Twotoz/C5VRX

ESP32-C5 Analog 5.8 GHz FPV Receiver

From live 5.8 GHz FPV video to real-time analog CVBS with one Seeed Studio XIAO ESP32-C5 and a passive resistor DAC.

Analog 5.8 GHz FPV receiver on one Seeed Studio XIAO ESP32-C5 with a passive 6-bit resistor DAC: live FM video in, composite PAL/NTSC out for goggles.

C5VRX-5 uses one live demodulator. Its receive-chain evidence is recorded in docs/FINDINGS.md; the U85Q study it is built on, with scoring corrections and failed experiments, is preserved in docs/pr190-key-findings.md and docs/recent-pr-receiver-findings.md. The IQLaneFusion research is host-only (its read-only firmware shadow J was removed 2026-10-12); it is not a demonstrated video or range improvement.

• U85Q demodulator in the TX BitScrambler: an untrained 32-phase tracker that keeps the picture locked across VTX deviations, off-centre carriers and multipath instead of dropping out. Board-validated on real video.
• Fixed ultrafine IQ (ADC bits 9,6,5,4) at 40 MS/s, no CPU in the sample path.
• Direct Gain V6: V5 gain control plus carrier memory across brief fades and same-context resets, with a learned, persisted gain map. GAIN selects saved AUTO (default), MAX or a fixed gain.
• Narrow analog filter, measured once per board at startup: the lowest noise bandwidth that still passes 12 MHz (plus a 10 MHz class for low-deviation VTXs and a 16 MHz class for wide ones; simulation-based, board test pending).
• Calibration only at startup: anything that interrupts the picture runs once after power-on (only what is missing) or from the menu item CALIBRATE, never during flight. The goggles show what it does and how long.
• VTX-matched profile (narrow for low-deviation VTXs, wide 16 MHz for wide ones) only from a fresh measurement: every boot starts with the default profile; the narrow profile is chosen only when this boot measured the VTX in the startup phase. Continuous VTX matching then measures the deviation passively every 30 s on live video (burst- confirmed sync, C/N >= 8 dB, no clipping, held gain; no AFC lock needed, a carrier offset does not change the sync-to-porch span). Each result is logged (PROFILE live), shown in T with agree/mismatch streaks and feeds the kp ladder's deviation class. With the live switch on (default, console A, NVS prof_live) a reliable result changes the profile in flight: wider (narrow -> DEFAULT -> WIDE) after one result, narrower only after two agreeing results and with the AFC locked (a narrow filter needs a centred carrier), each through a filter relock plus the verified tracker restart (a few ms, shared 5 s cooldown; board test pending). A off = report only: the result waits as pending for the next boot or CALIBRATE.
• Field diagnostics (serial D, R): what is really active, and why each dropout happened as far as the receiver can tell (docs/FIELD_TEST.md).
• DC correction for every gain row, also with the VTX on: missing rows are measured once at startup on a quiet channel, so far-out (high gain) reception has no uncorrected DC.
• Automatic receive-chain care: per-gain DC correction, verified live DC trials, passive sampling-phase monitoring and AFC carrier centring.
• Sync flywheel: saved ON/OFF, default ON (a saved OFF is kept). Camera/monitor acceptance remains experimental; compare OFF on the ground.
• Simple menu: band, channel, scan, video standard, calibrate, output level, gain, flywheel, exit.

XIAO ESP32-C5. DAC on D4..D9 (GPIO 23, 24, 11, 12, 8, 9; LSB first) with 8.2k / 3.9k / 2k / 1k / 470R / 240R and a 200R load, as in C5VRX-3/4 (schematic). The receiver uses the ESP32-C5's own Wi-Fi radio; no extra RF parts.

Six resistors, one per XIAO pin, all joined on one node: that node is the video output (CVBS). A seventh resistor from that node to GND is the load.

