Analog modelling

[Richard_Synapse]

When using a modelled 303 in a track alongside 30 other (virtual) instruments, reverbed, eq'ed etc. who cares if it's not 100% accurate as a real 303? It's the end result that counts. I've made tracks using the inherently shitty Korg Electribe ER-1. The thing is glitchy, sounds tinny, and yet, I think that's part of the charm. Hell, I even went to an acid party where it was used to great effect, despite being dispised by many.

Before the plugin age the MC-202 and other low-budget analog hardware have been used instead of the TB-303 as well, and some great tracks have emerged that way. So yeah, music production is about the end result. It's different though when you look at an instrument from the composition or performance point of view. I might opt for a cheap piano sound when producing a track, because it happens to fit perfectly. But when composing or performing something, in general I want the best piano and not the worst.

Richard

[mystran]

Real 303 seems to trigger a certain physical response in me (not 100% of the time, but pretty often) that practically everything else mostly fails at (and someday I'll figure out what the hell it is about that sound), but I still have to agree that whether or not you use a real 303 doesn't make or break a track.

[antto]

indeed
after (i think) i understood many of the important aspects of it, i decided to try and make a cheaper model, a more "fake" one, and to also lift some of the sonic limitations of the original, to see how many things i can strip away from my "better" model, and how many restrictions i can break, and still be able to achieve a roughly convincing end result
well, this cheaper model isn't bad, but stripping away the sequencer from it (even tho i know how to sequence patterns via MIDI) is a huge loss
on the other hand, i have a hardware 303 clone with sequencer, which sends MIDI notes, so when i couple them - the result is much more enjoyable (minus the MIDI jitter)
i will always recognize these two models in a solo A/B test of course, but it gets much harder on an actual track
so i can make pretty much the same track with two different software models and one real analog(ue) clone, and if i don't say what i used - it'd be hard to guess
however, the actual process of creating that track in the first place would be radically different, depending on which of those three things i choose to use

in actual practice, i use all of them
the hardware clone is the closest thing i have to the original, it has the sound and the sequencer interface, it looks a bit uglier so it's not perfect... i use it for jams and when i want the real tweakability
i use the model when in the DAW and when i want more control, plus it has faster way of pattern storage (vst preset memory)
and i use the cheap model.. mostly for weird non-303 sounds (because it can), but also for approximating famous 303 patterns from popular acid tracks.. because i use a brute-force method for that

this is one of the most cloned synths and there seems to be always room for another clone

[Richard_Synapse]

If you aren't sure where the magic is exactly, you could compare the hardware clone to the original and see if the clone falls short somewhere. May provide some useful insight which components are more/less relevant to that sound

As far as software clones are concerned, there is some truely excellent ones out there, though none of them sound exactly like my hardware TB-303 (and I'm not sure if such an emulation will come anytime soon, could be part tolerances, age etc giving it that special sound).

Richard

[meneervermeer]

When using a modelled 303 in a track alongside 30 other (virtual) instruments, reverbed, eq'ed etc. who cares if it's not 100% accurate as a real 303? It's the end result that counts.

you're missing a little something - getting to the end result

the 303 sounds a certain way on its own
it also behaves a certain way (not just the synth, but also the sequencer and interface)
this has a real effect on the stuff you would and would not do with it

IMO there is a great amount of "feedback" during the music creation process, at least in my case (disclaimer: i'm not a real musician, i make electronic noises)
with "feedback" i mean the process of messing about with a sound, hearing it, in an iterative cycle
you know when you twist some parameter and it sounds nasty - you move away from that spot, and when you find some spot or some combination which sounds nice - you may get inspired and make a track out of, or around that ... IMO this is all feedback between the instrument(s) and the person operating them

in this sense, a more accurate model has a better chance to hit more of the same areas which the original hits

these are tiny, hard to measure things, but they are certainly there

of course you could re-create an existing track with a very poor model by brute-force methods, and end up with an audibly similar end result, but the creation process of the original track vs the brute-force recreation process will have a vast difference..
i'd think most people would prefer to have "fun" till they reach the end result, but i know there certainly are people who would have "fun" while brute-forcing a cover, for different reasons

Good point, and I understand the need for perfect emulation because of various reasons. But then again, I think there is this flawed idea amongst producers that you can only make great music when using a perfect model of instrument X. Or the idea that, when I have this piece of gear, THEN I will make great music. But you can create amazing stuff only using a woodblock. But that's just my opinion.
And I understand that these classic synths became classics for a reason. Be it the sound, the workflow, the quirks or a combination of those.

