Principle and application scenario of ECHO cancellation

Abstract: Today, with the development of chip technology and high-bandwidth communications, developers are able to make a fuss about a more superior voice-calling experience, while the demand for hands-free calls is getting higher on the market. However, hands-free communication systems often need to be able to cope with various disturbances. One of the interferences that cannot be overlooked is acoustic echo in the Speaker Echo microphone field (loudspeaker Echomicphone System). This paper explores the causes of acoustic echo (acoustic echo) during a call and its basic acoustic echo cancellation (acoustic echo cancellation) application scenario.

Reasons for acoustic echo generation:

The structure of a VoIP call is shown below:

When the remote far-end has a speaker, the sound travels to the near-end (near-end) speaker, and the sound goes back to the far side through the space delay and transmission delay. This creates an acoustic echo. Near-to-far-end processes also have the same acoustic echo, which only describes the near-end processes.

In order to suppress acoustic echo, it is first possible to make a fuss from the spatial propagation path, including the use of a small or directional speaker by the proximal user using the headset and the sound field range, physically blocking the Echo's propagation recording. Another method is to suppress the echo by using an AEC module or an AES module in the near-end mic input signal. If you want to use a call scene in Speakerphone, the second way is necessary. Of course, I've also encountered some headphones recently. The more sensitive situations also require AEC processing.

The basic principle of the Echo cancellation module is to use the far-end signal to make the corresponding estimate, at the near end of the time to find this part of the signal and then to beautify the photo in the same way to pull it off-this is I can think of the most straightforward explanation. In addition, the effect of echo cancellation will be affected by noise and near-end speech, so that noise near-end voice as a source of pollution, interfering with the "cutout" effect.

Here's a look at the above image:

Subsequent analysis of the AEC algorithm if no special instructions are used, the symbol defined by the drawing is applied. Main symbolic meanings:

: Remote signal;

: Remote signal;

: Noise signal;

: Physical echo signal;

: Time points;

: Near-end signal;

: Error signal;

: physical echo channel;

: Estimating the Echo channel.

In general modeling, when the physical Echo channel is assumed to be linear, the echo-generated formula is:

, (1)

The near-end signal is:

。 (2)

Then the purpose of the AEC is 1) using an estimated echo channel to simulate the physical channel and then obtain the corresponding Echo:

。 (3)

2) Subtract the echo from the near-end signal to obtain an echo-free error signal:

(4).

This article does not do the actual procedure and the algorithm analysis, only introduces the AEC usage scene and the principle. Future articles will analyze the AEC algorithm and its performance in depth