Newly discovered Neovibe failure mode

[vince76]

Hi there,
I have a distorsion issue with my neovibe (only at high input levels), and when I found this thread I thought: here is the solution.
But... my neovibe was built using BC547 not 2N5088. The voltages seem ok. At the time of build I posted voltages here: http://www.diystompboxes.com/smfforum/index.php?topic=74406.msg604818#msg604818
...I had a cap placed in the wrong direction. Then the effect worked good until used a pick up with greater output.
Anyway I find that if I put a finger on Q2 the distorsion stops.
I think the issue is the same described here.
Do you think a transistor replacement is the right way to solve it?
No I haven't tried to stick another transistor yet because the board has experimented a lot of mod before and the traces are a bit damaged.

Thank you,
V.

[R.G.]

I have a distorsion issue with my neovibe (only at high input levels), and when I found this thread I thought: here is the solution.
But... my neovibe was built using BC547 not 2N5088. The voltages seem ok. At the time of build I
...
had a cap placed in the wrong direction. Then the effect worked good until used a pick up with greater output.
Anyway I find that if I put a finger on Q2 the distorsion stops.
I think the issue is the same described here.Do you think a transistor replacement is the right way to solve it?

I think it is an oscillation problem, until more facts can be found. The key symptom is "I find that if I put a finger on Q2 the distorsion stops". That is very much a likely result for a gain-phase oscillation happening.

Instead of doing a transistor replacement, I think a better compensation scheme would help. I found this in some simulation of the parts of the univibe, and it held true as far as I could tell when I tried it on a real prototype. The "compensation cap" in the preamp of Q1/Q2/Q3 is C4, 330pF.  This cap actually makes the thing prone to oscillation out at a few MHz for some transistors. I was mildly astonished at this and ran many sim runs to test out what I thought I was seeing. A much better way to make it stable is to remove C4, and instead put a 30pf - 47pF cap from the collector of Q2 to the base of Q2. This seemed to make the stability better for every version I simulated. And it had no ill effect that I could either hear or find with an oscilloscope.

I think you might want to try this.

Changing to a different transistor for Q2 might work fine, but I think that the different compensation might be easier. Just remove C4, then solder a small ceramic cap from collector to base of Q2. I'm pretty sure it won't hurt, and if it fixes your problem, it proves the theory.

[PRR]

> prone to oscillation out at a few MHz for some transistors. I was mildly astonished at this .... A much better way ... instead put a 30pf - 47pF cap from the collector of Q2 to the base of Q2.

"Pole-splitting". Solomon's The Monolithic Op Amp: A Tutorial Study has a discussion near page 13.
http://www.national.com/an/AN/AN-A.pdf (National re-print)
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.113.3480&rep=rep1&type=pdf (scan of original)

> I was mildly astonished....

"Mildly" because you know high-gain amps find ways to oscillate.

Without the 330pFd, Q1 and Q2 will have similar poles (offset a bit by Q1's low current). We should invite Brymus to compute the forward gain, but I'll guess total gain near 5,000 and corner near 5KHz.

Since feedback margin is something like 1,000 we may need to control phase shift over three decades of frequency. Each stage has 45 deg at some KHz, asymptotic to 90 deg at infinite frequency

At high frequency we have two 89 degree phase shifts, 178 degree total phase shift. This can't oscillate. But Q3 is in the loop. Emitter followers have small phase shift at high frequencies, but just a few degrees added to the 178 deg already will make an oscillator.

Pole-splitting shifts two points in pole-space, gives a broader range of stability. If you remember compensating the '709, even the LM318, you know what a wonder the pole-split '101 was (and how the 741 made things nearly idiot-proof).

And this Neovibe circuit is essentially the '101/'741 topology, minus frills. It even has a similar current in the input stage, and your 33pFd is essentially what the '741 uses.

And like the '741, slew-rate can be marginal. Probably moot here. We simply can't have large signals, nor fast signals.

Side-bar: that does seem like a lot of parts for a mere gain of 4. However it does give +4 and -4 outputs, handy. I do wonder if a TL072 could do the same function, with same or fewer parts and connections, and some little performance increase. But I gather the root design is quite old (may predate useful chips), and there's crates of existing PCBs, so nobody is inclined to re-layout.

[R.G.]

> I was mildly astonished....
"Mildly" because you know high-gain amps find ways to oscillate.

