Using Thetis with FreeDV

[JoeG]

Hi All,

The latest version of FreeDV has a new digital voice vocoder (RADEV1) which is now rivalling SSB for weak signal performance (-2 dB) providing virtually noise-free, 8 kHz audio bandwidth in a 1.65 kHz RF channel (using 33 x 50 Hz carriers).

By all accounts this digital HF setup works well on conventional transceivers with a sound-card (including IC-7300 SDR types with internal sound-cards etc) but it is reported, and I can confirm, that it does not work well using various SDRs together with Thetis, as follows:

Transmitting the FreeDV waveform using Thetis works fine and is well received by everyone, but receiving presents frequent SNR dropouts and bad decodes (sounding like gibberish) even on strong received signals. This is particularly prevalent during periods of fast, frequency-selective, ionospheric fading, probably caused by multipath propagation. Others, however, do not experience this problem during the same band conditions and signal strength. Then, at other times, the reception using Thetis is perfect.

My setup is: Apache Labs ANAN 200D - Ethernet cable - Intel i7 NUC PC - Windows 11 - Thetis 2.10.3.11 - VB Audio Cable using MME - FreeDV. The PC is running at 10% CPU. There are no VAC overflows or underflows. This setup works fine with WSJT-X and FLDIGI.

I have tried: Adjusting RX audio levels displayed on freeDV (low, mid, high) using Thetis AF and AGC gain controls. Adjusting: Thetis RX filter bandwidth; noise reduction settings; fast, slow and long AGC; 0 to 50 dB attenuation; Network Throttle Tweak, ANAN 200D Dither and Random options enabled/disabled; Filter low latency/linear phase; one and two VB cables; various VB cable gains, data rates, buffer sizes and mono/stereo select; all to no avail.

It appears that the main difference between the reported working and non-working systems is the use of Ethernet, Thetis and VAC. As general usage of this new digital mode is hotting up, I expect this may become a more significant problem. Can anyone provide an explanation, work-around or solution?

73, Joe VK3SRC

[w-u-2-o]

It appears that the main difference between the reported working and non-working systems is the use of Ethernet, Thetis and VAC.

Not at all. It's probably a Thetis set-up issue, and a subtle one at that.

Unfortunately I can't test this right now, as my station is in some disarray. However, I'm going to guess that it might have something to do with the filter type settings in Thetis > Setup > DSP > Options.

What filter type do you have set for Digital modes right now (this is automatically selected when using DIGL or DIGU)? Is it Low Latency? If so, try changing that to Linear, and set the filter size to 1024.

The reason I suspect this setting is the issue is because the low latency filter type has a less well-behaved impulse response than the linear filter type. This can have negative impacts on certain waveforms. For instance, when receiving analog fax transmissions, the low latency filter will cause ghost images ("echoes") in the received fax. Certain CW operators can hear ringing in the dits and dahs. So perhaps this might be an issue for FreeDV reception, as well.

[JoeG]

Hi Scott,

Thanks for the prompt response and an excellent solution!

What filter type do you have set for Digital modes right now (this is automatically selected when using DIGL or DIGU)? Is it Low Latency? If so, try changing that to Linear, and set the filter size to 1024.

I have set Filter Type to Linear Phase, but I did not try setting the Buffer Size to 1024. I have now set Buffer Size to 1024, as you suggested, and FreeDV appears to decode with out excessive errors! I will distribute your solution to others I know who are having the same problem and continue testing.

Thanks again,
73, Joe VK3SRC

[w-u-2-o]

That's interesting. Very steep ("brick wall") filter edges, even with the linear filter type selected, will provide a less ideal impulse response than less steep edges. However, the linear filter type is extremely well behaved in this respect. Nevertheless, it would seem that, empirically, going any steeper than 1024 does not work well for FreeDV.

It's worth nothing that, even at the minimum setting of 1024, the edges of the filter passband are much steeper than probably any other radio out there with the possible exception of Flex radios. In the old days some of us used to get a little crazy and run filter sizes of 4096, just because we could. But I can think of no case when it is truly necessary to run anything other than 1024 for any HF mode or waveform.