CEM3320 Filter designs

Like my SSM2044 page, this page is a look at how various synths implemented the CEM3320 filter, with the datasheet design as a reference standard. Whereas the SSM2044 is a dedicated lowpass filter, the CEM3320 is just a set of filter building blocks. In this respect, it is more like the earlier SSM2040. The CEM equivalent of the SSM2044 is probably the CEM3328, which is basically the same as the 3320, but wired up as a dedicated lowpass filter, thereby saving a few pins.

First, the datasheet designs.

Datasheet CEM3320 Lowpass Filter


The datasheet makes what is actually a pretty straightforward circuit look really complicated by showing the chip with its pins numbered in order. This makes the whole thing a tangle, as the layout of the separate stages inside the filter is not at all logical. I’ve put the stages in order here and just noted the pin numbers to make it clearer.

Datasheet CEM3320 Highpass Filter


This is a slightly simpler circuit than the lowpass design, and it is fairly easy to see the way that the “R”s and “C”s have been swapped around to turn a lowpass filter into a highpass. This filter is a fairly rare beast, a 24dB/Oct highpass.

Datasheet CEM3320 Bandpass Filter


This circuit is simply a combination of the two above. The first two stages are set up as highpass filters, and the last two as lowpass filters. This gives a 12dB/Oct bandpass response overall.

Sequential Pro-One


The Sequential Pro-One uses the filter in a circuit which is quite close to the datasheet. They’ve added a buffer op-amp on the output, which also affects the resonance path. They also use the virtual ground node at Pin 1 to sum signals from the two oscillators and the noise source. The filter capacitors have also been halved in value, which is going to raise the basic cutoff frequency of the filter by an octave.

Update: What is that buffer for, anyway?

It seems from experiments that the plain datasheet lowpass circuit has a couple of characteristics you might want to avoid. One of these is that the signal level when the filter is oscillating is much less at lower frequencies than at high frequencies. This also affects the filter’s ability to start oscillation at these frequencies. Adding a buffer like in the Pro-One (notice it has a gain of x3.4) solves this problem and gives the filter a more even response.

Oberheim OB-Xa

The Oberheim OB-Xa is a quite extraordinary synth in that it has two filters, but only lets you use one at once, which means one filter is always entirely wasted. Obviously CEM chips didn’t cost then what they cost now!

The idea of this was to provide a switchable response between 2 pole, 12dB/Oct lowpass and 4 pole, 24dB/Oct lowpass. There is one filter for each response, and you can switch between the two.

Oberheim OB-Xa 4-pole Lowpass Filter

The 24dB/Oct lowpass filter is a fairly standard design, although it uses a +15V/-5V supply. Otherwise, it is much as the datasheet suggests. I don’t understand why they specified 1% resistors for the Frequency CV input, and then went and put a 10K trimmer in – seems daft to me. Perhaps the 1% resistors had better thermal characteristics or something.

Oberheim OB-Xa 2-pole Lowpass Filter



The 12dB/Oct lowpass filter is a completely different story! This is nothing like the datasheet! Instead, Oberheim have used the chip to implement a 2 pole state-variable filter. Stage 1 is only used to mix the signals. This is usually done with a differential op-amp, but they’ve saved one amp by doing it like this. Stages 2 and 3 are the actual filter poles, and Stage 4 is unused. I think the fact that they haven’t got a proper differential amp accounts for the extra-complicated feedback and resonance paths, but I’m not sure. Given that this is a state-variable filter, it should also produce a highpass and a bandpass response, but these aren’t used here. It’s a pity, as that would have set the OB-Xa apart from the competition.

Elka Synthex Multimode Filter (LP/BP/HP)

The Elka Synthex has probably the most sophisticated filter circuit built on the CEM3320 to ever have been used commercially. It uses analogue switches to reconfigure the circuit and provide a variety of responses. The Oberheim OB-8 that followed the OB-Xa did something similar, but only provided the 12dB and 24dB lowpass reponses that the OB-Xa had.


I’ve left out some of the Freq CV input circuitry, which includes a CV mixer and some other components to allow an inverted filter envelope.

The filter looks pretty complicated at first sight, but once you break it up into its consituent units, it isn’t so bad. The first two stages both include a pair of analogue switches to enable them to switch between lowpass and highpass responses. For each stage, with the switches open, it’s highpass, closed it’s lowpass. The other switches select one of two output points, either after the first two stages, or after all four. Three switches are used for this to allow some adjustment of output and feedback resistances. The overall structure is shown below.