D4 --[8.2k]--+
D5 --[3.9k]--+
D6 --[2.0k]--+
D7 --[1.0k]--+-----------> CVBS out (goggle / monitor AV-in, video)
D8 --[470R]--+     |
D9 --[240R]--+   [200R]   (optional: 470 pF from the node to GND,
                   |        improvement unproven)
GND ---------------+------> GND (goggle / monitor AV-in ground)

• The pin order matters: the largest resistor (8.2k) goes on D4 and the smallest (240R) on D9. A swapped pair gives a scrambled, banded picture.
• Connect the output node to the video input (centre/tip) of the goggles, a DVR or an analog monitor, and the XIAO GND to its ground. A common ground is required.
• All values are standard E24; 1 % metal-film resistors are fine. Keep the wires from the pins to the node short.
• Too dark or too dim on some goggles: try OUTPUT in the menu (0.5-5.0x, saved). Higher levels can clip; check the loaded output on your display.
• The XIAO is powered over USB-C or its 5 V pin as usual; the DAC needs no separate supply.
• Check the build with the DAC self-test before chasing picture problems: a test pattern for the goggles and a multimeter check that names a swapped, open or wrong resistor. See Building and testing the resistor DAC.

1. Flash (below) and power the receiver, best with the VTX off, for a few minutes (at most eight): the goggles show CALIBRATING with the progress. A firmware with a new calibration generation discards the calibration of an earlier firmware and calibrates fully once. With a VTX on it calibrates the filter on a quiet channel and comes back. Results are stored; later boots skip what is already measured. The first DC measurement is the template for every channel: switching channels needs no new DC search, also with the VTX on. When a boot finds the own channel quiet (VTX off), it measures that channel and replaces the template rows with its own; tables are kept for six channels.
2. Switch the VTX on. Without a stored channel the receiver starts on A1 (5865 MHz); use the menu or scan to find yours.

Menu items: BAND (next band), CHANNEL (next channel), SCAN (all 48 channels, picks the strongest centred analog video), VIDEO (AUTO / NTSC / PAL for the menu picture; AUTO follows the camera), CALIBRATE (on the ground: measure again), OUTPUT (0.5-5.0x DAC level, saved), GAIN (AUTO / MAX / fixed G78..G24, saved), FLYWHEEL (ON/OFF, saved, default ON), DAC TEST (bench only: test pattern; hold for the multimeter steps, see the DAC guide), EXIT. The menu closes by itself after 30 s without a press (5 min on DAC TEST). It releases the flywheel before allocating its DMA chain and retries a smaller progressive raster if the interlaced chain does not fit. Menu entry/exit includes a 150-ms relock gap; verify the picture on the connected monitor.

• Web flasher: twotoz.github.io/C5VRX, tab C5VRX-5, Full firmware when coming from another generation.
• ESP-IDF 6.0.2: idf.py build flash in the repository root. The filter and DC correction need the exact libphy of IDF 6.0.2 (checked at build time); with another PHY they are disabled and the build warns.
• A running receiver enters USB download mode from the console key ` (tools/enter_download.py), so no BOOT button is needed for flashing.

T status, D range/health report, R per-second FIELD_DIAG lines on/off (tools/field_diag_compare.py), I IQ snapshot (tools/capture_iq_snapshot.py), o menu, c next channel, b next band, n/x menu next/select, s scan, F calibrate again (as the menu item), M GAINMAP lines (for a gain-map preset), L bench-only IQ lane comparison (video stops ~3 s), Z DAC self-test (pattern, then the multimeter steps; tools/dac_check.py), ` download mode, h help.

v cycles the video standard AUTO / NTSC / PAL using the same action as the on-screen VIDEO item. e closes the on-screen menu without opening it when already closed. Both are direct controls for the web receiver menu.

