uv-k1-k5v3-firmware-custom/App/apps/sstv

SSTV: send the boot logo, receive pictures (work in progress)

Overlay app for the Labs firmware (needs the sysinfo capability for sys_storage_read). It receives 14 SSTV modes, picked by the VIS: Robot 36 and 72, Martin M1 and M2, Scottie S1, S2 and DX, PD50, PD90, PD120, PD160, PD180, PD240 and PD290 (the ISS sends PD120 or PD180); it sends the boot logo in any of them (key 1).

Step	State
Modes (test/modes.py, Dayton paper as MMSSTV / QSSTV / slowrx)	14 modes; lengths = the published ones (PD120 126.1 s, Martin M1 114.3 s, ...)
Model: TX schedule, radio channel, integer receiver (test/model.py)	Done (sweep below)
TX sequence vs the PySSTV reference encoder (Robot36 class)	Same VIS and length (36.910 s); 0.8 % of the time differs, at pixel edges (128 vs 320 columns)
Radio app (sstv_app.c)	v0.3, 3676 B of the 4 KiB overlay (GCC 13.3.1; last measured before the key remap was final: to be confirmed by the user's build)
The built app in Unicorn (test/emu.py) vs the model	REG_71 writes = the model's tones (Robot 36, Scottie S1, PD120); pictures identical pixel for pixel (Robot 36, Martin M1, Scottie S1, PD120), with the real LCD blit times
On the radios	v0.1 (2026-10-03): a picture starts, then nothing (see below); v0.2 works (Robot 36, 2026-10-03); v0.3 (all modes) not tested yet

v0.1 timed everything from differences of SysTick readings, which lose 10 ms whenever a service runs longer than the 10 ms SysTick period: a full blit is ~11 ms (LCD SPI at 48 MHz / 64 = 750 kHz, 7 pages of 128 B). The first picture redraw put the receiver 96 samples late, at the edge of the sync window, and the picture was lost a few lines later; the TX lines carrying a redraw were 10 ms too long. v0.2 reads an absolute clock: the 10 ms SysTick interrupt count (ticks_ms) times 480000 plus the position in the period. test/emu.py charges 1.5 ms per LCD page and serves ticks_ms at its 10 ms resolution: v0.1 fails there (919 wrong pixels), v0.2 matches the model.

Using the app

1. Set the VFO, FM, on a simplex frequency (low power for the first tests). For the ISS: 145.800 MHz, wide FM (Doppler ±3.5 kHz), during an announced ARISS SSTV event.
2. Launch SSTV. It listens at once (the speaker as last left, off the first time: the decoder does not need it, as in APRS RX). A known VIS starts a picture, drawn row by row over the whole screen, status line included.
3. PTT (or MENU) sends the boot logo in the TX mode (key 1). The logo is shown while it goes out, each row inverted once sent. EXIT aborts. The long modes keep the transmitter on for minutes (Scottie DX 269 s, PD290 289 s): low power.

Key	Action
PTT or MENU	Send the boot logo in the TX mode
1 / F then 1	Next / previous TX mode (Robot 36 -> ... -> PD290 -> Robot 36), shown in the left capsule; saved
F	Arm the next key's backward direction (1), as APRS TX; F icon in the status bar (x = 70) while armed; F again disarms
2	Rendering: 1-bit (threshold, sharp for the logo; default) / dither (4x4 Bayer, for photos), from the next row on; right capsule; saved
3	Speaker on / off (off by default: the decoder does not need it); saved
4	Picture / info screen, once a picture came (before: No picture yet)
EXIT	Quit (during a TX: abort it)

Info screen: WAIT in the status bar (x = 40, as APRS RX) until a picture comes, the speaker icon at x = 59 while on; on line 0, 3x5 capsules: the TX mode (key 1) on the left, the rendering (1-BIT / DITHER, key 2) on the right; the status with the mode of the picture (PD120 123/248 (periods: two lines each in PD), PD120 OK, PD120 lost), or Logo sent, TX denied, TX aborted, No picture yet; the RX frequency; the keys.

Modes

Mode	VIS	Length	Period	Scans read (weight)
Robot 36	8	36 s	sync 9, porch 3, Y 88, sep 4.5, porch 1.5, chroma 44	Y
Robot 72	12	72 s	sync 9, porch 3, Y 138, 2 x (sep 4.5, porch 1.5, chroma 69)	Y
Martin M1 / M2	44 / 40	114 / 58 s	sync 4.862, then G B R, 146.432 / 73.216 each, 0.572 gaps	G 2, B 1, R 1
Scottie S1 / S2 / DX	60 / 56 / 76	110 / 71 / 269 s	sep 1.5, G, sep 1.5, B, sync 9, porch 1.5, R (138.24 / 88.064 / 345.6)	R 1, G 2, B 1
PD50 / 90 / 120 / 160 / 180 / 240 / 290	93 / 99 / 95 / 98 / 96 / 97 / 94	50 / 90 / 126 / 161 / 187 / 248 / 289 s	sync 20, porch 2.08, Y0, R-Y, B-Y, Y1	Y0 2, Y1 2

Times in ms. Scottie sends one more 9 ms sync after the VIS. Each mode is one 112-byte record in the assets (test/modes.py record(), mode_t in the C): the period's segments for the TX, and for the RX the scan offsets from the sync end, their weights, the period and the pixel step.

TX

The boot logo is read where ui/welcome.c reads it (external flash 0x011008, 1024 B, LCD layout: 8 pages of 128 columns, bit set = dark pixel). Each scan carries one logo row in 128 equal columns (scan / 128: whole cycles at 48 MHz, e.g. 45600 for PD120), the 64 rows spread over the periods (racc += A, a new row each time racc >= B: 4/15 for Robot, 1/4 for Martin and Scottie, 8/31 for PD120). Dark = black (1500 Hz), clear = white (2300 Hz); every colour scan carries the same row and the chroma is neutral (1900 Hz): grey. An erased logo sector (no logo uploaded) sends black.

