Replacing an MN3005 with 2x v3205 in Parallel

[YouAre]

Some basic background:

The every elusive MN3005 gives us (from my limited understanding) two things in our analog delays that Other BBD's lack. It has twice the inherent delay time of the MN3008, and runs at a higher voltage than the MN3205/V3205.

The former can be rectified by doubling up on 3008 IC's, but you're doubling your IC budget. One could argue that 2x MN3008's are cheaper than a single MN3005, but MN3008's will eventually have gone the way of the 3005 in terms of its lack of availability. It's my understanding that in terms of current output of a clock driver, 2x MN3008's can replace 1x MN3005 without any issues. So this does not seem to pose an issue for us, but this will come up later.

Replacing a 3005 with a 3205 variant would require the simple change in voltage configuration to the IC, and possibly a slight adjustment to the companding. The expense comes in the form of significantly lower headroom. This does seem to be a major reported issue, especially with the v3205.



What if...

Let's use Madbean's Aquaboy Delay as an example: http://madbeanpedals.com/projects/Aquaboy/docs/AquaboyDX.pdf

What if we were to take the filtered output of our usual compander (the 1uF capacitor C17) and run that to a phase splitter (using something like this http://www.learnabout-electronics.org/Amplifiers/amplifiers54.php). From that, we could rectify each signal so that we only have the positive signal on one side, and the negative on the opposite.

Those two rectified signals can then be run to parallel 3205 delay lines, coupled and biased so that each of the rectified signals is sitting in the middle of the optimal bias range for the v3205. This bias would also have to be adjusted to have some overlap between the two delay lines so that when these two signals are summed, there is no crossover distortion. I suppose this would resemble a Class A/B Amplifier, a push-pull summer of sorts. The output of this push-pull summer would feed into whatever the output of our delay would normally feed next (resistor R30 in the Madbean Pedal).

It's quite a bit of extra circuitry, yes. But this could be our workaround for headroom issues with the v3205. Additionally, remember the clock driver issue we faced before? Well the 3101/3102 will drive 2x 3X05 BBD's nicely, so we're save if we only have 2x BBD's. If we wanted to utilize 4x 3205's (to get effectively 2x 3005 in series), we'd have to majorly adjust the clock driver. It's definitely doable though.

Is there anything I'm missing in my assumptions for this or something crucial that I'm not considering? Does anyone have any other thoughts/experience on this concept?

[Mark Hammer]

Hi Murad,

Interesting idea, in principle.  My initial impression is that the crossover distortion one would get during the rectification would have a far greater impact on signal quality than any headroom limitations in the BBD itself.  It's those blasted diodes one needs to use to separate the yolk and the white in the audio signal.  You end up leaving a bit of each behind on the shell. 

No, I think one simply has to place one's faith in companding, decent lowpass filtering, and high sample rates.

I suppose one could consider using frequency splitting, and a pair (or quartet) of BBDs synced to the same clock, such that headroom limitations were less of a challenge.  So, perhaps the low end gets more seriously compressed than the mids and highs, both get processed, then expanded in corresponding fashion, and mixed.  You wouldn't have the crossover distortion issues noted above.  Just exactly how much additional signal quality that would get you, though, is anybody's guess.

[YouAre]

Hi Murad,

Interesting idea, in principle.  My initial impression is that the crossover distortion one would get during the rectification would have a far greater impact on signal quality than any headroom limitations in the BBD itself.  It's those blasted diodes one needs to use to separate the yolk and the white in the audio signal.  You end up leaving a bit of each behind on the shell. 

Salutations, Markus.

I had considered crossover distortion, but what about biasing it to ensure some overlap? It would be like adjusting from Class B to Class A/B. I guess my assumption is that this effectively reduces the crossover distortion to hopefully desirable levels (maybe it adds some MOJO??) Or is the selective biasing merely just a mitigation technique that doesn't entirely reduce the issue?

No, I think one simply has to place one's faith in companding, decent lowpass filtering, and high sample rates.

These are practical solutions. I am not a practical man. You should've realized this by now 

I suppose one could consider using frequency splitting, and a pair (or quartet) of BBDs synced to the same clock, such that headroom limitations were less of a challenge.  So, perhaps the low end gets more seriously compressed than the mids and highs, both get processed, then expanded in corresponding fashion, and mixed.  You wouldn't have the crossover distortion issues noted above.  Just exactly how much additional signal quality that would get you, though, is anybody's guess.

