[IDEA] Implement a four microphone-array (4-microphones) on the PCB similar to "Satellite1 Dev Kit" hardware design by FutureProofHomes?

[Hedda]

Suggest that you consider implementing a four microphone-array (4-microphones) on the PCB similar to "Satellite1 Dev Kit" hardware design by FutureProofHomes if is enough room on the PCB with proper distance to get a better result with four instead of two mics:

• https://futureproofhomes.net/products/satellite1-pcb-dev-kit

  • https://github.com/FutureProofHomes/Satellite1-Hardware

    • https://github.com/FutureProofHomes/Satellite1-XMOS

FutureProofHomes Satellite1 also makes uses of a variant of the XMOS XU316 similar to f the original Home Assistant Voice Preview Edition and designs that are is inspired by it like the home-mini-v1-drop-in-pcb and nest-mini-drop-in-pcb projects, but unlike those PCB designs it implements a four microphone-array (4-microphones) which can give it a leg up and some benefits in, epecially with acousic beamforming / microphone array beamforming technology which allow better extraction of desired speech signals in an interference-dominant and noisy environments, (even if not enabled in XMOS firmware today it could hopefully make use of such technologies in the future becuase it has four microphone-array).

• https://www.mathworks.com/help/phased/ug/acoustic-beamforming-using-a-microphone-array.html

[rauli2780]

I totally agree with the move toward future-proofing the hardware. Implementing a four-microphone array—similar to the Satellite1 design—makes a lot of sense for a more well-rounded and accurate range.
Even if the current XMOS firmware doesn't fully utilize it yet, having that physical overhead for acoustic beamforming would give this a massive leg up in noisy environments. If we can find the real estate on the PCB to maintain proper distance between the mics, I’m all for it. Let’s see if we can squeeze them in!

[iMike78]

I think it’s important to clarify that a 4-microphone array is not simply a matter of “adding two more microphones”.

For beamforming and spatial audio processing to actually work well, microphone placement, spacing, orientation, enclosure acoustics, reflections, and DSP tuning all matter. The physical geometry of the array is critical. A poorly implemented 4-mic setup can easily perform worse than a well-designed 2-mic setup.

The original Nest Audio itself only uses a 2-microphone configuration, and the microphone daughterboard inside the device is relatively small, which severely limits the available spacing and geometry options for a proper beamforming array.

On larger devices like Satellite1, the enclosure and PCB were designed around a multi-mic array from the beginning. Retrofitting that concept into an existing Google speaker enclosure is a very different engineering problem.

Also, the current XMOS/Home Assistant Voice firmware stack does not magically gain advanced beamforming capabilities just because four microphones are present. The DSP pipeline must explicitly support and tune those algorithms for the exact microphone geometry being used.

Personally, I think the real value here is making the device fully local and reusable in the first place. Reusing the original PA/power board and possibly the LED board already removes a huge amount of reverse-engineering effort and complexity. That alone is a major achievement.

Going from 2 microphones to 4 in this kind of constrained enclosure adds a lot of complexity for relatively questionable real-world benefit, especially compared to the engineering effort required to do it properly.

[josb25]

I totally agree, adding more mics than in the original Design is probably not worth it. But the Nest Audio actually has 3 mics (two on the UI board and one on the MIC board). The drop in PCB should probably route all three to the Xmos even If the software stack only ulltilizes two of them initially.

[Hedda]

I totally agree, adding more mics than in the original Design is probably not worth it. But the Nest Audio actually has 3 mics (two on the mic board and one on the mute board). The drop in PCB should probably route all three to the Xmos even If the software stack only ulltilizes two of them initially.

Also remember and keep in mind that if plan to use the exact same XMOS firmware configuration as the FutureProofHomes Satellite1
or the original Home Assistant Voice Preview Edition then you need to also really need to use the exact physical inter-mic spacing distance measurement between the microphones or else it will skews the algoritm it runs which could distort the Acoustic Echo Cancellation, Noise Suppression or Interference Cancellation that it does.

• https://github.com/FutureProofHomes/Satellite1-XMOS

• https://github.com/esphome/voice-kit-xmos-firmware

  • https://github.com/esphome/xmos_fwk_io

    • https://github.com/esphome/xmos_fwk_rtos

That is, if you change the physical inter-mic spacing distance measurement between the microphones that are used then you should really also change the configuration for the XMOS firmware and built it with that new configuration, which it technically written as a specific array geometry in the firmware source code. The DSP algorithms rely on this physical spacing to perform spatial filtering (beamforming) and direction of arrival (DoA) estimation.

I believe that the original Home Assistant Voice Preview Edition uses a 71 mm inter-mic spacing,

Other examples is the 2-mic XMOS XK-VOICE-L71 Evaluation Kit which also uses 71 mm inter-mic spacing according to its official documentation, so a good guess is that Nabu Casa modeled the original Home Assistant Voice Preview Edition after the L71 layout:

• https://www.xmos.com/xk-voice-l71

Yet I read nest-mini-drop-in-pcb project by @iMike78 was using 68 mm inter-mic spacing but use Home Assistant Voice PE firmware, however since it only has two microphones a sub-centimeter differences in mic spacing will probably not matter as much as when using more microphones with advanced beamforming (which require three or more microphones).

• https://github.com/iMike78/nest-mini-drop-in-pcb

When using an XMOS DSP processors with input for more microphones standard firmware releases from XMOS themselves include:

• Square/Circular Array: Typically 4 microphones with a 66 mm spacing.

• Linear Array: Typically 4 microphones spaced 33 mm apart (with all the microphones in a single row).

General Limits: While other spacings up to 100 mm are technically possible, they require firmware builds with customized configurations.