[linux-audio-dev] Latency and feedback problems: soundcard with live microphone pass-thru, optimal solution ?

Dan Mills dmills at spamblock.demon.co.uk
Fri Oct 28 18:07:55 UTC 2005


On Friday 28 October 2005 12:26, Benno Senoner wrote:
> Hi all,
> I would like to route a microphone through a sound card and back to
> powerful amplified speakers.
>
> As we know in analog PA gear you have the microphone feedback problem
> (usually it comes in form
> of high pitched whistle sounds).
>
> But if I route a mic from into the soundcard and out to the speakers,
> there will be a small delay
> due to the audio card buffers. Even if it's only a few msecs it makes
> the problem much worse
> than in the case of analogue gear because the feedback sound will come
> in chunks that's one audio card buffer at time.
> For example if I use 64 samples per buffer which gives me acceptable
> latency for a live singer, the feedback noise
> could possibly generate a much lower pitched signal/interference (I
> assume something like 44100/64 Hz) which is I think
> not easy to filter out compared to the high pitched feedback (is in the
> latter case sufficient to cut some high frequencies using an EQ ?).
>
> How can one solve the feedback problem in case of mic to speaker delays
> of let's say 5-10msecs ?
> Is an echo canceller algorithm needed ? If yes does this compromise the
> quality of the input signal (the singer).

Actually applying a little delay is fairly standard in conventional PA as 
well. 
Consider that 10ms is roughly 10 feet of acoustic delay, and that you usually 
want the direct sound to arrive at the audience a little (10 - 30ms) before 
the sound from the PA as this really helps localisation.
Thus even on a small stage you may be delaying the FOH by 40ms or so to help 
localise the singer. 

Monitors are a little different obviously, but even there, it just is not a 
problem at these levels. 

 As for feedback, you notch it out with a graphic (or better a parametric), 
the same as always. 

In fact it turns out that a constant delay makes the feedback slower to build, 
and may alter the exact frequency slightly (by adding a constant phase shift 
at any given frequency), but does not radically change the overall system 
response in the way you think. 
 
Just about every modern PA system has at least one stage of AD-DA conversion 
(Think things like the digital crossovers so common in modern systems) which 
probably introduces 5ms or so right there. 

Regards, Dan.



More information about the Linux-audio-dev mailing list