AetherSDR is a Linux-native client designed to operate FlexRadio Systems transceivers seamlessly without requiring Wine or virtual machines. Built using Qt6 and C++20, it natively supports the SmartSDR protocol, delivering a feature-rich experience tailored for amateur radio enthusiasts.
Key Features:
GPU-Accelerated Spectrum & Waterfall: Leverages QRhi rendering (OpenGL/Metal/D3D11) for efficient GPU-based visualization, reducing CPU load by up to 71%.
Multi-Slice & Panadapter Support: Offers independent TX assignment and diversity/ESC beamforming across up to 8 detachable pans with VITA-49 waterfall tiles.
KiwiSDR Browser Integration: Connects to public KiwiSDR receivers worldwide, enabling diversity receive and RX-only TX inhibit.
Aetherial Audio Processing: Includes a comprehensive DSP suite (gate, EQ, compressor, exciter, etc.) for enhanced audio quality.
Multi-Engine Noise Reduction: Supports NR2, RN2, NR4, DFNR, BNR (NVIDIA GPU AI), and MNR (macOS) for versatile noise suppression.
DAX Virtual Audio & IQ: Provides 4 RX + 1 TX channels, raw I/Q at 24–192 kHz, and WFM demodulation for satellite data.
AetherModem Packet Radio: Features KISS-over-TCP TNC, AX.25 BBS, APRS client, and Direwolf-derived VHF demodulator.
Audience & Benefit:
Ideal for amateur radio operators using FlexRadio transceivers on Linux, AetherSDR offers a fully functional, high-performance environment with advanced features like multi-device control, packet radio capabilities, and integration with public receiver networks. It enables users to enjoy enhanced audio processing, noise reduction, and seamless interaction with tools like KiwiSDR and APRS.
Additional Notes:
AetherSDR supports installation via winget for Linux users, ensuring easy setup and access to its robust functionality across supported FlexRadio transceivers, including FLEX-6000, FLEX-8000, and Aurora series devices.
README
AetherSDR
A cross-platform, open-source client for FlexRadio Systems transceivers
AetherSDR brings full FlexRadio operation to Linux, macOS, and Windows — each a native build, no Wine or virtual machines. A native aarch64 build also runs on Raspberry Pi and other embedded ARM devices. Built from the ground up with Qt6 and C++20, it speaks the SmartSDR protocol natively and aims to replicate the full SmartSDR experience.
> Native builds for Linux, macOS, and Windows — Linux AppImage (x86-64 + aarch64), macOS DMG (Apple Silicon + Intel), Windows installer and portable ZIP. Every platform is built, tested in CI, and released together.
Native. Open. Yours.
Highlights
GPU-accelerated spectrum & waterfall — QRhi rendering on the GPU (OpenGL/Metal/D3D11) with a per-pixel FFT trace at up to 60 fps, an optional 3D stacked-trace spectrum mode (perspective FFT history, floor-anchored ridges), ~71% CPU reduction over CPU paint, GPU-composited slice flags, and multi-GPU adapter selection
Multi-slice & multi-panadapter — colour-coded VFO overlays, independent TX assignment, diversity/ESC beamforming; up to 8 detachable pans with native VITA-49 waterfall tiles, with selectable S-meter / SmartMTR meter views per flag
KiwiSDR public-receiver browser — find and connect to public KiwiSDR receivers worldwide through an API-policy-aware directory (diversity receive with receive-only TX inhibit)
Aetherial Audio Channel Strip — a unified RX and TX DSP suite (gate, EQ, compressor, de-esser, tube, AetherVoice exciter, reverb, brickwall limiter) with a preset library and a per-side scope
Six client-side noise-reduction engines — NR2 (spectral), RN2 (RNNoise), NR4 (libspecbleach), DFNR (DeepFilterNet3), BNR (NVIDIA GPU AI — the Maxine denoiser in-process on a local NVIDIA RTX/GeForce GPU, Linux + Windows; see docs/nvidia-bnr.md), and MNR (macOS)
DAX virtual audio + IQ — 4 RX + 1 TX channels and raw I/Q at 24–192 kHz for WSJT-X / fldigi / VARA / JS8Call, plus a per-slice WFM demodulator for satellite data
