Calibration Guide: How to get stable, accurate results

[ruvnet]

WiFi DensePose requires room-specific calibration to produce stable, accurate results. Without it, the visualization will jitter and skeletons will jump between states.

Why Calibration Matters

ESP32 CSI signals are extremely environment-dependent. Raw CSI amplitude varies with:

• Room geometry, furniture, wall materials
• ESP32 placement (height, orientation, distance between nodes)
• WiFi channel noise, interference from other APs
• Frame-to-frame jitter inherent in ~10 FPS CSI sampling

The default threshold-based classifier uses static values for a "generic room" — your room is not generic.

How to Calibrate (ADR-048 Adaptive Classifier)

Step 1: Record Training Data (~2 minutes)

Record ~30 seconds of each activity state:

# Empty room (leave the room!)
curl -X POST http://localhost:3000/api/v1/recording/start \
  -H 'Content-Type: application/json' \
  -d '{"id":"train_empty_room"}'
# Wait 30s, then:
curl -X POST http://localhost:3000/api/v1/recording/stop

# Standing still
curl -X POST http://localhost:3000/api/v1/recording/start \
  -H 'Content-Type: application/json' \
  -d '{"id":"train_still_presence"}'
# Wait 30s, then stop

# Walking around
curl -X POST http://localhost:3000/api/v1/recording/start \
  -H 'Content-Type: application/json' \
  -d '{"id":"train_walking"}'
# Wait 30s, then stop

# Active movement (waving arms, exercising)
curl -X POST http://localhost:3000/api/v1/recording/start \
  -H 'Content-Type: application/json' \
  -d '{"id":"train_active_movement"}'
# Wait 30s, then stop

Step 2: Train

curl -X POST http://localhost:3000/api/v1/adaptive/train

Training takes <1 second for 3,000+ frames.

Step 3: Verify

curl http://localhost:3000/api/v1/adaptive/status

Multi-ESP32 Tips

1. Stagger TDM slots — each ESP32 needs a unique --tdm-slot (0, 1, 2, 3)
2. Consistent placement — keep nodes at ~1.2m height, spaced around the room
3. Same WiFi channel — all nodes on the same AP and channel
4. Re-train after moving — if you move any ESP32, re-record and re-train

Built-in Smoothing (No Training Required)

Even without adaptive training, the latest version includes:

• Adaptive baseline subtraction (EMA, ~30s time constant)
• Hysteresis debounce (4 consecutive frames to change state)
• Vital signs smoothing (21-frame median + trimmed mean + dead-band)

Make sure you're on the latest main branch.

Related

• ADR-048: Adaptive CSI Classifier
• Issue Using two ESP32-S3 modules, real-time human detection and blinking #141 — Multi-ESP32 flickering discussion

[jl2035]

I did all these steps, but I don't notice any difference in observatory. Am I doing something wrong?

[jl2035]

@ruvnet it's even worse on my 2nd attempt. Now the recordings are just empty files and the step 2 fails.