I am now contemplating PCB designs for my modular Model III EMS, in the form of PAiA FracRak. And I'm aiming to get some modules actually made by years end, because I want to actually start playing with this modular synthesizer after decades of development! So, this bias toward completion is forcing some priority decisions. Like what module to complete first, given a large collection of research and development in circuit designs; noted in this blog, but also at my Beyond Application Note 72 blog too.
The first thing that came to mind was my Trimless VCA. Just because it was so elegant, and has small parts count. But, that design used virtually perfect op amps and transistor arrays, which made it very expensive. After a bit more thought, my New VCA, based on application with the LM324, was much more intriguing. Because it was a more adaptable design, one based on inexpensive components, like late-modern discrete transistors with high β, and the LM324 quad op amp. Devices that tended to be generally available during the supply chain disruptions in the last several years. In fact the LM324 is still today an immensely popular device, and it's more than 50 years old.
I was torn: I liked how the low cost New VCA worked, but I wanted a tiny bit more performance out of it, and I also really wanted to have a trimless design: no trimmer resistors at all. I wanted to keep DC accuracy and DC coupling, reasonable bandwidth, moderate distortion. Because PCB designs needed to get started, some parsiminomy of parts made sense. For example, what would it take to obtain a dual VCA? So, I wondered if there was a way to apply the benefits of each VCA design.
In late-May, I put together the best of Trimless and Good Performance With Inexpensive Parts. The circuit design is shown in the engineering notebook schematic below, and it went through an evolution while bench testing on a pad-per-hole breadboard.
I'll first describe the design thinking that led to this design. Then following that are some annotated scope fotos from measurements I took during circuit characterization.
I was torn: I liked how the low cost New VCA worked, but I wanted a tiny bit more performance out of it, and I also really wanted to have a trimless design: no trimmer resistors at all. I wanted to keep DC accuracy and DC coupling, reasonable bandwidth, moderate distortion. Because PCB designs needed to get started, some parsiminomy of parts made sense. For example, what would it take to obtain a dual VCA? So, I wondered if there was a way to apply the benefits of each VCA design.
In late-May, I put together the best of Trimless and Good Performance With Inexpensive Parts. The circuit design is shown in the engineering notebook schematic below, and it went through an evolution while bench testing on a pad-per-hole breadboard.
I'll first describe the design thinking that led to this design. Then following that are some annotated scope fotos from measurements I took during circuit characterization.
- Keep the ±2.5V I/O voltage range for audio, 0-5V for control voltage. Other "standards" that are 2x larger exist, but these larger voltage ranges are not useful for audio that must ultimately be recorded. More signal swing requires higher-grade op amps, while reducing bandwidth, all for extra performance that is actually unnecessary.
- The topology of a differential transistor transconductance core driven by a two-transistor current mirror was retained, because this result in relatively linear VCA action. This as opposed to exponential gain control. And this topology is evidently relatively temperature stable. And stable with discrete transistors, not from an integrated transistor array. The topology is also such that one op amp handles DC gain, and the other AC + DC signals for audio. So, only half of a quad op amp is used, affording a dual VCA using only one IC package.
- As will be seen in the scope fotos, the prototype achieved good performance using randomly selected 2N5088 NPN transistors, which I obtained from Digi-Key in 2003. The prototype constructed in the pad-per-hole breadboard didn't even have thermally coupled transistors -- and it still worked well!
- Maintaining power supply voltage independence. The calibrated zener diodes provided reference voltages for both positive and negative supply as seen by the transconductance pair. These reference voltages also assist in the control voltage and voltage offset scaling needed to drive the current mirror so that a control voltage input of 0 turns off the VCA.
- Both the Trimless VCA and the New VCA had a feature where the VCA would be off until the input control voltage exceeded the Vbe of the current mirror. I wanted to narrow that 0.45V cutin voltage gap down to much closer to zero, so that the VCA would act more like an arithmetically perfect two-quadrant multiplication.
- Use the LM324A versus the LM324, to improve DC offset accuracy by about 2x. The offset voltage of the LM324A across a 5Vpp signal range is actually quite small as a percentage of full scale. This better DC accuracy helps with the performance of the output differential amplifier.
- Maintain modest DC accuracy, but still not requiring trimmers. Was there any part of the topology where use of 0.1% precision resistor would be essential? But are there other parts where lower precision 1% or 5% resistors could still play a role? The idea here being that really excellent external components preserve the best qualities of even a moderately good op amp like the LM324A. Similarly, relatively inexpensive but well-matched components like 0.1% precision resistors in a good topology could also preserve whatever matching might be achieved with randomly selected transistors.
- The above idea was key, and it worked! Using ultra-matched collector loads tended to provide DC and AC current balance between type-similar but otherwise unmatched transconductance transistors. And, as discovered during development of the Trimless VCA, where every component was a matched type, the input attenuator resistor networks need to be matched. But with single ended input, only the base resistors need to be matched to 0.1%, and the lone single ended input resistor can be a less expensive 1% type. This is because precision differential variable gain is not being attempted. Lastly, the output differential amplifier required 0.1% resistors for differential gain balance, but due to the previously researched VCA topology, only two such resistors are required instead of four. Use of low differential gain (2x) simplifies much as well, reducing the impact of LM324A Vos and also Ios. Additionally, the input control voltage gain scaling and offset for the current mirror had a very interesting fulcrum: choice of full scale peak to peak voltage for maximum control voltage versus minimum output at minimum control voltage. In other words, tradeoff choices for maximum dynamic range.
- In these configurations, very reasonable VCA performance resulted. Imperfections are mapped to areas that do not have deleterious effects sonically or in terms of DC accuracy.
- The final characterized circuit design had the following results. Good DC accuracy with the VCA off, feedthrough minimized without trimming. When off, the VCA offers a DC voltage offset < 10mV that is very consistent. A modest fraction of a 1V/octave semitone voltage interval. At full output the DC offset does grow to 2% full-scale (e.g Y-offset feedthrough), but you don't hear that with an AC signal modulated by envelope control voltages. The very distortions that Barry Gilbert's classic 1968 paper discusses minimizing using integrated circuit techniques. But here, fairly good performance is possible with four discrete transistors and commodity industrial op amps. Sinusoidal full power bandwidth under load to 20 kHz was seen with very shallow phase lag. Square wave performance is slew rate limited but edges are clean ramps of equal duration, essentially no pulse dynamics. Maximum dynamic range for the output Vpp matching the input can be 72 dB. The classic crossover distortion seen with the LM324/A is controlled very cleanly with a pull-down resistor. A very pleasant surprise is that distortion is primarily a second harmonic at < -55dBc. Lots of potential for good sounding overdrive. Not a precision two-quadrant analog multiplier, but a sonically good VCA.
Comments
Post a Comment