[sdiy] Roland SH-2000 VCO question
21pointy at tpg.com.au
Thu May 16 06:34:06 CEST 2019
>>> On 1 May 2019, at 5:42 PM, Oakley Sound via Synth-diy <synth-diy at synth-diy.org <mailto:synth-diy at synth-diy.org>> wrote:
>>> The actual VCO core is based around the SCR Q103. It's a linear VCO with the main timing cap C102. I believe the 555 is being used as a 'semi-fixed' monostable to produce an error CV, courtesy of IC101 pins 1, 2 & 3, that partly compensates for high frequency tracking droop and also allows some control over the overall pitch of the VCO core.
>> Many thanks for this Tony. I’m not familiar with SCR-based oscillators. I had a look around the net, and found a couple that indeed use a 555, but nothing really like this one. This is more complicated. I poked around with the scope to try to get an idea of what’s happening, I’ve put the new scope shots up here along with the VCO schematic...
>> https://mezzoauto.blogspot.com/2019/05/roland-sh-2000.html <https://mezzoauto.blogspot.com/2019/05/roland-sh-2000.html>
>> You can see there is a big change to the mark-space ratio between these frequencies at pin 3 of the 555. This pulse output is then integrated by the first stage of IC 101, it would seem (see pic)… but I wonder what is happening at the second stage where the negative key CV is introduced?
>> Any further enlightenment from anyone will be greatly appreciated!
> The keyboard is 3 octaves not 4.
> F0 is approx 1.5 kHz and F3 is approx 11kHz.
I found some good stuff in the archives about SCR based VCOs.
I didn’t realise the VCO in the 303 is one such thing - the SCR there is made from two transistors (thanks Richie!)
Neil mentions a Jim Williams App note AN299 that features a VCO with a similar idea.
Juergen Haible was fascinated by them. He wondered why they were considered inferior, when they were put to such effective use in the early Korg synths. This led him on to develop his “JH-720” which had the most temperature stable VCO of all his synths “by far”. (Funnily enough, Juergen DIDN’T like stair-stepped waveforms, so maybe he wouldn’t have liked the SH-2000. I think they sound great in this synth, but that’s probably because of the ladder filter.)
So I think after reading all this, what is happening in the SH-2000 at the top half of IC101 is that the pitch CV directly affects the rate the timing cap charges, hence the frequency. There’s a lot of other stuff going on that I don’t get, but that was the bit that was annoying me (cos I’m a control freak about control voltages.)
I’ve pasted some interesting posts below.
Sun Aug 10 19:23:58 CEST 1997
After reading about the thyratron VCO, I thought: "Why not an SCR
I have quite a pile of them of various ratings, and I wondered if
anyone has tried this? Assuming a saw waveform, the question (for me)
is will an SCR discharge the cap fast enough to make a VCO with decent
- Scott Gravenhorst (Synthaholic)
Mon Aug 11 14:02:24 CEST 1997
Most Korg VCOs actually use SCR circuits !
(MS-20, Polysix, Trident, to name just a few.)
Only that they don't use an SCR *component* - they
build their own SCR circuit from two bipolar transistors.
But it works exactly the same way: It starts to conduct
when a certain gk voltage is reached, and it stops when
the current thru the device reaches zero (i.e. when the
capacitor is discharged).
The first Moog VCOs used unijunction transistors - not
the same, but similar in operation.
An actual SCR component I have only seen in Roland
System 700 docs. Not in the VCO's, but in the envelopes,
if memory serves.
PS: The Korg circuit looks very elegant, as you don't need
any voltage comparator (it's buit in), nor hysterestic switch
or monoflop (the end condition is "current == 0").
Does this circuit also have some drawbacks ?
Tolerances of trigger voltage ?
Temperature dependance ?
Any idea why it wasn't used by everybody?
Mon Aug 11 19:18:06 CEST 1997
UJTs and four-layer devices (SCRs, triacs, etc.) have all
been used for building VCOs. Their major problem is one of
temperature drift--it's serious. Especially in the case of the
UJT, which is often referred to as a "junk part" by analog
design engineers. (If you can find a copy of ANALOG DESIGN
ENGINEERING: ART, SCIENCE AND PERSONALITIES by Jim Williams,
EDN Electronic Series, 1994, you will find a section in there
by Bob Pease about the agony of making a linear V/F converter
(VCO) with a UJT.)
