Pole mixing on the breadboard

Electric Druid shows a pair of partial designs for a pole mixing filter, one based on the V2164/AS2164 filter chip and one using the AS3320. I’ve been breadboarding a circuit based on the latter, and my results have… not been encouraging.

There are a couple of errors in the Electric Druid schematics. There should be a 51k resistor between the op amp and AS3320 pin 8, and there’s a copy and paste error in the mixer section where the 4P N resistances should be 30k, 15k, 15k, 15k, 15k. (These errors are acknowledged by Tom Wiltshire of ED in the comments and they write “I’ll get an updated schematic uploaded soon” but it hasn’t happened.)

Other issues have to do with the filter stages’ DC gain. As discussed in the preceding post, for reasonably accurate high pass and band pass outputs, you want the DC gain to be unity to within about 1% at worst — closer to 0.1% if possible.

Per the CEM3320 datasheet (which gives more detail than the AS3320 datasheet), the voltage gain in each stage is

where and are the coupling and feedback resistances, respectively. With the values shown in the datasheet, 91k and 100k, the gain is 0.999. In the Electric Druid design, though, 82k — which gives gain equal to 0.832. So that seems very wrong, as pointed out by commenters on the ED post.

I’ve breadboarded the LPF core of the datasheet circuit, and verified that 82k resistors give too low gain. I also saw gain differences between the stages of around 2%. So I think it’s best to replace the fixed 91k coupling resistors (and the input resistor) with fixed 82k plus 20k multi-turn trimmers.

But there are more gain related problems I don’t know how to solve (yet).

I’m measuring DC gain by putting in -5, -2, 0, 1, 2, and 5 V and measuring the voltage outputs of the four stages. (Before the AC coupling capacitor, of course.) I’ve done that for several values of the frequency control voltage, which at the AS3340 pin is supposed to be in the range -25 to +155 mV “for best results”. Then I take the ratio of output differences for pairs of input voltages to input differences. For instance if I measure LP1 at 7.23 V and 3.177 V for inputs of -2 V and +2 V the gain is (3.177-7.23)/(2-(-2)) = -4.053/4 = -1.013. (Each stage inverts so the gain is negative.) For the later stages the input is just the output of the previous stage.

One problem I find is that the gains depend on the frequency control voltage. As the CV runs from -40 to +100 mV (that bottom value is below the best range, but the results wouldn’t change much if I omitted that value) each stage’s gain pretty consistently decreases in magnitude by something like half a percent. I also recorded results with a CV of 150 mV and that’s really bad: from 100 to 150 mV the gains drop around 2%.

Of course the output amplitude for the fourth stage also depends on the gains of the first three stages, so it changes over the -40 to 100 mV range by something like 2%. Going up to 150 mV it changes by about 8%. That’s really bad if you want low DC leakage for your high pass and band pass filters.

The datasheet talks about control voltage feedthrough, which refers to a change in the DC offset of the output when the frequency control voltage is changed, and can be countered by adjusting the resistance on the VEE pin. Nominally it should be

which for works out to 1163Ω. (Electric Druid rounded that down to 1.1k, which is a difficult value to find, though you could use 1k and 100Ω in series, but 1.2k is closer.) However, the datasheet says making this resistance adjustable with a trimmer allows minimization of control voltage feedthrough. I’d hoped the CV dependence of the gains also could be nulled by adjusting this resistance, but no, it’s completely insensitive. I haven’t yet found any way to change, let alone eliminate, the gain variation.

There’s another problem which is very worrisome. I can calculate gains using different pairs of input voltages — I’ve used 0 and 1 V, 0 and +2 V, ±2 V, 0 and +5 V, and ±5 V. And the gains I get disagree! For any stage and any CV, these five gains jump around by 1–2%. The results are repeatable and the overall general behavior is similar for the five gains, but they disagree with each other. I note that input of ±5 V results in outputs at each stage of about 0 and 10 V, and the datasheet seems to say the maximum good output is about 9 V, so the last two possibilities are presumably out of spec — but they don’t look much worse than the other three in any respect. Maybe the inconsistencies are smaller if you only look at the first three, but they’re definitely still there.

Gain vs. CV. On the left, using ±2 V; on the right, using 0 and 2 V.

That seems to imply the outputs just are not linear in the inputs. Unless there’s just something wrong with the way I’m getting the DC gains, but I don’t see how.

I’m out of ideas at the moment, and not too sure the project is worth pursuing if I can’t get the gains under better control. Which is very discouraging, given how much time I’ve put into it.

There are a couple of other things I didn’t like about the ED circuit:

In the datasheet circuit the LP4 output goes through a 51k resistor to the resonance input pin. In the ED circuit they send the LP4 output and the input signals into a mixer op amp to subtract one from the other (with some coefficients), and send the output of that via the resistor to the resonance input pin. This is to offset the problem where the amplitude below cutoff drops as the resonance is turned up. Seems to work. However, I was worried about what would be the effect of a DC offset in the input signal. Per the datasheet, you can have control voltage feedthrough from the resonance CV, too, and the recommended fix for that is to add a small DC voltage bias to the resonance input. But doesn’t that mean an input with a DC offset could mess up the resonance CV feedthrough?

Well, I’m not sure it matters. Again, the feedthrough only (I think) affects the DC offset of the outputs, and we AC couple those, so who cares? I did do some tests early on in which adding a capacitor to AC couple the input signal going to the resonance compensation mixer seemed to improve the filter’s behavior, but I’m not sure I believe it. I may have been making some mistakes then. Still, I don’t think the capacitor does any harm.

Another thing I didn’t like was that the filter was going into self oscillation at rather a rather low resonance CV value — only about a quarter of the way up from minimum on the knob. If I bypassed the compensation mixer and just sent the 4-pole output through the 51k resistor to the resonance input, self oscillation didn’t start until I had the knob around 2/3 of the way up. But apparently the compensation mixer was resulting in a larger amplitude signal going to the resonance input. I found changing that resistor to about 120k resulted in better behavior.

Well, hooray, making improvements. But the core problems remain, and I don’t have a clue how or if they can be solved.