This document is also available in LaTeX printed form.
I recently decided, enough was enough. I've used $10 RadioShack soldering irons for the better part of my life. They're too hot, they're too cold, and the tip dissolves away in no time. I used them for so long because they're cheap enough not to matter, and, something mumbled about adversity (I've always been the kind of person who makes do with less). After the last iron dissolved down to a nub, I bought a Hakko FX-888 soldering station (which works nicely; in case you were wondering, the performance, construction and pricing are comparable to the Weller WES51; either is a worthwhile investment). As something of an inaugural project, I realized, hey, I haven't built a Theremin in a while. And since I've been building radio projects lately, it should be a cinch to put together something really nice.
So, I grabbed some copper clad off the shelf and started cutting and tacking pieces together. Whereas my radio breadboard projects are mostly discrete boxes strung together with wires, I decided to build this in a holistic manner. Also, to echo the standard styling of a theremin, the top has a sloped shape to it.
As you can see, the two variable oscillator coils are already in place. The tuning screw is detailed below:
This is my preferred method for prototype adjustments. It's assembled as follows:
In this example, "additional hardware" includes a stop nut, superglued in place near the end of the screw, and a nut at the very end to provide gluing surface on the end. A ferrite cylinder is superglued to the end. Make sure it's centered and straight.
Outside view, showing the locking nut.
As it turns out, in this project, I didn't need the range provided by the ferrite slug — so I broke it off, leaving just the nuts on the screw. This has a mixed effect: the brass screw itself will tend to reduce inductance slightly; you would imagine the unplated steel nuts should tend to increase it, but steel isn't very magnetic at high frequencies. The overall effect is, it tends to reduce the inductance. A competing effect is the increased capacitance from winding to screw, which tends to oppose the reduction of inductance. It turns out, the overall effect acts to raise the resonant frequency, so eddy currents in the steel nuts dominate. One or two brass nuts would be better here, and sure to have the same effect at any frequency.
I started this project by winding the coils. According to this excellent calculator, I got about 40μH, which is kind of low among theremin coils, and means my operating frequency will be on the high side. I don't think this is a problem; 1mH air-cored coils are certainly a pain to wind (I don't have any good RFCs handy), and a higher frequency will just make it that much more sensitive (and that much more unstable, but we'll see).
One problem I identified early: I didn't put taps on these coils, so I have to use a Colpitts rather than Hartley topology. That's fine, but I'd like one end grounded, so I can get maximum sensitivity by connecting the antenna to the opposite end. Also, any savings I can make on coupling capacitors or RFCs comes in handy.
I rearranged the standard circuit to the layout shown below. The circuit ends up a common-collector form, and requires an RFC in the emitter to develop voltage. The output waveform seems unusual because the emitter voltage is biased a little above GND, saturates to the +V rail, and swings some amount below GND.
A JFET follower was added to improve isolation between mixers and oscillators. Cs was added, in series with the tuning coil, to effectively reduce the tank capacitance, raising the resonant frequency. This helps compensate for the added capacity of the antennas, and any other differences between units. The volume oscillator (which had a slightly misshapen coil, and needs a larger frequency difference from the local oscillator) ended up needing a rather large adjustment, a 470pF capacitor in series (effectively reducing the total tank capacitance, raising the resonant frequency). The pitch oscillator got 1.5nF in series, while the local oscillator got none (Cs shorted). Note on capacitors: polypropylene or polystyrene film, C0G ceramic or silvered mica are required for all tuning components. Generic ceramic types or polyester films are much too drifty, even if you get lucky enough to have them all drifting in the same direction at any given point in time.
Complete oscillator board, sans tuning coil and external connections. Constructed on a chunk of copper clad. Nodes are isolated by carving a perimeter around the pad with a utility knife. I chose to build with THT components, even though they end up SMT'd anyway, for ease of assembly and rework (ease being a relative thing with this kind of assembly).
The general scheme of a theremin is to mix two variable frequency oscillators (VFOs) into a third, the local oscillator (LO) (or use two pairs, one LO for each VFO, though I don't really see the point of adding a fourth oscillator), obtaining two independent products. One product is detected for frequency shift somehow, while the other is used directly for pitch. Typical frequency detection methods include the slope detector or the ratio detector. A variable gain amplifier is used to control the pitch's amplitude, which might be the pitch mixer itself.
This is the schematic used for the volume mixer (the output ports from the volume VFO and LO connect directly via unterminated coax cables). A single balanced BJT mixer was chosen because double-balanced diode mixers require transformers, and the low coupling, power gain and current-source output are advantageous. The detector is also combined, being a simple slope detector and voltage doubler detector (aside from a capacitor or two, the BAT54S is the only proper SMT component in the project, as I don't happen to have THT schottky diodes handy).
