User avatar
jbeale
Posts: 3884
Joined: Tue Nov 22, 2011 11:51 pm
Contact: Website

advice for low-noise sensor performance

Sun May 30, 2021 7:11 pm

I figured this out before, forgot it, and re-learned it so I'm going to write it down here to help me remember. It might help someone else too.

In my case I'm working with a cheap 24 GHz doppler-radar module, the CDM324 (about $10 on ebay). It needs +5V and and it gives you a signal with a frequency proportional to the speed of a moving object. Its output is a very low-level audio signal, like a low-sensitivity microphone. You can get this into a Pi using the mic input on a USB audio interface, although it is so weak that it also benefits from a low-noise preamp. The problem is that because this is such a low-level signal with zero power-supply rejection, you end up getting any electrical noise on the +5V power supply added on top of the signal.

With any noise-sensitive device the best bet is direct battery power, and indeed that works well, but of course that's not a solution if you want to run it for a long time because you'll need to recharge it eventually. If you power your Pi from an external PoE device that gives you +12 V, with an additional DC-DC converter to get the +5V for the Pi, you might think that you can just add a separate 12V -> 5V converter to get a cleaner power supply for your sensor, rather than using the +5V going to the Pi which of course has lots of digital noise running around.

The key point I need to remind myself is that in this case it is NOT true, I believe due to ground loops. No matter how many stages of active regulators and passive RLC filters, with normal and common-mode filtering I used with that configuration, I saw a lot of noise on the sensor output as measured by the USB audio device. I got noise around 1 kHz, 2 kHz and many bands from 10-100 Hz, plus big broadband spikes whenever transferring data via ethernet.

What did work much better (to be quantitative, around 20 dB better) is the somewhat non-intuitive solution of using a 5V -> 12V DC boost device powered from a USB port on the Pi, which then goes into a 12V->5V DC buck converter with a simple L-C filter on that output. All narrowband noise and ethernet-related noise is now completely gone from the output spectrum of interest (5 Hz -> 4 kHz). Each DC converter adds some loss, but in my case since I only need 35 mA @ 5V for my preamp + sensor, it's still not a huge load on the Pi's USB port. The specific boost device I used was "USB 5V to DC 12V Converter Step Up Voltage Converter Power Cable" https://amazon.com/dp/B07T6487LW but there are many similar things sold.
Last edited by jbeale on Fri Jun 04, 2021 3:44 am, edited 3 times in total.

User avatar
neilgl
Posts: 3347
Joined: Sun Jan 26, 2014 8:36 pm
Location: Near The National Museum of Computing

Re: advice for low-noise sensor performance

Mon May 31, 2021 10:43 pm

Thanks

User avatar
jbeale
Posts: 3884
Joined: Tue Nov 22, 2011 11:51 pm
Contact: Website

Re: advice for low-noise sensor performance

Mon Jun 07, 2021 4:36 am

In case you're wondering what you can do with a sensor like that, here's one: point it straight up into the sky. On a clear day you see nothing except an occasional blip from a bird or insect flying inside the radar beam. However, when it rains you can tell not only how heavy the rain is, but also how large the raindrops are, because the terminal velocity of a raindrop is a function of the drop size. Here's an example with a brief afternoon shower today. I thought the plot was interesting because it shows in some detail how the rain changes when heavier or lighter clouds pass overhead.
RainShower-Compare.jpg
vertical doppler signal on rain
RainShower-Compare.jpg (69.38 KiB) Viewed 405 times
I got this plot and measured an apparent peak velocity around 9.8 m/s based on my sensor calibration, before I looked up what raindrop speeds actually are. It felt pretty good to read that the largest drops reach 10 m/s, and smaller drops are slower. https://gpm.nasa.gov/resources/faq/how- ... drops-fall

I'm guessing that the thin vertical spikes could be discrete impact events when a larger drop hits the plastic bag that is over my sensor, in turn splattering micro-droplets in all directions, but I don't know.

User avatar
Gavinmc42
Posts: 5851
Joined: Wed Aug 28, 2013 3:31 am

Re: advice for low-noise sensor performance

Mon Jun 07, 2021 7:29 am

Interesting, those antistatic bags would make a good 5G blocker ;)
I'm dancing on Rainbows.
Raspberries are not Apples or Oranges

twostage
Posts: 125
Joined: Sun May 07, 2017 6:31 pm
Location: Northumberland

Re: advice for low-noise sensor performance

Tue Jun 15, 2021 3:44 pm

How far away does it detect ?

Return to “Automation, sensing and robotics”