The gas thyratron is vaguely similar to a four-layer device
in electrical operation; it does not conduct until the control
electrode reaches the threshold, whereupon conduction occurs.
Shutting the device off requires shutting off or reversing the
voltage across it. In a relaxation oscillator, this is done by
the capacitor which sets the oscillator rate.
Where the difference comes in is in the utter un-transistor-like
behavior of the thyratron. It is dependent on a hot cathode and
gas ionization, not on avalanching of a 4-layer semiconductor structure.
The 4-layer device varies greatly with temperature, while the
thyratron shows almost no variation. Both devices are not very
linear V/F converters, unless many schemes are undertaken to make
them linear. Still, there is one advantage of the tube over its
analogous semiconductor device.
As for the sound and behavior, well, Bob Edgar got a chance to
try my modular tube synthesizer this past Sunday. Ask him if it's
real. Any comments Bob?
Svetlana Electron Devices
Portola Valley CA USA
Mon Aug 11 19:40:01 CEST 1997
Date: Mon, 11 Aug 1997 10:18:06 -0700
From: Eric Barbour
UJTs and four-layer devices (SCRs, triacs, etc.) have all
been used for building VCOs.
[Gas thyratron stuff...]
And neon lamps too!
Mon Aug 11 19:40:51 CEST 1997
>Their major problem is one of
>temperature drift--it's serious. Especially in the case of the
>UJT, which is often referred to as a "junk part" by analog
Ok, the UJT may be crap. But the thyristor structure
(composed of a discrete npn and pnp) is widely
used in Korg synths and apparetly doesn't cause too
much problems (?).
>Both devices are not very
>linear V/F converters, unless many schemes are undertaken to
I don't understand this. Nonlinear V/F, or better I/F, would mean
that the trigger point would depend on frequency. Just curious ...
Mon Aug 11 22:51:59 CEST 1997
The system 700 lfo uses a SCR too.
SCR VCOs (was: AW: 303 clone filter and VCO not reproducable
Thu Nov 5 14:37:37 CET 1998
>And I think that the SCR method is no good idea for a vco at all
I tend to agree, because some simulations I've done say
the same thing. But I'm still wondering about all these
Korg synths that use them - they are not exactly known
as not stable. Rather the contrary. It's still a mystery to me.
Thu Nov 5 14:53:27 CET 1998
Not to boost with it, but I deal with this SCR stuff every day now.
For ESD-Protection devices. I've just measured about 40 of such structures
@ 27C and @120C Tj. From this I can see that at higher temperature
the leakage gets very much worse for an SCR (implicit current gain,
this gain is quite high for currents in the leakage range). Also the
threshold voltage for the snap-back (this is over head triggering, not
gated trigger) varries widely with temperature. geometry and parameters.
And even worse: aging. I mean these devices see a lot of current, and
this causes parameter shift.
Excactly the kind of stuff I don't want to have in my osc. Principally
faulty. Ok, Korg may have used them, but what is the meaning of "they
are not exactly known as not stable" ? If we compare it with tempco
effects it may look not so important, because tempco is put into some exp
function, whereas trigger level and leakage are more "linear" effect,
but I don't want to call them "second order effects".
At low frequencys the leakage should be noticeable.
Thu Nov 5 16:14:36 CET 1998
What I meant is "what tricks did they use to make them useful
*despite* all these effects you mentioned". Their circuit is not
just a naked npn and pnp, but some passive circuitry around
It's clear that the 3rd transistor is a temperature compensation
of sort, but how good is it ?
They *could* compensate for constant leakage (even if its quite
high) in linear VCOs (where you need an offset trimmer anyway),
but not if ageing plays a role ...
I mean, I know that SCRs have some lousy specs, but I'm after
the design tricks to make them usable for such a delicate
application as VCOs nevertheless. It's done, and not too bad,
and my aim is to understand why and how it works.
(Sorry that I have been a little unprecise in my first mail.)
My new VCO
Wed Jun 2 01:12:54 CEST 1999
I have attatched the schematics of my new VCO.
It's "Poly Modular Compatible" (if anybody is still interested),
i.e. V/Hz keyboard input (from a MUX'ed expo converter etc.),
and it has a tempco-less exponential current-in summing node for
exponential Modulation / Detuning etc.
But I've actually built this VCO for different reasons, namely
a little performance synth inspired by the old Korg Monosynths.
