To celebrate the 2016 return of the CEM3340 chip, I thought I’d do another page in my series looking at how various synths implemented classic synth chips (the others being CEM3320 Filter designs and SSM2044 LP Filter designs). This is especially useful right now, since many people have bought a few CEM3340s for their own use and are looking around for circuits to use them in. If that’s your purpose, note that these are all synth voicecard circuits, so they’re used in a known situation. If you want a CEM VCO for a modular synth, you’ll need much more protection on inputs and outputs than most of these circuits offer.
First, the datasheet design.
The datasheet design uses pretty much all the features of the chip, and shows one or two of its quirks too. The CEM3340 is extremely unusual in that it offers both Hard and Soft Sync inputs (pin 6 and pin 9), and neither of them work the way you might expect. As we’ll see, this lead to them being widely ignored. However, the datasheet does offer alternative circuits for synchronization which can give you the classic sync effect, and manufacturers seem mostly to have gone with variations on this instead. The datasheet’s claim that their inputs “provide a wider variety of synchronized sounds than available through conventionally synchronized oscillators” seems to have fallen on deaf ears. Sometimes people just want what they know and love.
One of the quirks of the CEM3340 is the waveform outputs. For a start, these are all different levels. The Ramp is 2/3rds of the positive supply. The Triangle is 1/3rd of the positive supply. The Pulse/Square is the positive supply level minus 1.3V. For a typical +15V supply, this gives levels of 0→10V, 0→5V, and 0→13.7V respectively. Another quirk is that the Pulse is an open collector output, so it needs a pull down resistor to set the lower level. Usually this would just be grounded, but it could be taken to the negative supply. Most synths compensate for the differing levels in the following waveform mixer stage. This is easily done by changing the input resistors in an inverting op-amp mixer.
Another quirk is the supply voltage. Typical supplies for synth circuits would be +/-15V or +/-12V, but the CEM3340 can’t stand more than 24V between its supply pins, so even +/-12V is right at the limit. Curtis dealt with this by adding a zener diode to limit the negative supply connected to pin 3 to -6.5V, so this gives 21.5V across the chip with +15V positive supply. The zener diode needs to be protected from over-current, so you typically see a current limiting resistor hanging off pin 3 (820R with -15V in the datasheet circuit). Alternatively, you can run the chip from +15/-5V supplies, in which case pin 3 can be directly connected.
The circuit breaks up into various functional units quite neatly. Let’s have a look at some of them.
Starting numerically at Pins 1 and 2, this is the temperature compensation circuit. This works by generating a temperature-dependent voltage which is then multiplied on-chip by the incoming CV. Since the VCO and the temperature-compensation circuit are all on the same die, they’re all the same temperature, and this gives a highly accurate result. There are still some minor errors from the multiplication process (hey, this is analog, after all) but the datasheet gives details of how to trim those out too if you can be bothered.
Pin 3 we’ve looked at. Pins 4, 5, and 6 are inputs and outputs all labelled above, so no need to discuss those.
Pin 7 is more interesting. It’s the High Frequency Tracking. This pin produces a current which can be used to compensate for the way that the time taken for the internal comparator to switch becomes significant at higher frequencies (>5KHz or so). The lower half of the preset resistor serves as a resistor to ground which converts this current to a voltage, and the upper half and the series resistor (1M) feed a small portion back into the Frequency CV input at Pin 15.
Pin 11 is the next important one. It’s the VCO’s timing capacitor. Use a good quality capacitor here, with low leakage and low tempco. Polystyrene film used to be the best regarded, and is specified in many of the manufacturer’s schematics, but Mica is mentioned in the datasheet and there are other options now (C0G, for example). Note that the datasheet circuit describes the capacitor as “1000pF” rather than “1nF” and virtually all manufacturer’s schematics I’ve seen have followed both the datasheet value (logical, since they all want the same audio range) and the naming convention. I’ve saved space by breaking the convention and calling it “1nF” instead.
Pin 13 is the Linear FM input, with an associated bias network. I haven’t seen a schematic that uses it, so mostly you only see the bias network (1M5, 470R, 10n) connected to this pin.
Pin 15 is the Frequency CV input. This is a virtual ground summing node, and you usually see a bias network as shown (360K, 470R, 10n), although often with the values tweaked, followed by the summing resistors coming from the various CV sources.