As APRS TX: tx_set_params, REG_51 = 0 (no CTCSS/DCS), tx_tone(1900) (tone path on, mic off, 50 ms settle, level 66), then each tone is started by writing REG_71 at its time. REG_71 is written only when the frequency changes, in case a write resets the tone phase. The row inversion (one LCD page, ~1.5 ms) and the key scan are done in each sync (4.862 ms at least). VIS: 1900 Hz 300 ms, 1200 Hz 10 ms, 1900 Hz 300 ms, start bit 1200 Hz 30 ms, 7 bits LSB first + even parity (1100 Hz = 1, 1300 Hz = 0), stop 1200 Hz 30 ms.

RX

The RX audio is sampled as in APRS RX and EPIRB 406: PA4 (biased by the DAC), ADC channel 4, 9.6 kHz, continuously.

• Band-pass (APRS RX's biquad, 4x its gain), then per sample p = y1 (y0 + y2) and q = y1^2: for a tone of angular step w, p / q = 2 cos w whatever its level, so no AGC is needed. R = 256 P / Q: 1100 Hz 380, 1200 Hz 362, 1500 Hz 285, 1900 Hz 164, 2300 Hz 33.
• P and Q smoothed (1/8 per sample) flag the leader (1800-2020 Hz), the sync (below ~1360 Hz) and the VIS bits (below ~1210 Hz), by comparisons only.
• VIS: a 20 ms sync run begun right after a 40 ms leader is the start bit; each bit is a vote over its middle 20 ms; the byte (parity included) is looked up in the mode records.
• Period: the end of the sync (1200 -> 1500 Hz in every mode, whatever comes before it) within ±10 ms of the prediction starts the period. Its scans start at their offsets (less 6.8 samples of filter delay, tuned on Robot 36), scaled to the tracked period. A late sync restarts the period; a sync near the next prediction cuts the current period short (a fast sender clock, or a period that ran on the prediction); no sync: the prediction (20 in a row: lost). The period is tracked (1/8 of each error, ±6 %), and the pixel step follows it.
• Pixel: lum = 285 - 256 Σp / Σq over its samples, 0..255, times the scan's weight into the line buffer (sum 4). Periods are averaged per screen row, then thresholded at 128 (1-bit) or against a 4x4 Bayer matrix (dither).
• Only the luminance is decoded: colour pictures come out grey.

The keys and the screen take sample time (key scan ~400 µs, one LCD page ~1.5 ms): in a picture they are served after each period's last scan, before its sync ends (5.4 ms in Martin), one page at most: the new row, else the next page of the screen not refreshed since the VIS (a full redraw, ~12 ms, could hide a sync). Out of a picture: every 50 ms unless a VIS may be on: a leader, a start bit begun right after one, or the VIS bits (3 s at most). Up to v0.3 a sync run alone held them off too: receiver noise, mostly low-pitched, reads as a sync half the time, so half the slots were skipped and a normal key press (60-120 ms) was often missed (test/emu.py key test). In a picture the keys are read once per period: hold a key up to 0.5 s in PD120, 1 s in Scottie DX.

Model sweep (test/model.py, the test logo)

Channel: 300 Hz high-pass, 750 µs de-emphasis, 3 kHz low-pass, 3 kHz deviation ~ 195 LSB at the ADC, as in the APRS RX model. Wrong pixels of 8192.

Every mode, standard path: 0 in 1-bit (dither 122-507: the logo's 1-pixel strokes rendered grey). PD120 / Martin M1 through the cases (1-bit): std 0 / 0, no de-emphasis 0 / 0, TX clock +1 % 47 / 26, -1.5 % 78 / 57, deviation 1.5 kHz 0 / 0, SNR 20 dB 0 / 0, 12 dB 0 / 0, 6 dB 3 / 0. The built app, Martin M1 at -1.5 %: 163 (the slots taken by the keys and the screen in its 5.4 ms gap).

Robot 36:

Case	1-bit	dither	sync error (samples)
standard path	0	267	-2..2
no de-emphasis	0	289	-6..2
TX clock +1 %	5	241	-2..28 (period 1455)
TX clock -1.5 %	31	291	-46..2 (period 1419)
deviation 1.5 kHz	0	265	-2..2
SNR 20 dB	0	299	-2..2
SNR 12 dB	0	484	-2..3
SNR 6 dB	28	915	-4..5

Dither errors are the logo's 1-pixel strokes rendered grey: expected.

Test plan

1. TX into a phone: send the logo, listen on a second radio (speaker on), decode with the Robot36 app (Robot 36 only) or MMSSTV / QSSTV on a PC. Without a radio: test/model.py wav logo.wav PD120 [dump.bin] writes the same transmission as a WAV (any mode name of test/modes.py).
2. Radio to radio: SSTV on both, PTT on one, each TX mode (key 1). The logo should come back pixel for pixel in 1-bit.
3. RX from other software: a PD120 WAV (above, or a recorded ISS pass) played into a radio (VOX, or acoustic coupling) to the app.

If the RX picture is shifted left or right on the radio, Y_OFF (22) is the knob; if the VIS is missed, check the audio level at PA4 (the APRS RX pp reading, ~300 expected).

Files

File	Role
sstv_app.c	the app
gen_assets.py	texts, status table, mode names and records, dither thresholds, speaker icon
test/modes.py	the 14 modes and their asset records
test/model.py	TX schedule, channel, receiver model, sweep, WAV writer
test/emu.py	runs the built sstv.elf in Unicorn against the model