Co-Founder @ Stoneforge | Sidequests: Battlebots, Drones and Game Dev · Stoneforge
5d ago
RF EngineerStoneforge
Our calibration target is a stainless steel plate taped to a brick.
Early calibration is about iterating fast, so we do it at the office, in hours rather than days. The prototype goes on a tripod on our balcony, because that's the biggest open space we have.
A wall scan is only worth trusting if the scanner already knows two things: what it sees when there's nothing in front of it, and how it sees something simple and known.
Pointed at open space, it gives us the background we subtract from every scan. That's the scanner's own coupling plus whatever the surroundings add.
Then the plate. A flat steel plate is about as simple as a target gets: it reflects almost everything straight back. We step the distance between the scanner and the plate, and that tells us:
1) What a clean, flat reflection looks like, double bounces included
2) Where the response shifts from near field to far field
3) Where the antennas' phase center sits at each frequency
We found that the antennas behaved, and we now have an estimate of the noise floor. The surprise wasn't the RF. It was the lidar we use to measure distance, which turned out noisier than the radar.
We now fuse the lidar with the rough range we get straight from the EM signal. Together they're much more accurate than the lidar alone.
Why do this at all? Until now a lot of our work was empirical: take many scans, label them, and let ML find the patterns. Calibration maps out the physics, so our next models can use both.
The finer optimizations can wait for an anechoic chamber. Everything we can learn on a balcony, we learn on a balcony.
If you're an RF, DSP or hardware engineer who likes fast iteration and good measurements, we're hiring in India: https://lnkd.in/gpvv6A3N
tal saysstoneforge is hiring an rf engineer in india to build and calibrate radar and lidar systems
engineering
India
rfdspradarlidar
comments 3 anon