This structure can provide various responses, of which the four below are implemented in the Synthex.


Note that other configurations are possible that the Synthex doesn’t use. A 12dB lowpass filter could also be provided, as could a 18dB lowpass, combined with a 6dB highpass. Only the Oberheim Xpander/Matrix followed this direction to its logical conclusion.

Building these designs

Since I originally wrote this article, the CEM3320 VCF chip has been cloned as the AS3320. We have the AS3320 voltage-controlled filter chip available in the shop.

44 thoughts on “CEM3320 Filter designs

  1. Quoting your sentence above; ‘The datasheet makes what is actually a pretty straightforward circuit look really complicated.’

    So true. I’m coming to a point where I just re-draw some circuits before even trying to understand them. You can only set up so many memory buffers in your mind, and after that it becomes inhuman to go for more, and possibly even counter-productive.

  2. Thank you for this explanation. Can I clarify something?

    The switches above are labelled A, B, C, or D.

    Is it correct to assume that the three switches marked A operate together, the two switches marked B operate together and switches C and D operate separately from one another and switches B and A?

    So the truth-table is like this:

    A B C D
    24dB low pass 1 1 1 0
    6dB band pass 0 1 0 1
    12dB band pass 0 0 1 0
    12dB high pass 0 0 0 1

    1. Yes, there are four control signals, A, B, C and D. A and B operate multiple switches. C and D only operate a single one.
      Your truth table looks pretty good to me. The only real query is switch C. It’s not clear when that is used. I imagine if one builds the circuit, it becomes fairly obvious since it allows a level boost to the output.

      Good luck!

  3. The datasheet tells us that max voltage between Frequency Control and GND pins is ±6V and that max voltage between Resonance Control and GND pins is +2V, -18V. Great, I can keep things within those limits, but I can’t quite make out what would be the voltage range for a full sweep of the cutoff frequency (for instance in the four-pole LPF from the datasheet).

    0V to 5V?
    -5V to +5V?
    -6V to +6V?

    What about the resonance CV?

      1. Hey Sylvain

        Did you find the voltage range for the resonance control input? I can’t find it anywhere.

        1. The reason you didn’t find a voltage range is because the resonance control input is a current input. The datasheet gives a specification of 100uA maximum, but depending what series resistor you use, you can use more or less any voltage level to achieve that. The datasheet circuit uses a 100K resistor which equates to a 0-10V CV input.
          The Sequential Pro-1 synth uses a 100K linear pot between ground and +15V to control the resonance. This 0-15V CV is fed to pin 9 of the 3320 via a 200K resistor. This gives a maximum of 75uA, and suggests that 100uA is more than adequate. Hence we can go a bit lower than 100uA and still expect to reach full oscillation.

          1. Ok, i see. Thanks for the explanation. I’ll try to calculate with these informations. Thanks Tom!

      1. The input impedance of the resonance VCA signal input is only a few K, so having a buffer in there might be a good idea.

        Update: I’ve done some experiments. See my comments in the Pro-One section above.

  4. Just looking at the Elka schematic – Is there a typo? Shouldn’t the resistor at stage 3 (between pins 17 and 15) be 82k – same as the others in the filter stages..

    1. Yes, the Minimoog ladder filter is also famous for this “passband droop”. It does lend a certain character, and quite a few modular synth filters include a compensation switch, so you can choose the compensated or uncompensated sound.

  5. Hi Tom.. I’m looking to build the Elka filter, and I’m comparing the original synth schematic to the one posted here..
    As well as the 820k resistor between CEM3320 pins 15 and 17 I mentioned earlier, there are a couple of other differences..

    The 68k resistor and 1uF cap at the audio input – these are switched round on the Elka so the cap comes first..
    The 220k resistor at the emitter of the the transistor is a 10k, rather than the 220k on the schematic here..
    The voltage divider at the Freq CV input – the Elka has an extra 82k resistor tied to -12v right before the CEM3320 Pin 12..

    Are these changes intentional, and how are they likely to affect the function of the filter? Particularly interested in the CV input – since I’m looking to control this VCF with an MCP4822 at 0-5v..

    Best wishes,

    1. Certainly I didn’t make any intentional changes, so if you’ve spotted differences, they’re errors that I must have made when copying the schematic. I admit by the time I got to the Elka one I might have been going cross-eyed! My apologies.

      To answer your question about the voltage divider for the cutoff, the CEM3320 (or AS3320) needs about 200mV CV to cover the whole audio range (18mV/oct) or perhaps a handful more. The MCP4822 outputs 0-4.095V, so a divider of 39K/2K gives about the right range.