This seems like it could produce some kind of interesting double track sort of sound to the delay. If we introduce some sort of phase differences by utilizing an "imperfect crossover," it could create some interesting sounds. 

[YouAre]

Gentle bump for the gentlemen that can comment on combining half wave rectified signals with minimal crossover distortion.

Also...what if instead of fully rectifying the signal a phase split signal, we clip the lower halves of the outputs of the phase split signal with a low threshold voltage diode? I understand in the summing process, we'd be super imposing a "clean" half of a wave to the same, albeit clipped, signal. I wonder what that would sound like. Would the larger amplitude "clean" signal take over the lower amplitude clipped signal?

[anotherjim]

Would 2 precision rectifiers (diode in opamp loop) do better than phase splitting?

Or how about going plain differential? You get x2 boost for free when you recombine with a diff amp + removal of common mode noise. This is the same gain as splitting the signal into pos and neg halves without the hassle recombining.

So the simple 1 BJT or FET phase splitter - Upper phase AC coupled to one BBD input and the lower feeds another?

[YouAre]

Would 2 precision rectifiers (diode in opamp loop) do better than phase splitting?

I was thinking about that today. This would remove that .7V of "turn on" voltage before the diode starts conducting, correct? Using 2 opposing precision rectifiers would give each "half" of the cycle," with neither of them inverted. Would we recombine with a simple mixer?

Or how about going plain differential? You get x2 boost for free when you recombine with a diff amp + removal of common mode noise. This is the same gain as splitting the signal into pos and neg halves without the hassle recombining.

Can you please explain this further? The goal I had in mind wasn't to get any boost, but to only use half of the signal amplitude for each delay chip to maximize on usable headroom for the BBD. We would then recombine the circuits to give us our "original" signal, but now processed by the delay.

So the simple 1 BJT or FET phase splitter - Upper phase AC coupled to one BBD input and the lower feeds another?

Close. I'd rectify the output of each of those outputs, then bias them to get max headroom from the BBD.

[anotherjim]

Differential would be no more "boost" than going with separate rectified halves.

If your BBD max peak-peak signal headroom is only (say) 1volt, but you'd rather it was (say) 2volt, then you send non-inverted and inverted full wave signals to different BBD's in parallel. These signals still can't be more than 1V p-p. A differential amp with (otherwise) unity gain will produce x2 (2v p-p) output from the 2 channels while cancelling most common mode noise picked up along the way. You can now get away with reducing the input x2. Effectively, you've doubled the headroom of the BBD's - which I thought was the point? No, it isn't actually doubling the headroom, but the improved SNR should make it equivalent to better headroom?

Going with 2 rectified halfwaves, yes you could use the opamp rectifier bias as the BBD bias and trim the bias so that (say) the positive cycle is using all the headroom. Combining them is the tricky bit - can you just AC couple into a simple 2 input mixer? I think not. Probably we need 2 more opamps as DC level shifters. Each one has a bias adjust so we can get both halves back around the same centre line. Then we can mix them back into one? One for the breadboard I think, but you don't need to involve a BBD, just concentrate on the splitting, biasing and recombining. Say it was a 12v supply, then the + BBD will bias at (say) 4v while the -BBD is biased to 8v - would prove the principle?
BTW, the best precision rectifiers use 2 diodes (they can be schottky)...

Reverse both diodes for the other half cycle.

[bean]

A much less elegant, but functional, work around for the v3205 headroom issue is a large series resistance at the input of each bbd and then whatever recovery needed afterward. I found that to work pretty well in a DMM setup. In the next iteration, I'm going to try a couple of limiting LEDs at the inputs in place of or in conjunction with input resistance. Not nearly as clever but it might get the job done for minimal extra parts.

[YouAre]

Differential would be no more "boost" than going with separate rectified halves.

Understood. It's only if you sum INVERTED signals that you get a 2x gain. I guess it should be noted that our differential amplifier should be summing 2 perfectly in phase inverted rectified signals (positive + negative), or 2 perfectly out of phase positive rectified inverted signals.

