SSB demodulator too sensitive since 2.15.10

[AsciiWolf]

The SSB (and to a lesser extent also AM) demodulator seems to be too sensitive since Gqrx 2.15.10 and there is now too much clicking / crackling in the demodulated audio.

Here is an example recording of our local beacon received with Gqrx 2.15.10 on Fedora Linux 38.

Here is an example recording of our local beacon received with Gqrx 2.15.9 on Fedora Linux 38.

Gain in both cases was set to -10.0 dB. Lowering the gain more has a slight effect, however there is still audible clicking / crackling even with -20.0 dB when using Gqrx 2.15.10.

It sounds even worse with gain set to 0.0 dB:

Example recording with Gqrx 2.15.10.

Example recording with Gqrx 2.15.9.

[vladisslav2011]

Hello. Could you revert this commit https://github.com/gqrx-sdr/gqrx/commit/c8c1e3f1e81d0f564e317234a6f70d4b296c6f3f and recheck?

[AsciiWolf]

Thanks! I can confirm that reverting the commit did solve the issue. Here is also a screenshot from Audacity:

On the left is latest master with https://github.com/gqrx-sdr/gqrx/commit/c8c1e3f1e81d0f564e317234a6f70d4b296c6f3f reverted, on the right is latest master without any changes.

edit: Personal note for me, feel free to ignore. :-) To build Gqrx on Fedora 38, it is needed to run cmake with -DJACK_LIBRARY=/usr/lib64/pipewire-0.3/jack/libjack.so.

[vladisslav2011]

There is alternate AGC implementation, that solves audio clipping issue independent of demodulator (NFM/WFM may have this issue too): https://github.com/gqrx-sdr/gqrx/pull/1096 The PR sequence needs to be updated with at least these commits: https://github.com/vladisslav2011/gqrx/commit/dc84c413e83e77c7ff53162376a58c5a888a0636 https://github.com/vladisslav2011/gqrx/commit/ccfe3c2aaeed7c73331c9fa712440af14f34c43b https://github.com/vladisslav2011/gqrx/commit/1e84eed1d90a9e467b1874b78f71407849c65e63 https://github.com/vladisslav2011/gqrx/commit/18e56672b3b3ff0bc5dfebc5ba0be000f4f5f7ae You may try to rebase the PR seqence on top of current master, cherry-pick given commits and test...

[AsciiWolf]

I am sadly overwhelmed by other work tasks at the moment. I may try working on it later, however I think that for now it would be the best to just revert https://github.com/gqrx-sdr/gqrx/commit/c8c1e3f1e81d0f564e317234a6f70d4b296c6f3f and maybe also release a new bugfix version of Gqrx.

/cc @argilo

[vladisslav2011]

Reverting c8c1e3f1e81d0f564e317234a6f70d4b296c6f3f would bring back https://github.com/gqrx-sdr/gqrx/issues/1055 , so I don't think that it is proper solution...

[argilo]

The change made in #1151 was intended to reduce CPU utilization while leaving the AGC output unchanged. It's possible that a bug was introduced. I'll see if I can spot anything.

[argilo]

Aha, I think I spotted the bug: m_DelaySamples and m_WindowSamples are calculated from the old m_SampleRate before it is updated:

https://github.com/gqrx-sdr/gqrx/pull/1151/files#diff-04098d6b048bd5018fd5714016059258c98d0ab34f99aa6f3908abe1754d9508R142

@AsciiWolf Could you try replacing m_SampleRate with SampleRate on lines 142 and 143 to see if that resolves the issue? Thanks!

[argilo]

Another (potentially better) fix would be to move lines 142 & 143 after line 148.

[AsciiWolf]

I can confirm that moving lines 142 and 143 after line 148 did solve the issue:

[argilo]

Thanks for confirming! I'll get a fix in shortly.

[argilo]

Fix: #1230.