Cancellation

Hi all,

I have read up on using echo cancellation and training FIR filter
coefficients with various algorithms. All these systems rely on two
uncorrelated signals to find echo paths. Does anyone know if methods
exist to allow echo paths to be found using a single frequency turned
on an off or a frequency sweep(chirp) to allow delays to be
determined? Could a FIR filter be trained using such signals.

Also, i simualted a system using Matlab and i can determine a echo
path using the LMS algorithim trained on noise. However if i also put
a single frequency through the system the training algorithm goes
silly. Is there a way to ignor periodic signals?


Any thoughts or comments would be much appreciated.

Thanks

Hi Michael,

Single tone will not generally work. You will usually find there are
an infinite number of solutions for the adaption with a single tone.
Any of them work for that tone, but only one of those solutions works
for a comprehensive range of signals. Most broadband things are OK. A
chirp can work, but broadband noise generally gives the best results.

Regards,
Steve

mickwjones@hotmail.com (Michael) wrote in message news:<f10151fb.0403180209.6527336@posting.google.com>...
> Hi all,
> 
> I have read up on using echo cancellation and training FIR filter
> coefficients with various algorithms. All these systems rely on two
> uncorrelated signals to find echo paths. Does anyone know if methods
> exist to allow echo paths to be found using a single frequency turned
> on an off or a frequency sweep(chirp) to allow delays to be
> determined? Could a FIR filter be trained using such signals.
> 
> Also, i simualted a system using Matlab and i can determine a echo
> path using the LMS algorithim trained on noise. However if i also put
> a single frequency through the system the training algorithm goes
> silly. Is there a way to ignor periodic signals?
> 
> 
> Any thoughts or comments would be much appreciated.
> 
> Thanks

mickwjones@hotmail.com (Michael) wrote in message news:<f10151fb.0403180209.6527336@posting.google.com>...
> Hi all,
> 
> I have read up on using echo cancellation and training FIR filter
> coefficients with various algorithms. All these systems rely on two
> uncorrelated signals to find echo paths. Does anyone know if methods
> exist to allow echo paths to be found using a single frequency turned
> on an off or a frequency sweep(chirp) to allow delays to be
> determined? Could a FIR filter be trained using such signals.
> 
> Also, i simualted a system using Matlab and i can determine a echo
> path using the LMS algorithim trained on noise. However if i also put
> a single frequency through the system the training algorithm goes
> silly. Is there a way to ignor periodic signals?
> 
> 
> Any thoughts or comments would be much appreciated.
> 
> Thanks


Around 1993 time frame, ANSI produced a technical report on echo
cancellation.  There are a lot of issues that are raised by your
questions.  I would suggest looking at the ANSI technical report.