So why haven't we got a Moog transistor ladder envelope filter project then?

[StephenGiles]

It occurred to me this morning when throwing a ball for Bonzo, our 3 legged dog, that if the Microsynth start/stop frequency CV buffer output could be clamped to the peak voltage equivalent to the strength of notes played, it would do away with the need for a sample & hold .

[Earthscum]

anyone willing to give up at least the basics of how this works?

Is this using the trannies/diodes as variable resistors? I'm basically trying to figure out how you don't get extreme volume cut at the lower end of the sweep.

[mrslunk]

if it hasn't already been posted,
see this: http://www.timstinchcombe.co.uk/synth/Moog_ladder_tf.pdf

[Earthscum]

Dude, that's sweet. Thanks! I love articles like this.

[earthtonesaudio]

For 9V operation it might be worth substituting MOSFETs in for the BJTs, although that would definitely have some effect on the sound it makes.

[PRR]

> MOSFETs in for the BJTs

1) MOSFETs usually need 2V-5V "grid cathode" bias, and a similar amount "plate cathode". The BJT biases 0.6V at the input and below 0.1V across the output. In a stack like that, at 9V, several layers of the BJT is tight, the MOSFET is one-stage at best.

2) The emitter impedance of a BJT is VERY predictable and linear over MANY octaves of current (nA to mA). You don't even have to know what part-number. 28 ohms at 1mA, 28,000 ohms at 1 uA, etc.

The same impedance of a MOSFET is not known even roughly until you pick a specific part number, not known closely until you measure a part in hand, and the impedance varies along three different laws (linear at ultra low current, square-law at low current, asymptotic to a parasitic resistance at high current). Even if you found two parts the same at 1mA, they are awful sure to differ at other currents.

"Much of the usefulness of the filter is the ability to apply voltage-control to the cut-off frequency..." - Timothy E. Stinchcombe

While you could vary the frequency with an FET, there's no real predictable relation between voltage (or current) and impedance (frequency). With the BJT the R/I (thus V/F) error over 1000:1 range may be near 1%.... your filters TRACK your oscillators (better than a Moog's oscillators track each other...). Stinchcombe's plots (for a SPICE-faked sim) show 2% error over the middle of the audio band. Some of this is my squint and his plot interval; I can't read the curve to 1% accuracy.

3) Bob Moog didn't have good MOSFETs. Even JFETs were rare.
> I love articles like this.

Good where it goes. I don't see where he says HOW the silly thing filters. The use of Ebbers-Moll seems clunky: I see it by using Shockley's Relation (remembering that SR un-simplifies when signals are large). The assumptions of balance and base-current are not essential. And the filter action does NOT happen in the "long tail pair". I see he work "my" way in section 2.2.... maybe I am less attached to reality than Timothy is.

> how you don't get extreme volume cut at the lower end of the sweep.

More correctly: why the gain/volume isn't varying over the ENTIRE sweep. You know that, usually, reducing device current while all else stays the same usually reduces gain.

In Stinchcombe's Fig 3: the input is across Q1 Q2 and the output is at the _emitters_ of Q11 Q12. The emitters are not amplifier stages, they are "passive" resistors.

The gain of a simple common-emitter amp is the load resistor over the emitter resistance. With resistor load, lower current is higher emitter resistance and lower gain.

With the "emitter resistance load", the "load resistor" VARIES with current. And because Q1 Q2 current is equal to Q11 Q12 current(*), the gain is always unity (until the caps shunt the signal).

And some -consistent- gain is vital when we wrap the feedback "boost" loop around, We will want gains like 0.980 which are STEADY across the frequency band, so the dB bump stays the same size (at least does not collapse or go into howl).

(*) Actually the base currents add-up going down. If hFE is 100 and Q11 Ie is at 1.000mA, then Q9 Ie is 1.010mA, Q7 is 1.02mA, Q5 1.03, Q3 1.04, Q1 is 1.05mA. So Q1 runs 5% richer than Q11 and the gain is 5% higher than unity. Which is moot because internal signal levels are MUCH lower than we want at the jacks, so the ladder is always followed by amplification which can trim-out any 5% fudge.

[Earthscum]

Thanks a bunch, Paul. I was skimming that article and decided to pop back on the boards for a moment. You pretty much answered most of my inquiry.

One question (if I don't figure it out before), does having more 'steps' in the ladder have a similar effect to adding stages to a phase shift circuit (in the most basic sense of 'sound', I guess)? Could a good experimenting point to be to start out with 2 steps, or does a single step provide enough effect to get a good starting basis, is I guess why I'm asking.

This circuit actually reminds me of discrete UV meters.

[PRR]

> does having more 'steps' in the ladder have a similar effect to adding stages to a phase shift circuit

One step is equivalent to a simple tone control: 6dB/oct slope, highs are less, not cut-off sharp. However the corner frequency is varied, which is less often done in tone controls (usually the corner is fixed, highs fall then shelf-off a variable depth).

So you need Q1 Q2, a C, and Q11 Q12.

This will not "bump up" with feedback. You need at least two stages for "some bump", at least three stages to bring "infinite bump" into reach (you don't want infinity but not much happens until you get close).

Note that input impedance is low (as low as 1K) and input level must be kept down to 20mV-50mV or it is more a "fuzz". You got some in-out interfacing to do.

