Let's Design and Build a (mostly) Digital Theremin!

The Thereminworld forums have a search function. Use it with the keywords "Glasgow Theremin" and a long list of discussion threads on this topic will show up.

The short resume is: The aforementioned article is an excellent theoretic tool, but the included circuits are just proofs of concept and were never intended to be realized for practical application in a music instrument. A few naive people did not understand that and tried to build these circuits before they discovered that it made absolutely no sense.

Never build a theremin (be it from a schematic or from a kit) which has not been approved by an experienced renowned precision theremin player! Some circuits may be looking extremely interesting for engineers and have impressive technical data but turn out absolutely unplayable in the hands, eyes and ears of a professionally trained musician.

Thierry's advice is very good, particularly if you are building an analog Theremin.  But if you are building a digital Theremin you just need a decent stable oscillator, and then you can implement the remaining functionality with a processor or FPGA (and if you are using an FPGA you might be able to incorporate a portion of the oscillator in logic.)

Anyway, yesterday I breadboarded the simplest LC oscillator I could think of, a single inverter type with RC delay, which is a CLC Pi network, or tank_mode=2.  We've got two waves of family and company coming to stay so I'm not sure if I'll be able to hook it up to an FPGA moving average filter for a few days.  When I do I'll report back.

" The aforementioned article is an excellent theoretic tool, but the included circuits are just proofs of concept and were never intended to be realized for practical application in a music instrument. " - Thierry

The unfortunate thing is that, if this is true, it is not made clear. There are physical Glasgow Theremins (both an Analogue and a "digital" model) constructed at the time the thesis was compiled.

It is only the fact that there is so little real technical data on theremins which has made the University of Glasgow's theremins "noteworthy" - But this is only due to the thesis, and not to the "designs".. The "designs" are utter unworkable rubbish.

The other fact is that the "Digital" theremin is not, IMO, "digital" at all - Just because one uses CMOS gates for oscillators does not make a theremin digital - these are operating as an analogue front-end. The rest of the circuit cannot in any way, IMO, be called "digital" either - There is a frequency to voltage converter (essentially analogue) and a voltage controlled oscillator (analogue) - Again, the fact that some of these components are implemented in dedicated IC's with some switching components does not make it "digital".

The closest to a fully digital implementation that I have seen (a full digital implementation of most things "real" is impossible - we are analogue, and most things we use need to interface with us, so even if its only A/D and D/A, "fully" digital cannot be implemented) is Dewsters development here - Most "digital" theremins are not digital at all, they are "Mixed Signal" - An example of this would be a circuit using CMOS oscillators and mixing with an XOR.. Even this is not, IMO, "Digital" - There is no number crunching, the oscillators are analogue even if they output square waves (on the time domain they are analogue - period being directly proportional to sensed capacitance, and no quantizing involved) and the mixer, although "digital" only does what an analogue diode mixer could do, and its output is immediately filtered creating an analogue difference output.

Bottom line, for me I suppose, is that "logic level" does not always mean "digital".

What Dewster is doing is different - Even the AFE is part of a true digital system.. it is the "Analogue to Digital Converter" albieit the Analogue is a capacitance detection.. Thereafter everything is handled in the digital domain - numbers are crunched etc to provide a set of numbers representing the output waveform - these are then converted to analogue at the final stage, so we can hear it!

IMO, the above is the only way that makes sense. If you are going the digital route, then use the power of digital computation to do things which are difficult or impossible to implement in analogue - Things like using lookup tables for tailoring linearity and exploiting audio wavetable synthesis, changing effective sensitivity, register switching etc..

Fred.

Fred, I agree with you 100%.

Brain picking time: have you given any thought to the EQ coil being magnetically coupled to the tank coil?  The turns ratio could simultaneously give a voltage boost and higher impedance at the antenna.  Andrey Smirnov's D-sensor seems to employ this and I wonder if he's onto something critically important.  High-end microphone preamps used to use transformers for impedance transformation and voltage gain, the resulting SNR boost could be quite amazing.

I'm coming to the conclusion that open chokes like the Bournes 6300 series are perhaps just too open to environmental magnetic fields.  I've got a tank_mode=2 single CMOS inverter oscillator going on my bench hooked up to an antenna & scope.  The scope is set to look at the output ~1ms after the trigger to see what one might expect in terms of sensitivity to hand capacitance.  Most of the time there is around 200ns of noise on the edge, but every once in a while it's WAY more than that.  I think it might be working like an AM antenna.  I wish my scope had an FFT function.

"have you given any thought to the EQ coil being magnetically coupled to the tank coil?"

Yes, I played with Andrey Smirnov's designs quite a bit, both "in the flesh" and through simulation. I found that for simple "D sensor" type designs, his system is a good simple one, particularly for longer distance sensing - but I found the linearity inferior to uncoupled inductors.

