[Firmware] ESP32-C6 v0.6.7 pre-built: 0 bytes CSI during valid association, no reassociation after AP-initiated disconnect

Summary

The pre-built ESP32-C6 firmware in firmware/esp32-csi-node/release_bins/c6-adr110/ (v0.6.7 per release_bins/version.txt) exhibits two related failures on a stock ESP32-C6-DevKitC-1:

1. During a successful WiFi association, the firmware transmits 0.00 B in either direction for the entire session — no CSI packets ever leave the device.
2. After the AP disconnects the client for inactivity, the firmware never re-associates, even across multiple full reflashes and NVS --reset reprovisionings, and even after switching to a completely different AP.

Because the C6 build routes its console to UART0 (default) and does not enable CONFIG_ESP_CONSOLE_USB_SERIAL_JTAG, there is no on-chip log visibility over the board's USB-Serial-JTAG port, so the WiFi driver state at the point of failure cannot be observed.

Hardware

Item	Value
Board	ESP32-C6-DevKitC-1 (stock, no soldering)
Chip	ESP32-C6 (QFN40) revision v0.2, 160 MHz, Single Core + LP Core, WiFi 6, BT 5, 802.15.4
Base MAC	fc:01:2c:XX:XX:XX (redacted)
Crystal	40 MHz
USB mode	USB-Serial/JTAG (VID 303A:1001)
Enumerates as	USB Serial Device (COM5)

Host

Item	Value
OS	Windows 11, PowerShell 7
Python	3.13.14
esptool	5.2.0
esp-idf-nvs-partition-gen	0.3.0
Docker Desktop	29.6.1 (Docker image ruvnet/wifi-densepose:latest)

Reproduction

# 1. Shallow clone
git clone --depth 1 https://github.com/ruvnet/RuView.git .

# 2. Flash pre-built C6 firmware (4 MB layout)
python -m esptool --chip esp32c6 --port COM5 --baud 460800 `
  --before usb-reset --after hard-reset write-flash `
  --flash-mode dio --flash-size 4MB `
  0x0     firmware/esp32-csi-node/release_bins/c6-adr110/bootloader.bin `
  0x8000  firmware/esp32-csi-node/release_bins/c6-adr110/partition-table.bin `
  0xf000  firmware/esp32-csi-node/release_bins/c6-adr110/ota_data_initial.bin `
  0x20000 firmware/esp32-csi-node/release_bins/c6-adr110/esp32-csi-node.bin

# 3. Provision NVS (fresh --reset)
python firmware/esp32-csi-node/provision.py `
  --port COM5 --chip esp32c6 `
  --ssid "<SSID>" --password "<PSK>" `
  --target-ip <host-ip> --target-port 5006 --reset

# 4. Start sensing-server
docker run -d --rm --name ruview `
  -e RUVIEW_ALLOW_UNAUTHENTICATED=1 -e CSI_SOURCE=esp32 `
  -p 127.0.0.1:3000:3000 -p 127.0.0.1:3001:3001 -p 5006:5005/udp `
  ruvnet/wifi-densepose:latest

What actually happens

Session 1 — first boot after fresh provision (SUCCESS on association, FAILURE on CSI)

Attempted on AP-A: a UniFi network (WPA2/WPA3-mixed, U6-LR access point).

UniFi event log (times shown relative to t=0 = association):

• t = 0 — Connected to AP-A on Ch. 1 (2.4 GHz, 20 MHz), -47 dBm, DHCP lease granted.
• t = +7m 50s — Disconnected by AP. Time Connected: 7m 50s. Data Used: 0.00 B up / 0.00 B down. Last signal -47 dBm.

Corresponding server-side metrics for the same 7m50s window:

GET /api/v1/metrics → { "tick": 0 }
GET /api/v1/sensing/latest → { "status": "no data yet" }
docker logs ruview | grep -i "udp\|csi frame\|frames"  # 0 matches

So: excellent RF (-47 dBm), 100% signal, TX Retries 0.0% at the AP — yet the firmware never transmitted a single UDP packet to the configured target-ip/port during the entire connected session. The AP eventually disassociated it for inactivity.

Session 2 onwards — no reassociation, ever

After the AP-initiated disconnect, the ESP32 has not reappeared in the UniFi client list or in ARP on the host, despite:

• Multiple esptool --after hard-reset chip-id cycles (confirmed chip responsive: MAC returned each time)
• Multiple full reflashes (bootloader + partition-table + otadata + app, all hash-verified)
• Multiple provision.py ... --reset cycles (verified persisted state has SSID + password + target-ip + target-port)
• Physical USB power-cycle and 30-min cool-down
• Switching to a completely different AP — AP-B: an iPhone Personal Hotspot, WPA3-Personal, 2.4 GHz Ch 5, "Maximize Compatibility" ON, 100% signal, host confirmed associated with a fresh DHCP lease — same outcome, 90 s wait, NOT SEEN in ARP.