AetherModem packet radio — KISS-over-TCP TNC, connected-mode AX.25 BBS, a personal mailbox, and an APRS client (station map, GPS beacon, messaging) with a Direwolf-derived VHF demodulator
AetherSweep — in-panadapter SWR analyzer with log scale, threshold-band shading, and interpolated bandwidth at SWR ≤ 1.5 / 2.0
SpotHub — DX Cluster, RBN, WSJT-X, POTA, and FreeDV Reporter spots with auto-mode switch
CW operator suite — real-time Morse decoder, MIDI/keyboard straight-key & iambic paddles with full QSK, optional Quindar tones
Copy Assist (speech-to-text) — on-device transcription of received voice via whisper.cpp, docked under the waterfall with confidence color-coding; CPU or GPU (Vulkan/Metal, auto-detected), download-on-demand models, and an optional remote OpenAI-compatible endpoint. Not in the Intel macOS build — it would force a macOS 15.5 floor on hardware that mostly cannot reach it (see docs/asr-copy-assist.md)
FreeDV RADE — AI digital-voice codec with a client-side neural encoder/decoder
SmartLink remote + TCI v2.0 server — Auth0/TLS WAN operation, and CAT + audio + IQ + CW + spots over a single TCI WebSocket
Broad hardware control — rigctld + virtual-serial CAT, MIDI mapping, the FlexControl knob, serial PTT/CW keying, and Multi-Flex operation alongside SmartSDR/Maestro
Built-in demo mode — a synthetic backend that generates its own RX audio and matching panadapter, with a fault-injection harness, so you can explore the full UI with no radio attached (it cannot transmit)
How AetherSDR Is Built
AetherSDR is developed using an AI-augmented open-source workflow:
Project lead (Jeremy KK7GWY) + a core contributor team working primarily through Claude Code and a mix of AI development tools — every commit goes through the merge gate; nothing reaches main without human review
AetherClaude orchestrator bot auto-triages incoming issues, drafts implementation plans, and produces PRs for issues labelled aetherclaude-eligible
Contributors use a mix of AI tools (Codex, Copilot, Cursor, Gemini, Aider) — the project's Constitution (14 principles, structured per Cisco's Foundry Constitution spec) codifies the conventions every contributor and every AI tool follows
Branch protection enforces signed commits, CI green, and CODEOWNERS review — every change goes through the same gate regardless of which AI tool (or human) produced it
At active pace: ~50 PRs per week, ~15,000–30,000 lifetime downloads, ≥6 distinct AI tools touching the codebase
See AGENTS.md for the canonical project guide that every AI assistant reads first, and CONSTITUTION.md for the principles that gate the contribution model.
The full list of code contributors is auto-generated from GitHub commit attribution — see the Contributors graph.
Supported external devices include the 4O3A/FlexRadio PGXL (Power Genius XL)
power amplifier and TGXL (Tuner Genius XL) antenna tuner.
Active test target is FLEX-8600 firmware 4.2.18 (SmartSDR protocol v1.4.0.0);
earlier 4.x firmware works; v3.x is unsupported.
No radio at all? Demo mode runs the full UI against a synthetic backend
that generates its own audio and spectrum.
Tested Controller Devices
AetherSDR supports external station-control hardware through USB serial, USB HID,
MIDI, Stream Deck/StreamController plugins, and generic USB-serial adapters:
FlexRadio FlexControl USB tuning knob
Icom RC-28 USB remote encoder
Griffin PowerMate USB knob
Contour ShuttleXpress and ShuttlePro v2 jog controllers
MIDI controllers with learn mode, profiles, and relative-encoder support
Elgato Stream Deck devices through the bundled macOS/Windows Stream Deck plugin
Stream Deck devices on Linux through the bundled StreamController plugin
USB-serial PTT/CW interfaces for foot switches, straight keys, iambic paddles,
amplifier keying lines, and external sequencers
Apple Silicon (M1+). Intel Macs via Rosetta. Signed & notarized.
