As I mentioned in my previous post, harnessing the true power of Acoustic Echo Cancellation (AEC) lies in the hands of the integrator. Their knowledge and expertise directly affects the AEC performance, and ultimately the intelligibility of a conference session.
Let’s start with gain structure – you’ve got to understand it. Doing so will make life much easier for your AEC. Standard best practices are the rule and an AEC system should be gain staged similarly to any well performing sound system. We’ve made our Sona™ AEC more tolerant of improper gain staging, but doing it right will yield the best results.
One unique thing about an AEC system is the use of references. References are what get sampled – the far end signal coming back in to your local inputs – and having the proper reference signal is crucial to AEC performance. It must contain the far end mic feed coming into the local room speakers.
You also don’t want to pick the reference signal off in the wrong part of the signal chain, or else you’ll experience problems. The referencing should be done after any adjustments to level controls that affect the gain of the speaker. For instance, if someone is raising the volume on the local speakers, it’s best that the AEC knows that it’s happening. It’s building an adaptive filter and the amplitude needs to be tracked.
One thing you want to avoid is the local mic signal going to your reference because then it tries to cancel that signal. If you have this issue, it’s easy to identify from a troubleshooting standpoint because you can hear the algorithm canceling the mic audio. It will sound very choppy.
Always remember that AECs don’t like it if you have mics opening and closing. It’s best if the mics are on all the time so the AEC can model the room response. That doesn’t mean you can’t ever have them turn off and turn on. You can use PPT (push to talk) mics with the muting done in the DSP unit rather than prior. Automixers can also be used after the AEC inputs in the DSP. In both cases, the mics going to the AEC remain open all of the time even though they behave as if they are opening and closing as needed. Although we recommend not muting mics prior to the DSP, we have provided options within Sona AEC to handle that situation much better than earlier algorithms did.
A high quality AEC algorithm should also include non-linear processing (NLP) and noise reduction (NR). NLP removes residual echo that the adaptive filters miss. If you have a really challenging acoustic environment, NR can help the AEC work by identifying steady state noise – the hum from lighting for example – and actively reducing that noise.
Finally, recent advances in conferencing systems have led to the development and refinement of wideband AEC. Biamp provided wideband AEC in our previous algorithm for use with high performance video conferencing codecs, so it’s not that new, but it’s becoming increasingly important due to the growing demand for VoIP conferencing. Analog phone lines are bandwidth limited by nature, which is OK for private conversation with a handset or ear piece but not as good for conferencing systems. The loss of high frequency information makes it more difficult to hear some elements of speech, especially when broadcast over speakers. With VoIP conferencing you can have a much wider frequency range, which is important for comfort during a conferencing session because people don’t have to struggle to hear and understand. VoIP conferencing has driven and will continue to drive demand for wideband AEC.
You may also be interested to know that our Sona algorithm is powerful enough to handle a full football stadium.