Pleasant distortion without clipping

[artifus]

yeah but what happens when i bend down from a flattened ninth to an augmented seventh over a passing fourth? does it go to eleven? 

[R.G.]

What I think is missing is a dynamical way of controlling gain, compression or expanding the signal to keep it in that good region for long enough time.

Hmmm. Almost sounds like a case for companding before and after distortion, doesn't it?

Well yeah, when the ratio of your loudest to quietest discernable signal amplitudes exceeds 80dB (i.e. 10,000:1) then keeping your signal in a tenuous "sweet spot" is a challenge. 

Hmmm. Almost sounds like a case for companding before and after distortion, doesn't it?

experimentation suggests to me that it's about far more than simple static curves. Pre and post eq plays a huge role

I've been suggesting for years if not a decade or more that distortion experimenters go get a surplus/cheap stereo graphic equalizer and put one channel before their distortion, one after. It's an easy way to spend a whole weekend making ever-more beautiful distortion sounds like snowflakes - no two the same, and probably no way to get back to the last one you loved, however fleetingly.

[Thecomedian]

EQ stack pre-distortion = selective band distortion, more or less distortion for frequency range.

possibly.

[WaveshapeIllusions]

Companding sounds like a great way to get it to sound right. Compress the signal to keep it in the right region; impress the input signal amplitude on the distorted side. Of course, with enough stages of a few percent distortion the desired effect would be fine. I think four might be enough. My current pre has three (two common source and a follower) and it sounds like its almost there. It clips on every low note attack though, so I think it needs more interstage attenuation. Just enough to minimize clipping though. Those clipped transients add to the sound.

I actually don't get the soft clipping tube vs transistor comparisons. On the amp comparisons, tube amps traditionally have minimal feedback. Tranistor amps have lots, so of course they clip suddenly. If you use tubes with that level of feedback their clipping performance is almost identical. A single transistor is quite a bit rounder when it saturates.

On the EQ idea, I've been meaning to put something like that together. I was thinking that a state variable filter with variable frequency would be good on the input. Possibly two peaking parametric EQs for the mids. I'd say the output shaping has a bigger effect on the tone than the input, so it would need more dramatic shaping options. Probably two shelving, maybe three. Three or four peaking. All parametric of course. If I wanted to be really crazy maybe one peaking that tracks frequency so that you could accentuate certain harmonics of the note.

I need to draw up a schematic for this. Call it the most tunable distortion.

[tca]

I've been suggesting for years if not a decade or more that distortion experimenters go get a surplus/cheap stereo graphic equalizer and put one channel before their distortion, one after. It's an easy way to spend a whole weekend making ever-more beautiful distortion sounds like snowflakes - no two the same, and probably no way to get back to the last one you loved, however fleetingly.

Specially the mids... from 300Hz to 2kHz, I had the same experience with a simple two pots position/cut mids.

[gritz]

Hmmm. Almost sounds like a case for companding before and after distortion, doesn't it?

Have you ever actually tried it?

I have and it sounded pish (to me).  

It's all subjective, of course - we are subconciously "trained" by years of experience and cultural influence  to expect guitar to sound / behave in a certain way and dynamic range compression is all part of the distortion sauce. By contrast, in electronic music land one can apply all sorts of "wrong" (flabby, fizzy or just plain broken) distortion to a synth or sample and in the right context it can sound awesome.

Anyway, companders. These will add time-variant amplitude discontinuities (and noise, most likely). In software we have the luxury of tricks like Hilbert transforms and lookahead that makes level detection for (fairly) blameless companding less of an ordeal, but in analogue hardware? Still, perhaps the results of the compander "getting it wrong" on transients might make it more interesting. Harmonic distortion is just harmonic distortion, after all. More or less of the same odds and / or evens, with maybe more (or less) up top. Is there a Holy Grail? I'm not sure. Are there better places to look for one? Quite possibly, I'd say - unless you want to make a guitar sound less like the guitar we all know, or you're just curious. But there ain't no point in being curious if you're not going to do anything about it, so maybe it's time to stop speculating and get building...  

