- MainForm: TXOverlay главной вьюхи взводился по FTransmitting без учёта
источника TX → красной становилась полоса главного VFO даже при передаче
со слайса. Теперь TXOverlay = FTransmitting and (FTxSliceId = 0).
- VfoOverlay: публичный TxActive (FTx) — вьюхи знают, какой слайс передаёт.
- SpectrumView/SpectrumViewOpengl: полоса/края/несущая передающего слайса
красные (alpha 100, как у главного при TX), иначе цвет бейджа слайса.
- Фикс цвета TX-полосы в GL: $00E03030 при ColorToRGB (младший байт = R)
давал синий вместо красного; общая константа CLR_TX_BAND = $003030E0.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Полоса фильтра, края и линия несущей теперь берут цвет из
SliceColor(SliceLetter) вместо захардкоженного янтарного, который
совпадал с бейджем слайса B. Каждый слайс (B/C/D…) получает свой оттенок.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Replace OpenGLContextEx with OpenGLContext in MainForm, SpectrumViewOpengl, WaterfallViewOpengl, and WidebandView.
Switch the Lazarus project/program dependencies from LazOpenGLContextEx to LazOpenGLContext in ewsdr.lpi and ewsdr.lpr.
PixelScale was introduced in 092a202; restore the previous viewport/texture path: glViewport(0, 0, W, H), EnsureTexture(W, H), and UploadCurrentRow(W).
- SPECTRUM_PIXELS 1024→4096 + настройка display_pixels (лимит точек
анализатора, SetDisplayResLimit); web-зеркало остаётся 1024 —
max-децимация в SnapshotPixels
- палитры водопада Classic/Inferno/Turbo + регулятор gamma (в LUT),
настройки wf_palette/wf_gamma, страница Waterfall → группа Rendering
- GL-водопад переведён на общий LUT/AGC-уровни базового класса
(убраны дубли ColorForDB/UpdateLevels)
- явные glViewport в GL-вьюхах умножаются на PixelScale (HiDPI),
текстура водопада — в физической ширине
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Пакет LazOpenGLContext заменён на LazOpenGLContextEx (форк в
../lazOpengl): Qt6-бэкенд работает поверх настоящего QOpenGLWidget
(своя C++-либа libqlclglwidget) вместо GLX-контекста на winId(),
поэтому --opengl теперь работает в нативной Wayland-сессии, а под
ws=qt6 открывается путь на Windows/macOS.
- uses OpenGLContext -> OpenGLContextEx (класс TOpenGLControl тот же)
- удалена заглушка QtPlatformAllowsOpenGLControls с советом про
XWayland: больше не нужна
- рантайм-зависимость: libqlclglwidget.so рядом с бинарником
(сборка: make -C ../lazOpengl/csrc install-app APPDIR=bin/x86_64-linux)
Проверено на живом эфире (QO-100): спектр/водопад через GL под
Wayland и xcb, прозрачность FBO-альфы починена в обёртке (forceOpaque).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Инструментация (ВРЕМЕННАЯ, не мержить в main):
- PerfLog.pas: зонные таймеры, отчёт раз в 5с в stderr (core% = доля ядра),
EWSDR_PERF=0 выкл, EWSDR_PERF_FILE=путь — дублировать в файл
- Замеры // PERF: сеть (DDC IQ/wideband), Pluto (rx_conv/tx_resamp/tx_fill),
DSP (push_iq/fexchange/slices/tx), аудио, UI (тики/draw/paint/gl_swap,
разбивка draw_spectrum по фазам)
- doc/PERF_PROFILING.md: карта зон, методика, выводы
Оптимизация CPU-рендера спектра (по результатам замеров на эфире):
диагноз — дорога не пиксельная работа, а чередование raw(ScanLine)/
canvas(QPainter) доступа к битмапу на Qt6 (~6 переходов за кадр, каждый
синкает весь битмап; полоса фильтра в тысячи пикселей стоила 4x дороже
полного memcpy кадра). DrawSpectrum перестроен в фазы calc->raw->canvas->raw
(2 перехода): FillBandRaw вместо Canvas.FillRect, DrawSliceFilterMarkers
разделён на raw-полосы + canvas-линии, композиты флагов и альфа-OR в
финальной raw-фазе. Кадр 8.3мс -> 6.1мс; spec_bands 1.87мс -> 0.03мс.