Windows Installer
AetherSDR-*-Windows-x64-setup.exe
Setup wizard with Start Menu shortcut and uninstaller.
Windows Portable
AetherSDR-*-Windows-x64-portable.zip
No install needed. Extract and run.
Building from Source
Dependencies
Install all dependencies for a full-featured build. Optional packages are noted — the build succeeds without them but the corresponding features are disabled.
Qt 6.8 or newer is required. This is the same Qt the release binaries are
built against (6.8.3 LTS), so what CI compiles is what ships. Distro Qt clears
it on Debian Trixie, Ubuntu 25.10+, Fedora 41+ and Arch. It does not clear
on Ubuntu 24.04 LTS, which ships Qt 6.4.2 — build there against a Qt from
aqtinstall or the Qt online installer
and point CMake at it with -DCMAKE_PREFIX_PATH=/path/to/Qt/6.8.3/gcc_64.
On macOS the Qt does not come from Homebrew at all — see the macOS note
below the install commands.
> macOS note — Qt and qtkeychain do not come from Homebrew. Homebrew's qt
> formula (aliased qt6 and qt@6) is a rolling release — 6.11.1 at the time
> of writing — while the DMG ships 6.8.3 LTS like every other artifact. Building
> against Homebrew's Qt means testing a Qt no release ships. Install the matching
> one and point CMake at it:
>
> bash > # A venv rather than a bare `pip install`: a PEP 668 python3 refuses the latter. > python3 -m venv ~/.venv/aqt && ~/.venv/aqt/bin/pip install aqtinstall > ~/.venv/aqt/bin/aqt install-qt mac desktop 6.8.3 clang_64 \ > -m qtmultimedia qtwebsockets qtserialport qtshadertools \ > --outputdir ~/Qt > cmake -B build -DCMAKE_PREFIX_PATH="$HOME/Qt/6.8.3/macos;$(brew --prefix)" >
>
> clang_64 is the only macOS desktop build Qt publishes, and it is universal2 —
> there is no separate arm64 archive to pick. $(brew --prefix) stays on the
> path for fftw, portaudio and hidapi.
>
> Homebrew's qtkeychain is left out for a related reason: the formula depends
> on qtbase, so installing it pulls a second Qt in behind your back. Build it
> against the Qt you just installed instead — or skip it and build without
> SmartLink credential persistence:
>
> bash > CMAKE_PREFIX_PATH="$HOME/Qt/6.8.3/macos" bash scripts/setup/setup-qtkeychain.sh >
>
> Two Qt installations visible to CMake at once is a real failure, not a
> theoretical one — it is what #711 and #812 were, and CMakeLists.txt puts
> $(brew --prefix)/include on the global include path on macOS, so a Homebrew
> Qt is discoverable whether or not you asked for it. If you have one,
> brew uninstall qt (plus whatever pulled it in) before building. The release
> workflow asserts this; your machine will not.
> Linux Mint / Ubuntu note: If PC audio devices show as "Dummy Output",
> install gstreamer1.0-pulseaudio. For PipeWire systems, also install gstreamer1.0-pipewire.
>
> Ubuntu 26.04 note: If AetherSDR fails to start with a missing
> libOpenGL.so.0 error, install libopengl0. 26.04 stopped pulling it in
> by default for the desktop image; the build-deps line above includes it
> explicitly so this only bites users who install just the AppImage.
RADE-enabled builds use a vendored Opus snapshot, so no additional Opus download
is required during configure or build.
Windows 11
Prerequisites: Visual Studio 2022 (Build Tools, Community, or higher) with the
MSVC C++ workload, CMake 3.25+, Ninja, and Qt 6.8+ (msvc2022_64; both CI and
the release binaries use 6.8.3 LTS).
:: 1. Activate the MSVC environment. Adjust the edition (BuildTools / Community /
:: Professional / Enterprise) to match your install; run "vswhere" if unsure.
