Hi all, I'm working on acoustic echo cancellation, my question is; When I test an algorithm like LMS or NLMS should I change the scale of my wave signal? As pcm file are Q15 between 1 and -1, do I need to multiply my signal by 2^15 for test under matlab. Like input=2^15*wavread('input.wav'). My question is relative to the fact that signal level have an impact on algorithm performances. thank you in advance. Moctar
wav file
Started by ●July 3, 2009
Reply by ●July 3, 20092009-07-03
mmoctar wrote:> Hi all, > I'm working on acoustic echo cancellation, my question is; > When I test an algorithm like LMS or NLMS should I change the scale of my > wave signal? > As pcm file are Q15 between 1 and -1, do I need to multiply my signal by > 2^15 for test under matlab. > Like input=2^15*wavread('input.wav'). > > My question is relative to the fact that signal level have an impact on > algorithm performances. > thank you in advance. > Moctar
Reply by ●July 3, 20092009-07-03
> > >mmoctar wrote: > >> Hi all, >> I'm working on acoustic echo cancellation, my question is; >> When I test an algorithm like LMS or NLMS should I change the scale ofmy>> wave signal? >> As pcm file are Q15 between 1 and -1, do I need to multiply my signalby>> 2^15 for test under matlab. >> Like input=2^15*wavread('input.wav'). >> >> My question is relative to the fact that signal level have an impacton>> algorithm performances. >> thank you in advance. >> Moctar >
Reply by ●July 3, 20092009-07-03
Why should the signal level impact on the algorithm performance? On Jul 3, 9:35�am, "mmoctar" <mmoc...@gmail.com> wrote:> Hi all, > I'm working on acoustic echo cancellation, my question is; > When I test an algorithm like LMS or NLMS should I change the scale of my > wave signal? > As pcm file are Q15 between 1 and -1, do I need to multiply my signal by > 2^15 for test under matlab. > Like input=2^15*wavread('input.wav'). > > My question is relative to the fact that signal level have an impact on > algorithm performances. > thank you in advance. > Moctar
Reply by ●July 3, 20092009-07-03
> >Why should the signal level impact on the algorithm performance?The impact is in the choice of the step size you have to choose. When your signal is high level you must choose a small step size (for stability) and the smallest the step size is the slower your algo 'll be. And when the algo use a power estimation (transform or freq lms) the result is completely different.> >On Jul 3, 9:35=A0am, "mmoctar" <mmoc...@gmail.com> wrote: >> Hi all, >> I'm working on acoustic echo cancellation, my question is; >> When I test an algorithm like LMS or NLMS should I change the scale ofmy>> wave signal? >> As pcm file are Q15 between 1 and -1, do I need to multiply my signalby>> 2^15 for test under matlab. >> Like input=3D2^15*wavread('input.wav'). >> >> My question is relative to the fact that signal level have an impacton>> algorithm performances. >> thank you in advance. >> Moctar > >Moctar
Reply by ●July 3, 20092009-07-03
Suppose you have developed the floating point algorithm and it works well, then whatever Q format you use, just scale the step size proportionally, you should be fine. On Jul 3, 5:06�pm, "mmoctar" <mmoc...@gmail.com> wrote:> >Why should the signal level impact on the algorithm performance? > > The impact is in the choice of the step size you have to choose. When your > signal is high level you must choose a small step size (for stability) and > the smallest the step size is the slower your algo 'll be. And when the > algo use a power estimation (transform or freq lms) the result is > completely different. > > > > >On Jul 3, 9:35=A0am, "mmoctar" <mmoc...@gmail.com> wrote: > >> Hi all, > >> I'm working on acoustic echo cancellation, my question is; > >> When I test an algorithm like LMS or NLMS should I change the scale of > my > >> wave signal? > >> As pcm file are Q15 between 1 and -1, do I need to multiply my signal > by > >> 2^15 for test under matlab. > >> Like input=3D2^15*wavread('input.wav'). > > >> My question is relative to the fact that signal level have an impact > on > >> algorithm performances. > >> thank you in advance. > >> Moctar > > Moctar