Question regarding hardware requirements: Is the mmWave sensor mandatory for 3D Pose Estimation?

[nicolherlindomah-arch]

Hi, I'm very interested in porting this project to my custom ESP32-S3 board. However, while reviewing the firmware source code, I found something confusing regarding the hardware dependencies.

The project is named "WiFi-DensePose", but in firmware/esp32-csi-node/main/edge_processing.c (inside the send_vitals_packet function), the system heavily relies on an external mmWave radar (mmwave_sensor.c). The Kalman fusion logic explicitly assigns an 80% weight to the mmWave radar and only 20% to the WiFi CSI data (fused_hr = mw.heart_rate_bpm * 0.8f + s_heartrate_bpm * 0.2f;).

My questions are:

Can this system actually perform 3D skeletal tracking and vital monitoring using ONLY a single-antenna ESP32 (SISO) without any external radar module attached?

From an RF physics perspective, a single antenna lacks the spatial diversity to compute the Angle of Arrival (AoA). If the radar is disconnected, isn't the UI just rendering preset 3D animations triggered by simple RSSI variance thresholds?

Thanks for clarifying!

[AntwerpDesignsIonity]

Hi, I'm very interested in porting this project to my custom ESP32-S3 board. However, while reviewing the firmware source code, I found something confusing regarding the hardware dependencies.

1. The project is named "WiFi-DensePose", but in firmware/esp32-csi-node/main/edge_processing.c (inside the send_vitals_packet function), the system heavily relies on an external mmWave radar (mmwave_sensor.c). The Kalman fusion logic explicitly assigns an 80% weight to the mmWave radar and only 20% to the WiFi CSI data (fused_hr = mw.heart_rate_bpm * 0.8f + s_heartrate_bpm * 0.2f;).

My questions are:

2)Can this system actually perform 3D skeletal tracking and vital monitoring using ONLY a single-antenna ESP32 (SISO) without any external radar module attached?

3. From an RF physics perspective, a single antenna lacks the spatial diversity to compute the Angle of Arrival (AoA). If the radar is disconnected, isn't the UI just rendering preset 3D animations triggered by simple RSSI variance thresholds?

Thanks for clarifying!
#################
"Hi I am not a expert on this yet but please check :

*1)Make sure it is a Esp32 S3 *Duel core as each core is used for another purpose within processes you will bottleneck of you do not use that exact model. I do think that mmwave is a upgrade but explore the potensial.

*2)I would rely on densities check Ruv's repo for more on that.

*3)CSI is not related to direct RSSI with. Start with 3 Nodes minimum. The Science behind it works so it is possible to enchance over time even.

Thats a newbies insight maybe wait for other also.

[ruvnet]

To answer both questions directly:

1. Is mmWave mandatory?

No. The firmware auto-detects the mmWave sensor at boot via UART probe. If no sensor is found, it falls back to WiFi CSI-only mode — the 80/20 Kalman weight only applies when a mmWave sensor is physically connected. Three deployment modes are supported:

• CSI-only: ~$9/node (ESP32-S3, no radar)
• mmWave-only: ~$15 (ESP32-C6 + MR60BHA2)
• Fused: ~$24 (both)

2. What can single-antenna CSI actually do?

Your RF physics point is correct — a single ESP32 cannot compute AoA. What it can do from CSI amplitude/phase variance across 52+ OFDM subcarriers:

• Presence detection, motion classification, person count (1–3)
• Breathing rate, basic heart rate, fall detection

These work with a single node. True 3D skeletal pose requires either a trained model across multiple nodes (3+ recommended for geometric diversity via differential CSI) or the mmWave sensor for depth. The 3D skeleton shown in the UI without a loaded model is driven by signal heuristics — it demonstrates the pipeline is live, not a full pose reconstruction. See the Known Issues guide (#301, "What WiFi CSI Can and Cannot Do") for the full breakdown.

---

Generated by Claude Code