hamradiotools

Nexus 0.20.0

Released 28 July 2026. The current release is 1.24.0. Download it · Every version

Fixed: JT65 could crash Nexus outright, and it is transmitting again

On Windows, pressing Call CQ on JT65 killed the app the moment the transmit cycle came round — before the radio was keyed. Transmit was switched off in 0.19.17 as a stopgap. The cause is now found and fixed, and JT65 transmits again.

Nothing was ever wrong with the transmit path. The crash came from the decode that runs at the same instant, which is why it looked like a transmit bug and why it appeared right when you pressed Call CQ.

Nexus decodes a full minute of audio for JT65. When it has not yet collected a full minute — the first minute after you select the mode, or after the buffer is reset as transmit begins — it pads the front of that minute with silence. Past about 28 % silence, a brightness reference inside the decoder went to zero, everything downstream became "not a number", and a peak-search step then read from an essentially random memory address. On Windows that is an instant, uncatchable process kill. On Linux the same code happened to land somewhere harmless, which is why it never showed up in testing here or in CI, and why only one mode was affected: this sync code is JT65's alone, which is what kept Q65 at the same 60-second period working perfectly throughout.

Three fixes: the reference can no longer be zero, the peak-search variable can no longer escape unset, and a second variable on the same path with the same flaw was closed too. A partly-filled minute now simply reports nothing, quietly. Both defects are inherited from upstream WSJT-X, which never meets them because it only ever decodes a full window of live audio.

Added: native crash reports on Windows

When Nexus dies from a fault in the DSP layer rather than a normal error, Windows tells you nothing and the window just disappears. Nexus now writes nexus-crash.txt — beside the program, or in your %TEMP% folder — naming the component at fault and the call path into it. Sending that file with a bug report turns a crash like the one above from a multi-day hunt into a single look. It records only addresses and module names: no callsign, no log, no personal information.

Six more modes now transmit

Nexus decoded eight WSJT-X modes. It now transmits six of them: Q65, FST4, FST4W, MSK144, JT65 and WSPR, alongside FT8, FT4 and the Tempo tiers.

Every waveform was checked by generating a transmission in Nexus and having stock WSJT-X decode it, rather than by testing Nexus against its own decoder — both halves come from the same vendored source, so a shared misreading would pass unnoticed. That is not hypothetical: FST4 at the 15-second period was going out half a second late and every in-house test passed, because the transmit duration and the modulation start time are two different numbers in the upstream source. Stock WSJT-X reported the offset. Q65's waveform was additionally compared sample by sample against WSJT-X's own generator and matched at 0.9985 correlation.

JT65 is the exception: upstream's JT65 decoder depends on KVASD, a non-free component Nexus does not ship, so there is no stock decoder to check against. It is verified by round-trip against WSJT-X's own signal generator instead.

Each mode keeps its own operating rhythm rather than inheriting FT8's. MSK144 waits twelve transmit periods before giving up on a contact, against three for FT8, and its CQ runs are uncapped — on meteor scatter silence is normal rather than a sign the other station has gone, and FT8's settings abandoned live contacts. WSPR and FST4W never touch the QSO sequencer at all; they transmit on a percentage schedule, and below 40% avoid two transmissions in a row while still hitting the requested rate.

Every mode now lands on the right frequency

Mode frequencies are read from WSJT-X's own frequency table rather than typed from memory. Previously every new mode inherited FT8's list, which is wrong for most of them: MSK144 and Q65 have no HF presence at all, FST4 and FST4W are LF and MF, and WSPR on 20 m is 14.0956 rather than 14.074, so "20 m WSPR" was listening to FT8. Selecting a mode with nothing on your current band now moves the radio to that mode's own calling frequency.

Transmit safety

A review of the transmit paths before any of this reached a radio found four real defects. The most serious: entering the Phone, CW or RTTY section arms transmit for you, and the beacon path was being reached before the check that stops digital modes keying while those sections own the radio — so a configured WSPR beacon would key on schedule while the operator worked SSB, putting 111 seconds of data tones into the 20 m phone band.

Also fixed: the transmit watchdog did not cover beacons and could not bound a long transmission; "Transmit 0%" did not stop a beacon with a Round Robin slot configured; and switching modes mid-transmission did not release the radio.

Selecting a receive-only mode and pressing Call CQ used to report that calls were going out while nothing was transmitted. Modes that cannot transmit now say so.

Fixed

  • A second radio that was switched off could spawn a CAT process every second, forever.

    There was no retry backoff. On Windows this is expensive process creation plus a 12 MB driver library re-scanned by antivirus each time, so it appeared as system CPU rather than as Nexus. Retries now back off to once a minute and recover when the radio returns.

  • A decoder crash could silently stop all receive.

    The app kept running and the waterfall kept painting, so it looked alive while it had gone deaf until restart.

  • A slow decode could delay or prevent a transmission.

    Modes other than FT8 and FT4 waited for the previous period's decode before keying, so the over went out late or, on longer modes, not at all. All modes now key at the slot boundary, as WSJT-X does.

  • Mode settings now take effect immediately.

    Changing a Q65 period or JT65 submode did nothing until you switched modes and back, while the rest of the app reported the new value.

  • The Phone cockpit gained its ⊞ Panels menu

    , which CW, RTTY and SSTV already had.

Program tells you when the repeater list is missing a band

The Program section's default source, hearham.com, is an open directory with real holes in rural country. Around Bozeman MT it lists nine repeaters and not one of them on 2 m, which is not a true description of Montana. That is worse than a short list: a channel list with no 2 m on it looks finished, and you find out it wasn't when you key up and nobody answers.

Program now checks the results for a major band with nothing on it at all, and says so, pointing at the RepeaterBook token in Settings as the fix. It looks for a missing band, not a low count — genuinely empty country stays balanced across 2 m and 70 cm (Amarillo TX has three of each), so counting repeaters would cry wolf in the plains while staying quiet where the data is actually wrong. It also counts what the directory lists rather than what is on the air, so a town whose 2 m machines are simply off-air, as Fairbanks AK's are, does not trigger it. Checked against the full 22,574-record hearham feed at eight locations, it fires at one.

Fixed — the app and the README disagreed about where repeater data comes from

Settings told you the Program section "gets RepeaterBook data through Nexus's shared access automatically". It does not. Shared access is still pending RepeaterBook's approval, so every install has been using hearham.com, and the README described a third arrangement again. Both now say the same true thing: hearham by default, your own RepeaterBook token if you add one, shared access when and if RepeaterBook approves it.

Every version