AEC & Q-SYS Conferencing System (Part 1)

Acoustical Echo Cancelling, or AEC for short, is applied mostly in Q-SYS conferencing applications.

In this application meeting participants in at least two rooms need to be able to communicate with each other

and participants in other rooms.

Each room typically has a local audio/video conferencing system and the systems in each room

are connected by standard telecommunications infrastructure.

This could be simple POTS or VoIP telephony, dedicated codec (using the internet for transmission)

or soft codec applications such as ZOOM, MS Teams or BlueJeans.

The participants in a given room view themselves as the ‘near end’ in conferencing terminology.

They are transmitting their voices to the remote participants

and receiving from them via the linking technology being used.

This all starts of course with the microphones, which may be placed on the table as in this example,

or elsewhere such as the ceiling.

The microphones will be attached to the conferencing system,

which in this case would ultimately be a Q-SYS core.

Conferencing microphones are typically digitally processed to optimize clarity and sound quality,

then automixed to send an audio signal of constant level to the receiving rooms.

After the mic signals are properly processed and mixed,

the ‘transmit’ signals are then sent to the linking technology, which may be a dedicated video codec,

a telephone line or USB interface to a soft codec.

The room or rooms that receive the near end signals are called ‘far ends’.

The near end signal comes from the connecting infrastructure

where it is mixed with any required local sources, such as local presentation audio, etc.

The signals are digitally processed for audio quality and clarity,

then fed to an amplifier and ultimately loudspeakers which are typically overhead.

Here’s where we see the fundamental problem with audio conferencing systems.

The receive signal is broadcast into the room through the speakers.

There are microphones on the table or ceiling that pick up these signals.

The microphones send what they hear out to the remote rooms, which means that the talkers in the remote room

will hear themselves with a long delay as they try to speak.

If you’ve ever experienced this condition, which most of us have,

it’s almost impossible to speak and have productive communication when this happens.

To make things even more complicated, the microphones don’t hear the far end signal only once,

but potentially a number of times based on the reflectivity of the room.

There’s a ‘direct’ signal from the speaker to microphone, but…

There are also signals reflected off the surfaces in the room and back into the microphone.

These are known as ‘Acoustic Echo’.

Based on the reflectivity of the room, we end up sending

many time-shifted signals of varying levels to the microphones.

Removing these signals becomes a very difficult problem to solve.

This hard work is done by the ‘Acoustical Echo Cancelling’ algorithm, or AEC block.

We send a reference signal to the AEC algorithm, which tells it what to look for and remove.

We can conceptualize what’s happening here by looking at each part of the echo path as a separate signal.

Let’s look at a very simple case with a direct signal and two reflections.

Each reflection will naturally have a little more time shift and decay in level.

Now, real voice signals would be more complex than shown in this diagram,

but we’ll use these to keep it simple.

The AEC algorithm is a very advanced set of adaptive filters.

We introduce the signal we’d like the algorithm to remove as the ‘reference’ signal.

When audio is present at the input of the algorithm,

the adaptive filter begins to identify and try to remove the unwanted signals.

The reference signal reaches the algorithm first,

so the block always knows what to remove before the signal will be seen at the microphone.

The algorithm works to calculate what’s known as the ‘impulse response’ of the room to decide

the relative time shift and amplitude of each reflection.

The algorithm then works to scale and shift the reference signal

to be exactly out of phase with the incoming mic signal.

As we know, adding two audio signals that are completely

out of phase results in a cancellation of the signals in question.

The same operation is carried out for the first reflection

..and finally for the second…

In a perfect world with simple signals the cancellation would be perfect as seen in the diagrams.

As we know, the real world is much more complex. Perfect cancellation is rarely attainable.

This and other considerations add to the complexity of the AEC algorithm.

If we look at the algorithm in detail, we see all the parts

that make the right functionality for the conferencing application:

The incoming conferencing signal goes to the AEC reference,

is processed and ultimately goes into the room.

Notice that the adaptive filter set includes steady-state noise reduction as well as the AEC functionality.

The algorithm makes the necessary calculations and sets up the adaptive filter.

The filter then operates on the mic signal.

As an adaptive system, the algorithm uses the error signal, the amount of signal that’s not cancelled,

to continue to fine-tune the filter settings.

The nonlinear processing then suppresses the error signal that gets back to the caller’s ear.

Note the algorithm uses a double-talk detector to know if people at both ends of the call are talking at once.

The adaptive filter holds in the double-talk condition to make sure it doesn’t attempt

to remove any signal from local talkers that needs to reach the far end.

As you can see, this is one of the most advanced uses of audio DSP out there.

There are other features in the Q-SYS algorithm that add even more complexity…we’ll discuss those in detail later.

For now let's take a quick break, and continue along whenever you're ready.