What Stories Do You Have to Tell About Signal-Chain Noise?

Back in my early days we were building a very high gain amplifier to monitor microvolt DC signals. We took a lot of care to design a low noise amplifier - all the best components, low value resistors, lots of decoupling, careful grounding and shielding etc. The signal source had only low frequency fluctuations so we had carefully designed filters as well. The prototype ended up with some 60 Hz noise on the output which just wouldn't go away. Everything looked clean with the input shorted to ground, but connecting the signal, which was very low impedance, introduced the noise. Simulating the signal with a clean source still resulted in noise. We tried a lot of crazy things to kill the noise, including running off battery power, all without success.

Eventually we discovered that just waving a hand over the circuit caused the noise to change, but not go away. It didn't take long after that to figure out the fluorescent lights flicker was affecting the circuit, but it still took quite a while to track down the culprit components. It turned out to be the protection diodes we had on the input. They were not suspected at first because they were packaged in completely opaque black plastic. There couldn't have been much light getting in, but when you have a gain of over a million, it doesn't take much photo-diode activity to be noticed. Up to that point it hadn't sunk in that any p-n junction could act like a photodiode.

At the time that phenomenon was a big surprise. Now of course, such problems get sorted out in milliseconds.

That key circuit and its system went on to make the company (not me) millions of dollars and is still earning big bucks almost 4 decades later. So stick it out when you run into a troubleshooting nightmare - it could pay off.

Back when I was working, a friend of mine came up with something he called an adaptive algorithm.

It used radial basis functions to model the noise in an analog signal and then filter out that noise to reveal the signal hidden within the noise.

He built up an example where he setup random noise over a video signal. Initially, you could not see the video background due to the noise, but as the algorithm implemented the noise model you slowly began to see the video background emerge in the image.

We then took the algorithm to some people who were trying find objects in very noisy video and they were amazed when the objects popped out of the noise.

I was part of the patent application, though the company decided not to go all the way to a patent.

It was at this point I was finally convinced that the "AI" technology could actually be useful.

DAB

Thats a pretty interesting story, would never expect a common diode to act as a photodiode..

It isn't something you would see with a normal meter or scope, but we had enormous gain. Looking directly at the input we would only see shot noise with our best instruments. Removing the protection diodes cleaned everything up - although that was not a permanent solution.

Pretty much any pn junction will work, often badly, as a photo diode.

Here's a slightly harebrained paper from Mitsubishi (2003) about using the indicator LED on something as a receiver.

https://manualzz.com/doc/9969533/very-low-cost-sensing-and-communication-using-bidirection...

It can be done - I know because my son was asked to implement it as a Uni project - he did it in my lab and used an NXP ARM M0 processor - I think we even have some boards left !

MK

Good point - I've seen LEDs used as photodiodes too - their transparent packaging helps a lot.

Actually all of your solar panels are just photodiodes.

DAB

Maybe we can make chips that are efficient enough to run off light - where a part of the chip is dedicated to harvesting light.