SoylentNews is people
SoylentNews
from the a-feature-is-something-that-is-broken-in-a-useful-way dept.
That headline sounds suspect, but it is the most succinct way to explain why the Roland TR-808 drum machine has a very distinct, and difficult to replicate noise circuit. The drum machine was borne of a hack. As the Secret Life of Synthesizers explains, it was a rejected part picked up and characterized by Roland which delivers this unique auditory thumbprint.
Pictured above is the 2SC828-R, and you can still get this part. But it won't function the same as the parts found in the original 808. The little dab of paint on the top of the transistor indicates that it was a very special subset of those rejected parts (the 2SC828-RNZ). A big batch of rejects were sold to Roland back in the 1970's — which they then thinned out in a mysterious testing process. What was left went into the noise circuit that gave the 808 its magical sizzle. When the parts ran out, production ended as newer processes didn't produce the same superbly flawed parts.
(Score: 5, Interesting) by Rich on Friday September 14 2018, @11:21AM (5 children)
There's a discussion going on on Gearslutz about the issue. Atsushi Hoshiai, one of the lead designers of the 909 (but apparently not 808), chimed in and said he could not imagine that there was something about mystic spectra in those transistors. Still, someone seems to have obtained a number of samples of the transistor in different selections and is going to make spectral measurements.
I imagine that the selection process might have been more about noise levels. I had the noise circuit of the Jupiter 8 (which is about contemporary, but more complicated) on my bench and in original specs, it would overdrive to a no-longer-adjustable level with a random BC547 as noise source that I picked. Only when I reduced the feedback of an opamp in the noise amplification chain to a quarter, it would work well.
Note that there's an engineering change (whatever "6 lot ~" means...) for the 808 where they float the negative input of IC24A that has an extra SMD cap in the feedback (causing a very slight cutoff), so not even all 808s are created spectrally identical. I'm not into deep 808 research, so I haven't tried to fully understand the effects of dropping R127 for C200, but it looks it extends the adjustment range (which would support my theory) and might assist with dc offset.
(Score: 2) by RS3 on Friday September 14 2018, @01:48PM (1 child)
It will be interesting to find out the noise spectra of the original transistors. There are certain frequencies and harmonics that are more pleasant to most people's ears- more chordal, less dissonant. It may be that some kind of EQ could replicate the sound, but maybe not...
Was it simply clipping?
(Score: 4, Informative) by Rich on Friday September 14 2018, @03:25PM
Yes, iirc with some unpleasant TL082 phase inversion thrown in. The original "module controller" (distribution board for four voices) uses two primary noise amplifier stages (IC3 and IC4), of which the second has a 470k feedback resistor against a variable resistance from 2.2k to 49.2k. I had to lower the 470k R33 to 100k to get into a range where I could properly adjust the clipping range. Speaks for the quality of the original 2SC945 ("selected") vs what we now get as BC547. :) But I guess they've had it at some point, because a revised edition module controller uses an OTA with autogain, with the trimmer gone.
You can easily find all the old Roland "Service Notes" on the web. Seems the noise source was a continuing source of trouble for them, because the TR-909 uses two CD4006 shift registers in a digital feedback scheme (but with some rather odd unbuffered half-analog CMOS logic around it).
(Score: 3, Informative) by RS3 on Friday September 14 2018, @02:05PM (1 child)
Another thought: I haven't seen the circuit yet, so I'm speculating / theorizing, but it's possible the "faulty" transistors had a slightly worse high-frequency rolloff; IE: didn't amplify the highest frequencies as much as the lower ones in hearing range. Noise containing more amplitude of frequencies at the higher end of human hearing is generally harsher. Think: turn down the treble control a bit, but at a higher frequency than an average treble control (E.g.: 15K rather than 4K).
Also, in that time period many people were (still) using noisy carbon resistors, and maybe occasionally using metal-film in audio circuits, so maybe that is a factor?
(Score: 3, Informative) by RS3 on Friday September 14 2018, @02:08PM
Yeah, I just found a pic of the circuit board and I see lots of carbon resistors and no metal-film. https://www.sequencer.de/pix/roland_TR/tr808_offen_original.jpg [sequencer.de]
(Score: 3, Informative) by RS3 on Saturday September 15 2018, @02:46PM
Okay, I need to practice what I occasionally preach about doing less speculating (but it's so easy to write stuff online!)
After some searching: transistor gain, frequency response, etc. have nothing to do with this thing. It's not being used as a transistor (3-wire amplifying device). Q35's BE (Base-Emitter) junction is used as a diode in reverse-bias at the Zener / avalanche breakdown voltage, and amplifying the resulting avalanche noise. One simple reference: https://www.electronics-tutorials.ws/diode/diode_7.html [electronics-tutorials.ws]
The chatter is that the defective 2SC828s create more "sizzle" - so more high-frequency component. It should be fairly easy to measure the spectral density and replicate that, even in software as JoeMerchant wrote https://soylentnews.org/comments.pl?sid=27590&cid=734766 [soylentnews.org].
But sticking with the simplicity of the original analog circuit, http://secretlifeofsynthesizers.com/wp-content/uploads/2018/05/TR-808-transistor.jpg [secretlifeofsynthesizers.com] one could vary the bias resistor, 1M R128, which is certainly adding noise, and the load resistor, 300K R131, and of course op-amp gain pot (TM4), and change the sound characteristics. Lower resistor values should give better high frequency response, but a lower R131 will shunt away more of the noise signal, so more op-amp gain would be needed, but op-amp circuit high frequency response might suffer (more study needed).
It would be very interesting to compare the avalanche / Zener voltage of a defective part to a good one. My hunch is the defective one is lower, so noisier.
As a kid I learned the hard way that transistors and resistors get much noisier if overheated by a soldering iron. One could build a noise generator just using carbon resistors and an op-amp- the higher the Ohms the better. One could hurt some transistors with external heat, or carefully overdrive some reverse-bias on the BE junction and cause noisier behavior. Maybe I should do that and sell them on ebay / reverb...
Maybe someday if I'm bored of writing here I'll mock up the circuit, change some values, cook some transistors, see / hear what happens ...