DSP building blocks

A digital signal processor (DSP) is the brain of a room's audio. Microphones, codecs, PCs and media players all feed it; it routes, cleans up and balances everything, then feeds the loudspeakers and the far end of the call. Whatever the brand, DSPs are built from the same handful of blocks. Learn what each one does and you can read almost any DSP design.

Equalisation

EQ raises or cuts chosen frequency ranges. In installed AV it is mostly corrective — removing problems rather than adding colour.

Dynamics

Dynamics processors change gain automatically depending on the signal's level. Each is defined by a threshold — the level at which it starts to act — and how strongly it acts beyond it.

• Compressor — above the threshold, output rises more slowly than input. At 3:1, a signal 9 dB over the threshold comes out only 3 dB over. Make-up gain then lifts everything back up, so quiet and loud talkers end up closer together.
• Limiter — a compressor with a very high ratio: a ceiling the signal can't pass. Used to protect loudspeakers, amplifiers and the far end from sudden peaks.
• Gate / expander — the opposite: below the threshold, the signal is turned down. It keeps unused mics quiet so their room noise doesn't add up.
• Automatic gain control (AGC) — a slow-acting leveller that nudges a channel toward a target level over seconds, useful for talkers who sit at varying distances.

Automatic mixing

Every open microphone adds room noise and reverberation, and makes feedback more likely. An automixer manages a group of mics so that only the ones being talked into are fully open.

• Gated automixers open and close each channel against a threshold, much like a gate.
• Gain-sharing automixers distribute a fixed total gain among the mics in proportion to how much each is being used. The overall level stays constant however many people talk, with no audible switching.

Gain-sharing is the approach associated with Dan Dugan, who developed it, and is widely used in conferencing DSPs today.

Acoustic echo cancellation

In a call, the far end's voice comes out of your loudspeakers, bounces around your room and goes back into your microphones. Without help, the far end hears themselves a moment later — echo.

An acoustic echo canceller (AEC) fixes this. It is given a reference: a copy of the far-end audio being sent to the loudspeakers. It continually learns how that signal is changed by the room on its way to the microphone, and subtracts its estimate from the mic signal before it is sent back.

Many DSPs add noise reduction after the AEC to suppress steady background noise such as HVAC, and a non-linear processor that removes any residual echo the canceller couldn't.

Putting it together

A typical conference-room mic channel, though orders vary by manufacturer and design:

On the output side, loudspeaker feeds usually get their own EQ, a limiter for protection and, in larger rooms, delay to time-align speakers at different distances.