I.e., after the first AP-initiated disconnect, the C6 firmware appears to permanently stop trying to associate on this hardware — not even to a fresh, different AP that has never seen this MAC before.

Ruled out

Suspect	How ruled out
Wrong PSK	First association on AP-A succeeded → PSK is correct. State file confirms the same PSK persisted for later attempts.
5 GHz-only AP	ESP32-C6 is 2.4 GHz only; both APs tested were on 2.4 GHz (Ch. 1 and Ch. 5).
UniFi bad-client block	Reproduced on a completely separate AP (phone hotspot, AP-B) that has no history with this MAC.
Weak signal	-47 dBm on AP-A (excellent) and 100% signal on AP-B.
Wrong band-steering	AP association on Ch. 1 (2.4 GHz) confirmed by UniFi client detail.
Damaged board	esptool consistently talks to the chip on COM5 with the correct MAC and completes full-image flash + verify.
Docker port-forwarding drops CSI frames	UniFi client counter shows 0.00 B up from the ESP32 during the entire 7m50s session — packets weren't dropped, they were never sent.
ESP-IDF NVS write bug	Verified persisted state matches what was written; also happens after --reset full-wipe writes.

What would help debug

The ESP32-C6 overlay firmware/esp32-csi-node/sdkconfig.defaults.esp32c6 does not set:

CONFIG_ESP_CONSOLE_USB_SERIAL_JTAG=y

so the default console goes to UART0 (physical pins), which is unreachable on a stock DevKitC-1 without a USB-UART adapter wired to GPIO 16/17. This makes it effectively impossible for a user with only the USB-C DevKitC-1 in hand to see:

• The WiFi driver's association attempts and disconnect reasons
• Whether stream_sender_init_with(...) succeeded and whether CSI callbacks are firing
• Whether the app is stuck at boot, in wifi_init, in DHCP, in a retry loop, or actually running

Ask 1: could the pre-built C6 release binary enable CONFIG_ESP_CONSOLE_USB_SERIAL_JTAG=y (and matching CONFIG_ESP_CONSOLE_SECONDARY_NONE=y if needed) so ESP-IDF's esp_log output is visible over the board's native USB-Serial/JTAG port?

Concrete questions

1. Is anyone else seeing "0 bytes over an associated session" on the C6 v0.6.7 pre-built binary? (Would confirm/deny hardware-specific vs firmware-wide.)
2. What is the intended behavior when the AP disassociates the client — does the firmware auto-reassociate, and if so, is there a Kconfig knob for the retry policy on the C6 overlay?
3. Should the CSI capture pipeline (which drives all UDP traffic) start unconditionally after WiFi IP_EVENT_STA_GOT_IP, or does it wait for another event that might not fire on the C6 target?

Environment / artefacts

• Firmware version: firmware/esp32-csi-node/release_bins/c6-adr110/ — release_bins/version.txt = 0.6.7
• SHA-256s (matched SHA256SUMS.txt in the same folder):
  • bootloader.bin = b0fb1f217a39c80bc95b5eb8208a0b8572ae64efa0f6d580b76caff4affe0f4d
  • partition-table.bin = 4c2cc4ffd52641e23b779bd57b3908014083ac3c1aab395756478c89e70d81f0
  • ota_data_initial.bin = 7d2c7ac4888bfd75cd5f56e8d61f69595121183afc81556c876732fd3782c62f
  • esp32-csi-node.bin = 4764c5b20a353895f70122816adc98f861ec20e9a8ea9b344dc0648b6341073c
• Docker image: ruvnet/wifi-densepose:latest, digest sha256:14b2fd3347c16a0959aca37bf5c61f0fabf79cfe8b6d01a60edccb0ab3e24b6a
• Sensing-server logs (relevant lines):

  INFO sensing_server: UDP listening on 0.0.0.0:5005 for ESP32, MediaTek, Qualcomm CSI, and RTL8720F radar frames
  INFO sensing_server: Data source: esp32 (udp_receiver=true, simulator=false, wifi=false)
  

  No frames received / equivalent log entries appeared for the entire multi-hour session.

Happy to run any additional diagnostics if a follow-up build with console logging or a specific ESP-IDF flag lands.

Start with firmware/esp32-csi-node/sdkconfig.defaults.esp32c6 and the C6 release files in firmware/esp32-csi-node/release_bins/c6-adr110/. Reproduce the flash and provisioning steps, then inspect the WiFi and CSI startup paths around IP_EVENT_STA_GOT_IP and stream_sender_init_with(...). Done means identifying why CSI sends 0 bytes or reassociation stops, with console output available over the native USB port.