A New Discrete VCA

My Electronic Music System is an art project to build on what might have been possible with analog synthesis after the Moog modulars were introduced in the mid-1960s, by taking the track of using more advanced silicon transistors of even that time, but not taking the track of Integrated Circuits, nor that of computers and software. So, this system is to be constructed entirely out of discrete components. In addition to avoiding ICs, this development is expressly based on Through Hole Technology, and explicitly avoids all use of the designed-for-obsolescence Surface Mount Technology.

Over the Decemeber 2024 Holiday, I completed several transistor DC voltage adder circuits, because these are part and parcel to internal control over various EMS elements. This discrete transistor work had been going on and off since January 2001, and there had been some good designs, including designs that can provide gain for positive-only outputs. Among the things I had developed before was a discrete op amp.

But what I learned in more recent years is that transistor circuit design does not need to mimic IC building blocks, and that interesting combinations of transistors can do amazing things that op amps cannot. As well, not everything in a useful circuit design even needs or requires an op amp. The biggest "Aha!" I've had in the last quarter century is that transistor circuits can easily out-perform ICs. Generally, at a cost of power and area, but not always.

What followed after tidying up DC adders was nailing down a good circuit design for a VCA. I had been studying the Moog 902 VCA for a long time, because I liked the generality of that module. That it supported differential inputs and outputs, and that it allowed signal combining between VCA modules for ring modulation. And I like that it had voltage-controlled gain to output a signal larger in amplitude than the input signal, if desired. Plus could be overdriven. So, these sort of things I wanted to keep for my VCA, but I was not intending to mimic this circuit design within a ±15V power system, as versus the +12V/-6V system that Moog used. In addition, I have done a couple of trimless VCA designs with ICs and discrete transistors, including a trimless design with low-cost op amp ICs.

For several weeks into this New Year I experimented with discrete variations on the trimless designs, first playing with the gain block side, and then with the output stage side. Then finding out how to best join them together.

The Moog 902 uses a three-tier differential amplifier scheme, and a very clever voltage reference scheme that further takes advantage of the ability of differential amplifiers to reject common mode voltages. There is an input differential stage, which allows both single and differential inputs; then a differentially operated gain stage; then finally a coupled differential output stage. My trimless designs seemed generally able to combine these three stage functions into two, and this discrete variant shows this still works. This is likely enabled because Silicon transistors of the late-1960s and later have more gain, more bandwidth, and more particularly there are matched pairs available ... than as for the case in 1964.

A Gestalt or "Lessons Learned" while developing this VCA design, include the following:

1. There is a need for a topology inversion between input stage and output stage; it can be NPN then PNP, or the reverse, doesn't matter. But a NPN output stage offers the opportunity for higher current and lower output impedance.
2. The Moog 902 VCA design used 680Ω 5% resistors for the outputs from the PNP differential output stage; I chose 5.00 kΩ instead to align output currents for a modern PNP dual matched transistor (LS351), and for less power dissipation. In earlier work, I was able to successfully construct an output stage in NPN form for 604Ω 0.1% resistors using a 2N2223A dual, but to get even 5Vpp signals required nearly 200 mW of power, and the transistor got noticably warm to the touch. This, I did not like, as it would not bode well for long-term reliability.
3. A differential pair with a CCS for the emitter current tends to have high-impedance inputs which are naturally susceptible to hum and noise pick up. And I was seeing this on the scope.
4. A really great solution to hum and noise pickup is to bias both inputs to the differential pair with high-precision resistors, at a lower impedance, to a common mode voltage above (PNP) or below (NPN) that is needed for output signal compliance. Even doing this with 0.1% tolerance resistors resulted in very small (4mV) voltage offsets, which become DC inputs that change the balance of outputs. While a bit annoying, a small trimmer resistor enabled the balance to be zeroed precisely. 
5. The huge Aha! in this design was that the DC common mode biasing resistors for the inputs to the differential output stage could in fact also function as the load resistors for the output of the input (gain control) differential stage! Another key insight was that the differential output resistors could easily be zeroed against ground by inserting a trimmable drop resistance to the requisite power supply voltage, rather than creating a specific DC voltage that would need to be regulated.
6. In various testing with the DC biased form of differential amplifier, I quickly learned that interstage coupling with capacitors is a really bad idea.
7. All through development of both input/gain differential stages and output differential stages, it was clear that tremendous freedom exists for scaling: emitter current, collector resistors, and the attenuation used to obtain input signal drive into the tens of millivolts level -- any of these offered control points to leverage specific outcomes.

What follows is the Engineering Notebook page of the new VCA, as well as scope fotos. In the circuit design, the DC adder for voltage control is not included, because I could mimic that with my signal generator for testing the VCA core. The real VCA in a synthesizer module would include that.

Engineering Notebook page on New Discrete VCA

DC Control Voltage sweep from -5V to 0V with 100 Hz triangle ramp inputs at 5Vpp.

Same scenario with 100 Hz sine wave input

When "not patched" e.g. no Control Voltage input, the output auto-aligns to zero.

Easily supports 100 kHz sine waves with extreme phase margin

Full Scale output at maximum control voltage is +3.2 dB larger than 5Vpp input; precise balance between differential outputs was hampered by a single-turn 200Ω trim pot, where a multiturn was needed.

Full scale 7.25Vpp triwave output somewhat softens at 10 kHz

A bit more noticeable at 20 kHz -- but then I can't hear that high of a frequency

-3dB bandwidth is 400 kHz!

Still pretty good signal quality for triwave at 50 Khz!

The output is "very loud" at 7.25 Vpp full scale. For this 1 kHz signal, the main distortion peak is 3rd order, some -33.2 dBc. Operated at lower nominal outputs (like 5Vpp) there would be even less distortion.