On the other hand, if one wants a perfect model of the original, get the original. Now I know this is often not economically viable. Original 303's run you back a cool 4000 bucks these days. So I do understand the need for a cheaper, perfect emulation/recreation. But in this case I think it's more practical and effective to simply recreate a real piece of hardware using real components. I've listened to comparisons between the original TB-303, the TT-303, the TB-03 and TB-3 of which the latter two are virtual analog models. The TT-303 sounds far more closely to the original and costs about the same as the virtual models. Granted, the boutique and aira models have a slightly greater feature set, but if I wanted a 303 sound and sequencer, I'd pick the TT-303.

Don't get me wrong, I do think it's important that people try to create perfect models and emulations of classic gear. There is obviously a market for it. But my personal vision as an aspiring developer and designer is to combine the strengths of digital and the sound of analog into something new. Something that would not be possible or impractical in true analog hardware.

[mystran]

If you aren't sure where the magic is exactly, you could compare the hardware clone to the original and see if the clone falls short somewhere.

I have a x0xb0x but it doesn't quite do the trick properly (well, it sort of does and sort of doesn't, it's weird).

Sadly no cash right now that I'd want to dump into a real silverbox.

[antto]

Don't get me wrong, I do think it's important that people try to create perfect models and emulations of classic gear. There is obviously a market for it. But my personal vision as an aspiring developer and designer is to combine the strengths of digital and the sound of analog into something new. Something that would not be possible or impractical in true analog hardware.

well, analog modelling is about modelling an analog device
developers and designers certainly go for hybrid synths, but those don't fall into this category

in my case, it was a very important lesson (or bunch of lessons) the whole journey of trying to model an analog synth
it was the unknown
you can't really start combining the strengths of analog (or VA) and digital before you understand them at least a little bit

If you aren't sure where the magic is exactly, you could compare the hardware clone to the original and see if the clone falls short somewhere.

I have a x0xb0x but it doesn't quite do the trick properly (well, it sort of does and sort of doesn't, it's weird).

i've heard this many times and i still don't have a clear answer or explanation what's so different about the sound of the x0xb0x vs the 303
the argument that they don't sound the same usually comes from folks who have had one of each unit on a table next to each other, and are not able to explain stuff in a technical manner
and in these conditions, the visual and tactile difference between the two boxes makes a subliminal difference which is hard to overcome
i've also invited such people to record a dry audio of a 303 which captures that special thing, so that i can approximate it and provide my version.. and so far this hasn't went anywhere close to a conclusion

[JCJR]

The first synth I had, roland sh1000, had a (so far as I recall) four pole diode filter. Not long ago I listened to some youtube sh1000 demos and it sounded good for what it is. Overall better than my recollection of my sh1000.

I kept the sh1000 for a long time and had chance to compare it to transistor ladder and various OTA filters in other synths. Compared to a ladder or OTA filter the SH1000 diode filter seemed "less ballsy" or "watered down" but OTOH seemed more useful for some sounds, especially emulative "acoustic" flute or oboe or brass kind of sound.

It is the only diode filter I recall using. Dunno if all diode filters would have similar characteristic, or even if the differences might have been of tracking slopes rather than actual filter performance.

Back then I assumed it had more to do with filter frequency response shape than distortion, but didn't study on it. Warping a dsp filter to behave as an OTA, ladder or diode filter would seem a challenge to figure out what exactly is different between them or whatever.

[aciddose]

i've heard this many times and i still don't have a clear answer or explanation what's so different about the sound of the x0xb0x vs the 303

None of the modern rehashes faithfully reproduce the entire circuit including:

• Sequencer
• Discrete DAC
• Glide circuit interacting with DAC
• Envelope interacting with accent
• Gate + glide + accent pulses from sequencer (notable effect on other modules)
• Custom exponential current source, oscillator and waveshaper
• Specific filters (high-pass, low-pass) in signal path
• Additional components of the filter itself including effect of res accent filter, buffer and custom expo current source
• Complicated accent handling and high level input to VCA
• Input mixer in signal path introduces significant non-linearity
• Custom poor quality switched mode power supply (more important than people think)
• At some point you know enough to recognize what you do not know: elements are missing from this list.

