FreeDV
| FreeDV | |
|---|---|
| Type | Open-source digital voice for HF |
| Lead developer | David Rowe, VK5DGR, with a volunteer project team |
| Default mode | RADE V1 (Radio Autoencoder); first official release in FreeDV 2.0, June 2025; selected automatically since 2.2.0, January 2026 |
| RADE bandwidth | About 1.5 kHz |
| Other modes | |
| Legacy modes | 700D, 700E and 1600, hidden by default |
| Removed in 2025 | 700C, 800XA, 2020 and 2020B |
| Legacy voice codec | Codec 2 |
| Needs | An SSB transceiver and a computer with a sound-card interface |
| Equipment | |
| Software | FreeDV application for Windows, macOS and Linux |
| Hardware adapters | SM1000 and ezDV (legacy modes only) |
| License | LGPL 2.1 (application and legacy modes); BSD (RADE) |
FreeDV is a free, open-source system for sending digital voice over the HF bands. It runs on an ordinary computer connected to an ordinary single-sideband transceiver: the software turns speech into a digital signal, sends it as a set of audio tones through the radio, and turns the tones back into speech at the other end. Unlike most digital voice systems, every part of FreeDV, including the speech coding, is published as open source.[1] The project is led by David Rowe, VK5DGR, who handles the signal processing, with a small leadership team that includes Mooneer Salem, K6AQ, who maintains the application.[1]
How it works
A FreeDV station needs no special radio. The computer listens to the radio's receive audio and feeds transmit audio back to it, through a digital mode interface or a radio with built-in USB audio, in the same way as for FT8 or RTTY. The radio transmits that audio as a normal SSB signal, but what goes out is a data signal, not a voice.
FreeDV's older modes, now called legacy modes, use Codec 2, a low-bit-rate speech codec that Rowe developed.[2] A codec squeezes speech into a stream of bits, here at only 700 to a few thousand bits per second, and a modem then sends those bits over the air with error correction.[2] The best known of these modes are:
- 700D, released in 2018, which uses an OFDM modem (many closely spaced carriers sent at once) in about 1 kHz and copes with very weak signals.[3]
- 700E, a variant of 700D designed for fast fading, at the cost of about 3 dB more power.[3]
- 1600, with low delay but poor performance on fading paths.[3]
Codec 2 is also used on VHF and UHF by M17. That contrasts with D-STAR, DMR and System Fusion, whose AMBE voice codecs are proprietary.
Digital voice behaves differently from SSB. A decoded signal has no background hiss, but reception tends to be all or nothing: speech is either decoded or it is not. The developers rate the speech quality of the legacy modes well below that of RADE.[3]
RADE
Starting from an idea in late 2023, the project spent 2024 developing RADE, the Radio Autoencoder, which had its first official release in FreeDV 2.0 on June 4, 2025.[4][5] Since version 2.2.0 in January 2026 the legacy modes have been hidden by default and RADE V1 is selected automatically.[6][3] RADE was conceived by Jean-Marc Valin and David Rowe, and its development was funded by a grant from Amateur Radio Digital Communications (ARDC).[4] It does not use Codec 2. Instead it combines machine learning with conventional digital signal processing: a trained neural network maps speech features directly onto the radio signal, and a neural vocoder called FARGAN rebuilds the speech at the receiver.[4]
The developers report that RADE V1 delivers speech with 8 kHz of audio bandwidth in about 1.5 kHz of radio spectrum, roughly half the width of an SSB signal, and works at signal-to-noise ratios as low as -2 dB.[4] It needs more computing power than the legacy modes: a typical PC or a Raspberry Pi 4 is enough, but the SM1000 and ezDV adapters cannot run it.[4] As of October 2026 a second version, RADE V2, was in beta testing.[4]
In September 2026, ARRL News asked "Will RADE Be the Next SSB?", noting that RADE takes half the bandwidth of SSB, as SSB once took half that of AM, and reporting ARDC's word that manufacturers were beginning to build the mode into their radios, starting with a downloadable module for some FlexRadio models.[7]
Getting on the air
- Software. Download the FreeDV application from freedv.org. Version 2.4.0 was released in August 2026.[8]
- Radio settings. FreeDV follows the usual amateur sideband conventions: upper sideband on 20 meters and above, and lower sideband on 40 meters and below, except on 60 meters, where US rules require upper sideband. By default it sets the radio to its data mode (DIGU or DIGL).[3] Use the widest receive filter available, about 3 kHz, and turn off speech processing.[3]
- Transmit level. Set the computer's audio so the ALC is only lightly engaged. An overdriven FreeDV signal is hard to decode.[3]
- Finding stations. The application can report activity to PSK Reporter and to the project's own FreeDV Reporter website, which shows who is on the air and where. Both stations need reporting enabled to appear.[3]
- Activity times. A worldwide FreeDV Activity Day is held on the third weekend of each month, and regional nets meet on set days, mostly on 40 meters, with others on 80 and 60 meters. Their frequencies and times are listed on freedv.org.[1]
See also
References
- ↑ 1.0 1.1 1.2 FreeDV: Open Source HF Digital Voice for Amateur Radio, FreeDV project.
- ↑ 2.0 2.1 Codec 2, Rowetel.
- ↑ 3.0 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 FreeDV User Manual, GitHub.
- ↑ 4.0 4.1 4.2 4.3 4.4 4.5 RADE: Radio Autoencoder, FreeDV project.
- ↑ FreeDV 2.0.0 released, FreeDV project, June 2025.
- ↑ freedv-gui releases, GitHub.
- ↑ Will RADE Be the Next SSB?, ARRL News, September 2026.
- ↑ FreeDV 2.4.0 released, FreeDV project, August 2026.