If your BBD max peak-peak signal headroom is only (say) 1volt, but you'd rather it was (say) 2volt, then you send non-inverted and inverted full wave signals to different BBD's in parallel. These signals still can't be more than 1V p-p. A differential amp with (otherwise) unity gain will produce x2 (2v p-p) output from the 2 channels while cancelling most common mode noise picked up along the way. You can now get away with reducing the input x2. Effectively, you've doubled the headroom of the BBD's - which I thought was the point? No, it isn't actually doubling the headroom, but the improved SNR should make it equivalent to better headroom?

The idea is not that we're reducing gain. We're only sending half of each signal to each parallel BBD. So we take out 2V p-p signal, and split it into a positive 1V p-p signal and a negative 1Vp-p signal. Alternatively, we phase split, and apply identical rectifiers to each signal. I believe this would simplify biasing.

Going with 2 rectified halfwaves, yes you could use the opamp rectifier bias as the BBD bias and trim the bias so that (say) the positive cycle is using all the headroom. Combining them is the tricky bit - can you just AC couple into a simple 2 input mixer? I think not. Probably we need 2 more opamps as DC level shifters. Each one has a bias adjust so we can get both halves back around the same centre line. Then we can mix them back into one? One for the breadboard I think, but you don't need to involve a BBD, just concentrate on the splitting, biasing and recombining. Say it was a 12v supply, then the + BBD will bias at (say) 4v while the -BBD is biased to 8v - would prove the principle?
BTW, the best precision rectifiers use 2 diodes (they can be schottky)...

Reverse both diodes for the other half cycle.

You're right. I was planning on getting my signal generator built and prototyping the "processing" circuit. Taking a signal, splitting, rectifying, and recombining. I'm liking this setup for a few reasons: Phase splitter >> Precision rectifier >> BBD (to be omitted from testing procedure) >> Differential opamp. We're simplifying bias, and we're getting the noise rejection from differential amplifiers with this setup. So someone please tell me, what am I missing (if anything)?

Question about the precision rectifier, as I've never used one. If the concept is to create the "ideal" diode (i.e. no turn-on voltage), why does the diode type matter? Wouldn't different diode types simple change the Vdrop?

Thanks again for your input.

A much less elegant, but functional, work around for the v3205 headroom issue is a large series resistance at the input of each bbd and then whatever recovery needed afterward. I found that to work pretty well in a DMM setup. In the next iteration, I'm going to try a couple of limiting LEDs at the inputs in place of or in conjunction with input resistance. Not nearly as clever but it might get the job done for minimal extra parts.

The bean himself! I was just about to post the same topic on your board (username: BrownMagicMurad).

I thought about futzing about with the companding, but I wasn't sure what that would do in terms of noise.

I had definitely considered adding clipping/limiting LED's. It could add a nice character, and would definitely avoid having us rail clipping the v3205.

I remember you voicing your distaste for the v3205 in one of your build docs. Was it mainly the headroom issue that you didn't like about it?

Thanks for responding!

[Scruffie]

LEDs will have way too higher fV to be of any use to a v3205, check the datasheet for its input signal swing capabilities.

A limiter would be a better direction to go.

[YouAre]

LEDs will have way too higher fV to be of any use to a v3205, check the datasheet for its input signal swing capabilities.

A limiter would be a better direction to go.

Really? That is pitiful headroom!

[PRR]

> biasing it to ensure some overlap? It would be like adjusting from Class B to Class A/B.

B/AB *amplifiers* usually have overall negative feedback to reduce the crossover glitch.

You can NOT run NFB around a significant delay. The output is far too late to pre-distort the input for correct overall operation.

Yes I can think of vacuum tube AB and even B amplifiers with no NFB. The key here is very long "toes" on a tube's cut-off curve. And also, in many cases, lowered expectations. ("Yeah, the beach-radio is raspy when playing soft, but the battery life is SO much better!")

You are also increasing the already non-trivial "small" distortion. The BBD has an "S"-bend transfer function. By biasing one high and one low you are splicing two S-bends together. We don't have a letter for that but think of a snake. Instead of just simple 3rd harmonic there's now significant 5th harmonic and over a wide range of signal levels (not just increasing with increased signal level).

