Forums

Eco Canceling

Started by bruno-pb December 18, 2008
 Hi people.
 I am trying to implement an eco canceling system using a DSP (ADSP -
21160M) and a LMS adaptive algorithm, but unfortunity it runs just at max
with 6 taps, beyond this value neither the sound unprocessed nor the
processed one can be heard. I think this is a problem releted with
processing rate of the the DSP, but I would Like to know if any body here
had a similar problem or have any possible solution.

 Best regards.


bruno-pb wrote:
> Hi people. > I am trying to implement an eco canceling system using a DSP (ADSP - > 21160M) and a LMS adaptive algorithm, but unfortunity it runs just at > max with 6 taps, beyond this value neither the sound unprocessed nor > the processed one can be heard. I think this is a problem releted with > processing rate of the the DSP, but I would Like to know if any body > here had a similar problem or have any possible solution.
I'm probably the least qualified to comment but since no one has: It seems to me that a 6 tap FIR filter would be an easy load for just about any processor. So that can't be the problem if the software is OK - which I highly suspect it may not be. Adding the LMS load is likely important but I'm not sure how much. At least a factor of 3 or 4 times I should think. Have you counted up the critical operations in the loop? How many adds and multiplies? How does that compare with the processor throughput capability? Then, add a factor of 2 or 3 for overhead and room and you should have a quick guess at whether it's the processor limitations or a bug in the code. Fred
On Dec 18, 6:02&#2013266080;pm, "bruno-pb" <marcell....@gmail.com> wrote:
> &#2013266080;Hi people. > &#2013266080;I am trying to implement an eco canceling system using a DSP (ADSP - > 21160M) and a LMS adaptive algorithm, but unfortunity it runs just at max > with 6 taps, beyond this value neither the sound unprocessed nor the > processed one can be heard. I think this is a problem releted with > processing rate of the the DSP, but I would Like to know if any body here > had a similar problem or have any possible solution. > > &#2013266080;Best regards.
Well since the LMS algo is about as simple as it gets, the problem is likely something else. An important detail which I don't see in your post, is what is your sample rate. How many updates per second do you need? Are you processing RF, video, audio? Clay
On Dec 18, 5:02&#2013266080;pm, "bruno-pb" <marcell....@gmail.com> wrote:
> &#2013266080;Hi people. > &#2013266080;I am trying to implement an eco canceling system using a DSP (ADSP - > 21160M) and a LMS adaptive algorithm, but unfortunity it runs just at max > with 6 taps, beyond this value neither the sound unprocessed nor the > processed one can be heard. I think this is a problem releted with > processing rate of the the DSP, but I would Like to know if any body here > had a similar problem or have any possible solution. > > &#2013266080;Best regards.
At a sample rate of 8KHz, I have implemented NLMS algorithms on the ADSP processor with 512 taps, embedded USB interface code, 4-channel A/ D, 2-channel DAC, 5-channel true RMS power meter, and "glue" code. The processor was less than 25% loaded. You have something else wrong, or you have an extremely fast sample rate. maurice Givens