Audio processing at K2ENF

 Well, it took a while, but I've got the external audio processor running correctly.  There's still some tweaking to do, but it's mostly there. I've made some mods to the original design for a bit of simplicity and because of newer hardware now in the shack.

So, here's the path:

Whatever mic I'm using, going to a Gemini GEM-05USB mix console. This brings the level up to common line level, but it also has an EQ in it, which eliminates the need for the EQ I had designed into this stack. 

Then, we have a Behringer Composer Pro-XL (MDX2600) This is a noise gate / compressor/ limiter/ De-esser, This keeps a much hotter input level to the radio, on average, and takes away the heavy lifting from the ALC. The result is a nice dense audio, without the overshoots the ALC is frankly too slow to respond to. (I'll expand on this thought, shortly)

And of course the 991 has an internal EQ as well, so for the moment I'm not running a second EQ on the output of the 2600, though I may tinker with that later on.  The EQ's are set up mostly for bandpass, but also for a little punch.  

Aside: With the two EQ's one on the 2600 input and one on the output, 'm quite sure I'm getting some weird phasing within the band-pass, but not enough to worry about since most of that's being swallowed up in the phase shifting the 2600 is inducing... nobody's going to notice. And, anyway, there's some advantages to phase scramble, in terms of filling the RF envelope. 

 Here's why: The human voice is notoriously asymmetrical. If you look at most folks voices on a scope,you'll notice that the voice tends to swing either more to the positive side or more to the negative. This asymmetry is hugely inefficient when you're talking about maximizing a signal, particularly AM or SSB where the audio IS the power.  With most voices, you leave 25 to 30% of the RF envelope unfilled by audio.A bit of phase scramble over the width of the bandpass tends to counter that. tends to counter that.

Why would an AGC cause phase changes? There are very few phase stable AGC's, in my experience. Ask any broadcast engineer who has dealt with multi-band processing about that.
 

The result of all of this is quite encouraging... I'm seeing higher average power, on the order of a couple dB, and I'm also noting my bandwidth consumption has gone down somewhat, even with the higher amount of audio on air... to the point where I'm now easily within the 2800 hz bandwidth I've been shooting for, whereas before I was seeing ALC overshoots getting out from under the skirt, particularly when I was "getting on it. " Now, no such problems. I've said it before and I'll say it again... the ALC in most transceivers simply isn't up to the drill of maintaining max audio without getting nasty side effects.  I expect the de-esser on the 2600 is helping there, as well.

Next, now that the basic framework is laid out and working well...

I'm considering a DF4ZS RF speech processor. to be placed at the output of the 2600. It's essentially an RF Clipper.. For those who don't know what an RF clipper is, it's basically a transmitter receiver combination.What happens is, line audio is fed to the transmitter side and the audio is run to saturation level of the internal (AM) carrier..  The receiver portion detects the processed audio, and converts it back out to line level, which in my case would be fed directly to the 991a. 

Here's a nice techie explanation of the concepts involved. It explains that AF clipper does work, but has serious limitations.  Tom Kneitel, K2AES, (SK) wrote an article years ago on these topics that alas, now, I can't locate. He called the RF Clipper the "Poor man's linear", and he was right.

. Those beasts are very effective, but rather touchy once you venture outside of their level "sweet spot". Now that the 2600 is in place and working well I should have no trouble with keeping level in the sweet spot of the clipper.  I figure it'll add perhaps 5db worth of talk power.

(Yeah, they advertise up to 9db.... but I figure some of that ground has already been achieved by the stack as I've already described to you, and a bit cleaner, since the clipper only uses clipping as it's gain reduction, the 2600 doesn't. I won't need to drive the clipper nearly as hard for the same result. No danger of over driving the clipper with my current setup. .)

Why go through all of this?

For starters, consider the FCC's RF exposure paperwork, (Which the ARRL linked here) which suggests that conversational SSB has a 20% duty cycle. Supposedly heavy processing brings that up to 50%. In other words, even with the (rather lame) procs in most HF transceivers, the proc adds another 30w of average power in a 100w setup, better than doubling the power and raising the signal on the received end, what, about half an S unit?

Trouble is, as I've already suggested,  pushing most of the procs on even today's HF radios that hard for that rather modest gain, you'll end up taking more bandwidth than most consider wise. The problem is that ALC is used for most of the heavy lifting, and as implemented in even modern day transmitters, it's simply not up to the drill. The ALC lets a fair amount of signal peaks slip by it which results in audio clipping. Drive the ALC hard enough to get a 50% (Or more) duty cycle, and you'll be clipping the bleep out of your audio.... and the filtering following the ALC is, if it exists at all, is simply not up to the drill of keeping your signal "in bounds" under those conditions. That's without even mentioning the other artifacts caused by such a setup.