Ok, so here’s one of Roland’s several implementations. This seems to be virtually identical to the Jupiter 6, bar one or two minor value changes. It’s a bit hard to tell, because the Jupiter 6 schematic doesn’t tell you what values are inside the special resistor arrays Roland used in both synths. The layouts are the same for sure, so I’m guessing the arrays are too.
The SH-101 is a strange beast, in that it uses four different positive power rails; +5V, +9V, +14V and +15V. In addition there are Ground and -5V power connections.
So, what’s different in this one? There’s an emitter follower buffer on the Pin 4 Pulse output, and the Pulse Width control on Pin 5 is handled quite differently to other designs. They’ve added a trimmer to get an exact value on Pin 14. The recommended value is 1.8K, and the combination of 1.69K plus a 300 Ohm trimmer gives that value. Other synths have simply specified a 1% resistor in this location, so the trimmer seems a little like overkill. There are also trimmers for the “Range Width” (that is to say “1V/Oct”) coming in from the DAC, and a VCO Tune trimmer to set the basic pitch. Again, there’s no linear FM.
Sequential Prophet 5 Rev.3
As everyone knows by now, the early Prophets used SSM chips, while the later Rev.3 moved to CEMs, which Sequential then stuck with for all future products. You can argue about the relative merits of one versus the other, but the fact is that they were having tuning problems, and using the CEM3340 helped them solve them.
Interestingly, the High Frequency tuning trim on Pin 7 isn’t used. Presumably the P5’s internal autotune routine was able to compensate for the notes tending towards flat at the high end, so there was no need.
Notice the sync circuit. This is the “Figure 5” sync circuit from page 6 of the datasheet, and produces the standard Hard Sync waveforms.
Also note that this is Oscillator A, the Sync Slave. Oscillator B is the same, except it uses the Triangle output and doesn’t have the Sync circuit.
Sequential Prophet T8
Another Sequential design, this time their flagship Prophet T8, five years after the Prophet 5. There are several differences to note from the Prophet 5 design. The Triangle output is used, and the Sync circuit has been altered, no longer using the Soft Sync input, but still using a transistor to reset the triangle core.
The Pulse Width mixer has also gained a few extra components. There’s a resistor connected to -5V which gives a +5.5V offset on the output, and the diode acts as a half-wave rectifier which prevents the output from going negative, since the input range for Pin 5 is 0→5V. Presumably this situation couldn’t arise in the Prophet 5, making such protection measures unnecessary.
With the other inputs at 0V, the +5.5V offset will set the pulse wave to beyond 100% – e.g. it will turn it off. Whereas the Prophet 5 uses three 4016 CMOS switches to turn each waveform on or off, the Prophet T8 only uses two, saving a CMOS switch by using this alternate method of turning off the Pulse output. It’s a clever little trick to save hardware.
When Moog tried building a polyphonic synth (still the only “proper” voice-based polysynth they’ve built, even in 2016 – Come on, Moog!) they couldn’t fight the tide and used CEM3340s for the oscillators. Imagine trying to tune six Minimoogs all piled up together!
The Memorymoog’s reputation has been somewhat spoiled by its poor reliability, probably a result of it being rushed out of the door in an attempt to save the company from bankruptcy. It’s a shame they didn’t have a little longer to spend on it, because in many ways it’s a great synth, perhaps even a masterpiece. Imagine! A three oscillator polysynth with Moog filters! That’s got to be good, if it works.
The CEM3340 implementation is pretty standard, apart from the 1M resistor linking the PW input and Pulse output which
I suspect is to cancel the DC offset you get with pulse waves is to add a little hysteresis to the pulse comparator (Thanks Bob! – see comments), and the non-inverting op-amp buffer on the Triangle output. This amp has x2 gain, so it boosts the triangle to 0-10V, the same range as the Ramp wave.
One thing to notice is that the Moog design uses three trimmers, whereas the Sequential designs manage with only one. I imagine that the Sequential autotune routine provides high frequency tracking and range trimming and so forth, whereas it’s not exactly clear what Moog’s autotune manages to do, since everything gets a trimmer anyway. Probably it compensates for the synth warming up (the Memorymoog had a fan to keep it cool!) and that’s about it. It’s not a great design, since three trimmers per osc for each of three oscillators is nine trims per voice, 54 trimmers total – and that’s just for the oscillators! What a pain in the backside!
Update 11th February 2017
Heinz Weierhorst spotted that I’d drawn the connections to the high frequency trimmer incorrectly (below in the comments). This affected the Datasheet, MKS-80, and MemoryMoog schematics. This is now corrected.