> K Sai Prasad
> Door No : 8-3-976/112
> FLAT NO : 102
> SHALIVAHANA NAGAR
> YELLAREDDY GUDA
> HYDERABAD
> Mobile : 9849773892.
>

=20

(You need to be a member of echocancel -- send a blank email to echocancel-subscribe@yahoogroups.com )