> This circuit actually reminds me of discrete UV meters.

I do not understand. What UV meter? Light sensors can use many types of amplifiers. Light covers a wide range of intensity and it is often convenient to express and measure in log form. Log-converters are often based on diff-pairs, but wired differently.

[Earthscum]

I do not understand. What UV meter? Light sensors can use many types of amplifiers. Light covers a wide range of intensity and it is often convenient to express and measure in log form. Log-converters are often based on diff-pairs, but wired differently.

Visual unit meter, the "decibel" light meters on stereos. They work, however, by dividing voltage down a set of buffers driving LED's. I'm sure you know what I'm talking about now 

Now that I understand the filter response, this doesn't remind me as much of how the meters work.

[Earthscum]

Multi Stage Waveshaper

Just came across this tonight... this is pretty much what I was talking about.   (BTW, is this by the same Christian that came up with the Bazz Fuss?)

Edit to add: Oh yeah... getting ready to upload a little finding of mine that may help this push along...

[Earthscum]

Ok, so I was playing around with this (inverter pair?) and wondered if the sound would change if I put a diode in between the bases. Yep... started to sound like a distorted hollow tube... kinda kewl sounding. Then I added a bunch more and got a REALLY kewl sound... WAIT! It's doing some kind of filtering... and this thread popped into my head, so I dropped caps from the junctions to ground, and VIOLA! It filters. I tried different combinations of caps, and all have their own characteristic sounds.

Notes:

It has a natural envelope to it, slow, looooong... but, that goes away if you buffer it, or have actives.

So, I decided I should be able to adjust the "Q" if I do like the second circuit. I used a 5k pot, but only about 2k of it was really active.

This seems like it could be a good base for stompbox synth filtering with 9V supply. I'm spent for the night, but figured I'd get this up for people to play with. I'm guessing that a buffered signal may improve the response (but not the sweep).

This needs some kind of sweep generation. ~2k range seems pretty sensitive, yet very controllable within the sweeps I was getting... should be easy to implement some kind of voltage control to one of these.

The "quack" I was getting with the 5k was really muffled, with a gradual build up to about 2k when the filter became easily apparent without sweep. After 2k, the thing literally quacked with pot rotation... like a wah... actually, more like a synth-wah.

Now I'm kinda wanting to do a Fet controlled by a LFO to ground with the feedback resistor as a "Q Range" control. Then can use envelope, the LFO... all sorts of things!

Anyways, here ya go.



ps: This thing pretty much makes the wave square, so I don't know that it's going to react much better if you insert a square wave like I've noticed some filters say. Plug your guitar straight in and you'll see what I mean. This is just a base circuit... if ya think I should, I could just start a development thread on this.

[StephenGiles]

 A similar filter was used in the EH Crashpad, designed in 1980!

[frequencycentral]

Well that looks very cool and inspiring, such a low parts count too. There's clipping going on as well as some filtering? I'm gonna have to play with this when I find time. Nice work. I'd love to hear some soundclips if you can rustle some up.

[Earthscum]

A similar filter was used in the EH Crashpad, designed in 1980!

That's awesome! I wanna see the schem now, see how they implemented it... nothing that I can find online, though... bummer. New one to keep an eye out for.

Ok, I noticed basically, pot to the 3906 emitter, like I drew, gives a dampened Q, but if you drop it to ground, it works like you would expect it to. So, after breakfast I'm gonna try a couple things.

[PRR]

> Plug your guitar straight in

Input impedance is 10K clean (up to ~~20mV) down to under 100 ohms when clipping, with ~~0.1uFd in shunt and 0.2uFd in series.

This does "horrible" things to a guitar pickup signal. It's a messy filter, drooping bad above 200Hz. OTOH most guitar signal can slap those two collectors like my barn door in a windstorm, adding lots of distortion overtones and filling-in the treble that is lost at the input.

> goes away if you buffer it

For "consistency", you may want to buffer and then build-out a pickup-like input circuit. For a breadboard, you might just slap an old/tired pickup inside a steel box (for shielding) and use it series between buffer and "IN". Full "pickup emulation" needs another 200pFd-1,000pFd shunt, but your 10,000++pFd caps make that moot. (There's other ways to "fake" a pickup's inductance as a buffer detail {emitter followers tend to have a trace of inductance} but hard thinking is needed.)

[Earthscum]

Yar... I figured with the people watching this thread I didn't need to say much about 'why' it is getting a natural envelope. It's the same effect as what drives the Bronx Cheer to do what it does, essentially.

Whilst playing with this, I decided to put a parallel set of diodes to separate the caps and put a 3906 at the top of it. The emitter gets a diode from the output, and the base is driven from the output through a .1u cap. Been playing with parts around this, but gives ultra-high Q and much better envelope. Oh yeah, also driving it with a square wave makes a bunch of difference (big surprise there). Now if I could find that one control point, lol... I'm thinking the 3904 to ground was working better as far as having some kind of control, but wasn't producing as much of the sharp peaking.

[PRR]

The obvious control is the "+9V" or the current through the "10K" resistor.

With your posted plan, 3V turns it off and 15V doubles current (halves the impedances). You actually only need a couple Volts across a resistor there; how about 3.3K and 5V nominal varying from 9V or 10V or 12V down to 3V? Or use a PNP current source down from the rail; that could be controlled over many decades of current.