YuSynth Fixed Filter Bank

When planning Kosmodrome, I decided to include a fixed filter bank (FFB). Back in the 1960s and 1970s, pretty much all Moog modulars had an FFB, either the 907 module or its successor the 914. These days, not too surprisingly, Behringer makes one, in Eurorack format, based on the Moog 914. Doepfer has a Eurorack FFB, and there are a number of others.

Still, they’re not that commonly seen, and probably for a few reasons. Size, for instance. 28 and 20 HP wide, respectively, for the Behringer and the Doepfer — each has about 15 knobs, and those take up space. Eurorack-heads especially will resist giving up that much space in their case without a good reason, and I suspect a lot of people don’t see a good reason because they don’t get what an FFB is for.

As I see it, an FFB is a way of giving something like formants to a synth. Acoustic instruments and human voices have (more or less) fixed frequency resonances in their frequency spectrum. Known as formants, these resonances have a lot to do with the individual tone quality of the instrument or voice. I think it’s fair to say that when you hear a voice behind you and think “Oh, that’s my friend Fred, I know that voice,” a significant part of what you’re recognizing is the formant structure of Fred’s voice.

Synths, on the other hand, don’t have such resonances and hence such rich tonal character — unless you put them there. An FFB puts fixed frequency resonances into the synth’s frequency spectrum.

The one I built is a somewhat modified version of Yves Usson’s module from his YuSynth website. Usson’s FFB’s functionality and front panel design are based heavily on the Moog 914. Both have a low pass filter and a high pass filter, and in between, twelve band pass filters. The input goes in parallel through all fourteen filters and the resulting outputs are mixed to form the module output, with a knob for each filter to attenuate its contribution to the mix. (This sounds rather like a graphic equalizer but, as Usson writes, “it has marked differences. Mainly, the filter [resonance] is higher and the [bandpass] slopes are steeper than in a graphic [equalizer]. Furthermore when a control is set to zero, the corresponding spectral band is completely muted.”)

Internally, though, the Moog and YuSynth FFBs are quite different. The Moog 914 used passive filters made of resistors, capacitors, and inductors — the latter being generally large, hard to find, and expensive parts. Usson instead used active filters, based around op amps, which are much smaller, more readily available, and cheaper.

My version’s changes are:

  • Tuning. See below.
  • Addition of an auxiliary PCB for board mounted slide pots and jacks (instead of Usson’s panel mounted, wired rotary pots and jacks).
  • Relocation of headers on the main PCB to accommodate mating with sockets on the panel components PCB. Usson’s 14-conductor ribbon cable is replaced by a mated header/socket pair.
  • Use of dual bypass caps (one from each power rail to ground for each op amp chip instead of one between the two rails).
  • Omission of MOTM and MTA power headers.

At 20 cm in both panel dimensions, this is the second largest synth module I’ve built, after the 20 x 40 cm sequencer, and no doubt the highest part count. Gratifyingly, 20 cm is just wide enough for 14 slide pots with only small gaps in between. Moog and Usson (and Behringer and Doepfer and… well, everyone else, looks like) used rotary pots but I felt like going with sliders. Either way, I didn’t feel like wiring up 52 pot and jack terminals, so I did a panel components board.

Tuning

Usson’s filters are tuned to the same frequencies as in the Moog 914 module, except that for some reason he used 750 Hz instead of 700 Hz. (I suspect this was simply a mistake on Usson’s part.) The Moog filter frequencies are about six semitones (half an octave) apart, with each filter exactly an octave above the second previous one. But note the following observation from the catalog description of the somewhat similar vintage module, the Serge Resonant Equalizer:

Except for the top and bottom frequency bands, all other bands are spaced at an interval of a major seventh. This non-standard spacing avoids the very common effect of an accentuated resonance in one key, as will be the effect from graphic equalizers with octave or third-octave spacing between bands. Spacing by octaves will reinforce a regular overtone structure for one musical key, thereby producing regularly spaced formants accenting a particular tonality. The Resonant Equalizer’s band spacing are much more interesting, producing formant peaks and valleys that are similar to those in acoustic instument sounds.

They don’t mention the Moog FFB specifically but obviously they’re throwing shade at it. All the odd-numbered bandpass filters in the Moog 914 are centered near the note B in different octaves, and all the even-numbered ones are centered near F. Do you necessarily want B and F emphasized in every octave in every song you create? Probably not, and I can imagine a luthier, even a traditional one with no quantitative knowledge of formants, making a guitar all of whose formants are on B and F; they’d probably play it and say “oh, no, no, no, that’s all wrong” and pull it apart and start scraping until they liked the tone — because they’d moved the formants to non-octave separated frequencies.

Following something like this idea, I have retuned Usson’s module. Specifically, the first bandpass is kept at 125 Hz and the other filters are tuned at six and a half semitones apart. Thankfully, Usson’s web page supplies the formulas one needs to calculate the needed component value changes.

An unintentional consequence of this retuning was to reduce the rather large parts count. In Usson’s design the band pass filters, for instance, all have six resistors, with the same values for every filter; the frequency is tuned by using different values of capacitance. But when Usson computed the capacitance values needed, none of them corresponded to the standard E12 values most readily available capacitors come in. His solution was to use two capacitors for each capacitance, in parallel, whose values summed to approximately the desired non-standard value: For instance, a 68 nF and a 4.7 nF to give a capacitance of 72.7 nF. There are two capacitors per capacitance, two capacitances per filter stage, two stages per filter, and twelve band pass filters, so that makes 96 capacitors right there. Then more for the low pass, high pass, and input and output sections, not to mention the bypass caps for the eight TL074 quad op amps.

SUDDEN GLOBAL CAPACITOR SHORTAGE BAFFLES SUPPLY CHAIN EXPERTS

Purely by chance, the tuning I chose to use turned out to require, for five of the twelve band pass filters, capacitances that were E12 standard values! That meant, for those five filters, I could use a single capacitor instead of two for each capacitance, for a net savings of 20 capacitors. (Which dropped to 12 when I replaced Usson’s single bypass caps with dual ones.) I chose to leave the unneeded capacitors (marked DNF, “do not fit”) in the schematic and their footprints on the PCB, in case I or anyone else ever wants to use this board for a differently-tuned version. But for this build, those footprints were left unpopulated.

Deliberately empty footprints

As is evident from the PCB photo, this is a very modular circuit. All the bandpass filters are exactly alike except for their capacitor values, and there are two for each TL074 quad op amp. The low and high pass filters share another, and the eighth is used by the input and output sections. Each of these has its own little patch on the PCB. So I decided to build incrementally: I did the input and output sections and one bandpass first, put one slider on the panel components board, put it together, and tested. Then I disassembled, added the low and high pass filters and two more sliders, put it together again, and tested again. Finally I took it apart one more time and did the remaining eleven bandpass filters.

Fixed filter (singular) bank

For that first test, things went bad: It drew lots of current, and the TL074s got very hot. I’d done a multimeter check to rule out shorts and to verify power was getting to each chip — but neglected to check where it was getting to on each chip. I looked at my schematic and, sure enough, I’d drawn one op amp with the power connected the wrong way around. And then I’d copied and pasted that seven more times. Can you believe up until now I haven’t had a check for that in my PCB pre-fabrication checklist? I do now. Fortunately quad op amps are, except for the power pins, rotationally symmetric, so all I had to do was replace the fried chips with new ones installed “backwards” (something I’ve had lots of past experience with doing) and everything worked.

Schematics, KiCad design files, Gerbers, a Python script to compute component values for different filter tunings, and documentation may be found here. They’ve been corrected to have the power connections right!

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s