[R O Tiree]

EQ stack pre-distortion = selective band distortion, more or less distortion for frequency range.

possibly.

Like the RAT's reduction of bass in the distortion cell? I suspect that this was designed to try to stop the bass strings swamping treble freqs as the chord dies away some seconds after pick-strike.

As to the compression inate to "classic" distortion circuits, how about an envelope detector at the input to alter the gain of the output buffer/gain recovery stage? You would get the gritty sound we all know and love, but playing dynamics would be in there as well...

[R.G.]

Have you ever actually tried it?
I have and it sounded pish (to me).  

Yes, I have. It is quite tricky to get normal companders to get the amount of compression and expansion right for a good guitar sound. And as always "good" and "sounded like" are very, very subjective.

It's all subjective, of course

... as you note. 

Anyway, companders. These will add time-variant amplitude discontinuities (and noise, most likely).

A well-done compander will match its compression and expansion time constants to the source and the listener's preferences. This is (a) tricky and (b) subjective again.

And noise. I found I needed to bring guitar up to line level first, then compand to get the compander I was using - the 571 - to work well. Trying to use the compander all by itself injected noticeable noise, pre-amplifying much less so. Of course, since distorting the signal is level dependent, that required reworking the distortion, too.

And, after all, what I said was that the post I replied to sounded like a case for compansion, no? 

Still, perhaps the results of the compander "getting it wrong" on transients might make it more interesting.
...
Harmonic distortion is just harmonic distortion, after all. More or less of the same odds and / or evens, with maybe more (or less) up top. Is there a Holy Grail? I'm not sure. Are there better places to look for one? Quite possibly, I'd say - unless you want to make a guitar sound less like the guitar we all know, or you're just curious.

Practically everything we do to guitar today makes it sound less like the guitar we know. Distortion lops off most of the excursions and transients by its nature. The holy grail of most metalheads is "sustain for days" and "more gain", meaning "more distortion" as we now know. In fact, we have a paucity of new places to look for effects these days. We distort the waveforms, mess with the attack time and decay time with attack delays and compression sustainers, fool with the harmonic content with pre-filters, EQs and post filters, time delay, wobbling frequency peaks and notches, even deliberately adding noise in two cases I remember.

In retrospect, the idea seems to be to change something about a guitar's sound to make it sound less like the guitar we all think we know, but to do that in an interesting and arguably new(ish) way that's not too far from the bounds of "guitar as we know it", but far enough to sound novel.

So - are there better ways? I'm sure there are. But the DIY community is an expanding wavefront in all directions today. The trick is to be new, but not too new, different but not too different, a true "don't care whatever happened before" rebel, but with a "vintage" sound somehow. It's a little schizoid, no?

And "transparent". I forgot "transparent".   

But there ain't no point in being curious if you're not going to do anything about it, so maybe it's time to stop speculating and get building...

I heartily recommend that. Been doing that for nearly 40 years now.

[gritz]

Hi R.G. - I suppose the answer (to the original poster as well as everyone else) is "well, it depends on that you want to do."

Answers might include:

!) "I want it to sound like x"
2) "I don't want it to sound like x"
3) "I don't have the faintest idea what I want it to sound like, but if I'm gonna have to spend weeks / months on it then it better not suck"
4) "I'm just curious and I'll get around to it one day"
5) "I'm just curious, but I doubt I'll ever do anything about it"
6) "I want analogue, dammit (with proper, old - fashioned components)"
7) I'm happy to rough it out in digital first, just to see if it's worth pursuing"

etc.

All are valid and will help narrow down the approach (or lack of it!)

re the NE570 / 571 and beyond. A 2:1 compression / expansion ratio isn't going to get us very far at all. As a very clunky comparison the Tubescreamer uses a maximum of over 40dB of gain to send the signal past that "crossover kink" for a reasonable duration. Keeping a signal range of 40dB (or more likely 60+dB) within the sexy part of a transfer function will require something far smarter - maybe based on the THAT 218x VCA and 2252 RMS detector with the adaptive capacitor time constant hack - details available on the THAT website. They're not the easiest chips in the world to breadboard though - simply because they're so sensitive.

