Device-free localization: Spatial coordinates to midi controller

Track multi-person spatial coordinates using UWB radar to control music arrangement in DAW

Categories:
Hobbies
Tags:
Music
MIDI
Signal processing

I moved to a new city on the other side of the country. What better way to meet people than invite strangers to my apartment for a social dance event?

A work colleague said I was brave.

I wanted to make the event different from the multitude of low-turnout meetups and Facebook events.

Being the musician and tinkerer I am, and since I wanted to host similar events in the future, I wanted to create an experience.

What better experience could there be than having the music react to the people dancing?

I needed to map the movement and energy of people dancing to the music arrangement in my DAW.

This is a solved problem right? It's not.

Goals

  • Plot spatial coordinates (2D at minimum, 3D if possible) of each person in the room.
  • Each person's movements should have an effect on the music.
  • Do this without requiring them to carry any additional devices.

Initial thinking

  • LIDAR/Sonar won't work.
  • Accuracy within ~10-20cm should be adequate (with smoothing).
  • Use a wavelength that can pass through people (without harming them) somewhere in the 500MHz to 5GHz range.
  • Deploy 2 to 3 nodes (Tx/Rx pairs) statically from a few points in the room.
  • Do some signal processing to 'triangulate' (and smooth) coordinates of all the people in the room.
    • This is the equivalent of distilling a PhD-level algorithm into a one-liner during sprint planning.

Problems

  • There was very little reference material to go on. All of it was highly academic with controlled testing environments.
  • I didn't know anything about research-grade signal processing.

The event

My initial thinking might have worked for a sparsely populated room, but there were ~100 people confirmed for this event. Plotting dense/tight groups (like people dancing together) in a crowded room introduces several issues (interference, shadowing, occlusions… signals would get too correlated/complex).

Given the complexity involved, I wasn't able to get this running for the event. Instead, I hacked a few basic depth sensors together and used those instead (it was quite 'dumb', but at least it worked).

For me, it wasn't the 'Whoa, awesome!' moment I hoped for. How could it be when I had my sights set so highly? For everyone else, it was a blast.

Current research

Looks like I was (and am) on to something.

Five/six years later, there are papers coming out about using millimeter-wave radar for this purpose.

Current thinking (as of Jan 2026)

This tech has way more useful applications beyond a fun house party.

  • Deploy more nodes in a multi-static config.
  • Perform signal processing (range-doppler maps, background subtraction, localization).
  • Combine range/angle/doppler to assign unique positions for each person, and track position over time (keep mapping as long as they're in the room).
  • Enhance processing speed with AI (use neural network(s) for spectral distortion classification).