AEC & Q-SYS Conferencing System (Part 3)

Welcome back.

Now that we’ve covered the more generalized description of what AEC is and what it does,

let’s look at the specific application in the Q-SYS environment.

The most basic signal flow for systems without local reinforcement have two distinct signal paths.

The path to the far end includes the microphone input block, the AEC processing,

an automix of the conferencing mics and finally the transmit block to the far end.

In this example it’s a softphone transmit block, but this could be a USB transmit

or a line level output fed to an outboard codec.

In some systems it may be a combination of or all of these.

Remember the best practice is to have the ‘last mic on’ setting enabled

in the automixer so that some room noise is always fed to the far end.

This helps keep the far end participants from feeling like they’ve been disconnected.

The path to the near end mixes the far end receive signal

or signals with any program sources such as those from a presenter’s computer.

After the mix, these may be fed to a line output and on to an amplifier and speakers,

or directly to a Q-SYS amplifier via Q-LAN.

Of course in many cases we’d like to process the incoming microphone signals

to maximize intelligibility and make them sound as good as possible.

As a best practice, we’d like to apply that processing AFTER the AEC block in the signal chain.

We’d like the AEC to operate on the signal before any processing,

especially any processing that is non-linear in nature such as compression.

This makes it much harder for the AEC to converge.

Another consideration here is the privacy muting of the conferencing mics.

When possible, this should be done in processing AFTER the AEC processing block.

When the far end is talking the AEC block will still hear the reference

and the resultant echo even when the mics are muted back to them.

This keeps the block converged, avoiding echo when the mic or mics are unmuted.

There’s one case where it is acceptable to apply some processing ahead of the AEC block.

When multi-microphone beamforming arrays are being used such as the SHURE MXA910,

it's desirable to follow the published guidelines for EQ ahead of the AEC to remove excess low frequency energy.

This can often be done in the microphone processing itself,

but can also be done in the Q-SYS environment as shown.

Even in this case we NEVER place microphone dynamic objects, linear or non-linear before the AEC block.

As outlined in the Quantum level one topic on gain structure and signal flow,

all microphones should be calibrated to a -20dBFS nominal input level.

From there the system should be able to operate at unity gain structure through the signal flow.

If user level control is required, it’s best to place

an overall transmit level control just ahead of the conferencing transmit block,

as giving level control of individual conferencing microphones can cause inbalances in the system.

You’ll want to also make sure you set reasonable upper and lower limits for this level control such that

the user can’t send a clipped or distorted signal to the far end.

Those things considered, the most important factor in having a functional conferencing system

with AEC is composing a proper AEC reference.

To recap here, we’re looking to tell the algorithm what signal it is we’re trying

to remove from the microphone being processed.

In cases with no local reinforcement, the reference signal should be:

All far end signals, as it’s obvious that's our goal, to keep these signals from going back to the far ends.

And any program feeds.

If we’re not sending program sources to the far end directly,

we don’t want them to hear it passing through the microphones.

If we are, we don’t want them to hear direct program feeds

and a delayed version that comes back through the conferencing mics.

In short, as long as there’s no voicelift in the room,

the AEC reference signal should simply be the signal that’s going to the room speakers.

Notice: we NEVER put a microphone signal into its own reference either directly or indirectly.

This results in a pumping, underwater like sound that reaches the far end.

When placing user controls in the system,

it often seems simple to place a single gain control directly before the output block.

Imagine a user in a call.

Let’s say the softphone receive signal is quiet, so the user increases the gain.

Then a program source starts and it’s incredibly loud because the single output gain control is set so high.

So, it’s best practice to place user gain controls with reasonable limits ahead of the mixer so that the user

is controlling the level of ONLY the source that needs to be changed.

In addition, this is great for AEC systems.

The red tag in this diagram represents the reference signal.

If I increase the level of the softphone receive in the room,

I also increase the level to the AEC reference by the same amount.

Placing controls in this manner reduces the amount of reconvergence required

by the AEC algorithm as compared to changing only the level in the room.

Alright, let's take a quick break right here and come back whenever you're ready.