I monkeyed with companding as part of research into harmonic excitement - feeding a compressed signal into a bendy transfer function in order to create a (hopefully) constantish level of harmonic content that could be eq'd, expanded and then mixed back in. Great for adding a bit of shimmer to certain sounds, but imo not stellar with guitar or bass in most situations. IMO, of course. It didn't start out as a guitar project, but ineveitably it went that way and I tried allsorts...

A hot dog is still a hot dog, no matter how much mustard you put on it. It's still a hot dog if you bury it in chilli sauce or tzatziki... Perhaps the question is "should I try another brand of hot dog? What do I want this hot dog to taste like anyway?"

Hungry now.

Still, I recently spent an evening re-amping a TS808 clone with seven different flavours of opamp and didn't experience any kind of epiphany whatsoever (just odd harmonics, mainly). So I guess I just don't get it.

[R.G.]

re the NE570 / 571 and beyond. A 2:1 compression / expansion ratio isn't going to get us very far at all.

Nope. That was one of the issues - needing far more companding. Hacking the 571 got a lot more compressed than the stock 2:1, and in the end it was complicated and only used the VCA. And it still needed more work.

Still, it did make the problem smaller.   

As a very clunky comparison the Tubescreamer uses a maximum of over 40dB of gain to send the signal past that "crossover kink" for a reasonable duration. Heeping a signal range of 40dB (or more likely 60+dB) within the sexy part of a transfer function will require something far smarter - maybe based on THAT 218x VCAs and THAT 2252 RMS detectors (with the adaptive capacitor time constant hack - details available on the THAT website. They're not the easiest chips in the world to bradboard though - simply because they're so sensitive.

Actually, I think the problem is not so much squeezing the signal range into the sensitive area of a soft clipping curve. I believe it's the practicalities of making a compressor and expander work quietly and track well enough through the whole process, including with time constants. I messed with the THAT chips a bit for another project, and they are fairly complex to work with.

I monkeyed with companding as part of research into harmonic excitement - feeding a compressed signal into a bendy transfer function in order to create a (hopefully) constantish level of harmonic content that could be eq'd, expanded and then mixed back in. Great for adding a bit of shimmer to certain sounds, but imo not stellar with guitar or bass in most situations. IMO, of course.

On the other hand, I think most guitarists aren't particularly wedded to getting a no-compression result. Real distortion heads LIKE the "sustain" that they get from just smashing the amplified signal through a clipper. It may be that a more usable compander would have non-matching compression and expansion.

It's one of those areas that are ripe for experimentation. 

Still, I recently spent an evening re-amping a TS808 clone with seven different flavours of opamp and didn't experience any kind of epiphany whatsoever (just odd harmonics, mainly). So I guess I just don't get it.

The company I work for did that with an audience of pro guitarists, then made the vote for which opamp they liked best - but blind, not knowing what opamp was what. The result was that no one opamp, including the JRC4558 and a video opamp as a ringer, had any particular advantage. I think you just reconfirmed our experiment. Not that True Believers would believe it, of course.

[R O Tiree]

"But Duuuude... it hasn't got a 4558 in it - it can't be any good!"

"OK, I'll put a 4558 in it. That'll be $20 extra."

Kerching!

[ashcat_lt]

I know that we're all about analog stuff in this particular forum, but it seems to me that if one wanted to experiment with this type of stuff there are a number of decent wave shaping plugins out there.  If you've got a decent computer you'd ought to be able to experiment with everything discussed in this thread without all of the fiddling to get the analog circuit to behave as intended.  Once you find a combination of digital modules which make a sound that you like, you can look into making it happen on the breadboard, and then...

I built some JS effects for Reaper specifically for the purpose of playing with these sorts of things.  I'm hack, of course, so they are only kind of half assed.  They don't do proper oversampling, and I don't think that the "diode curves" are technically mathematically accurate.  One of them, though, allows you to adjust the curve.  I used gain = 1/(x^n), where n is the input and and you can set x as you please.  I would be delighted if it actually did anybody any good.