There are many that implement most of that, but none of them implement 100% of it with 100% accuracy. I'm not even talking about having to use identical transistors or anything remotely as stupid as that. I'm simply talking about accurate reproduction of the circuits themselves and all their natural interactions.

It must be recognized that the tb-303 is not a general purpose synthesizer. It is designed to make a single "sound" with a few minor tweaks available. The entirety of the all the circuits are designed together as a single unit and can not be broken down into individual modules and replaced or modified. If the entire circuit is not reproduced because a single "unimportant" element is replaced or removed you end up with an incomplete "that sound" minus "supposedly unimportant element".

One of the biggest mistakes people make is to focus on the oscillator pulse waveform (waveshaper) and filter while ignoring all the controlling elements and the PSU which are far more important.

If you reproduce a "tb-303 oscillator" and a "tb-303 filter" and wire them together in your synthesizer it'll sound nothing like a tb-303! In fact the individual modules are considered garbage on their own and mostly useless in a general purpose synthesizer.

[Richard_Synapse]

It must be recognized that the tb-303 is not a general purpose synthesizer. It is designed to make a single "sound" with a few minor tweaks available. The entirety of the all the circuits are designed together as a single unit and can not be broken down into individual modules and replaced or modified. If the entire circuit is not reproduced because a single "unimportant" element is replaced or removed you end up with an incomplete "that sound" minus "supposedly unimportant element".

That's why I'm hesitant to work on yet another TB-303 emulation, it's just too hard. Since emulations that sound very good/close to the real thing already exist, making another one of those seems pointless.

If you reproduce a "tb-303 oscillator" and a "tb-303 filter" and wire them together in your synthesizer it'll sound nothing like a tb-303! In fact the individual modules are considered garbage on their own and mostly useless in a general purpose synthesizer.

Sure but once you leave the "exact emulation" path you can tweak the diode ladder filter into any direction you like, say an EMS diode ladder or whatever, such a filter is certainly useful for some sounds.

Richard

[aciddose]

There is no doubt that "diode ladder" filters ("four series differential lossy integrator; npn LTP input; trans-conductance via PN junction; sans buffers; single npn current mirror long-tailed npn output buffer") are useful but the specific circuit used in the tb-303 is not useful anywhere but the tb-303.

Mostly due to the fact the design depends upon near impossible to source components with near impossible tolerances (simply a bad general purpose design; very sensitive) but also due to the fact the things that make it sound good as a tb-303 make it sound bad when applied for other purposes.

Not that someone wouldn't want a similar or even identical filter in their modular... just that I doubt it would be their favorite or most used filter module.

In that sense even considering the tb-303 or "diode ladders" rather than just focusing on a "four series lossy integrator" filter really wouldn't be productive at all. This is because the circuit is very specific and limiting oneself to only that very small subset of possible implementations would make 90% of the other filters in that class (4x lossy integrator negative feedback) unavailable.

[meneervermeer]

in my case, it was a very important lesson (or bunch of lessons) the whole journey of trying to model an analog synth
it was the unknown
you can't really start combining the strengths of analog (or VA) and digital before you understand them at least a little bit

Very true. In fact, this is my first step as a developer; getting the analog stuff right. That's why my first project is to build a synth lightly modelled after the mini moog.

[antto]

i've heard this many times and i still don't have a clear answer or explanation what's so different about the sound of the x0xb0x vs the 303

None of the modern rehashes faithfully reproduce the entire circuit including:

• Sequencer
• Discrete DAC
• Glide circuit interacting with DAC

well, there is a 303 sequencer approximation available for the x0xb0x (it's a firmware + custom front panel)
what's different about the DAC and glide? it's a resistor network with a buffer and a switched lowpass filter what interaction is there?

• Envelope interacting with accent

other than the accent being derived from the main envelope, what other interaction is there?

