Maximum number of taps

Jeiko,

Unfortunately, it cannot.

The AEC block on SigmaDSP is intended to be a low cost AEC solution for speech. As such, the length of the adaptive filter is limited primarily by resources. However, it’s not much of a limitation in practice for most speech applications. The FIR represents an approximation of the impulse response from the speaker to the near end mic. The required length of the filter is primarily dictated by the reverberation time of the room and by ringing at low frequencies.

Since we’re processing speech, frequencies below approximately 150 Hz are not relevant. The AEC runs at 16 kHz so 1000 samples would correspond to an IR length of 62.5 ms and a low frequency cutoff of 16 Hz. Therefore, low frequencies are not really a problem. However, if your reverberation tail is such that you have strong reflections after about ~50 ms, then this is, admittedly, not a suitable solution. It does not work well for large reverberant spaces.

We do have much better (ITU P.11xx compliant), wider bandwidth algorithms that are more appropriate for larger rooms, but they require considerably more resources and run only on our SHARC family of DSPs. That AEC is part of our LISTN suite, which also includes much better noise reduction, excellent proprietary beamforming, etc.

Ken