Single Power Supply Stable Antilog Generator

Engineering Notes from a 2nd revision of the Antilog Generator

The LM3900 and the LM324 are the first "Quads," 4x op amps in one package, and the first intended for both bipolar and unipolar power supplies. Rounding out the "Quad Power Trio" is the LM339 Quad Comparator. These devices were developed by National Semiconductor in the early 1970s. I guess because I'm a nostalgic and perhaps also a stubborn SOB, I keep finding new and novel ways to use these 50-year old ICs in through-hole designs for a modular synthesizer. The "Quad Power Trio" has an industrial design emphasis, these are not high-performance professional audio devices. They are also not late-modern, rail-to-rail, fast, or low-offset op amp and comparator devices. Yet their feature sets presage the modern era, while also having unique features, the LM3900 particularly so. But these devices are still available a half-century later, they are widely used, and they are inexpensive. I also prefer bipolar technology more so than CMOS, though clearly CMOS does have a role. For analog synthesizers I'm finding ways to use good-enough performance to build a synthesizer with industrial devices -- which even CMOS CD4000 logic is considered as.

As just one preliminary word on performance: the degree of optimality obtainable by circuit elements like op amps does depend a lot on the electronic characteristics of those devices. However, op amp applications, for virtually any of these devices, are mainly limited by the perfection of the external components used. There is a lot of great circuit design performance space available using industrial analog IC devices with Vos ≤ 3mV, and Ib ≤ 100nA when 0.1% precision resistors are available for ≤ $0.05/each!

Working on my LM3900 circuit designs for electronic music applications, I've amassed a library of +15V single power supply designs. I've also been exploring use of the LM324, including using that device in a design for an inexpensive yet good performance VCA that is trimless. Lately, I've been building a higher performance version of the classic AN-72 VCO for the LM3900, and this has involved creating single power supply (+15V only) Voltage Controlled Current Source (VCCS) to drive it, which has lead back the LM324. These two devices, the LM3900 and LM324, esp. for a single supply voltage operation act together like peas in a pod, for current and voltage inputs, respectively.

A kind of break-through circuit element I bench tested with the VCCS is a precision LM324 adder-subtractor that operates on a single power supply, and only requires one op amp. Most variants of an adder-subtractor op amp topology require bipolar supplies, and not infrequently are two op amps used. The LM324 single power supply VCCS adder-subtractor will take a +10.000V reference like the LM4040 and allow that to be subtracted from a variable offset voltage that has ≤ 10V range; this shallower range offset sets the current through an emitter resistor for a PNP-based precision VCCS. Of course, newer single supply op amps will also work in this same adder-subtractor topology depicted in the LM324 data sheet, but it's not as easy to get very late modern devices that will do that for power supplies up to 15-36V.

But bench testing the VCCS with the adder-subtractor got me thinking about the other part of the typical synthesizer VCO: the antilog generator. The general topology of offset-and-inversion is de rigueur in the LM3900 world of design, due to the Norton (current) mode of operation. Combining a reference voltage with an offset with precision would enable generating a ±180mV offset voltage against a BJT tied to that reference voltage. Which is perfect for setting up an antilog generator design on a single power supply voltage, esp. if the output is already intended to be used with inversion -- such as for a VCCS.

Contrary to the vast majority of recorded history in electronic music that reliable antilog generation takes two matched BJTs that are of alike polarity to eliminate thermal effects due to their Is, in fact this Is cancelation can also be done with two matched BJTs of both polarities, NPN and PNP. The traditional dual-matched NPN or dual-matched PNP used in an antilog generator is a parallel circuit, whereas Is cancellation will occur with a NPN and matching PNP transistor when they are used in cascade -- a serial circuit. Such things are possible with modern transistor arrays from That Corporation, which offers ultra-matched PNP + NPN transistors on the same chip; devices like the THAT 340.

I had an intuition that such an ancient abandoned idea ("this is clearly wrong and not possible") might have other favorable thermal properties such as not requiring compensation for temperature drift effects of the matched transistors. I've had a string of good luck with complementary-symmetry circuit designs, part of why I very much prefer bipolar devices. The antilog generator in the engineering notes and scope fotos appears to not require external compensation against temperature drift. I don't have an environmental chamber for calibrated electronic testing, but I tested with a heat-inducing flashlight and also an electronically controlled hot-air source, and this PNP + NPN circuit did not move at all. Whereas, with every unbalanced transistor and LM3900 circuit I've ever worked with you can immediately see thermals on the scope waving your hand or adding hot breath to the test circuit. The low working voltages aross BJTs, and in relation to audio signals, are also why I wanted a 6.5 digit DMM, and a 12-bit oscilloscope.

The antilog convertor shown in the engineering notes above is an exercise in simplicity. Here, I'm using the LM324A Quad op amp for the improved Vos ≤ 3mV across temperature, vs the LM324. An adder-subtractor combines a buffered precision +10V reference from a LM4040 with a ±5V (10 Octaves, in V/OCT) input that is precision attenuated to ±180mV. The 10V±180mV is then sent to an NPN emitter follower which uses a 50µA current reference generator, instead of a resistor. The output of the NPN emitter follower is tied to base of a PNP antilogging transistor which has it's emitter tied to the same buffered +10V precision reference voltage. Due to the ±500µV matching of the NPN and PNP transistors on a monolithic device, -Vbe of the PNP device is cancelled by +Vbe of the NPN device and "zero" at the output of the adder-subtractor matches "zero" for the antilog PNP device, with precision. The antilog current provided by the PNP device is converted to an output voltage with an op amp, which is creates an "upside down" antilog signal against an alternative reference voltage. For the available voltage compliance here in a +15V circuit with a +10V reference voltage, this alternative reference voltage was set to +7.5V, so the antilog output swings down from 7.5V. To eliminate supply voltage variance from the design, the alternative reference voltage is derived from the primary +10V reference voltage. The voltage used is application-dependent, but it cannot be as high as +10V due to compliance with the antilog transistor collector; and the lower it is, the less signal swing is available for the antilog voltage output. +7.5V is a good compromise value, and others are possible.

Thermal stability was very much a greatly desired feature of this NPN + PNP experiment. The circuit design is however not (yet) absolutely perfect. Due the serial topology of the monolithic and matched transistors, thermal effect compensation is built-in. However, the transistors are operated such that there is a base-collector voltage across them, which means that the Early Effect will have some impact in the antilog ideality. What I've seen so far is that the circuit design is holding across 6-7 octaves, but above that, the accessible curve starts to linearize to less than one octave (output) change per volt (input). Additional tweaking of the design may yield more -- but the performance from this one-weekend bench test prototype is already good enough for a 61-note keyboard!

Representative scope fotos now follow.