Acoustic Echo Cancellation: 4kHz-peak, convergence time

Hello ADI-team,

I'm testing the SigmaStudio AEC-algorithm in one of our communication products (ADAU1452) with a speaker-microphone-combination where duplex communication should be achieved.

I'm using the wideband setting (8kHz) and basically it works.

But I have 2 questions:

1. In measurements with white noise there is a huge peak at 4k where no cancellation happens, which can be clearly heard in speaking tests too.

Here a picture of the measurement: (green - transfer function from loudspeaker to mic with AEC+RES inactive, yellow - AEC+RES active)

Can you tell me why and how to remove it? (I can send you my .dspproj-file if you tell me where to send.)

2. In our speaking test and measurments (laboratory conditions with almost no background noise) it takes up to 30-60s for the AEC/RES to converge or to reach a proper attenuation of the loudspeaker signal in the mic-path.

Is there a chance to speed this up?

Thanks in advance

Christian