Consider Collins and the work they did years ago.Example, "Single-Sideband Systems and Circuits by W E Sabin and E O Schoenike", ( ISBN 0-07-054407-7)

(I gather they were working for Collins at the time of publication.)

As dated as that work is, most transceivers, even to this day, do not employ their advice on design and implementation of processing. In short, there's a LOT of room for improvement. And of course with technology improving rather a lot since those days... (including better output filters) the improvements in performance can (and, I think should) be even larger. In short, your signal is not nearly as efficient and clean as it could be.

I suggest reading the work of Leif Asbrink, SM5BSZ for some background. I found this article, particularly interesting.

Asbrink makes mention of that same book in his article, by the way. It, too, is a little dated but since so little has changed in audio proc design on HAM transcivers since then, it's still valid. One of the big points made in this article is that only about 2 or 3 db of ALC action is required for their to be undesirable out of passband products. Again, ALC overshoots. The reason is most ALC's simply are not fast enough to prevent overshoots and thereby clipping which results in splatter given the weak passband filtering is most radios, even today. You simply cannot get the best, most dense audio signal out with that setup. (and by the way, I know of blessedly few HAMs who run only 2-3 db of ALC as a rule.)

Understand me; Part of the object of adding external processing, particularly peak limiting, is to eliminate those overshoots, by removing them before the ALC even gets involved, which in turn allows for "filling the envelope" more fully without the problems ALC overshoots cause, thus allowing for a more effective signal for a given power level.

(I should, I suppose, note that SDR  radios with their digital bandpass filtering (IC7300, FT991, etc) have a tendency to handle such overshoots a bit better than units made even 20 years ago. Still, why allow such overshoots at all?)

And yes, I've heard it before.... SSB has a horrible S/N ratio, you're never going to get broadcast quality audio from that setup, etc. The. thing is that idea ignores that with decent processing at (or in front of) the transmitter, one can attain a much higher S/N ratio. Even doing what Sabin and Schoenike suggested 35 years ago adds 3db to the S/N ratio, or a half an S unit on average.... and that's without any serious compressing, peak limiting and so on, before the audio even gets to the radio. With that combo, with that external processing, I see no reason the average power couldn't be much higher, while still maintaining intelligibility.

That last word... intelligibility.... is critical to understanding what I'm about, here. I'm not in pursuit of broadcast quality audio. For openers, that's flat out impossible in a 3KC bandwidth and it's also counter productive to the goal of maximizing weak signal intelligibility.

What I'm about with my audio experimenting is making the most of available RF power while adding as few undesirable artifacts as possible, which in turn makes the difference between making the contact and being buried in the noise level.

So far, it's working.

73, K2ENF

 

Update:

 As of 8/6/23

I'm pleased with the progress I've been making. I've had a few comments that it tends to sound a bit harsh when in full cry, but it's now made the difference in making the contact or making noise several times. It does add some significant average power, but at this point I'm approaching diminishing returns. 

I must say that this is the first time I've been somewhat disappointed in the documentation surrounding the 991. I would have assumed that the internal audio processor used some kind of audio clipper as well as the internal compressor and ALC. Trouble is, I'm not seeing any clipping on any but the strongest peaks, which for the most part the external processor I'm using eliminates. A little clipping, perhaps up to 3dB or so would be desirable as a wave shaper, and watching the waveform I'm not seeing it on the 991's output.  That wave shaping is the only thing my audio lacks....

The other issue is that once the average audio power being applied to the radio goes up, by means of the external proc, the internal compression reacts too harshly. I'm seeing 15dB or so worth of compression  on the internal compressor, which would be a serious issue were it not for the noise gate I'm running on the input.... all you'd be hearing is background noise.  

I've tried turning the output of the stack down, and even cut the radio's mike gain to near nothing. But, while that lowers the amount of internal compression applied, while still giving a nicely dense audio signal,  it produces an actual loss of audio in the RF envelope less watts to the antenna.  

 Frankly, I wish there was a way to adjust the parameters of the compression in the radio.  Instead it's a sort of black box I can't access. So far, the working solution has been a rather strong setting on the noise gate on the external proc. This cures the issue of the internal compressor riding on the noise level during voice pauses, but the level ends up being a little jumpy as a result, precisely what I've been trying to avoid using this system in the first place.

It's all a learning process, but as I say I'm pleased with the gains I'm making so far. 

Addendum: 9/16/23
Well, I got rid of a lot of the crud I had been chasing. While I had thought it was the 991's clipper, it turned out to be a grounding issue with the mix console. All set now, however. Getting good reports.