RADE Digital Voice

RADE Digital Voice (FreeDV Radio Autoencoder)

AetherSDR includes RADE v1, an AI-based digital voice codec from the FreeDV project. RADE uses a neural network to encode speech into an OFDM waveform for transmission over HF radio, then decode it back to speech on the receiving end.

Contributed by @pepefrog1234

How It Works

The radio is set to DIGU or DIGL (SSB passthrough) — it just transmits the raw OFDM waveform. All the encoding and decoding happens on your PC:

TX path: Mic (24kHz) → downsample to 16kHz → LPCNet feature extraction → RADE encoder → OFDM modulation (8kHz) → upsample to 24kHz → VITA-49 → radio transmits

RX path: DAX audio from radio (24kHz) → downsample to 8kHz → RADE demodulator → neural decoder → FARGAN vocoder (16kHz) → upsample to 24kHz → speaker

Enabling RADE

Select RADE from the mode dropdown in either the RX applet or the VFO widget. The radio automatically switches to DIGU (≥10 MHz) or DIGL (<10 MHz) for passthrough.

RADE mode appears in the mode list only when AetherSDR is built with RADE support (enabled by default on Linux and macOS).

Requirements

• Both stations must be running RADE — one station transmits RADE OFDM, the other decodes it
• Compatible with other FreeDV RADE v1 implementations (FreeDV GUI, etc.)
• Internet access required at build time (Opus is auto-downloaded during cmake configure)

Technical Details

• Codec: RADE v1 (Radio Autoencoder) by David Rowe VK5DGR
• Library: radae_nopy (BSD-2 license), bundled
• Dependency: Opus with FARGAN/LPCNet (auto-downloaded and built via ExternalProject)
• Resampling: r8brain-free-src polyphase resampler for all rate conversions
• Frequency selection: DIGL below 10 MHz, DIGU at 10 MHz and above
• Filter: 0-3500 Hz passband for DIGU, -3500-0 Hz for DIGL (set automatically)

Platform Support

Platform	RADE Support
Linux	Yes (bundled, auto-built)
macOS	Yes (bundled, auto-built)
Windows	Not yet — Opus build requires autotools (see #87)

Sync and SNR

The RADE modem uses pilot symbols for synchronization and can acquire signals down to approximately 0 dB SNR. A visual sync/SNR indicator in the VFO widget is planned (#88).