Well.. the project wasn’t intended to be an acoustic recording device for biologists. It was intended to be a microphone addition to a security project I have that would dynamically merge audio with video. Something you can’t do with microcontrollers.
But along the way, I added a GPS to keep the time accurate and then with the sub-microsecond time accuracy I thought of sound localization. So I developed it further into a sound localizing recorder with in principle enough time accuracy to localize bats that are close even.
With the power of a pi and sub-microsecond clock timing, coupled with a linkage into jackd that can support multiple consumers, you can both record and still have enough CPU to on principle perform real time bird recognition and localization. That is not going to be possible on a microcontroller.
With this project I have been able to localize explosions to less than 20m accuracy from microphones located more than 3km away. Again not something possible on microcontrollers.
The setup of the author is at his own home. Which means that in principle he’s not constrained by power. With that’s setup by adding icecast2 and darkice. Whilst still recording and not needing a big host host computer. Likely even better with go2rtc. Next on my list to try.
Why do you need to merge these items at the point of collection? Why not just stream the sensor data off with temporal correlation data and assemble them elsewhere?
Either way, galvanically isolating 5W isn't too terribly hard either.
Streaming it away and then merge is exactly how it's intended to with jackd streams. You create a dummy slave off the first one, then you have clients off that.
The point being that if I had take a potentially easy way of just recording directly from the device to write the files, then I would not easily at the same time be able to perform real time analysis of the stream. So the way I've done it you can indeed do this.
Note, with manual examination of the sound files with raven lite, I'm able to reliably determine the start time of a sound to less than a millisecond of error (Jitter in the pipeline, otherwise it would be a lot more accurate).
Making it entirely possibly to locate the position of bat calls in real time to quite a good degree of accuracy if you wanted.
The chip you point to suggests a much simpler approach. Is it really that simple? There are a great many more bits on the board I point to. Why would they take a hard way if one simple chip can do it ?
The board you linked isn't a galvanic isolator, it's just a capacitor bank covering a bunch of different frequency ranges.
The ADUM isolator I linked as well as flyback converters use transformers to provide a high-voltage isolation barrier between the source and device being supplied.
Transformers are generally larger, more expensive, and less efficient than non-isolated DC circuits.
The POE supply itself is isolated (with a transformer) and doesn't need to be noisy necessarily - but if you want even more quiet, another level of isolation can't hurt.
But along the way, I added a GPS to keep the time accurate and then with the sub-microsecond time accuracy I thought of sound localization. So I developed it further into a sound localizing recorder with in principle enough time accuracy to localize bats that are close even.
With the power of a pi and sub-microsecond clock timing, coupled with a linkage into jackd that can support multiple consumers, you can both record and still have enough CPU to on principle perform real time bird recognition and localization. That is not going to be possible on a microcontroller.
With this project I have been able to localize explosions to less than 20m accuracy from microphones located more than 3km away. Again not something possible on microcontrollers.
The setup of the author is at his own home. Which means that in principle he’s not constrained by power. With that’s setup by adding icecast2 and darkice. Whilst still recording and not needing a big host host computer. Likely even better with go2rtc. Next on my list to try.