TB-303 'Facts' a Myth?

[ENV1]

Hi Everyone,

According to information on the internet, the 303s Square Wave is derived from the Saw Wave via a Waveshaping Circuit. However, after doing some tests in order to refine something im currently working on, i find this statement somewhat unlikely because while it is in fact possible to waveshape a Saw into a waveform that is reminiscent of the 303s 'Square', there exists another much simpler way of creating this strange 'Square' (and the Saw too) which simply cannot fail as long as you are working with the right 'input' waveform.


We can discuss this further, but first let me show you what im talking about, i.e. how i got that idea. I will use z3ta+ to demonstrate it because its very easy to show it with, but in general any other suitable synth should give the same result. If you want to replicate this 'live', load z3ta+ and do the following:

- INIT z3ta+ (my skin has a pack of INITs included)

- Set Osc1 to 'Vintage Square 2'

- Click the 'Shape' button to open the Waveshape window

- Drag the 'Highpass Filter' slider to ca. 50%. Theres your 303 Square.

- Next drag the 'Window' slider to 100%. Theres your 303 Saw.


click to view animation



Going by these observations, it would appear that the statement that the 303 Square is derived from the Saw is false, and that it is actually the other way around. (If anything.) Further the so called 'waveshaper' alleged to create the 303s Square more likely is just a highpass, because using and highpassing a 'SawSquare' (like z3tas Vintage Square 2) is what instantly gives you the unmistakable 303 'Square' waveform, and doing to this highpassed waveform whatever z3tas 'Window' parameter is doing also will give you the 303 'Saw'.

And as you can see, unlike other methods, which require lots of tweaking, this method just cannot fail. You can alter the sharpness/intensity of the bending by using more or less highpass, yes, but other than that it is downright impossible NOT to end up with the 303 Square because its quite simply the only waveform that can ultimately result from it. I would therefore hypothesize that this is exactly what is really happening in the hardware as well, because it is simple, probably cheap to implement, and gives you the very waveforms we are talking about.



So much for my hypothesis, feel free to discuss it but please lets do it in a civilized manner. (I.e. lets hold the drama and focus on the educational aspect.) If you have proof that the conclusions im drawing here are wrong, im looking forward to hearing it.

[EvilDragon]

It might be true. Most analog oscillators are generated via waveshaping either from a sawtooth core, or a triangle core.

[chk071]

Nice, thanks for the tips. Happy to try that out.

[ENV1]

It might be true. Most analog oscillators are generated via waveshaping either from a sawtooth core, or a triangle core.

If we assume this to be true in this case as well, and the raw osc really is just a 'regular' Saw, then the waveshaper that others were talking about might simply be resposible for turning that 'regular' Saw into a waveform that looks like z3tas Vintage Square 2, which is then sent into a highpass which in turn gives the waveform its unique 'squeezed' character.

Yes, this would actually seem very likely. It would perfectly explain why there would be an actual waveshaper involved, and to what end.

Nice, thanks for the tips. Happy to try that out.

If you do, be sure to play with the width of the raw wave too. Because with a fully symmetrical waveform it will be way to 'hollow' regardless of the highpassing and thus too far away from the 303 flavor. Depending on the synth, (or better said its pulsewave), a ratio of circa 55:45 is typically a good point to start at.

[EvilDragon]

I think it's rather true. Sawtooth is much MUCH easier to do in an analog (keyword analog) oscillator circuit than a square.

[Gamma-UT]

I had this feeling of deja vu, searched and found this: http://www.kvraudio.com/forum/viewtopic ... 4#p3732204

The 'waveshaper' is Q8 in the schematic (the first link in the post) towards the top left - there's a little picture of the square waveform and its raw voltage levels on the output line from the VCO.

It is, pretty much, a thresholding circuit, so it outputs the high voltage level until the output from the saw-generating transistor falls below a certain level and it switches to the low level needed for the rest of the square wave.

[subterfuge]

Going by these observations, it would appear that the statement that the 303 Square is derived from the Saw is false, and that it is actually the other way around. (If anything.) Further the so called 'waveshaper' alleged to create the 303s Square more likely is just a highpass

The TB303 VCO creates first the saw wave and from that the square. That's a common order for other old analog synths as well. The waveshaper is not just a highpass, although pretty simple. Just two caps, four resistors and one transistor. There are plenty of original and modified circuit schematics around to show all this. Adafruit forums have more talk about anything x0xb0x tech (303) related.

[1wob2many]

This is fun:

https://www.youtube.com/watch?v=k5gacnQouSU

[ENV1]

I think it's rather true. Sawtooth is much MUCH easier to do in an analog (keyword analog) oscillator circuit than a square.

Yes, it makes sense if you assume a regular normal Saw Oscillator. But if you look at how it is worded in places like Wikipedia, (Quote: "The square wave is derived from the sawtooth waveform using a simple, single-transistor waveshaping circuit."), its as if they were saying that the Square is derived 'from the 303s Saw', i.e. from the Saw that you hear, not from a raw Saw oscillator.


But i get it now.

So what i now think is happening in the hardware is this:

Saw Oscillator -> Waveshaper -> HighpassFilter

The waveshaper only plays a minor role: It simply shapes a 'raw' Saw into a shape like that of z3tas Vintage Square 2, which is then going into a highpass filter giving the waveform its final shape. The waveforms unique shape is therefore not actually created by the waveshaper (as some sources would suggest), but by a highpass filter, just as my tests indicate.

Make sense?

[ENV1]

@ subterfuge/Gamma-UT

Saw your posts just now, the forum wouldnt respond for a while.

