Chewborg Envelope Filter

Introduction

On my quest through re-imagining effects pedals, I felt like envelope-controlled filters deserved some attention. As it is usual, my motivation comes from the unsatisfaction with existing circuits: from the widespread use of OTAs, with their drawbacks in terms of distortion and noise, and limited availability in some cases; to one-trick pony circuits; to circuits that require matching; circuits whose complexity isn't reflected in the features offered. Ripple is another common issue, at least in my experience with the Mu-Tron Micro V, and I find it very noticeable on a resonant filter. The Mu-Tron III and its improvement, the Q-Tron, get a mention for almost doing things right: LDRs weren't my first choice, and there are still some things worth changing (the Q-Tron fixed the awful input stage of the Mu-Tron). Still, I liked the choice of a State-Variable Filter, which is the filter topology I've ended up using. There are many other possibilities even just between two-pole filters, like Sallen-Key types or Multiple-Feedback, but the ability of the SVF of providing three filter modes while being easily tunable independently from Q and gain are hard to beat.

Another dilemma was in the control element to use; almost everything has been tried to voltage-control analog filters: LDR, JFET, BJT, diodes, PWM, switched capacitors, compander chips... and MOSFETs! While the limitations in terms of distortion of this last option are well known, my previous experience with the PUP was encouraging, and the advantages are just as obvious: when packaged in an IC, CMOS FETs are a cheap and easy source of matched devices. After this bit of background, it's time to introduce the circuit and continue from there.

The circuit

Here's my realized example of an envelope-controlled two-pole SVF with CD4007 FETs as variable resistors. Since only two VCRs are necessary, this classic chip is the perfect choice, and comes with one to spare. As I mentioned, I have some first-hand experience using CMOS chips for VCR applications with the PUP, but more specifically there is one documented use in VCFs already. This last schematic served as inspiration for my own inverting SVF filter, but even without considering the sidechain, there was still a lot to be done to adapt it to single supply, to the measly 9V usually available for stompboxes, and in the lack of a parallel resistor to the FETs, which makes it more of a proof of concept than a working circuit.

The signal path

The first thing in the signal path is, alas, a buffer, which is made necessary for the filter input and for the rectifier, which require both a low impedance source. The SVF topology that follows is a pretty common one, with two N-MOSFETs in parallel to resistors tuning the frequency from 159 Hz to about 10 kHz. The lower limit can be made lower by increasing the parallel resistor, up to 47k, but has been chosen to be limited to a more useful range. The upper limit depends on the minimum Rds achieved, but is high enough for most purposes. Of course both ends of the range depend also on the 100n integrator capacitors, but all values have been chosen for using a small parallel resistor, which improves the control law by limiting it to a shallower segment of the pseudo-hyperbolic curve, while exploiting the full available gate drive to still reach high frequencies. Speaking of the parallel resistor, its presence is necessary for the filter to work at all, or it will latch up for lack of DC feedback, and too large of a value will cause excessive distortion.

Knowing the headroom limits of MOSFETs allow for working around them: the passband gain of the filter, excluding resonance, is -20dB; this amount is then recovered at the output of the filter. Attenuating by 10 might seem a lot, and you pay for it with a proportional increase in the noise floor, but it's what I found necessary for handling line level and the line-level peaks of instrument sources without an undesirable amount of distortion. Don't despair yet, though: this is the low-frequency gain; high frequencies (+3dB at 1 kHz) aren't attenuated because of the pre-emphasis network, and are then brought back to the same level at the output by the matching de-emphasis. This approach allows to have reduction of distortion at low frequencies, where it matters, without surrendering 20 dB of high frequency hiss by output amplification. In practice, the filter is very quiet: a fixed high-frequency full-resonance peak might be noisy, but you'll find that most of that noise comes from the source itself.

The "Q" potentiometer allows independent control of resonance. The series resistor limits the action to the more interesting upper half, and the 3.3k allows for plenty of resonance (you can decrease it if you ever wanted more). A linear potentiometer works well, but a C taper might allow better control over the higher settings.

The sidechain

This is where things get really interesting. First, there's a full-wave precision rectifier followed by an AR envelope generator. I've explored these circuits in depth while developing that compressor, but here I've chosen a different variation of rectifier, still all referenced to half supply, to more easily adjust gain. This variation also needs less care put into the diode currents to obtain symmetry, at the small cost of requiring an odd 20 kΩ resistor. I'm lucky enough to have found tens of those in assortments and grab bags, almost making it an "honorary E6", but if you don't have any don't worry: within E6, 68k and 33k feature a remarkable ratio of 2.06, so you can use those in place of R13 and R14, 1M for the feedback resistor and 10p for the capacitor. The small asymmetry is unlikely to be relevant, and performance will still be way better than with an half-wave rectifier.

The rectifier has a gain of 16.5 for two reasons: first, the amount has been chosen to be able to get a full-range sweep even with small inputs; second, if the rectifier output clips for large input peaks, this results in a compression of the dynamic range of the filter sweep which helps to make it more consistent. As a bonus, any gain added here makes the voltage drop of the envelope generator diodes relatively smaller. An alternative value of 220k for R15 is suggested in case you find the 330k to be too much.

