Each Acoustic Echo Cancellation (AEC) block has an associated AEC Reference block. The purpose of AEC is to process the microphone audio so that audio coming from the loudspeaker is removed from the microphone signal. To do this, the AEC needs to know what audio signal is being sent to the loudspeakers. This signal is provided to the AEC Algorithm through the AEC Reference input. The AEC Algorithm uses the AEC Reference signal to generate a model of the Room Impulse Response, which is used to program the AEC Adaptive Filter to remove the loudspeaker audio from the microphone signal.
It is important that any changes made to the signal being sent to the loudspeakers are also reflected in the AEC Reference signal. This is especially true of dynamic changes. If a dynamic processing object, such as a compressor or limiter, makes changes to the signal being sent to the loudspeaker, or if a volume control changes the loudspeaker volume without also changing the AEC Reference, this difference between the loudspeaker signal and the AEC Reference will cause the AEC Algorithm to need to adjust the Room Impulse Response model and reprogram the AEC Adaptive Filter. This is commonly referred to as a loss of AEC convergence. If the far end talker is speaking while these changes are happening, the far end talker will hear an echo of their own voice until the AEC Algorithm has had time to re-converge and update the AEC Adaptive Filter to match the new conditions. If the gain state is being constantly modified by a compressor, but the AEC reference is not being modified in the same way, it becomes impossible for the AEC algorithm to ever catch up with the changes. Providing the AEC Reference with processing that exactly mirrors the loudspeaker processing is critical in these cases.