Альфа-OR проход на Qt6 подтверждён нужным (без него canvas-пиксели теряют
альфу и пропадают при блите), сам цикл ~0.08мс — оставлен; тумблер
EWSDR_ALPHA_FIX=0 только для экспериментов. Кисть перед AGC-подписями
теперь ставится явно (была остаточной после перестройки фаз).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Flags (A + B+) now lay out in one pass: each sits beside its own filter
band as before, but a left neighbour pushes the next flag right so they
never cross (forward pack + backward pull-in on right-edge overflow).
LayoutFlags replaces the per-flag positioning; PositionVfoOverlay/
PositionSliceFlags are thin wrappers.
Identity is shown by a letter drawn at the top of each filter band in
the spectrum (both CPU + GL renderers), so a flag pushed away from its
band is still traceable. Letters and the flag letter-badge share one
per-slice palette (SliceColor); F is violet, not red, to avoid clashing
with the TX indicator.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Independent RX slices sharing one panadapter: each has its own
freq/mode/filter/AGC/NR-NB-SNB-ANF, S-meter, volume and sound card.
VFO A = main receiver; B+ are soft fan-out WDSP channels within the
captured span (ideal for Pluto/QO-100; HW DDC path is Stage 3).
- WDSPEngine: soft fan-out (TSliceChan, ProcessSlices in DSP thread,
per-slice NCO/mode/filter/AGC/NR/NB/SNB/ANF + S-meter, OnSliceAudio),
FSliceLock serialises add/remove/set vs DSP; reopened on rate change.
- RadioController: TCtrlSlice with its own TAudioOutput/device,
Add/Remove/SetSlice*, RefreshSliceShifts keeps slices on target Hz
across centre changes. Slice letters start at 'B' (A = main receiver).
- UI: per-slice flags (VfoOverlay) with letter badge/close, filter-band
markers in BOTH GL and CPU renderers, drag-retune, wheel; flags placed
beside the filter band with flip, like the main flag. Ctrl+click adds.
- Main flag (slice A) visible whenever the left panel is hidden OR any
slice exists; synced with panel-driven DSP/mode/AGC changes.
- Perf: S-meter decoupled from flag chrome cache (FMeterDirty +
meter-only RefreshMeter via EnsureRendered); InvalidateVfoOverlay
avoids global overlay-cache churn on meter ticks.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add a thin colored strip along the bottom of the spectrum (above the
ruler) showing the QO-100 narrow-band transponder bandplan: mode
segments (CW ONLY / NB DIGI / DIGI / SSB / MIXED MODES) plus the three
beacons (CW / PSK / EXP) as muted-red cells at their ranges.
- New unit BandPlanOverlay.pas (TBandPlanOverlay), modelled on
VfoOverlay: content rendered once into a cache bitmap, per-frame cost
is just a constant-alpha composite (CPU) / one textured quad (GL).
Cache invalidates only on view change (center/span/width) so it idles
while the QO-100 downlink display is static.
- Wired into both render paths: SpectrumView (CPU composite) and
SpectrumViewOpengl (texture, re-uploaded only when dirty).
- MainForm: SetView fed from SyncSpecViewFreq; visibility gated by
InQO100 (Pluto + full-duplex transponder), like RX MUTE / BEACON.
- DPI-robust: strip height scales with Screen.PixelsPerInch and the font
height is tied to the strip height (not point size), so labels always
fit at any Windows scaling.