I've been bringing up the topic of thyristor-based VCOs on the
list some time ago, and I remember there were quite some
critical comments. But I wondered why Korg used them over
such a long span of time, in so many different products from their
earliest stuff up to the Trident. I thought they cannot be *that* bad
Though they were also used in true wide range VCOs (MS-20),
I guess that they really have their biggest advantage in the
upper frequency range: Very short discharge time. There is no
delay from a high impedance buffer, no comparator and monoflop,
just the two complementary transistors that immediately discharge
the capacitor when the voltage across them gets over the threshold.
So it's probably the best idea to use them in a configuration
like the 700 (Minikorg), 800DV etc.: Run them in the highest
footage (2'), and use a digital divider to generate the lower
octaves. This is no crappy staircase stuff, mind you. A SAW
wave is reconstructed for every footage by adding just the right
amount of the top octave SAW to the staircase. Apart from
*very* short glitches (inaudible ?) you get very nice waveforms
for all octaves. A bunch of cheap 1% resistors do the job. (Sorry
Jorgen B. - no E12 values; even with E24 I still needed some
With the VCO running at rather high frequencies, a (possible)
leakage of the thyristor (is there any ? Martin ?) would not be
Remains the infamous Interlock problem. Everybody who played
an MS-20 knows how bad this can be. (Maybe it's part of the
MS-20's "thin" sound ?). In order to avoid the use of the unregulated
voltage for the buffer FET (MS-20) or a negative reference voltage
for the exponential converter (Polysix), I chose to generate two
positive auxiliary voltages, +5V and +10V, anyway. Now I
carefully designed that part to get a precise DC voltage, and very
strong AC decoupling at the same time: A slow dual opamp is used
for the reference voltages, with a tantal across the outputs to provide
the HF current spikes as good as possible. (the cap is large enough
to form a dominant pole with the opams' output resistance in order
to preserve stability.) In case this decoupling was not enough, I
have RC-decoupled *both*, the GND and +15V supplies of the
dual opamp. Only the resistor divider for the reference voltage is
connected directly to the rails to provide exact DC values.
I can proudly report that this prevents any interlock. I've built two
of these circuits in close proximity on a single Eurocard (16x10cm),
and I have very pleasant fluctuations of slow beat rates in all
octaves, but no unwanted sync. Needless to say that a pair of these
VCOs has a built in "Linear Detune" feature without extra tricks:
You can set the constant term of the beat rate with the Offset trimmers,
and the frequency proportional term with the Tune trimmers (or with
a tiny DC voltage at the Expo Mod input).
No waveform converters in the drawing - of course you need these
in addition, depending on the desired VCO waveforms.
(BTW, if you don't like V/Hz, just use a tempco resistor and a better
More notes about the circuit, especially the components, in the
Hope you'll like it,
Mon Jun 7 16:30:30 CEST 1999
this reminds me of my very first "serious" DIY music project, the
Tuenker "Sound Organ". Oscillator was with SCR too, no VCO though. The
four sawtooth footages were generated the same way with resistor
networks. No ICs, two frequency dividers on one PCB.
Anyway, I have some questions (sorry if some of these have been covered
What is the usable frequency range of the VCO itself (without frequency
divider), using the V/Hz input? Are the frequency dividers just to
select the footage or do their outputs have to be switched by the
keyboard, when a larger (88 keys) keyboard is used?
I think it's very useful when the footage outputs could be mixed too.
Especially if there were tri/sine converters for each footage, a organ
drawbar thing could be done.
However, if only one footage needs to be available at a time there's no
need for a complete resistor network for each footage. To get the next
lower footage just add the next lower square wave with double the
amplitude of the previous one. The Reset input of the frequency divider
IC could be used to disable/enable the squares. With a GAL as frequency
divider, the footage could be selected with static digital signals;
useful if several VCOs are ganged for a polyphonic system.
Of course the amplitude of the saw is twice as high then (could be
compensated with another switch scaling down the output level).
Instead of the resistor network, a multiplying DAC IC could be used. Not
necessarily cheaper, but maybe space saving. The top saw had still to be
added with a discrete resistor.
Even if all footages are available the same time, with an output buffer
for each footage, only two voltages need to be added for each footage:
the saw of the next higher footage and the square.
Another question: Is the + input of the TL071 really connected to two
Mon Jun 7 19:56:03 CEST 1999
>this reminds me of my very first "serious" DIY music project, the
>Tuenker "Sound Organ". Oscillator was with SCR too, no VCO though.
