Acoustic Echo Cancellation
Echo cancellation is accomplished by monitoring an incoming channel (in the case of AG, we monitor the RX from Codec) and the loudspeaker channel (or another output). If a signal exists in the microphone channel that also exists in the loudspeaker channel, the device eliminates this signal from the microphone channel. Echo cancellation settings for each microphone can be accessed from the "AEC" button on each channel in the "Input 1-8" window.
The Non-linear Processing setting determines the intensity of the cancellation. Experience has shown that the default(Soft) setting works well in most environments.
The PA Adapt and AEC Reference can be set to any output or virtual output. This should be set to the loudspeaker output for most configurations.
To cancel echo, we want to monitor the microphone signal for audio coming from an input source. If audio from the input source is present in the microphone channel, the XAP800 creates a cancellation signal by copying the input source signal, shifting the phase 180° then adding the shifted signal to the microphone channel. This removes the input source signal from the microphone signal. The microphone signal is then sent to the matrix where it is routed to the appropriate output. Below is an illustration of what goes on inside the AEC box shown in the flowchart diagram above.
mic signal produced by local participants. | |||
Signal in mic channel due to bleeding of loudpeaker audio into the mics. |
| ||
Input Signal from remote site heard through loudspeakers. | XAP800 detects the input signal in the mic signal. | XAP800 copies the input signal and shifts the phase 180°. The phase shifted signal is then added to the mic signal. |
Input signal cancelled and mic signal produced by participants sent to output. |
Generated signal 180° out of phase from Input Signal from remote site. |
Getting Local Sound Reinforcement using a virtual reference
Normally, patching the microphones into the local room loudspeakers causes echo and/or feedback. But, using a Virtual Reference, you can send an attenuated signal to the loudspeakers giving your participants an idea of the audio they are sending to remote sites. This is useful when participants decide to move the microphone or crinkle their potato chip bag right next to a microphone. The participants will hear the noise and are more likely to stop before you have to interrupt the meeting and chastize them.
To set up a virtual reference, select Virtual Ref 1 from the PA Adapt and AEC Reference list. This will open the Virtual Reference configuration window:
Here we have selected "RX from Codec"(audio from the AG) as our Input Source to be cancelled. In order for the AEC to track level changes you must specify a hardware output for "Output Level Tracking". We could have selected the "TX to Codec" output, but then only the "RX to Codec" -> "TX to Codec" level changes would be tracked and other inputs would not be cancelled properly. "To Speakers" is selected so we can add other inputs to the virtual reference later on. This selection assures that the proper amplitude signal is applied to cancel the Input Source signal.
Click "Close". Now open the AEC dialog for each microphone and select "Virtual Ref 1" for the "PA Adapt and AEC Reference". This will open the "Virtual Reference" dialog each time, but, it is not necessary to make any changes.
Now that the virtual reference is selected for each mic, the mic processing channel can be routed to the speakers. You'll notice above that some attenuation is necessary to avoid feedback.
In the "Matrix" screen, right-click the cell that routes the mic processing channel to the loudspeakers. Select "Cross Point". This brings up a small dialog that lets you set the attenuation level for that particular route. The room for this setup needs -18db attenuation. The amount may vary depending on the room. You should adjust the attenuation level down until you no longer hear any feedback or ringing. Be sure to test with several consonants, pops and grumbles at a fairly loud level to assure there are no frequencies that will feedback. Also be aware that if participants are in various locations in the room, some frequencies may be excited necessitating additional attentuation. It is recommended that the node operator monitor this cross point closely as feedback can be introduced quite unexpectedly.
