[sdiy] Synthex Oscillator

[Tom Wiltshire]

Brian,

Your idea that the falling counter acts to linearise the RC curve of the capacitor discharging is very clever. However, I can't believe it's true. After all, why not just [ut the cap in the feedback loop of the op-amp (there is one there already, so it's no big deal) and linearise it like that. Alkternatively, why even bother with all that when you've *got* a highly linear ramp from the counter. It doesn't make enough sense to me just yet, but it's a very interesting notion, and it goes a long way to explain what Mario might mean.

Thanks,
Tom

==================
       Electric Druid
Synth & Stompbox DIY
==================


On 9 Jul 2017, at 05:07, rsdio at audiobanshee.com wrote:

> 
> On Jul 8, 2017, at 1:16 AM, Tom Wiltshire <tom at electricdruid.net> wrote:
>> The Synthex is a somewhat revered, certainly highly thought of synth. And it has one of the most unusual oscillator designs you're likely to see. I've been studying it in the hope of understanding what exactly is going on. My interest here is to be able to clone it using modern technology, since using a boardful of LS series logic seems a bit like overkill, although it'd be cheap enough to do and all the chips are available.
> In my mind, a boardful of LS series logic begs to be put into an FPGA, but only if the cost is right. So far, every design I've done has turned out to be cheaper as discrete LS logic, simply because those parts are so damn cheap and prevalent. The equivalent FPGA has always been more expensive. There's also the inherent advantage that LS logic does what it's supposed to do as soon as it's soldered on the board, whereas programmable logic has to be programmed before it will do what you want.
> 
> I mention this because you should consider making your design first, and after you're settled on the logic you could take a look at how it would affect size and price to squeeze all or part of the logic into a CPLD, FPGA, or other device.
> 
> 
>> There's another little red box containing a single NOR gate. Now this NOR gate means that when pulses come in from the first divider, the clock pulse to the second (blue) counter) disappears. I understand this is known as a "pulse swallowing counter". Note that the second counter is not *clocked* by the first one, but is rather *cleared* by it (LS393, pins 2 and 12). It's clocked only from the 4MHz master clock via that NOR gate. If I'm getting it right, instead of giving you 1/255th of the input frequency, it gives you 254/255ths of the input frequency (for example) because every 255th pulse gets swallowed.
> I think it might be very important to pay attention to the two AND gates at 3G. These are controlled by what I would call a "Mode" signal that is set by D4 of the latch at 1E. Depending upon whether that mode bit is high or low by default, you might be chasing a red herring. In other words, if D4 of 1E is always low (except in some weird synth option), then it doesn't matter that much. At the very least, I assume the D4 is not changing during a note - if it changes at all then it probably only changes before Note On because the processor would have a hard time synchronizing its updates to D4 to coincide with the waveform.
> 
>> That determines the basic pitch, I think. From there we move to the green box, which is the wave shaping part of the oscillator. This consists of another 8-bout counter (2 x LS193) fed to a weighted-resistor DAC. The DAC output can be inverted by the 8 XOR gates.
>> 
>> There's a lot I don't understand here. My best guess is that this is used as an 8-bit counter to generate ramps, and as an 8-bit up/down counter to generate triangles, and the XOR is also used by VCO2 to provide a basic digital Ring Mod function. What I thought was going on was that this was then followed by a smoothing capacitor (93, 10nF) and a transistor to remove that cap from the circuit so ramp edges aren't affected (BC559).
> […]
>> This contradicts my understanding of the circuit on many ways. He's suggesting that the DAC is used as a digitally controlled current source for a DCO, whereas I was seeing it as a digitally generated audio output, although with smoothing.
> Ok, despite being very interested in this, I still haven't managed to finish my analysis. That said, I think it's most important to pay attention to the node connecting C93, R89, T90, and the resistor ladder "DAC" - plus some consideration of 1Nb and the analog switch at 2N.
> 
> Kirchhoff's current law should be able to tell us everything we need to know about C93.
> 
> T90 (the BC559) can quickly charge this cap to +5V when the AND gate at 1H tells it to.
> 
> The op-amp at 1Nb should ideally have no current flow into or out of pin 3.
> 
> Assuming 2N is open and T90 is off, and also assuming that the initial state of C93 is +5V at the beginning of each cycle, then the resistor ladder made by R92 and R94-R100 will control the rate of discharge of C93. Any of the NOR gates in 1L and 2L that are outputting a "0" logic level will be combined in parallel to discharge the cap. I'm assuming that the NOR gates that are outputting "1" would initially have no effect because there would be +5V on both sides of the resistors, but as C93 discharges those resistors will start conducting current and slowing the discharge of C93. I can't quite predict the exact consequences, but see below (*).
> 
> Note that if there was an active stage between the resistor ladder "DAC" and the cap, such as with an op-amp follower, then it would surely be a DAC because the current flowing through the resistors would depend entirely upon the DAC and not the voltage on the cap. If C93 were acting as a smoothing capacitor, then I would expect a fixed resistance in series with the DAC output, making a simple RC low pass filter. I don't see C93 as a smoothing capacitor at all.
> 
> The analog switch at 2N is bizarre, because it either grounds the cap or connects it to +5V, depending upon the MSB of the "DAC" so I'm not quite sure what to make of that without seeing circuitry off the provided schematic.
> 
> * Here are my thoughts on this interesting digital controlled current source. The standard ramp / sawtooth oscillator is an op-amp integrator. Because an op-amp is an active circuit, an integrator has a linear charge or discharge rate, unlike the typical RC circuit which is logarithmic. In this Synthex circuit, the capacitor is not inside the feedback loop of an op-amp, so it's not really a standard integrator. My wild-ass guess is that the changing current flow of the resistor ladder ends up changing what would normally be a logarithmic discharge rate (as seen in an RC circuit) to perhaps something that is more linear. I'd have to build the whole thing in SPICE to see the curve, but since you're already planning to simulate then I can possibly get by with being lazy and waiting for your results! In light of my theory, perhaps you could start by simulating just the components around C93 and see what you get.
> 
> 
>> If it is a current source, I can't see how it gets a set value - it's a counter, not a latch.
> That's the biggest mystery in my cursory glance at the circuit. I would have expected that a value would be set according to the programmed frequency, but perhaps we can figure it out.
> 
> Returning to my wild-ass guess, perhaps the counter adjusts the discharge resistance slowly over time to linearize what would normally be a logarithmic discharge.
> 
>> So, what insights can you offer? Is my "discrete digital oscillator driven from divide-down technology" view of the circuit right at all? Is it really a discrete DCO design, like Mario seems to be claiming? What on earth is going on?!
> Mario does refer to C93 as the central part of the sawtooth DCO, so I'd focus on that part of the circuit first. Keep in mind that the resistor ladder "DAC" is not buffered, so it's not really outputting a Voltage. Instead, it's a bizarre, digitally-controlled discharge network for the sawtooth cap.
> 
> Now I need to focus on the LS193 parts and see what's up there. Or maybe I should move to SPICE already and stop the wild-ass guesses.
> 
> Brian Willoughby
>