aethersdr/AetherSDR v0.8.19

v0.8.19 — Community landfest + macOS MNR + quick memory actions

on GitHub

[v0.8.19] — 2026-04-20

Community contributions landfest + macOS MNR + quick memory actions

This release is mostly community work. Ten community PRs landed, plus an
AetherClaude-draft fix batch covering bandwidth-correction echoes, stale
slice-click tuning, Windows rigctld IPv4, serial port persistence, DX
cluster reconnect, compression-meter gain reduction, and macOS audio
device override.

Features

macOS MMSE-Wiener spectral noise reduction (MNR) (#1672, Chaosuk97)

• New macOS-exclusive client-side NR mode accelerated via Apple
  Accelerate (vDSP / AMX on Apple Silicon). Decision-directed
  MMSE-Wiener filter with 25-frame minimum-statistics noise-floor
  tracking. 512-point real FFT at 24 kHz, 50% overlap-add with
  sqrt-Hann window, per-bin gain with temporal smoothing to suppress
  musical noise. ~0.3 ms per hop on M-series chips, one-hop startup
  latency (≈10.7 ms).
• MNR button appears in both the VFO DSP tab and the spectrum overlay
  DSP panel. DSP Settings dialog gains a new MNR tab with a strength
  slider (0-100). Mutually exclusive with NR2/RN2/NR4/BNR/DFNR.
  Persists enable state + strength across restart.
• Non-Apple builds: #ifdef __APPLE__ gated throughout — zero MNR
  overhead, buttons hidden.

PooDoo™ TX Reverb (#1741) — carried forward, chain now complete

• Already shipped in v0.8.18 final release notes; listed here for
  completeness since it rolls into the 0.8.19 post-release chain view.

Quick memory browse + save on slice overlays (#1781, jensenpat)

• Two new actions in the slice overlay rail between Display and DAX:
  • MEM▸ opens a compact 252×430 browser drawer, frequency-sorted,
    pre-scrolled to the active slice's nearest memory. Click-to-recall.
  • MEM+ opens a small save dialog for the current slice on that
    panadapter. Captures mode, filter, step, repeater offset, tone,
    squelch, RTTY/DIGL offsets — same field set as the main Memory
    dialog via the new shared MemoryCommands helper module.
• Pan-aware routing: save/recall targets the active slice if it lives
  on this pan, otherwise the first slice found on the pan. The MEM
  buttons now show the target slice letter (e.g. MEM▸ A) so users
  can see which slice will be hit before clicking — makes the
  multi-slice fallback visible.

Improved TNF notch interaction (#1547, rfoust)

• TNF markers no longer extend into the waterfall — they stop at the
  bottom of the spectrum pane so the waterfall stays a historical
  signal record.
• Depth visualisation via hatching (12/8/5 px spacing for depth 1/2/3).
• Floating "RF Tracking Notch" popup on hover shows frequency + width.
• Width/depth submenu items show a checkmark on the currently-active
  value.
• Bidirectional TNF drag: horizontal tunes, vertical resizes via
  2^(-dy/48) octaves, clamped 10-12 000 Hz.
• Preferred-TNF hit-testing for overlapping notches — dragging a TNF
  preserves the target across its sibling's overlap zone.
• Tune guides suppressed while a TNF is hovered or dragged.
• TnfModel: optimistic local updates in setters so UI responds
  without waiting for radio echo; width parser tolerates both Hz
  ("width=100") and fractional-MHz forms for firmware-version variance.

RAC Canada Band Plan corrected + expanded (#1709, VE3NEM)

• VHF / UHF bands added (6 m, 2 m, 1.25 m, 70 cm) — prior file was
  HF-only, so 4 new ham-band coverage tiers arrive in this update.
• 60 m channels updated to match the Feb 2026 ISED regulation change
  (new 5.3570-5.3598 channel).
• 12 m frequencies corrected.
• FT8 spot at 14.074 added back, one duplicate spot removed.