"Precision Rectifiers" are a very tough problem at high audio frequencies. With a dumb rectifier you have a 0.6V dead-band. This can be reduced by the gain of the op-amp. But TL072 at 20KHz has excess-gain of only 400. 0.6V 600mV deadband is reduced to 1.5mV dead-band, which is just 66dB below a 3V peak signal, and MUCH more obnoxious than 1.5mV of random crap because it correlates with signal in un-obvious way. There's also delays in the diodes, and any good Precision Rectifier has two or more op-amps thus a bigger loss-budget. Yes, I suppose that lower-Vf faster diodes, 5KHz bandpass, it may be acceptable for rock-n-roll. With a fuzz in the chain, the crossover distortion may be masked.

"Plain Differential" at least has the benefit of simplicity. S/N improvement won't be large, and THD probably rises (since both are S-bend devices), but simple is good.

I'm not sure what pre-clipping does for us. At least the earliest BBDs, they clipped (for-sure!) but not offensively, no worse than diode-clippers (and less than some).

The Way Forward "should" be digital delay lines. RAM is cheap-cheap-cheap. ADC/DAC at the 12-bit level are cheap, yet far better fidelity than any BBD. If you must have BBD-like bentness, function that into the digital; add hiss too if you want.

[anotherjim]

The switching speeds of the diodes may matter in a precision rectifier, but most schemes show plain 1N4148 or 1N914.

I've messed about with signal rectifiers at 40Khz (ultrasonics), it becomes very hard to get nice symmetry at that sort of speed -  all those little differences in stray capacitance here and there really mess things up (for full wave, the product is up at 80Khz + new harmonics).  In a guitar effect, the BBD bandwidth shouldn't need to be so high that it should cause such problems.

PRR makes a good point about this putting all of the signal in the non-linear extremes of the BBD, usually it's just the peaks that go there. If the object is to make a soundalike clone, this could be a killer. If that isn't the object, it could be good! After all, if we wanted delay without the "character" of BBD's, there's a DSP near you ready and waiting 

[YouAre]

> biasing it to ensure some overlap? It would be like adjusting from Class B to Class A/B.

B/AB *amplifiers* usually have overall negative feedback to reduce the crossover glitch.

You can NOT run NFB around a significant delay. The output is far too late to pre-distort the input for correct overall operation.

Yes I can think of vacuum tube AB and even B amplifiers with no NFB. The key here is very long "toes" on a tube's cut-off curve. And also, in many cases, lowered expectations. ("Yeah, the beach-radio is raspy when playing soft, but the battery life is SO much better!")

Well, global NFB in tube amps is mainly relevant because of output transformers, correct? A differential amplifier will have localized NFB, which I'm sure helps things a bit.

You are also increasing the already non-trivial "small" distortion. The BBD has an "S"-bend transfer function. By biasing one high and one low you are splicing two S-bends together. We don't have a letter for that but think of a snake. Instead of just simple 3rd harmonic there's now significant 5th harmonic and over a wide range of signal levels (not just increasing with increased signal level).

Can you please elaborate more on S-Bend transfer functions? I'm wholly unfamiliar with them, and my initial searches didn't turn up anything relevant. Instead of giving me a detailed explanation, I would totally appreciate if you could point me in a good direction.

"Precision Rectifiers" are a very tough problem at high audio frequencies. With a dumb rectifier you have a 0.6V dead-band. This can be reduced by the gain of the op-amp. But TL072 at 20KHz has excess-gain of only 400. 0.6V 600mV deadband is reduced to 1.5mV dead-band, which is just 66dB below a 3V peak signal, and MUCH more obnoxious than 1.5mV of random crap because it correlates with signal in un-obvious way. There's also delays in the diodes, and any good Precision Rectifier has two or more op-amps thus a bigger loss-budget. Yes, I suppose that lower-Vf faster diodes, 5KHz bandpass, it may be acceptable for rock-n-roll. With a fuzz in the chain, the crossover distortion may be masked.

Thank you for sharing that constraint of precision rectifiers. But how high are "high audio frequencies?" Are they above what would normally be filtered out by the standard analog delay LPF before and after the BBD?

"Plain Differential" at least has the benefit of simplicity. S/N improvement won't be large, and THD probably rises (since both are S-bend devices), but simple is good.

