Rhythm Wolf Percussion

Overview

Original schematic

This has three main sections. The first is a standard pinged resonant filter (U13A,B) for the low and high percussion sounds. The second is a noise source gated by a transistor (Q37), and the third is a compressor built around a few opamps (U19,24) and transistors (Q2,17,8). The pinged filters work as they should and sound fine, but are limited by D50/51 such that accent does change the volume much. The gated noise source sort of works. It turns the noise on butonly moderately follows the envelope, and it doesnt really shut off. The JFET has a nominal on resistance of 60ohms, so with the 13k drain resistor, that should give -45dB attenuation for "off", which it might do, but that's not really enough. You can definitely still hear it at the ouput, even when the voice is not triggered. This is compounded by the fact that it goes through a compressor, which amplifies it when its "off". If you use the "howl" function, this noise is completely unreasonable. If you have the percussion voice turned up at all, the noise is constantly at the output, and its distorted, so it splatters all across the audible range.

The compressor is the most sonically interesting part of the circuit. It removes the usual exponential decay of the envelope, and instead keeps the amplitude constant for the majority of the range. This also has the effect of extending the apparent length of the notes, increasing the gain as the signal decays out. This makes a more "hollow" sound as compared to the typical "ping" of the resonant filter.

The compressor is driven by three signals: feedback from the output, a short attack spike, and the envelope. The envelope is a simple RC decay set by C149 and R279, and this is intended to serve as both an envelope, and a gate to turn the compressor off when no notes are played (U22A). Whether due to JFET variations, or just poor design, it doesn't do either on the unit i tested, as the level is too low to turn the compressor off. The attack spike (U22B) also has a failing, in that the blocking capacitor C216 has no discharge path, so it charges once on the first gate, and then never passes the attack signal again. The compressor's main input is from the fullwave rectifier (U19B) which is fed from the output. This has a gain of five, and a lowpass filter on the output. The heart of the compressor is a VCA composed of Q2,17,8, where Q8 is JFET used as a variable resistor to vary the gain of the VCA.

The inclusion of JFETs is a bit strange as they could have used a standard diffpair with a current source. The JFET in the noise gate also doesn't really do its job, and they would have been better off using the gating scheme from the hi-hats. If those introduced too much distortion, another diffpair could be used. It suprises me to see a six voice analog synthesizer with no trimmers inside. I have to imagine a bunch of things are just out of spec.

And now on to the electrical oddities:

1. There is a cap on the mixing pot wiper. Either end of the pot is driven by an opamp, so when its at the extremes it puts a hefty capacitive load on the opamps. This makes them less stable. A series resistor on the wiper would have fixed this.
2. The gate of Q39 is driven directly by an opamp output. This means that any excursion above 0.6V will destroy the JFET. Normal operation precludes this, but who knows what happens at startup. A series resistor would have fixed this, and reduced capacitive feedthrough on the attack phase.
3. D54 discharges C126, but there is no return current path, so it only happens once on the first trigger, and then never effects the circuit again. A current drain resistor is need on D54. This pathway is intended to gate the compressor open for the attack phase, but this is already done with the D49 pathway, and could have been replicated with a capacitor in parallel with D49.
4. U11B is completely redundant, and off in a corner of the board above the digital section. U5A is not used. i'm guessing they originally ran the signal all the way over to the corner and then back to the mixing section, but later decided this was not worth it, and added another opamp (U5), but then did not unroute the old opamp. So you have half an opamp going unused and a signal being double buffered.

Mods

Modified schematic

i made the decision for my box to completely cut the noise tone. It didn't sound good, was a lot of work to get sounding ok, and duplicated the same sounds i got from other voices. Instead, i used the mix pot to mix between the hi/lo tones. This gives a much wider range of sounds. This involved disabling the noise source so it wasn't splattering all over the circuit, and cutting the trace to the mix pot. On mine, i cut both ends of the pot, and wired directly from the outputs of U13. But, its easier to just cut one side and adjust the volume on U8B. C120 is removed to keep the opamps from having to drive a large capacitive load when the pot is at the extremes. R296 needed to be reduced as the levels were much hotter now. 100ohms was a good value to put it just at the edge of distortion across the whole range. You can go a bit smaller if you want it perfectly clean, but the SNR gets worse.