My First JS's
And of course if anybody can see a major flaw in the program, or offer suggestions for improvements, I'd love to hear it!

[tca]

What I think is missing is a dynamical way of controlling gain, compression or expanding the signal to keep it in that good region for long enough time.

Hmmm. Almost sounds like a case for companding before and after distortion, doesn't it?

Actually when I wrote that I was thinking also (and mainly) to move the relative position of the knee voltage, using a peak detector  and getting some kind of duty cycle distortion. Instead of manipulating the input signal amplitude (by companding) to alter the clipping level based on that amplitude value.

[gritz]

On the other hand, I think most guitarists aren't particularly wedded to getting a no-compression result. Real distortion heads LIKE the "sustain" that they get from just smashing the amplified signal through a clipper. It may be that a more usable compander would have non-matching compression and expansion.

It's one of those areas that are ripe for experimentation.  

Yes, Definitely! IMO some form of dynamic range compression is generally desirable - if only so we can make ourselves heard over the drummer   but many guitarists eschew "transparent" compression (the sort that doesn't leak transients or apply tonal variation or harmonic content) because it doesn't sound like it's doing anything. So we tend to desire "something" to give us the impression of dynamics when the actual signal has a very squashed dynamic range.

Putting on my synth hat for a moment (if only to cover my bald head) I have to say that static waveforms are deadly dull and don't cut it in a mix*, which is why we apply envelopes, filters and sometimes distortion and all manner of other whatnot.

* The one exception to this is sub-bass which - while enveloped in an on / off manner - tends just to be a sine. But it's more "felt" than heard. A buttress, if you will - rather than a timbre in it's own right.

So, progress. But how to accentuate the good and toss out the bad? And how to decide what is good and bad in the first place? We're allowed to struggle here, because I think that the whole concept of timbre in a psychoacoustic sense (i.e. how the human brain perceives the messages that the ear sends it when confronted by musical noises) is still very much open to speculation. A lot of reseach suggests that identification of a particular instrument relies heavily on the note onset and if that note onset is sawn off then test subjects find it harder to identify an instrument from the sustain portion of a note alone. However, as musos we treat this stuff as a way of life and perhaps are more attuned to those little differences. Or more subject to the power of suggestion. Or both. A quick story:

Years ago I swore I could hear even order harmonics burping out of my Danelectro Fabtone distortion pedal. I thought this absolutely couldn't be the case, because I knew the gain part of the pedal was just a couple of opamp gain stages followed by a symmetrical diode clip to ground. So I ignored it, but it just kept nagging at me. Eventually I attacked it with the spectrum analyser - and those even order harmonics were indeed there. Puzzled, I pulled the pedal apart, attacked it with the scope and found that when your average bipolar output opamp (i.e. almost every opamp we use in pedal land) rail-to-rail clips, it gets about half a volt closer to V+ than to V- and that's enough to produce audibe even order harmonics - even when the asymetric swing is brutally clipped by a pair of silicon diodes afterwards. I pulled out the opamp (a JRC4558D) and socketed, so I could try some of the other usual suspects (a TL072, LM833, NE5532, OP275, etc...). Not a blind bit of difference between 'em in this most testing application- not even in the frankly woeful noise floor.

This is why I have a (possibly unreasonably) low cork - sniffing tolerance. If (for instance) two opamps sound vastly different in circuit, then either:

1)  at least one of them is complaining about something and research should be aimed at why in order to extract maximum goodness, rather than just crediting Mr Mojo, or:

2) It's a product of of a flawed test, or one's own imagination.

I did however find that the rather cruel addition of a 4k7 resistor between clipping opamp output and V+ gave an audible (and measureable) increase in even order products (and was quicker than lowering Vref). It's still there - and it's still one of my go-to pedals, having had a bit of tuning to the frequency response of various parts.

Sorry, went a bit off-topic there...