The envelope generator is a standard, low ripple one, that only discharges the capacitor when the input goes low. The values of attack and release have been tweaked for an envelope filter, and I've shown suggestions for replacing pot and series resistor with fixed values, even though I recommend having control over those. You can use either linear or log pots here. The suggested connection should make wiring more efficient. A buffer isolates the envelope capacitor from the "Range" attenuverter! You might have seen similar circuits having switches for higher or lower cutoff range, input sensitivity and upwards and downwards filter sweep: this control efficiently accomplishes all three functions, deciding the envelope polarity, the width of the sweep and, together with "Tune", the range of the sweep. How it works is that "Range" in the middle means no envelope modulation, so you can tune the filter just with the "Tune" know or external CV; turning "Range" to the right means increasing amounts of positive modulation (the filter sweeps upwards); turning the knob counterclockwise increases negative modulation (downwards sweep). This is already accomplishing two things at once, but since the circuit is capable of sweeping almost the full audio range, together with the "Tune" settings you can control how much, in which direction, and across what frequency range the filter sweeps (for a given input) with just two knobs.

As you might have guessed by now, the next stage sums the envelope output with a variable offset given by the "Tune" pot: because of offsets at the lower end, variability of Vgs for the higher end, and the standard values used, this control has a bit of a dead zone at each end. I've tried to make as small as possible, while still being conservative to allow full adjustment with variations in supply voltages and Vgs of different chips. This knob alone can span the full range of the filter (as defined by capacitors, parallel resistors, and Rds at the op-amp saturation voltage), and so can the envelope with sufficient input. This is the reason the two are summed actively instead of passively. The choice of using the non-inverting input for the offset is just to use one fewer resistor and avoid unintentional tapering effects.

The sidechain output goes into the FETs with the usual and indispensable drain-gate linearization network. This time this has been AC-coupled, which allows the use of the full control voltage to drive the gates, at the cost of some speed, which isn't as necessary as in a compressor (the envelope attack can still be very fast).

With five op-amps, this sidechain might seem inefficient, but it's a result of necessity without sacrificing performance. The necessity is mostly in the order of the operations: rectify, envelope, attenuvert, offset. It was possible to offset before the attenuverter, at the rectifier output, but this was undesirable because this offset would have been not only inverted (weird), but also attenuated together with the envelope in some interaction terrible for usability.

The rest

Starting from the supply section, the reference voltage is lower than half supply: as in the PUP, this has been done to increase the available gate voltage respective to the source (which is at this voltage), and doesn't result in premature distortion because the FETs are the limiting factor.

You can see pin 7 of the 4007 being tied to Vref, despite this FET not being used. This is because all NMOS substrates are tied to this pin, and because of something called "body effect" attempting to lower the substrate voltage with respect to the source results in inflated Vgs to reach the same resistance, something we can't afford on a 9V supply (even if this could result in increased headroom in other circumstances). All 4007 that aren't shown are left floating.

Finally, the output recovery stage boosts the signal back up with a variable amount of gain, from 3 to 33, allowing plenty of output even from the quietest filter settings. This arrangement of the gain pot allows smooth control even with linear taper.

Unused tricks

For the return of the "unused tricks" section, there are two things: the first is a servo implemented with the last NMOS; since I wanted manual control over the quiescent point, its purpose wasn't the same as in the PUP, even if the circuit is derived from that one. The purpose here is to linearize the FET in hope of getting a smoother control law than the original. It looked like this after tweaking the input voltages (mind that this is with DC gate feedback):

Sadly, this made things worse, with the filter dwelling more on lower frequencies until it shot up with a slightly larger signal, but it's still worth mentioning because it could be the way to go to implement V/oct with an expo.

Second trick was to put diodes across Q 1 and 2, including the series resistor. This should have the effect of taming the amplitude of the resonance while keeping the same effect, but in practice the voltages in that path are so small that you need schottky diodes to even start to hear a difference. I decided the effect was too small to be worth keeping them.

Demo

How does it sound? This time I've made 4 demos to show only a few of the possibilities of this circuit. The rest is up to you.

Layout

I've ended up home-etching Lani's layout on FSB. I have to admit ordering it is preferable, but I've managed the feat. Sadly, there seem to be some issues with leakage from the sidechain into the signal path that I didn't experience on breadboard, and they persisted even when I ordered a manifactured PCB. You're welcome to come up with a different layout and tell me how it goes, now that you've been warned of the danger. I think perfboard is possible, maybe still within a 125B; try it if you want.

Build

Here are a few pictures of my own build:

Conclusions

I'm happy with the result: it sounds like an analog filter with plenty of character, clean enough to feel that way, but with some inevitable and flavorful distortion at the loud resonance peak. More in general, I think I have brought something fresh to envelope filter pedals, with an original sidechain and a very versatile filter with three modes and a wide frequency range. While this circuit does a lot, there's still plenty of room for different filters, both envelope or CV using this approach with CMOS VCRs. This was my first time using the CD4007 for it, and I'm happy to say it's perfectly suitable. Aotmr deserves the usual thanks for the discussion and feedback and the crucial ideas in a moment of difficulty. In particular, their contribution resulted in the use of the emphasis networks and AC MOSFET feedback, but their suggestions were crucial for the development of the whole circuit.