Here's the finished circuit. Instead of carving out pads (except for the detector diodes), as on the oscillators, this was built using a combination of self-supported components and "Manhattan" construction. A lot of resistors and capacitors go to ground, so these act to anchor nodes that are otherwise floating in space (a good analogy for their electrical purpose, too). "Manhattan" construction refers to the technique of soldering down small cut pieces of PCB, acting as the ground floor of a skyscraper, supporting a node. The supply node (green wire, bottom right, and associated components) is supported in this manner.
This is the pitch mixer. The mixer itself is essentially identical (some resistor values changed for convenience, with little effect on operation), with the major change that, instead of the VFO port being biased from a fixed voltage source, its voltage divider is supplied from the volume detector's output voltage. In this way, when the volume VFO is tuned away from the detector's resonant peak, no bias is developed and the mixer remains off. Approaching resonance, a small voltage is developed, biasing the mixer on (it remains partially in cutoff, resulting in a somewhat distorted waveform at the output). The pitch VFO's output is also attenuated with a series resistor, which minimizes distortion, most significant at full output, where a nearly pure sinewave is produced.
Since the output is a "downward" current, a load resistor, diode and transistor mirror this current back down towards the ground rail. This voltage is AC coupled to the volume control potentiometer, and from the pot to the amplifier.
Finally, to buffer the audio output and bring it up to enough drive for a cable run, or headphones, a small power amplifier. This one circuit uses almost as many transistors as the rest of the circuit (seven transistors, oddly enough), just because it's a fully formed operational amplifier.
A pretty standard class B amplifier. Only 5V is available in this particular project, so a single-supply approach was used. This has disadvantages due to all the coupling capacitors required, reduced PSRR (power supply rejection ratio) and the turn-on "pop" that occurs when first charging those capacitors. Still, the results are good.
The construction method follows from the mixers, with a flatter layout this time. Minor testing reveals response up to about 1MHz (1Vpp, slew rate limited), reasonable linearity (no obvious sloping or kinks on the waveform) and enough output capacity to do what I need. It drives common 32-ohm headphones very nicely.
And here's an overall view of the completed project. It plays very nicely, though I'm not going to even try waving out a tune for a general audience!
That said, here's a quick demo of the unit, including the output waveform, which as you can see is an excellent sine wave.
Putting it all together, this is the full schematic:
This is, of course, easier if you have a frequency counter or something like that, it's just more tedious the regular way. The procedure:
First, you need to get all three oscillators in the same range. If you leave the LO alone, you can add or remove capacitors from the VFOs to adjust them in large steps, or use a variable trimmer capacitor to cover a wide, continuous range. Tank capacitance can be increased by adding capacitance from the antenna to ground, while it can be reduced by placing capacitance in series with the coil. The more in series with the coil, the smaller effect it has, so the 1440pF with the pitch coil causes a small shift (about 3% higher frequency), while the 660pF with the volume coil causes a larger shift (about 6.2%). The approximate shift can be calculated from:
Where Cs is the series capacitor added, and C is the equivalent total before the change (which, for a Colpitts type oscillator, is the equivalent capacitance of the two tuning capacitors in series).
If you don't have a frequency counter, you can determine the direction of adjustment by making two assumptions: 1. proximity to the antenna always drops frequency; 2. the tuning screws (without ferrites) tend to increase frequency. If you can observe a blip of a beat frequency on approaching the antenna, then it has to be brought lower. If you don't observe a blip, it's too low and has to be raised by adding series capacitance.
The volume VFO has the special behavior that it can have a double-humped response if tuned to the higher peak. This is probably undesirable (for a given volume setting, there are up to four frequencies you can choose!). It should be tuned to the center of the lower-frequency peak, so that, on approaching the antenna, you can only push it further and further from resonance. Some people prefer a passive-off state, which can be achieved by tuning to the valley between peaks; this circuit can be used that way, but if you go past the peak, the amplitude drops again. The best way to do that would be to change the slope detector from bandpass to a notch type filter.
Since the pitch oscillator might be way off initially, defeating an audible tuning attempt, a volume indicator LED was added so adjustments can be made visually. First, find the volume oscillator peak (maximum LED intensity), then adjust the pitch oscillator until you can hear the beat frequency. Adjust it near zero, then tighten down the locking nut partway. Holding the nut with a pair of pliers, use a screwdriver to turn the screw ever so slightly until it's good. Final adjustments in place are best made by moving metal objects around the antennas — this won't impact the sensitivity of the antennas, and environmental sensitivity is obviously important regardless. Might as well take advantage!