  6. What is the point of the 3kΩ resistor from pin 8 (Vres) to ground, in the Pro-One? There’s a 3k6Ω resistor to ground internally. I assume it’s a voltage divider along with the 51kΩ. 3k6 and 3k make about 1k6, which halves the signal fed back, compared to only using the 3k6. Is that to compensate for the 3.4 gain? Couldn’t a gain of 1.7 do the same thing? Should I go make coffee?

    I have the datasheet LPF running at ±9VDC with a STOMPLFO and a bare bones envelope follower. It sounds great, and is a quarter the size of the LM13700 rat’s nest it replaced, not to mention the LFO, thanks Tom!

    1. Yes, it’s cutting down the input level to the resonance input. The Res VCA is an OTA, so can’t cope with a high signal level. You *could* reduce the gain of the buffer, but then the signal output would also be quieter, so this way has an advantage.

      Good work with the filter/StompLFO/env follower combination. Sounds like fun.

  7. How important do you think CV reject trim is in these circuits? My friend and I have built a bunch of AS3320 based circuits. We always include the CV reject trimpot but not sure if it’s actually necessary.

    1. If leaving it out is good enough for Oberheim and Sequential Circuits, it’s good enough for me!!

      I notice the datasheet circuit ignores it for Lowpass, but includes it on the Bandpass and Highpass circuits – could it be that feedthrough is more noticeable in those situations? You’d be more likely to hear a “thump” if there’s less low-end content to mask it.

      1. Yes, we’ve noticed more thumping problems with the high pass. Of the ones listed here the Elka Synthex Multimode seems to be the only one that isn’t just a low pass and excludes the cv reject.

  8. Hello, Thank you so much for this page and article. So I just got into music synthesis, However I am an electrical engineering student. I am embarking on this ambitious journey to build a semi analog polysynth with my design teammates. Please I would like to know if it is possible to build this with an EDA. I’m specifically asking because I woulds like to know if there is a parts component library for this CEM IC’s, because it would be tricky trying to create a component library as I don’t know how to. Also Please I would like to know if there are any additional resources to consult as we embark on this journey for a class project. Thank you very much for your help.

    1. You probably won’t find a component library that includes these parts or their modern clones, but that shouldn’t stop you. You will find that is the case for quite a few parts you will need, and the ability to create new components in the software you are using is an important skill that you’ll need to know. I’d recommend that you find out how and go for it. Good luck!

  9. Has anyone found that in highpass mode the resonance doesn’t behave nicely?
    I’ve tried several different arrangements. At best, I can only get a slight resonant peak. At worst, I’m getting some kind of build up of squeaky noise, almost like static. Not sure what I’m doing wrong.
    Note that I am using the pro-1 buffered resonance.

  10. hi, i was analyzing the OB-Xa 4 pole Lowpass filter circuit and i’m a bit confused on the freq control input (pin 12)
    would you be so kind to help me understand?

    so..the Pin 12 is a voltage dependent input, so it doesn’t matter the current flowing into it (anyway it looks like to be max 250 uA) but the voltage..
    The Offset (+15v/0V) Voltage range is 0/150 mV whereas the Freq CV range (given a 0 /+5V source) is 0/100mV. This should give a max final range 0/250mV.
    Should’t the total summed range be 0/200mv max?
    Even if not so, in this case.. would a sort of clipper circuit or clamping the output voltage to the desired range be a solution?

    1. Yes, the Freq CV input is a voltage input not a current input. The absolute max for the chip’s input is in the datasheet, but it’s *way* above what you’d ever use on the Freq CV, so it doesn’t matter much. If the Freq CV is 300mV rather than 200mV all that happens is the filter cutoff goes up above audio where we can’t hear it and it makes no difference. Nothing blows up.

      The Offset (“Init Freq”) trimmer is intended to make sure that the sweepable range of the filter from the Freq CV input covers the part of the spectrum we’re interested in – the audio range from 20Hz to 25KHz. If the trimmer is well designed, the typical required value will be close to the centre of the trimmer (to give you plenty of adjustment either way) so we can probably reckon on a fixed output of around 75mV from the trimmer.

      Are you sure about the Freq CV source being only 0-5V? The 3320 datasheet tells us the frequency CV response is 60mv/decade, which is 18mv/octave. So 100mV is only 5.5 octaves. That doesn’t seem like enough to me! I suspect a 10V CV or +/-5V CV for a 0-200mV range at the chip, but you’d have to get a voltmeter inside the synth to be sure!