>four sawtooth footages were generated the same way with resistor
>networks. No ICs, two frequency dividers on one PCB.
Real saw wave outputs or just staircase waveforms ? (This makes a big
>What is the usable frequency range of the VCO itself (without
>divider), using the V/Hz input? Are the frequency dividers just to
>select the footage or do their outputs have to be switched by the
>keyboard, when a larger (88 keys) keyboard is used?
It runs down to 0Hz. The closer you come to the offset voltage of opamps,
the larger will the relative error be, however. That's why Korg 700 etc
used the freq. dividers, and MS-20 etc switched different resistors for
fottage selection (to change the current, but keep the voltage the same !)
I've tested the circuit over approx 6 Octaves. No problem here. You have
to adjust the offset voltage of the opamps, of course. It's easy: zero
input voltage -> adjust for zero frequency (in practice, adjust for one
click every few seconds for the 2' footage)
LF411 will be better than TL071 in terms of drift.
[good divider ideas snoípped]
>Another question: Is the + input of the TL071 really connected to
The GND connection here.
Understanding thyristors (?) (was:AW: My new VCO)
Wed Jun 2 17:53:56 CEST 1999
> is this chinese or what? :)
Japanese, actually. Designed by the clever people of Korg. (;->)
But seriously, I'm not sure if I have understood the circuit myself.
I'm not that familiar with thyristors at all.
What I *think* to know is this:
If you connect a npn and pnp to a typical scr circuit (base to collector,
and vice versa), it would normally not conduct, even if there is a voltage
applied between the two emitters (what you'd call anode and cathode
on a scr). Only if you have a trigger pulse of the right polarity on either
base (what you'd call gate then), one transistor would conduct, and
make the other one conduct as well, even if the gate pulse is not there
anymore. The circuit would lock up in a conductive state. Only when the
current thru the scr would get close to zero because of external reasons
(e. g. the external voltage breaking down because a capacitor is
This would end the locked state, and even when the external voltage is
growing again, another gate pulse is needed to start the action again.
Now there would be another way to trigger the scr - I think it's called
"über Kopf zünden" in German (is that "overhead trigger" (?) ): When
an external voltage would exceed the BC reverse breakdown voltage of one
of the transistors, the current would start to flow even without a gate
(Please note that this is more a *question* than an explanation - I really
don't know for sure.)
Anyway, if you look at the VCO circuit, neither of these mechanisms would
happen, would it ? No external gate pulses, and no excess voltage.
So I suspect that the leakage currents of the transistors would be *used*
for the desired operation. If you remove Q3, wouldn't the tinyest current
from the collector of Q2 be amplified in Q1, and again amplified in Q2,
and the circuit would lock up at once ?
This would be prevented by Q3. I'd say there is a *current* divider built
from Q1 and Q3, the current distribution being determined by the voltages
on the emitters (E) of Q1 and Q3. As long as E(Q1) is more positive
than E(Q3), the vast majority of current would flow thru Q3 and prevent
the positive feedback described above. Only when E(Q1) - the capacitor
voltage ! - comes down near +5V, the current amplification circle would
Now the cap becomes discharged, E(Q1) rising again. But as we now have
*high* currents to distribute between Q1 and Q3, the resistor R6 would
the most of the current to flow thru Q3, and the Q1/Q2 circle is not broken.
You'll get a voltage pulse across R6, however, which is used to trigger
the CMOS divider.
Does this make sense ? And is it different from "normal" thyristor action ?
(What prevents the leakage currents in a normal tyristor from forming
a positive feedback without external gate voltage, btw. ?)
And what is the purpose of the diode ?
I don't know if this shed some light. If not, I hope it will "trigger"
else to put it right or explain it further. (I admit when I did my first
with these circuits, I was looking for a reverse breakdown to trigger it
Thu Jun 3 01:47:11 CEST 1999
How I think it is triggered: the "Gate" of the thyristor, the point where
D1 and Q1s base meet, is held at 5volts (well almost there is a drop at Q3
and R6), the current sink sinks towards 0V. Initially the cap is
discharged, both plates at 10V. When the cap reaches (5volts - Ube) of Q1
it starts to conduct. (A thyristor can be triggered from the anode too.)
Then the current thru Q1 turns on Q2 which in turn acts to raise the gate
potential, further increasing conduction in Q1 and so on... The two
transistors are very soon fully saturated, and now the cap can get
discharged very fast.