• Gate + glide + accent pulses from sequencer (notable effect on other modules)
• Custom exponential current source, oscillator and waveshaper
• Specific filters (high-pass, low-pass) in signal path
• Additional components of the filter itself including effect of res accent filter, buffer and custom expo current source
• Complicated accent handling and high level input to VCA
• Input mixer in signal path introduces significant non-linearity

i honestly don't understand, and i feel like it will be counted as off topic if we discuss the details on these, but i don't think i can get the answers to those things from anyone else, i don't even understand some of those points in order to ask someone else
you used to be on IRC, pls come back

• Custom poor quality switched mode power supply (more important than people think)

yes, the power supply, this is the most famous difference, but i still don't know exactly what it affects in the sound/behavior of the synth

At some point you know enough to recognize what you do not know: elements are missing from this list.

yes, and i may add the actual PCB layout being significantly different, which is also a famous difference
but i also ask the question - which of these differences result in a "significant" audible difference, and where
for example, i know (i've read it somewhere) that in general, the x0xb0x audio is less noisy than the 303 - (assuming this is true) is that a bad thing?
i'd personally say "no" unless someone convinces me that the noise in the 303 is important or desirable

There are many that implement most of that, but none of them implement 100% of it with 100% accuracy. I'm not even talking about having to use identical transistors or anything remotely as stupid as that. I'm simply talking about accurate reproduction of the circuits themselves and all their natural interactions.

there is (since a year or so) a PCB replica project (re-303), it is said that it's a copy of both PCBs (on modern FR4 material) and has formed a comunity of folks.. basically, they are trying to build complete replicas.. does that not count as 100%?

It must be recognized that the tb-303 is not a general purpose synthesizer. It is designed to make a single "sound" with a few minor tweaks available. The entirety of the all the circuits are designed together as a single unit and can not be broken down into individual modules and replaced or modified.

yes, absolutely! i've been telling this to other people.. many think that if they take out the filter and feed some signals thru it, it'll surely give em the "303 sound" ... but the 303 sound is not in any one of its sub module alone

there's also the other extreme - modifications, adding more things to the synth and/or extending it's parameter ranges
i've been saying that those things will only result in a "less 303" sound
i still had the desire to add mods to my "fake" software model, just to see exactly how hard it is to add things on top without breaking anything severly, but also to experience how those mods change the thing from a user perspective.. the result is: when you have so many extra knobs, you tend to tweak them away from their "neutral" positions too often, so keeping this model into "303 territory" is a mind battle for the user.. i mostly use this model to make it produce completely non-303 sounds

[antto]

i mean, obviously, if you want to do it properly, you'd just try to copy the whole thing 1:1 as much as possible
starting from the PCB (that's what re-303 did, except the PCB soldermask is.. blue), the components, the cpu, the plastic case and buttons/knobs, etc..

but the thing is.. if you do that - you wouldn't be required to actually learn how the thing works, and why

i came from a different direction.. i wanted to know the answers to all those questions and understand why that sound is so good
this opened a bag of even more questions like:
- which of the things are absolutely vital and cannot be changed
- what can be changed and considered an "improvement"
- what can be added on top (and how) without breaking anything or looking like an alien tumor

i think there should be a clear difference between an attempt to replicate something 1:1, and an attempt to approximate it but not down to molecular level
i used to think there are sepparate words for those things: replica vs clone, but i'm not so sure anymore
in any case, what i wanted was the latter, because a 1:1 copy is easier to achieve (especially today)

[aciddose]

what's different about the DAC and glide? it's a resistor network with a buffer and a switched lowpass filter what interaction is there?

The glide time is slightly dependent upon the R2R network and buffer because it is unbuffered. It connects directly to the FET and capacitor and the "glide" pulse can introduce spikes into the CV signal. It is obviously arguable that this is only a very minor difference but just as I said: ignoring it does make a difference even if small on its own.

The effect of poor matching in the R2R network or current from the buffer is more likely to slightly throw off the tuning than influence the glide time as is commonly believed. That effect (feeding an R2R into a capacitor) is not actually present at all if the network is matched (1% or better = near zero effect.)

The small pulses from switching the glide as well as poor tuning though may have a far more significant effect on the sound than people assume. It is easy to hear C2 being slightly out of tune with C3 especially when you mix instruments. In poly synthesizers this often creates that "fat" detuned/drifting sound.

other than the accent being derived from the main envelope, what other interaction is there?