      Hope that helps!

  11. Just to clarify, on the switches, the markings “A” etc are just markings an not like a “Global Label” in KiCad all “A’s” are not connected. They are simply multi pole switches in the same group?

    Is that correct?



    1. No, I think all the A’s are connected. The letters represent control signals provided by the main microprocessor in the original synth. Certain switches use the same control and switch at the same time. In terms of implementation, I think they’re all done with 4016/4066 SPST switches, but it’s true that they wouldn’t have to be – you could use a DPST switch for the two ‘A’ switches.

  12. Regarding ELKA’s original schematics I also noticed that 220k resistor and diode placed in parallel between pins 7 and 4 are also switched around.
    What’s the potentiometer just after C switch for?

  13. hi,
    excuse my question if it´s nooby, (just a guy poking around, havind astonishing results)
    but: how can i gain acess to hp bp filter types on the 12 db oberheim obxa filter? this would be great to have in a modular setup

    1. You could try tapping pin 7 for a highpass response, and pin 6 for a bandpass response. Both pins have a heavy DC bias, so you need an electrolytic DC-blocking cap of several uF, positive end to the 3320’s pin.
      Let us know if it works. As you say, it’d be great to be able to get at the extra responses.

  14. Can someone tell me how to calculate the frequency range, respectively the cutoff frequency of the LPF. Especially the Pro-One. I only see differences in the caps used (150pf/300pf/220pf Digisound version). However, i have a hard time figuring out how to calculate it. It doesn’t seem to be as easy as a 1-Pole RC Network…Would really appreciate it, if someone could tell me the range or even point me in the right direction on how to calculate it.
    The reason why I want to know is, that we breadboarded the Pro-One LPF Version, but feel it’s not going dark/bassy enough when fully closed. Feels like cutoff is around 1khz instead of i.e. 400Hz or something…Thank you very much in advance.

    1. There’s an equation given for the pole frequency on page 3 of the CEM3320 datasheet. However, it’s worth noting that some of the values in the equation come with -50%/+100% tolerances, so while it’s possible to get a rough estimate, you can’t calculate it exactly for a particular chip. That’s why it’s common to see some kind of Freq Trim on a filter circuit.

      1. Hi Tom. Thank you for your reply. I found the CEM3320 Datasheet with the formular mentionend. But I have a hard time wrapping my head around it. Do you know by chance the frequency range of the above mentioned filter designs (Pro-One et.al.)?

        In the mod wiggler forum someone mentioned, that a higher CV on Pin 12 will lower the pole frequencies of the filter. (Also referred to in the CEM3320 datasheet.) https://modwiggler.com/forum/viewtopic.php?t=243331

        My input section on Pin 12 is: +15V going through a B100K Pot (Filter Pot) through a switched TRS Jack (Expression pedal takes over Filter Pot when plugged in) through a 150K in series with a 56K/1K voltage divider into Pin 12. The resistor values are taken from the Digisound 80.6 version and will give me ≈72mV which is comfortably in the VCFI range of -25mV to +155mV stated in the AS3320 datasheet. The voltage divider from the AS3320 datasheet (100K/1.8K) will give me 265mV (@+15VCC). Taking out the 150K and going only with the 56K/1K Voltage divider will roughly do the same: ≈263mV.
        These values will work with the max. input voltage at Pin 12 (±6V) but are beyond which is recommend (+155mV). Do you think it’s worth trying to lower the overall frequency cutoff/range? Does the Sequential Pro-One also uses +15V as Freq. CV > 100K/1.8K > Pin12? (I assume so…)

        1. Tom, nevermind. Just letting you know, the voltage divider did the trick. I remove the 150K from the equation and now the filter works as expected.

  15. Three things. First of all, the Sequential Pro-One LP filter output buffer: would this be equally beneficial to include with HP and BP filters also?

    Secondly, is it mostly a matter of convention that Bandpass designs usually cascade the Highpass section before the Lowpass section, or is there some good reason to have the HP before the LP?

    Thirdly, breadboarding these filters is a bit messy and I am wondering if anyone has made two-sided PCB layouts, or whether it is going to be up to me?


    1. 1) I don’t know, if I’m honest. Try it and see.
      2) It’s done that way for noise reasons. Noise is more perceptible at high frequencies, so if you put the highpass filter first, the lowpass filter removes more objectionable noise coming from the highpass.
      3) I don’t know of any. Does anyone else out there have something they can share please?

Leave a Reply

Your email address will not be published.