Do you think what i posted just now is about right or am i still missing something?


Thanks folks, this was very helpful so far.

[DJ Warmonger]

Here you go, square-shaping circuit based on linked article (need Java to run):
http://www.falstad.com/circuit/circuitj ... 25+1+-1%0A

What it does is to basically truncate saw from bottom and up. Capacitor values seem a bit off, I never saw multi-milifarad condenser. But I'm a digital mind

[antto]

Saw Oscillator -> Waveshaper -> HighpassFilter

The waveshaper only plays a minor role: It simply shapes a 'raw' Saw into a shape like that of z3tas Vintage Square 2, which is then going into a highpass filter giving the waveform its final shape. The waveforms unique shape is therefore not actually created by the waveshaper (as some sources would suggest), but by a highpass filter, just as my tests indicate.

Make sense?

not exactly
first of all, the saw is pretty much an ideal ramp wave..
there is only one highpass filter between the oscillator and the filter, but it is tuned very low, as it should be to play the role of a DC-blocker..
there are like 9 more highpass filters afterwards, including 4 or 5 in the actual filter (and its feedback path) some of which are tuned much higher than they should be for DC-blockers, and that's why the output waveforms look bent like crazy

but that's nothing, the two important things going on here are:
1) the highpass doesn't contribute to the square wave, it contributes to the lack of resonance in when the filter cutoff is low, in the bassy frequencies.. this makes the 303 very suitable for distortions, unlike most other synths with resonant filters
2) the sound of the square wave is mostly in the very typical frequency-dependant waveshape
the waveshaper is not memoryless, there are two capacitors involved, and as a result, at low frequencies the square's middle edge has a soft knee, it gets sharper at higher frequencies, and the pulse-width also varies dramatically
the most audible aspect of the square is the frequency-dependant pulse-width

[antto]

Here you go, square-shaping circuit based on linked article (need Java to run):
http://www.falstad.com/circuit/circuitj ... 25+1+-1%0A

What it does is to basically truncate saw from bottom and up. Capacitor values seem a bit off, I never saw multi-milifarad condenser. But I'm a digital mind

the caps are 1uF and 0.01uF, and there is +12V, the ramp goes down too and is biased
here's a slightly better version:
http://www.falstad.com/circuit/circuitj ... 25+1+-1%0A

check out the "knee" at 33Hz.. then edit the oscillator and try a higher freq like 100Hz to see how things change

[ENV1]

First of all sorry for the late reply, yesterday the forum had a glitch which kept some members from posting while posts that actually got through got inserted in the wrong position. (See do_androids_dream's post, it should have been above this one.) This post is therefore going to be a bit longer than it had been otherwise.

Here you go, square-shaping circuit based on linked article (need Java to run):
http://www.falstad.com/circuit/circuitj ... 25+1+-1%0A

Thanks. I tried to check this out but the site doesnt open in my browser.

I take it its some kind of circuit simulator?

first of all, the saw is pretty much an ideal ramp wave..

You mean the raw oscillator, right? Because i know for a fact that the audible 303 Saw is nowhere near a perfect ramp, (far from it), it looks very much like in the clip in the first post after applying the 'Window' parameter. (Not 100% like that, but the difference is negligable.)

there are like 9 more highpass filters afterwards, including 4 or 5 in the actual filter (and its feedback path) some of which are tuned much higher than they should be for DC-blockers, and that's why the output waveforms look bent like crazy

I think there can be no doubt that it is the highpassing that is ultimately responsible for the odd waveshapes. After all when you test this you can clearly see that a HPF is always giving exactly this unique result, thus the probability that this could be mere 'coincidence' is practically zero.

the highpass doesn't contribute to the square wave, it contributes to the lack of resonance in when the filter cutoff is low, in the bassy frequencies..

Well it 'contributes' at least in so far as that it is the decisive factor with respect to the waveforms ultimate shape, no?

the sound of the square wave is mostly in the very typical frequency-dependant waveshape
the waveshaper is not memoryless, there are two capacitors involved, and as a result, at low frequencies the square's middle edge has a soft knee, it gets sharper at higher frequencies, and the pulse-width also varies dramatically
the most audible aspect of the square is the frequency-dependant pulse-width

Id be lying if i said that i completely understood this, (no offense), but the part about the varying pulsewidth is interesting, especially since the 'driver-oscillator' really appears to be just a bog standard Saw. Is this actual research-data gathered from hardware-measurements or a hypothesis based on interpreting schematics? Id like to know more about this.

This is fun:

https://www.youtube.com/watch?v=k5gacnQouSU

Well he just seems to be repeating what many other sources are saying.

He also says that the 303s Filter and Amp are controlled by an AR envelope, (he even draws an Attack/Release graph when he says it), so im not sure he can be considered a 'reliable source'.


At any rate, i myself prefer a more scientific approach, i.e. observe, hypothesize, then try to prove the hypothesis. (And go back to step 1 if you cant.) Absent the option to take measurements from an actual 303 its the best one can do, and so far it would seem that im not entirely on the wrong track because the tests repeatedly give the same result, indicating it cant be just random chance or mere coincidence.

Heres another test setup. The signal goes Saw -> Waveshaper -> HPF, and sure enough im getting the exact same result again:

click to view animation


Interesting, isnt it. But dont get me wrong here, i really couldnt care less if im right or not. All im interested in is whats really going on, so if there is irrefutable proof that the 303s odd waveshapes are not a direct result of the highpassing i would absolutely like to see it.

[deastman]

I really don't see how playing around with z3ta+ is going to reveal anything about the inner workings of the 303. Just look at the schematic... The answers are all right there!