My recent experiments on the Lev oscillator have been focussed on having a seperate coupled winding for the antenna - take the Lev circuit, add a winding to the existing tank transformer, one end of this to Gnd, the other to the antenna via the EQ coils. By changing the windings on this one can vary the sensitivity, and it seems that linearity is good and almost constant regardless of selected sensitivity. (actually, one does not need an entirely seperate winding - one can just add an extra winding or two to one of the existing tank inductors and connect the antenna to one of these - I am using small relays to select the sensitivity)

"Most of the time there is around 200ns of noise on the edge, but every once in a while it's WAY more than that."

Sorry - No idea about what could be causing that.. But I wonder - could you have another inductor as a sensing element, just to observe what sort of signals are being picked up that may be causing the problem? The other thing I wonder is whether the frequency you are operating at could be making life more difficult - I understand the reason for the low frequency, but if you were to run at a higher frequency and count the clocks, looking for cumulated phase shift after every n clocks, would you not be able to effectively get the results you are seeking but have an averaged phase shift rather than a cycle-cycle phase shift?

One other trick I have used in my experiments is to shield the inductors - Fridge magnet material is great for this - Put the inductors centred (I used foam sheet to keep them in the centre) in a plastic tube (the bigger the diameter the better) and wrap the fridge magnet on the outside of the tube.. I have also played with iron oxide mixed with silicon sealant, which I painted onto the inside of the theremin box.. mu metal would be an easier solution, but it is hard to get and damn expensive... I dont mess with the above anymore as I find that a correctly designed analogue front end (capacitors in series with the EQ coils etc) work fine with an analogue theremin - It was the digital front-ends which caused the problems and which drove me to desperate measures!

Fred.

"Yes, I played with Andrey Smirnov's designs quite a bit, both "in the flesh" and through simulation. I found that for simple "D sensor" type designs, his system is a good simple one, particularly for longer distance sensing - but I found the linearity inferior to uncoupled inductors."  - FredM

Very, very interesting, thanks for that info Fred!

"My recent experiments on the Lev oscillator have been focussed on having a seperate coupled winding for the antenna - take the Lev circuit, add a winding to the existing tank transformer, one end of this to Gnd, the other to the antenna via the EQ coils. By changing the windings on this one can vary the sensitivity..."

Also quite interesting.  You seem to be using the winding to optionally reduce sensitivity (quite clever) but have you noticed if this arrangement can increase it over the standard LC tank & L EQ?  I'm really interested in increasing sensitivity.  If I read you right, if the separate winding has the same number of turns as the tank L you get similar sensitivity as just connecting the EQ | antenna to the tank winding like usual?

"But I wonder - could you have another inductor as a sensing element, just to observe what sort of signals are being picked up that may be causing the problem?"

Thanks!  I tried your suggestion today, lots of splatter on the scope now and then (infinite persistence pays off) but I'll have to compare it to a smaller coil to know how bad it is.

The other thing I wonder is whether the frequency you are operating at could be making life more difficult - I understand the reason for the low frequency, but if you were to run at a higher frequency and count the clocks, looking for cumulated phase shift after every n clocks, would you not be able to effectively get the results you are seeking but have an averaged phase shift rather than a cycle-cycle phase shift?

Yes, I have the strong feeling that lower frequencies are more problematic for interference.  Though the 1mH inductors I was using before are electrically shielded.

Unlike Tank mode 1, mode 0 doesn't seem to simulate or function well when the tank inductor is significantly smaller than the EQ inductor - keeping them in the same ballpark means the tank inductor is bigger than what most designers use.  The up side is no tuning or tweaking at all and pretty good stock linearity, the downside is a lack of inductors to choose from due to the low frequency operation.  I feel like I'm back where I started, looking for a large mH transformer with a closed, low tempco ferrite core and high self resonant frequency.  I can see how people turn to IF xfrmrs and the like, but I don't need the adjustability.

What I don't get is why the Bournes | Miller EQ inductors don't seem to overly bother analog Theremins - hanging a long string of them off the tank seems to be asking for it.

[EDIT] 100mH electrically shielded inductor seems to be perhaps worse than the unshielded 50mH in terms of picking up environmental noise; 1mH shielded seems to be a little better.  Hmm..

"Also quite interesting.  You seem to be using the winding to optionally reduce sensitivity (quite clever) but have you noticed if this arrangement can increase it over the standard LC tank & L EQ?  I'm really interested in increasing sensitivity.  If I read you right, if the separate winding has the same number of turns as the tank L you get similar sensitivity as just connecting the EQ | antenna to the tank winding like usual?" - Dewster

Actually, I am using the extra windings to increase sensitivity - The RCA (well, I am now using the Clara-Min topology) has quite low sensitivity - but the principle could be used for either increasing or decreasing sensitivity I think ..