The next thing i did was mod the hi/lo tones so they are louder with more accent. i think it was originally done as a fixed volume because the mute circuit also did envelope control. But, once i got the mute function going, i decided i liked the variable volume effect anyways. This is done by removing the diodes at the inputs of U13, and replacing the capacitors with resistors. i picked the values to give a bit of distortion at the highest level. the resistors can be increased by 20% or so to eliminate this, or reduced by 20% to increase this distortion. Due to these capacitors being replaced with resistors, the decay time was made shorter. Its ok, but the notes lost a bit of that hollow sound, so i removed R279 and added another 1uF cap to bring the decay time back to where it was. Using an SMT capacitor, i was able to just replace R279 with this cap.

Next up was the accent circuit, which just needed a 10k drain resistor for C216 to discharge through. This gives a click at the beginning of the notes, which is louder with more accent. i'm not sure i fully like it, but it can be easily switched out by disconnecting the drain resistor (or any other method of blocking the signal).

Finally, the mute circuit, which probably worked on their bench but not in production due to the wide spread of JFET parameters. i measured the JFET active range for my VCA compressor, and it was -4V for full off, and -3V for 1/2 gain. Full gain was up at -0.5V. i modded R253 so the output was around -8.5V with no signal. This, mixed with the 0V output of the fullwave rectifier, gives -4.2V, which ensures that the VCA is off when there is no signal. This greatly reduces the background hiss of this voice. Next, i decreased R6 to 36k to increase the envelope. This allowed the compressor to open up, giving longer notes that were louder. This is very similar to the original sound, as the envelope was not working at all before modding R253. If you put a 1.2k from the inverting pin of U22A to ground, the gain increases and gets rid of the envelope effect entirely, just gating the VCA.

Other things i tried, but didnt want to add switches/pots for at this time:

1. Put a pot or switch on the signal going from the output to the compressor. A pot here is great. For testing, i just wired from the volume pot wiper, and was tempted to leave it like this. This allows for a really wide range of sounds from the voice. Without the compressor, they are very much just resonant ringdowns, very sharp and short. With the compressor, they get a nice hollow sound to them.
2. Put a diode across R304. This makes the compressor turn on instantly, which sounds pretty good when there is no accent added. It also increases the compression effect, but distorts the beginning of notes some.
3. i tried higher compression gain settings (increased R33), and they were fun, but ultimately i decided it was too much for a fixed setting, and they distorted a lot. A pot here could allow for some fun variations.
4. Fixing the noise source. This is possible, but complicated. Basically, the JFET needs to be wired differently. Currently, the envelope is sent the gate, and noise is at the drain. But, due to the nonlinear nature of the gate input, there is only a small window where the JFET is active, and the envelope just gates the signal as it doesn't spend much time in this regoin. The noise source should go to the gate, and the the envelope should go to the drain. Since JFETs vary a lot, you will need to test values to get the response you like.
5. First, start by removing R8 and R199 and connect a 100k resistor from the envelope at C149 to the drain. With a 100k drain resistor to C149, the envelope will shorten by ten percent or more, so the capacitance might need to increase by 220nF or more. If the attack time is too long a bit at this point, short out R295. Next, cut the trace between the the drain and U21A and put in a 0.1uF capacitor and 10k resistor to ground. This will act as a highpass filter and lessen the DC thump. Using a larger capacitor or resistor here will pass more low frequencies, but will also pass more DC thump. Finally, take a 10k resistor and tie one end to the noise source and the other to the gate. Then connect the gate with a 100k pot to -12V. Adjust the pot until the noise source is sounding right. Check the value of the pot and solder that value in place of the pot.
6. Second, check the output volume. It will most likely be distorting or too loud. If too loud, remove R187. If it is still too loud or distorting, decrease the value of the 10k resistor in the highpass filter on the drain. You will most likely need to increase the capacitor value accordingly. To further shape the noise, a capacitor can be placed between the gate and ground to cut out high frequencies.