I have been looking into digital ways to reduce or eliminate feedback in a reinforced speech system (PA). In my findings, it seemed AFC was a little different than AEC. Is that actually the case? If so how could the AEC module be used to perform AFC? Whether is it adding an additional "feedback" path or making AEC a subsystem of a larger configuration.
thanks
mike
3:33pm
Hi Mike,
Audio Weaver's AEC module should eliminate feedback just fine. Conceptually, you give the AEC two audio feeds: the microphone feed and then a "speaker reference" - which is what to expect (and eliminate) from the mic signal. The AEC scans the microphone feed and eliminates that reference signal if it finds it - which means you're not adding layer upon layer of the same sound.
Note: for best performance, you want your speaker to have as little distortion as possible - which is to say, you want to the reference signal to closely match how the speakers actually sound to the mics. e.g. If you tell the AEC to look for some nice, clean, full-range music, but then play back through a crappy little speaker that distorts and only plays down to 600hz... the AEC can't do a great job.
So, to get the most performance, you need to process your output such that you're only asking your speaker to reproduce audio it can handle - i.e. that's it's not too loud (use an AGC limiter) and that's been properly filtered (e.g. don't ask a tweeter to play bass).
Hope that helps?
Matt
11:22am
Hi Matt,
I believe I understand it more now. Basically, if the intended signal going to the speaker is found on the mic input the AEC module will remove it. Will different aspects like gain/volume effect the performance. I wouldn't expect the mic to pick up at the same level as the output. I am asking because it seems important where you pick for the intended speaker output. For example, audio weaver sends DAC output to a codec which allows for gain and volume adjusts. Audio weaver would only know about the information before the DAC. is that a problem?
11:45am
The AEC will smartly handle any gain differences - in fact, that's actually it's specialty: The AEC doesn't just find and eliminate one copy of the speaker in the recording. The 'E' in AEC is because it has to find and cancel multiple instances of the speaker signal - each with a different delay, attenuation, and frequency response. The direct path Speaker->Mic might be loud and highly coherent. The first echo off a nearby hard surface might nearly as loud and also fairly coherent. The second echo off a distant surface will be delayed, quieter, and have a different frequency signature. The AEC deals with all that.
2:13pm
awesome, thanks for the explanation.
11:04am
I'm working on a similar problem and unfortunately there's a lot of confusion but what exactly is meant by feedback in the literature. So far we've been able to use the provided AEC module to eliminate a lot of the effects from the acoustic coupling that facilitates the feedback between our speaker and mic. I'm not sure where you are in your research but I found this paper helped clarify a lot of things for me: ftp://ftp.esat.kuleuven.be/pub/SISTA/vanwaterschoot/abstracts/08-13.html
11:25am
Hi Graham, thanks for the information. I will definitely give it a read. Hopefully, it will help clarify things for me. Thanks again.