This past month, I focused mainly on getting version 2.4.0 released. This involved resolving remaining bugs, as well as enabling Profile Guided Optimization (PGO) on Windows and macOS to improve performance. Some bugs that were fixed include:

• Fix FreeDV Reporter column-order corruption and Last TX column width
• Remove CLIP indicator from main window.
• Fix waterfall/spectrum filling solid colour during half-duplex TX
• Windows/macOS: Zero audio on startup.
• Fix issue preventing TX thread from sleeping.

In addition, work on improving the behavior and usability of the Level meter for the upcoming version 3.0. This involved having the Level meter take audio directly from the microphone (previously, it was taken post-AGC) as well as having the meter use log scale instead of linear scale.

For the next month, work on integrating RADEV2 with FreeDV and tests will continue.

More information can be found in the commit history below:

• freedv-gui
• freedv-backend
• freedv-integrations

(Note that all commit logs above were generated with the following command line:)

git log --author="member@email" --after "Month 1, 2026" --before "Month 31, 2026 23:59:59" --all

A quiet month for me, as I spent much of August hiking the Larapinta trail, a rugged 230 km walk in the middle of Australia.

However I did have some time for a little FreeDV work. I spent some time finalising an upcoming QST article on RADE, working with the good people at the ARRL.

Most of the time was spent tracking down a subtle issue thrown up by verification procedure – the “loss” metric of the C version of RADE V2 was found to be sensitive to the timing offset (thanks Mooneer for finding this one). Turns out this was an AGC algorithm error, which I have been working on correcting, testing, and documenting. I’ve had this experience before – when e.g. when manually porting between Octave and C for previous FreeDV modes.

These days with agentic coding the port is (e.g from Python to C) quite fast (~ 1 day compared to months doing it manually) – but for these subtle bugs a human guiding the AI was definitely required to track down the root cause. Claude, however, deserves the credit for suggesting we try running without AGC which finally trapped the issue.