Hmmm....there's potential. I don't know about in a daisy-chain configuration, but maybe it would work in a [source > distributed network > summed signal] configuration. I will confess to having no direct experience with what I'm about to descrie; it's only something I've read about.

Because of the random quality of Brownian Noise, a signal that gets split, fed to matching electronics, and subsequently summed can have an increase in signal to noise ratio (from a purely electronics standpoint that is...I'll touch on that in next paragraph). A pair of signals from the same source that are summed, after nominal gain in each set of electronics, achieves a 6dB increase in level...while the noise floor from the electronics only receives a 3dB increase in gain. That means you've gained 3dB in the signal-to-noise ratio (note, this is only "self noise"). This was a trick that was used occasionally in analog recording. To my knowledge, this will not work in the digital domain, because there is a finite number of variations that can be present in a digitally encoded system. This will affect exactly how Brownian in nature the system noise is. The higher your bit and sample rate, the farther you might be able to go. But there will be a point of diminishing returns, and that point will be far earlier than the point that is capable with an analog signal.

Any sound not generated by the electronics is, in this case, signal; whether it's something you wanted to capture or not. Noise, as I was using it in the previous paragraph, refers only to that which is inherent to the electrical ciruits. I think the possibility of recording those commonly "imperceptible" sounds would be reliant on three factors:

• Can the transducer actually react to those sounds?
• What is the signal-to-noise ratio in the recording environment?
• The number and quality of the components within the analog circuit.

If the transducer cannot react to those vibrations or [compressions + rarefactions], the sound cannot be recorded. If it can...then even if the sound is in the system's noise floor (remembering that many devices are still technically reacting to "signal," even if it is in the noise floor) then it is theoretically possible to get some information using that method mentioned above.

Of course, this would be a pointless exercise if any other sound in the environment is going to mask those commonly imperceptible sounds. There's a point of sound level separation where spectrum proximity is irrelevant to masking. If something is loud enough, it will mask the quieter sound (even if they occupy radically different locations within the sound spectrum). We won't be able to hear that target sound on reproduction...even though it is being recorded.

As to the number and quality of components...well...no real-world passive or active component operates at "ideal" potential. This would affect just how close to the ideal Brownian Noise the system comes...which would in turn affect your ability to achieve that ideal 3dB gain in signal-to-noise ratio.