*This article is a Japanese translation by Macnica of a blog written by an engineer at DSP Concepts.

Full-duplex communication allows people on both ends of the line to talk and hear each other without interfering with each other. Whether you're using an Ethernet connection or an old analog phone line, the technology for full-duplex communication is well established and has been proven for decades - even before you hooked up a speakerphone.

With a standard smartphone, interference isn't an issue with full-duplex communication. Your voice goes into the phone's microphone, through the carrier, and then directly out the phone's speaker to the other person's ear. The phone's microphone doesn't pick up much of the sound from the speaker, so there's no need to worry about interference. The other person's voice doesn't interfere with your voice, and vice versa.

In other devices used for conferencing, such as speakerphones or video bars, conference room displays, and smart speakers with video screens, the microphones can "listen" to the speaker, so not only do they pick up your voice, they also pick up the voice of the other party on the line (called the "POOL") coming in from the speaker. Conferencing devices can solve the problem to some extent by blending a slightly delayed, phase-inverted version of the other party's voice with the microphone signal, canceling the audio coming directly from the speaker to the microphone.

But what about the sound that leaves the speaker, bounces around the room, and enters the microphone? This sound is an echo that has a delay of anywhere from a few ms to 100 ms or more, depending on the size of the room, making it much harder to cancel. Of course, the more reverberant the space, the more likely the sound will bounce back to the microphone multiple times. Also, by the time the sound returns to the microphone, its spectral content has been altered by the acoustics of the space, making it harder to cancel.

Without an effective way to process the signal from the microphone to remove these echoes, the other person's voice will mix with your own in unpredictable ways, making the sound cluttered, unclear and often unintelligible.

The solution is to cancel these echoes. But it's not easy. A digital signal processor must distinguish between what's coming into the microphone (your voice) and what you don't want (an echo of POOL's voice). To do this, it analyzes the nature of the original signal from the other end of the call, then analyzes the sound coming into the microphone to determine if it's your voice or an echo of POOL's voice.

The problem becomes even more difficult when you consider that high-quality conferencing devices typically use an array of three or four microphones to "steer" the pickup pattern: echoes of the POOL's voice arrive at these microphones at slightly different times and amplitudes, so the digital signal processor must analyze and correct for four inputs, not just one.

If you are interested in the issues that can occur with full-duplex audio conferencing, there is a video posted by DSP Concepts below.