There is a small effect on decay time by the accent pulse changing the output current. The circuit uses a darlington but it is far worse than a FET buffer and the effect on timing gets significant especially at the maximum decay time setting.

• Gate + glide + accent pulses from sequencer (notable effect on other modules)

Small pulses are introduced due to EMI as well as the varied current from the PSU. This can be very significant due to the 5.33v reference generator.

It's very common to ignore the PSU and simply assume the voltages are static and have infinite current supply potential at all frequencies of interest as well as that all necessary supply bypassing is accomplished in the circuits but this is not 100% true.

• Custom exponential current source, oscillator and waveshaper

The exact structure of the current source (including input clipping) and its temperature coefficient may contribute to how the oscillator drifts as well as how it responds to different changes in CVs and parasitic signals from the supply rails.

• Specific filters (high-pass, low-pass) in signal path

It is easy to assume these passive filters do not significantly influence the timbre. Again, little things can add up to make a more significant difference.

• Additional components of the filter itself including effect of res accent filter, buffer and custom expo current source

This should be obvious. The accent circuit generates an additional envelope based upon the res setting. This is probably one of if not the most essential part of the "303 sound".

• Complicated accent handling and high level input to VCA

Matching various factors such as the simple amplitude compensation for the res control can be critical due to their influence upon non-linearity later in the circuit. The way the gate + accent envelopes are mixed into the VCA current source is easy to over-look but has a non-trivial effect on the result. Without actually building your own (lm13700, etc) and testing it you might never realize the exact scaling of the envelope signals and the non-linearity of the current source. As we all know: you can definitely hear the difference between different amplitude envelope shapes. Again a very small influence that can add up to others.

• Input mixer in signal path introduces significant non-linearity

This one is simply obvious: if you put an amplifier in the signal path you should probably account for any additional non-linear effect it has as well as its effect on low/high frequency content.

yes, the power supply, this is the most famous difference, but i still don't know exactly what it affects in the sound/behavior of the synth

It allows for signals to be transmitted along the supply rails. Just because the rail is "v+" or "+5.333" doesn't mean it isn't like any other wire: it carries a voltage and current which are both variable and influenced by every circuit they are connected to.

Changes in the amount of current drawn due to the pulse buffer circuits switching (slide, accent, gate) can introduce small impulses into the other circuits.

yes, and i may add the actual PCB layout being significantly different, which is also a famous difference

but i also ask the question - which of these differences result in a "significant" audible difference, and where for example, i know (i've read it somewhere) that in general, the x0xb0x audio is less noisy than the 303 - (assuming this is true) is that a bad thing?

Yes because the PCB layout influences EMI and how effective the supply bypassing is which has a significant and very audible effect on which signals are transmitted across the rails between circuits.

For example with two oscillators connected to a poor quality supply with the main rails correctly bypassed but something like a reference rail "5v / 5.333v" not bypassed as much, the oscillators may use the reference as: surprise! A reference voltage. If small spikes are introduced on this rail by the oscillator sync (reset) pulse it can cause oscillators tuned near fractions of each other (1/2, 3/4, etc) to sync together on their own!

This sort of effect is absolutely essential to accurately reproduce a synthesizer which suffers from it. Again, all small "insignificant" effects that add up to produce notable differences in how the entire unit functions.

there is (since a year or so) a PCB replica project (re-303), it is said that it's a copy of both PCBs (on modern FR4 material) and has formed a comunity of folks.. basically, they are trying to build complete replicas.. does that not count as 100%?

It may or may not, depending upon how perfect the reproduction is. I doubt a duplicate PCB is needed but in the very least all these "minor" effects need to be taken into account and reproduced by any modern circuit.

Reproducing the PCB 1-to-1 is just an easy way to have zero understanding of how/why and get a near perfect replica, assuming the components used are matched correctly.

when you have so many extra knobs, you tend to tweak them away from their "neutral" positions too often, so keeping this model into "303 territory" is a mind battle for the user.. i mostly use this model to make it produce completely non-303 sounds

Absolutely: The tb-303 is an interesting sound but really not applicable to the majority of places you'd want a general-purpose synthesizer instead. That's because the sound is so ultra-specific and unique to the 303 that people almost instantly know "that's a tb-303 or similar".