When the cap is discharged only small currents can flow and the thyristor
resets. I *guess* D1 is there to prevent the thyristor to keep conducting
on the current that the expo convertor sinks, by altering the turn off
Forgot to mention, that the leakage of Q2 cannot forward bias the BE
junction of Q1, since the emitter of Q1 is at a higher potential than the
base. With that diode back biased the current is forced to flow out of the
gate, or if the gate is open it cannot flow!
VCOs (time to change the subject line (;->) )
Mon Oct 2 12:35:50 CEST 2000
> Me either. A few years ago I asked a question here about what kind of
> temperature dependencies the Korg-style SCRs might have and I still
> haven't gotten or reasoned out the answer. I'd start by guessing that
> any temperature dependencies must cancel out in the various devices,
> otherwise it wouldn't be a practical VCO
In practice, of all the synths I've built, the JH-720 with its MS-20 style
thyristor core is the most temperature stable by far. Switching it on
it is spot on tune to a few cents (compared to divide-down string
ensembles and organs at least). As you normally need 2 BJTs to emulate
a SCR, the 3rd BJT would be for temperature compensation of the
Temperature stable as it is, the general "stability" may be inferior to
designs. It needs some "headroom" to start oscillating (That is, the fist
saw cycle after power on is larger then the normal ones), the series
resistor between the cascode FET and the composite SCR is for some reason
important, and using different BJTs makes a difference too.
I could spice it, I guess.
I did, and reporting PSpice resuluts with the usual grain of salt, I can say
reset times are really impressive. I don't have the numbers here, but from
memory: it was not as fast as Ian's latest saw VCO with 319 comparators,
but in a similar order of magnitude (and therfore better than many other
designs). Just think of it, in the JH-720 (and in the Mini/Maxi-Korgs) it
always runs in the highest octave (2'), with *no* means of HFT compensation.
Speaking of HFT compensation: I'm aware that it's not as crucial in linear
VCOs as in exponential ones. We don't have the Rbb error, but we still
have the reset time error, and (depending on the implementation) we also
have a latency time before a comparator or SCR or hysteretic switch
Now how can we measure that latency time ??
From: rburnett at richieburnett.co.uk
Subject: [sdiy] Roland 303 VCO questions??
Date: 15 August 2011 6:37:09 PM
To: synth-diy at dropmix.xs4all.nl
Reply-To: rburnett at richieburnett.co.uk
i have been looking at this schematic all day and I would really
like to understand how it works.
anyone want to take a shot at it?
the schematic I am working from is HERE:
I am interested in parts in the VCO, basically IC11B to a little
bit past Q24
IC11B is the VCO CV buffer from the R-2R ladder DAC. It also forms part of the
portamento circuit with analogue switches IC12C/D and capacitor C35 to limit CV
slew rate during programmed slides.
Matched transistor pair Q26 and it's servo amplifier IC11A convert the linear
pitch CV from IC11B into an exponential current sink. This achieves the
approximate 1v/octave law for the oscillator. The expo converter continuously
drains current from capacitor C33 in order to make it's voltage ramp downwards
linearly. Q24, 25 and 27 form a thyristor. This triggers when the voltage on
C33 falls below a preset level determine by the 5.333V applied to Q24 emitter
via R101. When this thyristor implementation fires Q27 and Q25 both turn on,
with each providing the base current to keep the other turned on. This results
in C33 being rapidly recharged up to approximately 12V via Q27 and Q25. They
stay in the conductive state like a thyristor would stay latched until the
charging current into C33 falls below their holding current (typically a mA or
so.) Then both transistors drop out of conduction and the expo current sink
causes the capacitor voltage to ramp downwards again at a rate determined by the
So in summary, you really have an exponential current sink centred around
Q26/IC11A, and a crude voltage comparator/reset circuit based around Q24, Q25,
and Q27. This gives you a raw downwards ramping sawtooth with fixed amplitude
and exponentially conforming frequency.
Q28 acts as a high input-impedance unity-gain buffer to buffer the voltage on
the integrating capacitor before passing the sawtooth output on to the VCF or
square wave shaper. Q8 and it's associated circuitry form an over-driven
common-emitter amplifier and act to shape the sawtooth waveform into something
that is occasionally vaguely square in shape at some pitches! Then it's onwards
to the VCF...
I hope this helps,
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Synth-diy