- Segment/beacon frequencies and colors are a single editable table.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Builds on the beacon-lock baseline: keeps the proven display-FFT control
path (hardware-LO trim folded into the active transverter LOError), and
hardens it. The narrowband NCO/FLL scaffold (BeaconLock.pas) is removed —
it was never in the control path.
- RadioController:
* Persist live: each correction folds straight into the active xvtr
LOError (visible in Settings, survives restart as LNB calibration);
disk writes throttled. FBeaconLockHz is now only a session indicator.
* Robust tracking vs nearby signals: narrow ±2 kHz gate while locked
(±6 kHz only for acquisition), slew-rate outlier rejection (a sudden
centroid jump = interferer, not slow LNB drift → ignored), and lobe
shape validation in MeasureBeaconFFT (width + power symmetry) so a CW
carrier / one-sided neighbour isn't mistaken for the BPSK beacon.
Prominence measured over the in-window noise floor (gate-width robust).
* Fast convergence: feed-forward (predict post-retune position so the
window follows the beacon through the jump) + near-deadbeat correction
on the manual click → one click locks instead of repeated tapping.
* Single central BPSK beacon (10489.750); dead CoarseBeaconOffset and
unused throttle counters removed; stale NCO/FLL/4 Hz comments fixed.
- WDSPEngine: drop the per-sample TBeaconTracker RX-IQ tap (tracker gone).
- MainForm: BEACON is now a plain button, shown only on Pluto in a
transverter. Lock status (off / click / search / LOCK + correction +
prominence) moves to status-bar field 7 in place of PLL on Pluto;
openHPSDR keeps PLL there.
- SettingsForm/StatusBar: widen LO Error field to ±10 MHz (QO-100 LNBs
drift hundreds of kHz) and widen status field 7 for the beacon text.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Click the beacon on the spectrum; the controller tracks it and continuously
trims the transverter LO so the whole downlink stays put as the LNB drifts.
- BeaconLock.pas: TBeaconTracker scaffold (narrowband NCO/FLL) — kept but no
longer in the control path (FLL diverged on the suppressed-carrier BPSK
beacon + Pluto DC offset). Measurement now uses the proven display-FFT path.
- RadioController: FBeaconLockHz folded into XvtrTranslate (on top of LOError).
MeasureBeaconFFT finds the beacon by POWER-WEIGHTED CENTROID of the lobe
(BPSK has a suppressed carrier → broad symmetric ~2.4 kHz lobe, not a line;
a peak-finder jitters, the centroid sits on the carrier). Track position in
ABSOLUTE display Hz (FBeaconTrackHz) so scrolling the waterfall/centre never
moves the search window off the beacon. Closed loop: e = tracked - ref,
d(e)/d(LockHz) = -1 -> LockHz += gain*e; settle cooldown after each correction
(display-FFT lags the LO retune -> avoids overshoot); deadband 75 Hz +
centroid/prominence smoothing + lock hysteresis for a steady lock.
Correction applied only after the user clicks (manual seed), so a crowded
band can't pull a neighbour to the reference.
- WDSPEngine: SetBeaconTracker + RX-IQ tap (tracker idle now, ~1 check/sample).
- SpectrumView + GL: beacon markers — green = reference (10489.750), orange =
tracked centroid.
- MainForm: BEACON button (RX block, XVTR-only). Flow: BEACON -> "CLICK BCN"
-> click beacon -> "L<corr> /<prom>". ServiceBeaconLock driven by MeterTimer.
Reference defaults to the middle BPSK beacon 10489.750. Tested on-air against
Es'hail-2: locks and holds, survives waterfall drag.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
On macOS Cocoa, TBitmap.pf32bit stores pixels in ARGB order
(byte0=A, byte1=R, byte2=G, byte3=B) rather than BGRA used on
Linux/Windows. All raw ScanLine pixel blending and channel mapping
is now guarded with { DARWIN} to use the correct offsets.
Also fix PortAudio library discovery on macOS: probe .dylib names
(including Homebrew paths) before falling through to .so/.dll.
ew8bak