"C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Auxiliary\Build\vcvars64.bat"
:: 2. Point at your Qt kit once, with forward slashes (CMake reads the path
:: literally, so backslashes would be taken as escape sequences). Change the
:: version/edition here to match your install; both steps below reuse it.
:: setup-qtkeychain.ps1 (step 4) reads QT_ROOT_DIR; on CI that variable is
:: exported by install-qt-action, so a local build has to set it explicitly
:: or the script exits with "Qt not found".
set "QT_KIT=C:/Qt/6.8.3/msvc2022_64"
set "QT_ROOT_DIR=%QT_KIT%"
:: 3. Generate the single-precision FFTW import lib (needed by NR4/libspecbleach)
powershell -File scripts\setup\setup-fftw.ps1
:: 4. Build qtkeychain (needed for QRZ/SmartLink credential persistence).
:: Downloads source and builds it against your Qt kit into third_party\qtkeychain\.
:: Skip this step and the build still succeeds, but QRZ/SmartLink passwords
:: won't be saved between runs.
powershell -File scripts\setup\setup-qtkeychain.ps1
:: 5. Configure. Ninja is required: the default Visual Studio generator is
:: multi-config (it ignores CMAKE_BUILD_TYPE) and takes a different
:: manifest-embed path. Point CMAKE_PREFIX_PATH at your Qt kit so
:: find_package(Qt6) resolves.
cmake -B build -G Ninja -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_PREFIX_PATH="%QT_KIT%"
:: 6. Build
cmake --build build --target AetherSDR
GPU Spectrum Rendering
GPU-accelerated spectrum/waterfall rendering requires Qt 6.7 or greater (QRhiWidget). Since the build now requires Qt 6.8 as a minimum, no build is held back by the Qt version any more — the aarch64 AppImage included. What decides whether a given binary renders via QRhi is the AETHER_GPU_SPECTRUM build option, and for a source build whether Qt's private GUI headers are installed: CMake turns the option off with GPU spectrum rendering disabled — Qt6GuiPrivate not found when they are missing (install qt6-base-private-dev / qt6-qtbase-private-devel).
The CPU QPainter path is a build-time alternative, not a runtime fallback. AETHER_GPU_SPECTRUM selects SpectrumWidget's base class — QRhiWidget or QWidget — and SpectrumWidget::paintEvent(), which is what draws the spectrum on the CPU, is compiled only into the QWidget build. (A GPU build still uses QPainter, but only to rasterise overlays into textures QRhi then composites.) Of the shipped artifacts only the Intel macOS DMG is built the other way, and deliberately: QRhiWidget misbehaves on older Metal/OpenGL hardware.
Having no GPU is usually a non-event, because in practice "no GPU" means a software rasterizer rather than nothing. QRhi comes up on whatever the platform provides — llvmpipe or softpipe (Mesa), WARP or Microsoft Basic Render (D3D11), SwiftShader — and the app detects it and says so: Help ▸ About shows a Renderer: line reading CPU QRhi (…) rather than GPU QRhi (…), naming the backend and device. Rendering is correct, just slow.
If QRhi cannot initialise at all — no usable GL/D3D/Metal, as on a headless host, in some VMs, or behind a broken driver — there is nothing to fall back to. The spectrum does not draw, and the failure is reported by Qt rather than by AetherSDR: the log records QRhiWidget: QRhi is not supported on this platform. or QRhiWidget: No QRhi, and QRhiWidget::renderFailed() fires with nothing listening, so there is no notice in the UI. The rest of the app (controls, audio, radio I/O) is unaffected.
AETHER_NO_GPU=1 forces software OpenGL on an already-built binary, without a rebuild:
AETHER_NO_GPU=1 ./AetherSDR-*.AppImage
That is the escape hatch if a GPU or driver renders the spectrum incorrectly — worth trying first on Raspberry Pi and other systems whose Mesa driver is newer than its hardware.
Wayland and XWayland
On a Wayland session AetherSDR asks Qt for wayland;xcb — native Wayland when
the platform plugin is available, XWayland otherwise. Native Wayland avoids the
GLX BadAccess crash that XWayland can produce when opening child dialogs on
some compositors, and renders correctly under fractional scaling instead of
being bitmap-scaled by the compositor.