Frameless Window toggle (View menu, Ctrl+Shift+F)

• View → Frameless Window applies Qt's FramelessWindowHint so users
  gain ~32 px of vertical screen space at the cost of the OS title bar.
  Works cross-platform (X11, Wayland, Windows, macOS) — KDE window
  rules don't strip Qt's client-side decorations reliably on Wayland,
  so we drop the frame in code. Move/close via WM shortcuts or
  taskbar. Persists across restart; applied before first show() so
  there's no flash-of-chrome on startup.

CHAIN-driven TX DSP (v0.8.18 rebase — carried forward)

• Already in v0.8.18 release notes; called out here because subsequent
  releases continue to build on the CHAIN ordering.

Bug fixes

NR2 crackling on strong signals (#1696, Chaosuk97)

• Cap gainMax default 1.5 → 1.0. The MMSE-LSA estimator is a noise
  reducer — the old 1.5 default allowed 50% amplification on strong
  signals, which pushed output past ±1.0 float32 full scale and caused
  QAudioSink digital-clipping crackle.
• Belt-and-suspenders std::clamp(sample, -1.0f, 1.0f) in
  processNr2(). A separate Bessel overflow (issue #1507) can still
  produce NaN-burst crackle on Gamma-method NR2 under strong-signal
  conditions — tracked for follow-up, not in this release.

Audio pan broken with client NR active (#1685, Chaosuk97) + per-slice AF mute persistence (#1560)

• Client-side NR (NR2 / RN2 / NR4 / DFNR) mono-mixes L+R before
  processing, destroying the radio-applied pan. New
  AudioEngine::setRxPan atomic + static applyRxPanInPlace helper
  re-applies a linear pan law to the NR output so the pan slider
  continues to work while client NR is active. No-op at pan 50
  (centre) — zero overhead when user hasn't panned.
• Per-slice AF mute now persists across sessions. VFO widget emits
  audioMuteToggled; MainWindow saves SliceAudioMuted_{A,B,C,D} to
  AppSettings on every user toggle, restores on reconnect. Uses the
  slice letter (not numeric id) for the settings key so state survives
  sessions where the radio assigns a different numeric id.

VFO / TNF / filter-edge ±1 px jitter during tuning (#1703, Chaosuk97) + cursor-snap-to-VFO (#1369)

• mhzToX() switched from integer truncation to std::round so the
  same frequency always maps to the same pixel column regardless of
  pan-center fractional drift during continuous tune. One-line fix
  that ripples through every vertical marker — VFO, TNF, filter edges,
  RIT/XIT, band plan, spots, grid.
• Cursor frequency readout snaps to the exact VFO frequency when
  within 8 px of a slice marker (GPU path already had this; software
  paintEvent path added for symmetry).
• setSliceOverlayFreq() now calls markOverlayDirty() and
  early-exits on unchanged value.

Zoom +/- can't reach full panadapter bandwidth (#1697, Chaosuk97)

• Replace early-return guard in emitZoom / keyboard zoom with
  std::clamp(newBw, minBwMhz, maxBwMhz) so the final click/keypress
  snaps to exactly the radio's min/max — matching the mouse-drag
  behaviour which already used clamp.

PC Audio button shows wrong state on connect (#1695, Chaosuk97)

• onConnectionStateChanged now reads PcAudioEnabled once and calls
  both m_titleBar->setPcAudioEnabled(...) (sync button, with
  QSignalBlocker) and audioStartRx() from the same decision.
  Previously the sink and button could diverge after a profile import.

Waterfall filter-passband fill removed (#1701, Chaosuk97)

• The coloured filter-passband fill in the waterfall was only rendered
  by the software-fallback paintEvent path — the GPU renderer already
  skipped it. Dropping the software fillRect brings the two render
  paths into agreement. Filter edge lines still render in both panes.