K cycles OFF / AUTO kp / AUTO kp+LP. k steps kp calmer, j faster; P manually toggles U85Q/U85Q-LP. P, k and j persist AUTO OFF, including across reboot; repeated presses do not rewrite unchanged NVS. Fresh settings default to AUTO kp+LP (operator decision 2026-10-12); an explicitly saved OFF remains OFF. Kp and program changes use a brief pipeline restart with a fully verified LUT; they are not seamless. All live changes share a five-second minimum interval, including manual P/k/j, K mode, OUTPUT and live profile changes. AUTO runs on every profile: a profile sets the filter and the kp it starts from (MATCHED 0.5 / 0.6, else 0.85), the ladder then owns kp and U85Q-LP, with its own curves on the narrow MATCHED filter. With AUTO OFF the profile's start kp stays. AUTO also requires the same kp/program target for two seconds of usable measurements. Rapid P presses are refused with wait_ms; they are not queued for delayed programme changes.

Test on the ground and check T: kp_verify_fail, prog_verify_fail, kp_restart_fail, prog_restart_fail, transport faults and measured kp_restart_us / prog_restart_us. switch_wait_ms shows the remaining cooldown. Live TX uses an owned fixed DMA ring and has a bounded 20-ms readiness wait with task delays, so a failed start cannot keep the control task spinning indefinitely.

g and menu OUTPUT cycle 0.5x..5.0x in 0.5x steps. A successful, verified application is saved and restored at boot; the default without a valid setting is 1.0x. Higher levels may help weak sync but can clip a wide VTX; check both sync and white with the actual 75-ohm load. Only DAC bits 0..5 are transformed, preserving all phase/token bits across kp changes, LP changes, menu return and rollback. The menu raster retains its own levels. T reports the setting, clipped LUT entries (not an observed waveform clip rate), recovery errors and a latched failure.

G and menu GAIN cycle AUTO (Direct Gain V6), MAX, then fixed G78 / G72 / G66 / G60 / G54 / G48 / G42 / G36 / G30 / G24. The choice is saved and restored at boot; AUTO is the default. V6 keeps the last healthy gain through same-context tracking resets, holds brief fades for 10 ms and then climbs gradually. This is host-tested behavior; extra range is unmeasured.

U and menu FLYWHEEL toggle the saved sync flywheel. Without a valid saved setting it is ON; a saved ON or OFF restores at boot. Check ON/OFF, menu return and the actual camera/monitor picture on the ground. Pulse counters do not prove monitor lock. Switch it OFF for a baseline RF comparison.

A failed restart cleans up before one attempt at the last verified program and LUT, including its OUTPUT setting. Failed rollback latches live OFF; the USB status/download tasks remain available. AUTO stops after a restart failure and reconciles to the verified configuration instead of chasing the failed target. This recovery coordinator is fault-injection tested on the host; actual SDK/peripheral error behaviour and picture acceptance still need hardware qualification.

Automatic OUTPUT is research only: research/output_level.py compares manual scaling with an ideal sync/white-referenced gain and pedestal. Its reference levels are ideal source measurements, not a proven live estimator. See research/data/output_level.json and the evidence notes in docs/FINDINGS.md. It does not establish HDZero compatibility or range gain. The follow-up passive research/output_probe_study.py uses the production AFC/Direct Gain helpers; its conservative white reserve and wide-VTX confidence remain unqualified. Multi-Rescue comparisons against MATCHED are in research/rescue_study.py and research/data/rescue_study.json. The earlier acceptance list remains in docs/PR196_TASK_PLAN.md; PR #212 records the current receiver changes and merge checks.

twotoz.github.io/C5VRX (Chrome, Edge or another browser with Web Serial) has three tabs:

• C5VRX-5: the newest versioned build first, then unmerged PR builds (marked, with a confirmation before flashing).
• Local File: flash a .bin you built yourself.
• Legacy builds: archived C5VRX-3 releases, no longer developed.

Firmware is mirrored into the GitHub Pages site and downloaded same-origin.