@ ashcat_lt: I just saw your post as I hit "submit". I agree that realtime approximations on computer are worth far more than guesswork and static spice-type stuff for getting a feel of the problem - I use Bidule a lot for applying maths to audio in real time. I haven't done Reaper in a long while and I'm rather up to my @ss in alligators at the moment, so I can't promise I'll be able to take a look any time soon, but I would encourage anyone with a computer to take a lead from the approach of ashcat_lt and make some noise / bust some myths.

[Thecomedian]

EQ stack pre-distortion = selective band distortion, more or less distortion for frequency range.

possibly.

Like the RAT's reduction of bass in the distortion cell? I suspect that this was designed to try to stop the bass strings swamping treble freqs as the chord dies away some seconds after pick-strike.

As to the compression inate to "classic" distortion circuits, how about an envelope detector at the input to alter the gain of the output buffer/gain recovery stage? You would get the gritty sound we all know and love, but playing dynamics would be in there as well...

spitballing off that, what about a pass filter to ground that is kept closed during the attack, and as decay sets in, the voltage level drops, which starts allowing more and more bleed-off of the lower frequencies, for a dynamic reduction in lows during decay while mid/high retain full strength? I guess like a gate?

[gritz]

EQ stack pre-distortion = selective band distortion, more or less distortion for frequency range.

possibly.

Like the RAT's reduction of bass in the distortion cell? I suspect that this was designed to try to stop the bass strings swamping treble freqs as the chord dies away some seconds after pick-strike.

As to the compression inate to "classic" distortion circuits, how about an envelope detector at the input to alter the gain of the output buffer/gain recovery stage? You would get the gritty sound we all know and love, but playing dynamics would be in there as well...

spitballing off that, what about a pass filter to ground that is kept closed during the attack, and as decay sets in, the voltage level drops, which starts allowing more and more bleed-off of the lower frequencies, for a dynamic reduction in lows during decay while mid/high retain full strength? I guess like a gate?

Funnily enough, I'm working on noise reduction (in software) at the moment - which is pretty much the opposite of what we're talking about here - i.e. I'm expanding the dynamic range of the signal before the distortion stages. As the signal drops below a certain level, lows below about 500Hz and highs above about 1000Hz are progressively rolled off, then the signal is "faded out"as it drops further. Because software amp sims don't create noise (in the form of hiss and hum) of their own, but simply amplify what you feed them, it works well and is pretty transparent, so that suggests that "going the other way" - i.e. compressing the signal heavily before the non-linear stage might not give any Holy Grail moments.

If the amp sim in question is prone to flab then your take on rolling off the bass as the input signal drops cleans it up nicely, but tbh it's just as effective to filter the lows out the old fashioned way (with a capacitor and resistor) and it doesn't hugely affect the note attack, because that contains a high proportion of high frequencies anyway.

[puretube]

fig.2 read page 8 (us-pat.#4899115)...

other food for thought... (us-pat.#4571548)

[Thecomedian]

I was imagining some diode or biased transistor that keeps the door closed if there's enough voltage, but then opens up more and more to a "lows-to-ground sink" as voltage starts dropping.

[puretube]

I was imagining some diode or biased transistor that keeps the door closed if there's enough voltage, but then opens up more and more to a "lows-to-ground sink" as voltage starts dropping.

as opposed to the "series-diodes simple noisegate-thingy"?

[Thecomedian]

I was imagining some diode or biased transistor that keeps the door closed if there's enough voltage, but then opens up more and more to a "lows-to-ground sink" as voltage starts dropping.

as opposed to the "series-diodes simple noisegate-thingy"?

I have honestly no idea how either would work, but I feel like using a transistor would allow much finer control. They are essentially like auto-potentiometers.

I have to withhold saying yes or no to either until I probe the limits of both designs.  

edit:
a glance at what noise gates are: the support circuitry in the noise gate with the IC probably simplify the process of frequency selectivity and bleeding. It would take a fair number of support components to build an analogous discreet transistor-based device that behaves similarly.