TCI DAX duplicate stream causing 2× audio speed (#1679 / #1678)

• Two code paths both subscribing to dax_rx for the same channel
  made daxAudioReady fire twice per period, doubling WSJT-X's
  apparent sample rate and killing FT8 decode. Registration now
  evicts stale subscribers and TCI uses a borrow protocol.

TCI TX no RF output (#1680)

• Always emit transmit set dax=1 on PTT regardless of the previous
  Low-Latency DAX menu toggle. The dax=0 path was advertised for
  external FreeDV over virtual audio cable but never actually worked
  because that route silently drops dax_tx VITA-49 packets. The
  Low-Latency DAX menu has since been retired (#1780).

Low-Latency DAX consolidated into RADE mode (#1780)

• The "Low-Latency DAX (FreeDV)" menu item was removed. RADE mode's
  activation now emits a daxRouteRequested signal that MainWindow
  routes to transmit set dax=N, flipping the route atomically with
  the RADE button. One control, one place. Stale
  DaxTxLowLatency AppSettings value is no longer read.

Bandwidth corrections silently dropped during pan-follow (#1729)

• The stale-echo guard in setFrequencyRange bypassed any echo
  during an in-flight pan-center animation when the center matched.
  If the echo also carried a bandwidth correction (e.g. post-resize
  xpixels re-sync), it was thrown away. Guard now also checks the
  bandwidth field so legitimate corrections pass through.

Pan-follow VFO edge-tracking after manual drag (#1643)

• After a manual spectrum drag, PanadapterModel::centerMhz lagged
  the visible pan center by up to one frame. panFollowVfo now
  reads the live center from the spectrum widget instead of the
  model, so edge-fence checks fire against the user's actual view.

Propagation overlay click tunes VFO by accident (#1647)

• Clicking the K/SFI propagation overlay was also triggering the
  release-to-tune path. Set m_spotClickConsumed = true after the
  propForecastClicked emission so the subsequent release event
  can short-circuit.

Off-screen slice click tunes wrong frequency (#1772)

• Same pattern as the prop-overlay fix: clicking an off-screen slice
  indicator now suppresses release-to-tune so the tuning action
  doesn't pile on top of the slice-activate.

rigctld TCP server not accepting IPv4 on Windows (#1737)

• QHostAddress::Any sometimes binds IPv6-only on Windows, leaving
  IPv4 clients unable to connect. Use QHostAddress::AnyIPv4
  explicitly.

Serial port parameters (data bits, parity, stop bits) not persisted (#1610)

• RadioSetupDialog's serial tab built the combos with defaults but
  never read from AppSettings or connected the save handler. Added
  both.

Compression meter shows raw dBFS instead of gain reduction (#1682)

• COMPPEAK is the compressor output level; AFTEREQ is the input.
  Meter was displaying raw COMPPEAK, which just shows "how loud is
  the output" instead of "how much is the compressor reducing". Now
  shows output - input (gain reduction, 0 = no reduction, negative
  = compressing).

macOS audio output device silently overridden (#1705)

• On macOS, AudioEngine replaced the user's chosen output device with
  QMediaDevices::defaultAudioOutput() whenever Qt reported 48 kHz
  as unsupported — but that happens on some normal CoreAudio outputs
  with newer Qt, not just Bluetooth telephony routes. Narrow the
  override to devices that genuinely can't handle the native 24 kHz
  format.

Internal

Issue template tightened

• Bug-report reproduction steps are now required. AetherSDR version,
  radio model+firmware, and OS are required fields with platform-
  specific guidance (Linux distro + Qt / macOS version + chip /
  Windows edition). Improves triage quality on incoming reports.

Contributors

• Added VE3NEM, Ian M7HNF (Chaosuk97), and jensenpat to the About
  dialog contributors list.

Reverted

• The spectrum/waterfall alignment tweak from #1694 was landed as a
  partial fix (alignment math only, skipping an unrelated
  SMOOTH_ALPHA tuning change) then reverted when the trimmed fix
  introduced misalignment rather than correcting it. Issue #1690
  closed — the originally-reported bug did not reproduce on main.