If a compositor misbehaves under native Wayland, force XWayland:
QT_QPA_PLATFORM=xcb ./AetherSDR-*.AppImage
Setting QT_QPA_PLATFORM yourself always wins — AetherSDR only supplies a
default when the variable is unset.
On a distribution whose Qt is older than the required 6.8 (notably Ubuntu 24.04 LTS at 6.4.2), install a newer Qt manually:
Option 1: Using a PPA (Ubuntu/Mint)
The kubuntu-backports PPA may provide a newer Qt — verify the version it ships before relying on it.
Option 2: Using the Qt Online Installer
Install Qt into your home directory (e.g., ~/Qt/6.8.3/gcc_64). Because CMake otherwise defaults to the system-provided Qt, point it at the newer install with -DCMAKE_PREFIX_PATH:
Make sure the qtshadertools and qt5compat (or equivalent) modules are selected in the Qt Online Installer along with qtbase.
Note: GPU rendering also needs the private QtGui headers (qt6-base-private-dev on Debian-family, included by default in the Qt Online Installer).
Install (optional, Linux)
sudo cmake --install build
Roadmap
Currently in flight:
aetherd — a vendor-neutral IRadioBackend seam so radio-family logic
lives behind a stable interface. Three backends ride it today (Flex, HL2,
and the demo simulator); the remaining step is the versioned protocol that
splits a headless engine from thin UI clients.
Hermes-Lite 2 — an experimental non-Flex backend on that seam, now
running four independent receivers, the SSB voice chain, CW/RTTY decoding,
AX.25 packet, band switching with hardware filters, memory channels and
per-radio operating-state restore. Not yet a supported radio family:
remaining work is wider mode coverage, panadapter parity with the Flex path,
and hardening the raw-IQ DSP chain.
AppSettings nested-JSON refactor — the storage layer moved to SQLite with
per-radio versioned feature documents; the remaining work is migrating the
legacy flat keys.
TX DSP chain visual rebuild and the Flathub submission.
See ROADMAP.md for the full picture and the community backlog,
and the issue tracker for
everything else.
Contributing
PRs, bug reports, and feature requests welcome! See CONTRIBUTING.md for guidelines.
Development environment: AetherSDR is developed using Claude Code as the primary development tool. We encourage contributors to use Claude Code for consistency. PRs must follow project conventions, pass CI, and include GPG-signed commits.
Not a developer? Click the lightbulb button in AetherSDR's title bar to create an AI-assisted bug report or feature request.
Related projects
Aether-gate — put any radio into
AetherSDR. A bridge that presents an Icom/Kenwood/Yaesu/Elecraft CAT rig (via
Hamlib), an Icom LAN rig, or a SoapySDR dongle to
AetherSDR as if it were a FlexRadio — live panadapter, waterfall, and
frequency/mode control. Receive + control today (no transmit yet). By Nigel
Fenton (G0JKN); GPL-3.0-or-later. (aethersdr/Aether-gate tracks upstream
nigelfenton/Aether-gate.)
AetherSDR integrates radios that earn deep native support directly in-engine; the
gate covers the long tail of legacy/CAT radios and dongles.
Verifying Downloads
Linux and Windows binaries are GPG-signed. macOS artifacts are Apple notarized. Each release includes .asc signatures and SHA256SUMS.txt.
Bundled third-party libraries retain their own licenses (see each third_party//LICENSE), all GPLv3-compatible. Notably, on-device speech-to-text uses whisper.cpp and ggml (MIT) — vendored under third_party/whisper.cpp/ (Vulkan/Metal GPU backends included); Whisper model weights are downloaded on demand and are MIT-licensed, not redistributed in this repository.
AetherSDR is an independent project and is not affiliated with or endorsed by FlexRadio Systems.D-STAR is a registered trademark of Icom Inc. AetherSDR is not affiliated with or endorsed by Icom Inc.