I understand that there is no free lunch. But I'm willing to douse my ramen with tons of hot sauce to get a discounted price lunch. (Does that metaphor make ANY sense?)

I'm not sure what pre-clipping does for us. At least the earliest BBDs, they clipped (for-sure!) but not offensively, no worse than diode-clippers (and less than some).

I wouldn't use them with this parallel setup.

The Way Forward "should" be digital delay lines. RAM is cheap-cheap-cheap. ADC/DAC at the 12-bit level are cheap, yet far better fidelity than any BBD. If you must have BBD-like bentness, function that into the digital; add hiss too if you want.

Sure, I can use that delay into my Class-D amplifier (which replaced my obsolete tubes), but where's the fun in that.
I totally understand that this is a helluva lot of effort for something that can be effectively replaced with a fantastic sounding digitech Pedal (or Strymon if you're feeling fancy). But the fact is that there is enough of a demand for Analog Delay BBD's that someone is attempting to recreate a relatively archaic technology. Unfortunately, it's widely believed (agreed?) that the current widely available offerings fall short of the spec's we're used to. This is an attempted solution for the guy that wants to make an analog delay today, without having to resort to finding NOS chips. Sure it's a practical solution to an impractical desire, but since when are Tubes and BBD practical anymore?

You're really keeping me in check, and I very much appreciate that. Thank you.

[YouAre]

The switching speeds of the diodes may matter in a precision rectifier, but most schemes show plain 1N4148 or 1N914.

Good point. Will see which diodes switch faster.

I've messed about with signal rectifiers at 40Khz (ultrasonics), it becomes very hard to get nice symmetry at that sort of speed -  all those little differences in stray capacitance here and there really mess things up (for full wave, the product is up at 80Khz + new harmonics).  In a guitar effect, the BBD bandwidth shouldn't need to be so high that it should cause such problems.

PRR makes a good point about this putting all of the signal in the non-linear extremes of the BBD, usually it's just the peaks that go there. If the object is to make a soundalike clone, this could be a killer. If that isn't the object, it could be good! After all, if we wanted delay without the "character" of BBD's, there's a DSP near you ready and waiting 

I think I touched on this in my response to PRR. Thank you for your help and support!

[anotherjim]

YouAre...very welcome.

Back in the stone age, I made a few Digital Delays. 8bit converters and the biggest SRAM I could I could get were 8KB chips. I still have the 16KB one I made somewhere. I could just about get the bits to make another today (before DIL parallel SRAM disappears altogether).

[YouAre]

YouAre...very welcome.

You cheeky son of a...

Back in the stone age, I made a few Digital Delays. 8bit converters and the biggest SRAM I could I could get were 8KB chips. I still have the 16KB one I made somewhere. I could just about get the bits to make another today (before DIL parallel SRAM disappears altogether).

Honestly, I tried the Strymon, and it just sounds fantastic. Their Timeline will probably be my be-all end-all. But that won't stop me from building a damn fine analog delay. Over engineering is just too much damn fun.

[analogguru]

So you want to invent "the wheel" ?

Don´t tell it anybody, but "the wheel" has already been invented in 1975:



Simply replace the two halfes of the MN3001 with two MN3205 (and adapt for different power supply of course).

[Lurco]

http://www.diystompboxes.com/smfforum/index.php?topic=78698.0
plain differential: http://www.pmerecords.com/Docs/Archer_SAD-1024_Tech_Data.pdf  fig.10

[anotherjim]

YouAre is talking about Class B operation with 2 BBD, Both Parallel Multiplex and Differential (the schemes just above do both!) don't address the lower headroom with the newer chips. I do suspect that it may in the end be best to take the benefits of "Diff-Mux" to offset having to run a weaker signal. Designers have always thought BBD "clean" headroom and bandwidth was too low - hence devices like the Eventide.

I had a horrible thought about the Class B idea. The half wave signal will be carrying new harmonics due to the flat top or bottom (these should cancel once you get them back together) - BUT, you are surely going to get aliasing with the BBD clock? If you run the halfwaves thru a Nyquist filter, won't it destroy the  DC level and waveshape making re-combining impossible? Maybe there's is a reason it hasn't been done already?

As to the Wheel, I for one am still working on that. So far I can say for certain that it needs AT LEAST 5 sides.