[PD] feedback control

[sven]

At 13:48 21.03.2005, matthew jones wrote:
>>if so multiply that bands which an adjustable value <1
>
>but wouldn't that have weird effects when the signal you are attempting to
>remove the feedback from is the sound of you whistling??
>Or is this method the one that many feedback supressors use?  I thought it
>was something like mimicking the effect of constantly moving the mic around
>so that the frequencies at which the system is unstable actually move
>around, done via slow variable delays or something like that.  Thats just a
>guess though.

well, ofcourse a feedback suppressor would effect the sound. just as any 
noise/click or
whatever suppressor also filters some of the wanted sound.
the fft idea was just a very simple solution. a better one would be to find the
freq at what the feedback happens and then filter w/ a very narrow (high q) 
notchfilter (FIR)
let's say feedback freq is 2khz, you filter 2khz -/+ 10Hz (hanningwindowed 
or similar)
so you filter a band which is a most 20Hz wide, which is just 1/50 (down) 
or 1/100 (up) of an
octave. the process is bearly hearable.
and ofcourse you don't need to cut out the band completely.
the suppressing factor just had to be a little lower than the speed with 
which the feedback builds up.
but before thinking about feedback suppression you should always try to setup
your equipment that feedbacks can't happen.



>You could use an adaptive line enhancer, if it was feasible to regularly
>calculate the inverse of a relatively large matrix and apply long FIR
>filters... tho on a laptop this is maybe a bit too hopeful.

sounds too computationally expensive. filtering w/ a simple notch should
be enough. what's important is just to find the frequency of the feedback
as exactly as possible.

>The true way would be to keep the mic fixed in place and measure the
>response from loudspeakers to the mic, then calculate the optimal response
>to cancel the signal as it arrives at the microphone with a second source
>(maybe the right side of a stereo pair) and use an adaptive filter..... but

if you keep the mic fixed in place and the speakers are not moving it should
be easy to set it all up that feedback won't happen anyway.


>as Thomas says it would require near zero latency and heck, a lot of cpu
>power!

the latency is the biggest problem imho. just did a quick test with testtones
and guessing the right frequency with some accurary from the fft needs
a window of at least 512 samples which is ~ 11ms @ 44.1kHz/s.


>Any ideas/improvements anybody?
>
>Matt