FreeDV will once again have a booth over at Pacificon in San Ramon, CA (in the Contra Costa Room) on October 16-18, 2026. Mooneer K6AQ will be there (as well as give talks at the Zero Retries Digital Conference and Pacificon itself), so come by and chat about FreeDV! More information at https://www.pacificon.org/ and https://www.zeroretries.org/p/conference.
NEW FOR 2026: The Pacificon special event station (W1AW/6) will have FreeDV installed on their 15 meter station. If you’ve ever wanted to make QSOs as (or with) W1AW using FreeDV, now’s your chance. FreeDV will be active at 2pm on Friday and Saturday of that weekend, and there may be at least one more time slot (TBD). Hope to hear you on the air!
Every so often a mode comes along that genuinely threatens to reshape how we talk on HF. Fifty years ago, that mode was single sideband, which overtook AM as the dominant voice mode largely because it packed a signal into roughly half the bandwidth. Now there’s a new contender aiming to do the same thing all over again.
The excellent Zero Retries newsletter often covers FreeDV and this week is no exception.
The voice quality of FreeDV RADE is arguably superior to the voice quality of some of the digital voice modes intended for VHF / UHF. Even the Amateur Radio digital voice mode with the narrowest bandwidth – D-Star using a 6.25 kHz channel seems inefficient when you consider that FreeDV RADE over SSB uses a 1.5 kHz channel. FreeDV over SSB is 4x more bandwidth-efficient, and 4x better power spectral density, and probably better battery life compared to FM or digital voice.
Thanks Steve and Tina for your excellent ongoing coverage of Amateur Radio.
This past month, I was finally able to verify the official reference FreeDV application (“freedv-gui”) with RADEV2. With this, we’re now able to begin testing of RADEV2 with volunteers running this version of the software. So far, there have been mainly minor issues with the application itself and not the performance of RADEV2, which is encouraging. However, it’s still fairly early in the testing process, so there is still time for issues to come up.
Additionally, with the help of Claude, I’ve been able to finally do some optimization of the FreeDV application that I haven’t had time to do perviously. For example, it was able to attain an over 50% improvement in CPU usage on macOS in rendering the various GUI plots and other UI elements. Smaller improvements were also made for Windows and Linux (when using Wayland directly instead of Xwayland).
As another example of what AI is able to do, I also pointed it at FreeDV’s audio pipeline and it was able to find some additional optimization opportunities there. For instance, libsamplerate (which is what we use for audio resampling) has traditionally used as much or even more CPU than RADE itself for transmit. Claude simply analyzed the code, suggested some fixes, ran benchmarks to confirm and as a result, libsamplerate no longer shows up as using significant CPU usage.
Next month we’ll be at Pacificon talking about the new things going on with FreeDV as well as continuing testing of RADEV2.
More information can be found in the commit history below:
This month I spent some time carefully performing over the cable (OTC) and over the air (OTA) tests using the latest C version of RADE V2. These tests involved the use of real SSB radios and built confidence that RADE V2 is working over real radio hardware and channels. To ensure these tests were decoupled from any application bugs I used the radae/ota_test.sh test script which employs command line sample play/record utilities. The OTC test results are written up in a verification test report.
One key take away is that with a verified application and careful SSB radio set up – there is very little distortion when the RADE signals are passed through real world SSB radios.
Here is a comparison of SSB/RADE V1/RADE V2 at very high SNR (software simulation), representing the best sound quality possible:
SSBRADE V1RADE V2Spectrogram. From left to right: calibration chirp – SSB – RADE V1 – RADE V2. The narrow bandwidth and clean spectrum of V2 can be clearly observed.
Here is the same sample sent over the air at about 10dB SNR, over a Ground wave plus NVIS path from my home to a nearby SDR:
SSBRADE V1RADE V2
Spectrogram of off air signals – note interfering carriers around 1900 Hz, and light NVIS fading.
During my tests I discovered the transmit band pass filter was not implemented correctly on the C version (somehow omitted during the agentic AI C Port). So I set about implementing and testing the filter. No difference in performance (as expected), but the Python and C are now consistent, and the C Tx spectrum slightly improved.
I also worked with Peter and Mooneer to review their verification reports that demonstrated successful integration of freedv-gui and FreeDVNeo with RADE V2. By working together we managed to fine tune the verification procedure, find some bugs in RADE V2, and (I presume) their applications and test procedures.
This verification work was the gateway to starting our RADE V2 beta test campaign, which is now in progress. In this campaign we are testing real time use of RADE V2 in over the air PTT conversations.
Through a dialog with a Google Gemini AI, I gained a better understanding of several properties of RADE V2 that had emerged through training, but were puzzling to me. One example is the benefit of the inherently low PAPR and bandwidth of V2 compared to SSB and traditional OFDM when driving typical Ham PAs which typically have modest inter-modulation performance. The take away is less interference on the bands with RADE V2 than compressed SSB due to “splatter” – another bonus.
The combination of my professional knowledge (e.g. I had to correct the AI on several points) and an AI that has “read every text book and paper” was quite powerful, and helped me better understand how RADE V2 works, and tease out some new benefits. Without the professional knowledge, the converse is true – use of AI can accelerate work down unhelpful paths. This is new technology, and how to use it effectively is a work in progress.
I have been updating and editing a draft paper on RADE V2, including the latest changes in the algorithm and the new take-aways from the AI discourse.
Finally, the rade_c library has been extended to build and test for win32 and macOS, in addition to the reference Linux.