I think it works like this..

The antenna tuned circuit is operating in its inductive zone - change in antenna capacitance 'transforms' into a changing inductance which is in parallel with the tank coil / transformer.

The variation in this "virtual" inductance, as you know, is a complex function based on the resonant frerquency of the antenna LC and the oscillator frequency driving this - But for a given change in capacitance over a given frequency span, the function (and therefore the variation in XL) is constant..

Now, if we take this varying "virtual" inductance, and strap it across a 160uH tank coil, we get a resultant change in the oscillators frequency (and the complex relationship caused because the change in oscillator frequency changes the value of the "virtual" inductance) .. If we strap this "virtual" inductance across 320uH (a tapping further up the coil) the proportional effect of the virtual inductance on the tank transformer is increased, and sensitivity is therefore increased.

I have yet to work out exactly what is happenning - its not a simple relationship, as in, the number of extra turns (or increased coil inductance) to increased sensitivity is not what one gets from simple inductances - the coupling being to both windings seems to be introducing anomalies.. I have now reverted back to having the extended windings connected to the original antenna connection point (the Anode in the case of the Clara-Min, and the Grid in the case of the RCA - I am using the Clara-Min topology now) rather than having a seperate winding which is grounded at one end, and this seems to be more stable / predictable.

But I will say no more until I am sure that its doing what I believe it is, that its repeatable, and I have checked my theoretical hypotheses against the test data and can be reasonably sure I am not talking BS ;-)

Thanks!  I tried your suggestion today, lots of splatter on the scope now and then (infinite persistence pays off) but I'll have to compare it to a smaller coil to know how bad it is.

If you trigger the 'scope from the "noise detector" you should be able to confirm whether this is whats causing the phase shift errors ?

"What I don't get is why the Bournes | Miller EQ inductors don't seem to overly bother analog Theremins - hanging a long string of them off the tank seems to be asking for it."

Analogue is a lot less fussy than digital - Just look at the filter on the mixer output - one can have 100us of highly distorted HF signals ariving at the mixer and being processed, and hardly even notice a distortion on one cycle of the audio output waveform... Try getting a PLL to stay stable when it loses its signals for 100us, or a digital counting system to give accurate output when signals go erratic and its a whole different ball game! .. With analogue heterodyning, there is an intrinsic "noise filter" or "averager" at its heart - individual HF waveforms have little or no impact on the audio output - one could probably lose 10% of all the "valid" HF and throw in 10% of random stuff, and still end up with audio that was not too objectionable.

Not that I would advise anyone to be cavalier with their analogue HF signals ;-) - But compared to the care required to keep a digital system happy, Analogue is, IMO, Easy! (I have certainly been bitten by the rabid digital theremin "dog" - And its fear as much as anything else which keeps me from returning to his garden! ;-)

Fred.

Oh, just FYI - I am using air core for my transformer - an EFD-25/13/9 former without the ferrite - had been using plastic sewing machine bobbins, but its a pain to mess with flying leads.. these Epcos EFD parts (PN: B66422W1010D1) have 10 pins are £1 and 180 turns on the first 2 layers gives 200uH, then 150 turns on the next 1.5 layers gives 160uH .. I strap a tunable IFT inductor across the first layer (between grid and ground) to trim the total series inductance, the 2nd layer is between "Anode" and +V, and the anode (normal antenna circuit connection point on the Clara-min) is taken to my additional windings.. I switch the antenna connection between the Anode (lowest sensitivity) to tappings from the 'extended' winding (the more turns, the greater the sensitivity) - Phase is obviously important.. If one has the 'phase dot' at the +V and the other main windings phase dot at the grid, then the extended winding continues in the same direction from the anode 'downwards' and tappings are taken from this. 

dewster wrote: ...schematic of a proven oscillator...Moog Etherwave schematic..my tank mode 2...Do you have a solderless breadboard and access to test equipment?

Yes, I do have everything necessary. 

What concerns oscillators, first of all I would like to test the AFE schematics You posted on the page 2. But right now I'm busy exploring some theoretical issues.

From theremin simulation spreadsheet I figured out that currents in the AFE circuit are relatively low, so circuit's NAND gates may be replaced by FPGA logic. What are the drawbacks of doing so?

I think it would be nice to have the choice of output signal's waveform. Possibly Look-up table method can be used for synthesis. But how is it possible to create a sample which will imitate the sound of for example some particular teremin?

Fred,  have you noticed much of a difference (in performance, range, sensitivity, etc)  between using the different the RCA and Clara-min  antenna-coil - connection-points?    (make sense?)

Hi Charlie,

So as not to hijack Dewsters thread, (more than I already have done, LOL - Sorry Dewster! )  I have replied here.