aethersdr/AetherSDR v0.9.1

v0.9.1 — Local iambic CW keyer + CW transmit reliability

on GitHub

[v0.9.1] — 2026-04-27

Local iambic CW keyer, unified sidetone controls, and CW transmit reliability

A focused follow-up to v0.9.0. The headline feature is a software
iambic keyer that turns any MIDI / serial paddle into a sub-5 ms
sidetone source via the new PortAudio backend (issue #2079) — the
radio still produces the on-air signal, but the local sidetone
gate fires the moment the paddle moves instead of waiting for the
radio's keyed-back signal. Three latent bugs in the netcw protocol
path that prevented CW from ever transmitting on FLEX-8600 v4.1.5
firmware are also fixed. Plus the CW panel collapses three sidetone
widget groups into one set of controls and finally wires up the L/R
pan slider that's been a placeholder since the panel was first added.

Features

Local iambic keyer for sub-5 ms paddle sidetone (#2079)

• Software iambic state machine that runs alongside the radio's RF
  iambic engine. Both engines see the same paddle inputs at the same
  WPM and produce identical Morse timing — but the local keyer drives
  the sidetone gate directly, avoiding the 50–200 ms round-trip
  through the radio's keyed-back signal.
• Modes A and B implemented (Ultimatic / Bug / Straight follow in a
  later phase). Hooked into both MIDI Gate params (cw.dit,
  cw.dah) and serial paddle paths (DTR/CTS via SerialPortController).
• Driven by the existing radio Iambic toggle — no new UI clutter.
  The keyer mirrors the radio's iambic state, mode, and WPM via the
  TransmitModel phoneStateChanged signal.
• Dedicated worker thread with std::chrono::steady_clock timing
  and a lock-free atomic key gate on the audio side. 9 unit tests
  covering single dit/dah timing, squeeze alternation, inter-element
  gap, mode A release behaviour, WPM scaling, paddle swap, idempotent
  start, and idle behaviour.

Sidetone controls unified (#2079)

• The CW panel previously had three separate sidetone widget groups:
  the radio's "Sidetone" toggle/volume, a "Local STn" toggle/volume
  for the local PortAudio sidetone, and a "Follow" pitch row with a
  manual override slider. All three are now collapsed into the
  single existing Sidetone button, which drives both engines in
  lockstep. The volume slider drives mon_gain_cw on the radio and
  the local sidetone identically. Pitch always follows the radio's
  cw_pitch.

CW pan slider wired up (#2079)

• The L/R pan slider in the CW panel was a dead UI element with a
  TODO comment. Now drives both the radio's mon_pan_cw (radio-side
  sidetone pan within the RX audio stream) and the local sidetone
  with constant-power pan law (cosine/sine for equal-loudness sweep,
  no center dip). Double-click on the slider recenters to 50.

Slice capacity notification (#48)

• Status-bar warning when adding another panadapter would exceed the
  radio's slice limit. Three guard points cover the pre-flight check
  in the layout dialog, the runtime check in applyPanLayout, and
  the async fallback when the radio rejects a panafall create.
  Includes the radio model name and slice count in the message.

PortAudio sidetone via JACK on Linux (#2075 follow-up)

• The PortAudio sidetone backend (introduced in v0.9.0) now prefers
  the JACK host API on Linux when available, with paFramesPerBuffer = 128 + suggestedLatency = 0 for sub-5 ms quantum. PipeWire's
  ALSA shim silently breaks callback-mode streams on some setups
  (Pa_StartStream returns success but the audio thread never
  schedules the callback); pipewire-jack delivers reliable callbacks
  at the device's native sample rate. 48 kHz is now the universal
  preference, with Pa_IsFormatSupported guarding fallback.

Bug fixes

CW transmit: invalid paddle command form

• RadioModel::sendCwPaddle was emitting cw key 1 0 (a 2-arg paddle
  form) which the radio's protocol does not accept — FlexLib only
  ever sends single-state cw key 1 or cw key 0 and expects the
  client to do iambic timing locally. The 2-arg form was silently
  dropped, so paddle keying produced no RF. Now collapses to a
  straight-key form when the local iambic keyer isn't running, or
  routes through the new sendCwKeyEdge + sendCwPtt primitives
  when it is.

CW transmit: lowercase hex in netcw payload

• FlexLib formats time=0x... and client_handle=0x... with C#
  ToString("X") (uppercase), and the radio's status messages do
  too (e.g. S23A59BDF|...). Firmware v4.1.5's netcw parser is
  case-sensitive on these fields and silently dropped lowercase
  packets. Now formatted explicitly uppercase.

CW transmit: dead TCP fallback

• The post-UDP TCP fallback was sending the netcw-decorated form
  (cw key 1 time=0x... index=... client_handle=0x...) which the
  radio rejects with 0x50001000 ("command syntax error") on TCP.
  Removed when the netcw stream is up; the no-netcw fallback (for
  firmware that doesn't support netcw stream creation) is preserved
  separately.

Optimistic updates for CW model setters

• setCwIambic, setCwIambicMode, setCwSpeed, and setCwPitch
  on TransmitModel previously sent the command to the radio
  without updating local state, on the assumption that the radio
  would echo the new value back. Firmware v1.4.0.0 doesn't reliably
  echo iambic flags, WPM, or pitch in transmit status messages, so
  any code reading these properties after a UI toggle saw stale
  values until the next periodic transmit status arrived (or never).
  Now follow the same optimistic-update pattern as setCwBreakIn.

Block on graceful disconnect (#1996, openstreem)

• RadioModel::disconnectFromRadio now uses Qt::BlockingQueued Connection for both the gracefulDisconnect lambda and the
  fallback disconnectFromRadio call, matching the destructor's
  pattern. Without this, the queued work could be cancelled before
  it ran during app teardown — leaving the radio with a stale
  session that required a power cycle.

Memory recall: restore repeater offset and tone (#1871, #1965, jensenpat)

• Recalled FM repeater memories properly restore repeater_offset_dir,
  fm_repeater_offset_freq, derived tx_offset_freq, and CTCSS tone
  state. Previously, switching from a repeater memory to simplex
  could leave stale TX offset state on the slice.

m_activeTxSlice initializer mismatch (#2076)

• Default-init was 0 but clear() sets to -1. Now both default
  to -1 so a fresh MeterModel and a cleared one have identical
  activeTxSlice() state.

Per-slice compression meter resolution (#2073)

• TX-chain COMPPEAK, AFTEREQ, and SC_MIC meters are now
  resolved per active slice instead of last-match-wins, fixing
  Multi-Flex setups where the wrong slice's compression value was
  surfaced in the UI.

ContainerWidget restoreState displaces children

• restoreState re-inserted children at indices 0..N–1 even when
  they were already in the layout, displacing non-saved children
  like the CHAIN widget. insertChildWidget is now a no-op when
  the child is already present.

Build and CI

• M_PI_2 replaced with a local kPiOver2 constexpr in
  CwSidetoneGenerator.cpp so the Windows MSVC build doesn't
  fail (<cmath> doesn't define M_PI_2 without _USE_MATH_DEFINES).

Acknowledgements

Thanks to the operators who tested CW workflows on real hardware
and surfaced the netcw protocol issues that had been silent since
the netcw backend landed, and to jensenpat and openstreem
for the community fixes bundled in.