The Terminal view has a visual receiver menu for BAND, CHANNEL, SCAN, VIDEO, CALIBRATE, OUTPUT, GAIN, FLYWHEEL and EXIT. GAIN/FLYWHEEL actions require firmware advertising G/U in its help. Their readouts show reported mode and requested/active flywheel state separately. Selecting a browser menu page only changes the view; its action button sends the command. It identifies C5VRX-5 from firmware help/status and displays the last reported band, channel, frequency, signal, video standard, output level, screen-menu state, scan result and calibration state. It updates by itself: the firmware sends a short STATUS line every 2 s (control_print_beat), which the menu reads and keeps out of the log. Firmware without that beat gets a T request after a command, when a scan or calibration finishes and every 10 s while idle. The firmware lists also refresh by themselves (every minute and when the page is shown again), keeping the selected build. VIDEO/EXIT direct actions require firmware advertising v/e in its help; older C5VRX-5 builds can still use the on-screen menu controls. Legacy builds do not enable the C5VRX-5 action menu. #terminal opens the Terminal view directly.

While a status screen owns the picture (first-boot calibration, CALIBRATE, lane test) the firmware ignores menu keys; it now answers each with COMMAND <key> refused=busy and reports busy= in CONTROL and the HB heartbeat, and the web menu greys out those buttons until it ends.

Diagnostics and bench controls are collapsed by default. Commands send one case-sensitive byte without Enter. Commands are sent at once without a browser popup; what an action changes is shown as the button's tooltip. Connecting the terminal releases the flasher, and connecting the flasher releases the terminal; they never open a port together. Download log saves the displayed text, including FIELD_DIAG and raw IQ captures. The display retains the latest 120,000 characters; download before reconnecting or clearing.

.github/workflows/c5vrx5.yml runs the host tests and research checks, builds the firmware and publishes it: c5vrx5-v5.0.N for every firmware change on main (immutable), and a mutable c5vrx5-pr-N prerelease per same-repository pull request, removed when the PR closes. Unchanged firmware inputs (tools/firmware_input_hash.py) keep the existing release, so a documentation or website change does not mint firmware. A successful build refreshes the Pages mirror (deploy-web.yml, always from trusted main).

Before merging a firmware PR, require green firmware and web checks on its final head. Flash the matching artifact (verify its COMMIT and checksum), then check boot, menu entry/exit, saved settings and flywheel ON/OFF on the board. The PR channel is an unmerged build; successful host tests do not establish display lock or RF range. PR #212 validation tracks the cleanup and remaining board check. After merge, verify the main release and Pages mirror contain the resulting firmware.

main/        C5VRX-5 firmware (U85Q, Direct Gain V6, calibration, menu, console)
test/        host tests (CI)
research/    simulations, generators and data behind the design choices
docs/        findings, field-test protocol, task plan, DAC schematic
tools/       console/log tools and the CI/Pages helpers
web/         web flasher (GitHub Pages)
legacy/      archived generations, never built:
             c5vrx1, c5vrx2, c5vrx3 (former root firmware), c5vrx4
             (former experiments/c5vrx-4), legacy-issues

The archived generations stay as reference: their findings explain why C5VRX-5 looks the way it does (see AGENTS.md for the research ledger).

Questions, build photos, feedback, or development discussion? Join the C5VRX Discord.

C5VRX-5 is part of C5VRX by Twotoz and the C5VRX contributors; the official website (twotoz.github.io/C5VRX) has the project and its Discord invite. It builds on C5VRX-2..4 (the continuous MODEM_DIAG to BitScrambler video path, Direct Gain V3-V5, the PAIR schedule and the U85Q study of PR #190), on zerowidth/C5VRX PR #3 (interference-burst gate, sampling-phase probe, off-grid channel tuning), on Louis Hitchcock's staged overload recovery (C5VRX #182/#184), and on ESPARGOS esp-sdr (GPL-3.0) for the C5 RX filter control. Espressif supplied the GDMA EOF fix (espressif/esp-idf#19091).

Like the rest of this repository, C5VRX-5 is licensed under GPL-3.0-only (see LICENSE). The GPL allows commercial use, including selling receivers with this firmware, provided the source (with your changes) is passed on under the same license.

The C5VRX name and logos have separate terms; see assets/BRANDING.md.