Voice Squelch Theory and Operation

[w-u-2-o]

Introduction

The new voice squelch feature introduced in Thetis 2.9.0.8 (final release) is, at the time of this writing and as far as I know, completely unique in any amateur (not commercial) radio transceiver. It is now a fully supported part of the WDSP software library (versions 1.21 and later) and thus could be incorporated in any SDR that uses WDSP.

The genesis of the design involved your's truly looking at some prior art that was available on the internet, all of which was realized in hardware, and discussing it with our esteemed Warren Pratt, NR0V, author of the WDSP library. I tried modeling some of these designs in LTSpice, and Warren tried a bit with software. Initially we could not get anything to function satisfactorily. In addition, all of the designs claimed to measure the presence of spoken syllables and measure the "syllabic rate". We found that was universally untrue, and thus avoided the term "syllabic squelch" in our discussions, preferring the term "voice squelch".

After some study, a workable design was found that provided excellent performance. Warren was able to quickly render that design in software for inclusion in WDSP.

Theory of Operation

Somewhat simplified, the voice squelch as implemented in WDSP works generally as follows:

- First stage: a simple thresholding function with gain (aka "squaring amplifier") to obtain only the frequency content of the signal, thus making it independent of amplitude.

- Second stage: a linear frequency to voltage (F-to-V) converter with a high frequency limit of 2KHz. Since normal band noise has content generally above 2KHz, part of the noise discrimination is the fact that the output of this stage is time varying with speech (and speech-like) signals, and is generally unvarying with noise.

- Third stage: a 10Hz low pass filter. Again, this helps to distinguish between the slower time varying output of the F-to-V converter commensurate with speech and the faster time varying output commensurate with noise.

- Fourth stage: peak-to-average ratio measurement. This further discriminates between speech and non-speech signals. Speech signals will cause larger variations in their frequency content than non-speech signals (like carriers or tones). Large variations equate to larger peak-to-average ratios on the low pass filtered F-to-V output.

- Fifth stage: thresholding. The mute/unmute threshold is set to a certain peak-to-average ratio.

- Sixth stage: mute/unmute time constants.

Operation

- The squelch button is now a three state button similar to the NR button. It can be off (SQL non-illuminated), in the traditional AM squelch mode (SQL illuminated), or in voice squelch mode (VSQL illuminated)

- The squelch threshold control represents an arbitrary threshold value in VSQL mode, and of course the normal received power level value in AM squelch mode.

- The metering is suppressed in VSQL mode as it has no meaning in that mode.

- Right clicking the squelch button will open the DSP > AM/SAM settings page for adjustment of both AM and voice squelch tail and time constant settings. For all settings I recommend the minimum value of 0.1 seconds (100 ms), however this can certainly be adjusted as you prefer.

Performance

The more steady the background noise level is the better voice squelch will perform. Performance on a relatively quiet band with NR2 activated is remarkable. With very small threshold settings in the range of 2 to 6, signals with as little as 6dB of SNR can be reliably processed. With slightly more noise, or without NR2 activated, squelch levels may need to be a bit higher, perhaps 10 or so, and this might prevent weak signals (SNR < 12dB) from unmuting.

With good band conditions and NR2 and VSQ enabled, performance is very good, almost FM-like. With ESSB signals, maybe better than FM!

For noisier conditions requiring thresholds in excess of 20 I have not found voice squelch to be very transparent, even with large signals.

If you have a very noisy band, for example 160 or 80M with a "heaving" noise baseline and static crashes, voice squelch is not really usable.

Finally, some non-voice signals will leak through. Typically ionosondes sweeping the band, data signals and that sort of thing. For carriers just turn on ANF as you normally would.

73,

Scott

[Joe]

Thanks, much work and thought put into this feature. Thanks Scott and Warren!

Thanks,
Joe
WD5Y

[Trucker]

Scott, I noticed last night on 20 meters several Ionosonde signals sweeping up and down the band. Listening to an ongoing QSO I never heard the familiar blip sound as the signal passes thru the passband. No disruption of the audio. Same thing with static crashes from thunderstorms that have been moving around and through north Texas.
I have been throughly impressed by the Voice Squelch used in conjunction with NR2.
It almost makes turning on and listening to my other radios annoying!
As you say,nothing else like it in Amateur Radio.
James
WD5GWY

[w-u-2-o]

Hermes Lite thread drift posts moved to: viewtopic.php?f=9&t=4682