Yeah. Actually, I wasn't mildly astonished that it could find a way to oscillate. I was mildly astonished that Mieda-san, the rumored designer, would have not done a simpler and in many respects better compensation scheme. I respect his work, and found it odd to find a hole in it, within the available technology at the time the Univibe was designed.

Without the 330pFd, Q1 and Q2 will have similar poles ... feedback margin ...asymptotic to 90 deg ... Emitter followers ... just a few degrees added to the 178 deg already will make an oscillator...

Yep. All that. Brymus does need to take on some analysis. 

Side-bar: that does seem like a lot of parts for a mere gain of 4. However it does give +4 and -4 outputs, handy. I do wonder if a TL072 could do the same function, with same or fewer parts and connections, and some little performance increase. But I gather the root design is quite old (may predate useful chips), and there's crates of existing PCBs, so nobody is inclined to re-layout.

Yeah.

I've actually redesigned it several times for this, that, or the other. The problem with this is that guitarists won't believe it sounds the same. When I first did what would become the Neovibe board back in the 90s, I was roundly castigated on usenet by a bunch of people who believed that my use of a regulator IC would make it impossible for the Neo to ever sound the same as an **original** circuit, warts and all. It made me wonder about "vintage" ripple.


The most current Neovibe level at General Guitar Gadgets is version H as I remember. It's really not that much trouble to re-lay a mod to that board. I have probably half a dozen versions I haven't released for one reason or another. This includes a tube version, an eight-stage version, a small-as-I-can-make-it version, and the recent version which is designed to mate up with the hole mounting holes in an older Crybaby shell. I had a couple of those protoed, and a volunteer has helped me find the couple of bugs I put in to try my own cleverness...  It'll be out soon.

But I don't mind modding the board if there's a reason to. A TL072 would do it. One side with a buffered gain of four, that's five resistors and one cap, then an inverter, two resistors; bias supply, two resistors and a cap. That's 9 resistors, two caps, and a dual opamp. The pin count on the opamp is only one smaller than three transistors, I'd call that even. The existing preamp has ... (!) nine resistors, three electros, and one film/ceramic cap. I'm dropping off the 22K/22K/47K on the input that serves primarily to cut the available treble from a guitar. Hm. I'd have expected the difference to be bigger than a couple of caps just by gut feel. It's probably an easier layout, though.

As I said, I don't mind doing the layout. It would probably take an hour or so, start to finish. The TL072 version might make a 9v only version easier, although I did an unreleased 9V only that's more like the original. There's a 9V only commercial product somewhere as I remember. it.

But I'm rambling again.

[PRR]

> Mieda-san, the rumored designer

There's guys like Solomon who dig the math, and guys who make stuff that works.

Analysis and test up past a MHz is not easy for old-time audio guys. Rapid (productive) audio design is a lot about short-cut simplifications. My newest 'scope is 475KHz tops, 'cuz I hardly ever had any joy from the top-decade of my 15MHz 'scopes.

I can see how a workable (at the time) design may go astray as 2N5088 actual parameters shift on new processes and foundries, layout changes, long-term moon-phase. So that what worked 99.99% back then (no problem in large runs) drifts to work 99.9% (you see rare trouble).

> won't believe it sounds the same. ... made me wonder about "vintage" ripple.

All Ripple(C) is vintage: Gallo stopped selling it. They even seem to be censoring it off the Web. {(C)"Ripple" was and is Gallo's trademark for a juice/wine liquid in a textured flask.}

Vintage "weed" pot makes music sound different than that headbanger herb/drug available the last few decades.

Such things probably make more difference than '5088 versus '072.

[Brymus]

You two really crack me up 
I am waiting for a couple of suggested books ,then read and more questions. 
Googling incessently for small bits of info to sort through and put together in context was quite tiring
I do appreciate the suggestion,I will try a crack at the neo stuff soon.

[01370022]

Vintage "weed" pot makes music sound different than that headbanger herb/drug available the last few decades.

Such things probably make more difference than '5088 versus '072.

That there is a very valid point that not only cracks me up, but got me thinking about how something seemingly unrelated is quite possibly the cause for a fair chunk of stompbox forum flame wars.

I wonder if there is a market for vintage weed, much like pedals? - "Dude, get a load of this, it's got that cool 60's mellow feel to it. It actually makes you feel like it's 1969"

[vince76]

Just remove C4, then solder a small ceramic cap from collector to base of Q2. I'm pretty sure it won't hurt, and if it fixes your problem, it proves the theory.

yeah, it worked. You're an ace.
I only notice a little drop in the high freq but I think it's only suggestion.

