AEC Introduction

As stated earlier, in addition to the desired near-end speech, the local microphone picks up a portion of far-end speech from the near-end loudspeaker. AEC reduces echo by subtracting this unwanted portion of the microphone signal before sending it to the far-end. We refer to this additional active echo reduction as Echo Return Loss Enhancement (ERLE). Total Echo Reduction (TER) is the combined echo reduction achieved with ERL and ERLE. The remaining audio consists of echo-free speech and some miscellaneous acoustic sounds, such as room noise. For this reason, AEC often includes ambient noise reduction to further improve the signal sent to the far-end. The following diagram shows our original two room conference application with the addition of AEC. Note that to remove echo from both rooms, each room requires AEC processing.

note: When echo is heard, it is the far-end audio system that is the culprit and needs work, not the local room.

How does AEC accomplish this task? First, an AEC algorithm models the room. It does this by comparing the reference signal (typically the same signal sent to the near-end loudspeaker) to what the microphone picks up. The model includes everything in the signal path between the AEC Reference and the AEC Local Mic In. Once the AEC algorithm has modeled a room, it is able to look at the AEC Reference, estimate what will appear at the AEC Local Mic Input and subtract the estimated signal from what the microphone picks up.

note: An AEC algorithm’s ability to remove echo is dependent on the accuracy of its room model.

Because the ability to remove echo is dependent on the accuracy of the room model, it is important to avoid unnecessary changes during operation that vary the relationship between AEC Reference and AEC Local Mic In. We’ll talk more about this later when we discuss where to locate signal processing in a system and the necessity of applying equal processing to room and reference signals. (See Basic Halogen Conferencing with Voice Lift)

One misconception is that AEC simply subtracts the reference from what is received from a microphone input. This is not true. AEC only subtracts the portion of the reference signal that the room model estimates will be picked up by the microphone.