V.

[R.G.]

Just remove C4, then solder a small ceramic cap from collector to base of Q2. I'm pretty sure it won't hurt, and if it fixes your problem, it proves the theory.

yeah, it worked. You're an ace.
I only notice a little drop in the high freq but I think it's only suggestion.

Hmmm. You may have re-upped my faith in Mieda-san.

It is **possible** as I think about this that there is a loss in treble as the highs are rolled off for compensating against oscillation. It may be that the peaky response from the C4 "compensation" makes up for that. I'd have to do some more modelling and parameter tinkering to be sure. But it is possible.

In any case, if you're losing treble, you can make the value of the capacitor you used smaller. That may help. It's a balancing act between preventing oscillation and losing treble for discrete circuits with no special high frequency layout, for which this qualifies. You already know what happens when you get it too small, probably. But it is also possible that there is enough stray capacitance in the layout to be a good enough compensation cap without there being a real cap in there. This is NEVER a good thing to count on, but sometimes it happens anyway.

It won't hurt to reduce the new collector-to-base cap a bit, increasing it a bit if it starts oscillating. 

[PRR]

> it is also possible that there is enough stray capacitance in the layout to be a good enough compensation cap without there being a real cap in there. This is NEVER a good thing to count on...

And yet it often does work well. This basic plan is used in a million phono preamps. They don't have two outputs but that's a frill. They have high gain in bass but low gain above the audio band, so the supersonic stability issue is similar. Quite a few manage with no compensation cap.

> Brymus does need to take on some analysis.

I dunno. This stuff was hard back in 1970(?). While perhaps it is "easier" now with network analyzers and such, it's still foggy. Brimus will soon be able to do a midband analysis (assuming all caps are infinite or zero) but a stability analysis is serious work.

So I asked the Idiot Assistant. A tool not in wide use when this circuit was born.

Three plots: 300pFd to ground, 30pFd C to B, and no comp-cap at all.



Top curve in each plot is the in/out response. It is FLAT across the audio band to 100KHz. None of these "drop in the high freq"; that's in your head (or beer).

The green plot is at Q1 collector and reflects the internal gains. The big bass plateau is because C3 is far too small to bypass to the bottom of the audio band. The side effects of this "may" color the bass in ways not easily emulated with opamps.

The 330p to ground plot shows a HUGE bump near 1MHz. Modest on the in/out plot but Q1 collector has gone nuts. This might be predicted (in hindsight): Q1 and Q2 are similar but Q1 runs at much lower current, has lower bandwidth. The 300p cap slugs Q2 to lower bandwidth. Now Q1 and Q2 have nearly the same bandwidth. And two poles at nearly equal frequencies is more than 2/3rd of the way to oscillations. Also note that above 4MHz the Q2 Q3 stages actually show loss. Some worst-case variations might ripen this into outright oscillation.

The 30p C to B plot is classic and does keep Q1 in control. The internal bump is 7dB and the in/out response is lovely smooth.

The "nothing" plot has only SPICE device capacitances; a more realistic model would have pFd caps at every node. I do not think that would change things much. It is also well behaved; not as beauty as the 30p C-B but not obviously on-the-edge.

I also ran some phase-plots. Between the short-bass and the top-rolloffs, the internal phase is  crazy. Too crazy for me to grok. OTOH, as would be hoped, the in/out phase is flat +/-5 degrees 3Hz to 100KHz. Feedback is a wonderous thing.

[PRR]

Despite the "poor" open-loop gain below 1KHz, the (simulated!!) THD numbers are "dead clean":

0.8V peak in, 2V peak out
(just below clipping)
50Hz 0.01% THD
1KHz 0.004% THD
5KHz 0.02% THD
50KHz 0.3% THD

I don't trust the specific numbers. But my gut says it is probably this good or better. This order of performance was state-of-the-art then, and hardly shabby today, considering it is just a dB or two below clipping.

This is with the 300pFd-to-Gnd compensation. The other plans will be the same at and below 1KHz, maybe a bit better at 5KHz.

The 30p C-B compensation is mildly slewing at high level at 50KHz; the 300p-to-Gnd and no-comp aren't. If you actually needed 2V peak at 50KHz.... but we probably don't.