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

You completely deserve to toot your own horn a little bit. You put a lot of work into the D-LEV and it sounds amazing!

What do you mean by relaxing the pitch field?Having an option for less octaves? 

Does the D-LEV have an option for reversed fields?

I tried to read the paper but my device freaked out when I clicked the link as I got redirected like 5 times. 

"What do you mean by relaxing the pitch field?Having an option for less octaves?"  - Trymeinchess

Yes, you can set the overall sensitivity of both fields to be just about anything you want.  I personally set the pitch field to about 1/2 analog sensitivity.

"Does the D-LEV have an option for reversed fields?"

Yes, you can reverse either or both of the fields.  I can't imagine a use for a reversed pitch field, but a reversed volume field feels more natural to me, and as an added bonus it doesn't wail when I leave it alone.

"I tried to read the paper but my device freaked out when I clicked the link as I got redirected like 5 times."

The link in the old post is wonky.  Try this [LINK].

XFMR Sensing

I've been messing around with this simulation for a few days (Cc is parasitic capacitance between the coils):

Which motivated me to construct the transformer.  I 3D printed a 12mm diameter 25mm long tube of white PETG.  Wound 55 turns of AWG30 on it, then put a piece of aluminum foil around the coil, leaving a gap so it wouldn't act like a shorted turn:

Laid a thin bare wire on the foil for an electrical connection, then put another 55 turns of the same AWG30 over all that.  Finally covered it in heatshrink tubing (not shown):

I also made an identical coil without the capacitive foil shielding.  The coupling factor k was measured to be 0.95, and very slightly lower for the shielded xfmr due to the slightly larger gap introduced by the foil.  

It turns out that the shield really messes with things, so that was a rather surprising dead end.  Moving on to the unshielded transformer, the circuit is very sensitive to which end of the sense coil is grounded and which end is being measured: the grounded end had to be the same as the driven end of the drive coil, otherwise the sensed signal is a crazy roller coaster of harmonics (using square wave drive).  Also necessary to avoid the carnival ride is the resonating 220pF capacitor C1, which also conveniently raises the sense Vpp.  Anyway, here's as good as I was able to make it (blue trace is the jig antenna voltage, yellow is the xfmr):

The phase is great, but it all strikes me as kind of a bust unfortunately.  I think there's just too much harmonic energy in the coil.  Sine drive would certainly work better, but I'm rather attached to square drive at this point.

This activity has inspired a transformer-less AFE sensing circuit idea that I hope pans out on the bench.  The idea is to filter the sense signal at the C divider point with RC high pass and low pass (so there is little to no phase delay around resonance) and not filter the reference at all.

Thanks for the link! It was a good case study to read. I can definitely see how the reversed volume field would feel more natural. 

What motivated you to do square drive in the first place? Do you like the sound or is it just that they're cheaper? Regardless, I'm curious to see how much the filters would help.

"What motivated you to do square drive in the first place? Do you like the sound or is it just that they're cheaper?"  - Trymeinchesss

The D-Lev uses FPGA logic to generate the coil stimulus, and for such a digital device square waves are the natural thing to produce.  The fields just produce numbers which are fed to a DSP synth, so the audio isn't impacted in any way by the choice of the coil drive waveform.

Coil Embiggening

As I posted over on another thread, the kit has been on hiatus for over a year now due to some health issues, though they are hopefully largely past now.  I'm reluctant to make more kits in their current form because incremental improvements could be made.  I want to:

1. Increase the inductance of the coils, perhaps 4x.
2. Improve the AFE if possible.
3. integrate the coil, AFE, and plate antenna as a unit.
4. Switch to CAT5 interconnect for the AFEs.
5. Use 5V LCDs if possible.
6. Use PCB material for the control and tuner panels, as well as the logo plate.
7. Beef up the ESD protection.
8. Mount the dedicated mute LED on the control panel.
9. Rearrange the expansion port, possibly separating it into two.
10. Possibly reduce the encoder interconnect and handle them more in SW.

Right now I'm working on #1 thru #3.  The goal of #1 is to lower the operating frequencies of both the pitch and volume fields, ideally to get both out of the AM broadcast band, and the lower frequencies would help ease timing demands on the oscillators.  The LC operating frequency is unfortunately inversely proportional to the square root of the inductance, so e.g. shifting the resonant frequency down an octave requires a 4x increase in inductance.  Previous coil experiments indicate that aspect ratios of 1 or larger are desirable from the standpoint of maximizing Q.  Previous coil winding experience has me avoiding wire any finer than 34AWG.  So these seemed like a good place to start.  I wound two 8mH coils, one with 34AWG on a 60mm diameter 3D printed form, and another with 32AWG on a 70mm diameter form:

Above from left to right: Kit 1mH 38mm/30AWG, test 8mH 70mm/AWG32, test 8mH 60mm/34AWG, kit 2mH 38mm/32AWG.

In terms of winding, finer wire always presents more difficulty, though the smaller form diameter somewhat makes up for this.  As usual, for both coil forms I set the 3D printing z-step to slightly more than the width of the wire, which really helps to guide the winding process.  Obviously, finer wire requires a smaller z-step, which directly lengthens the printing time.

On the test jig the 60mm/34AWG Q measures around 158, whereas the 70mm/AWG32 Q measures around 198, so the larger coil is the clear winner here, and it has the highest Q that I've seen so far.  I went ahead and designed and printed a plate mount for it:

Above: 70mm/AWG32 with plate mount.

The mount places the "hot" end of the coil 2/3 of the form diameter away from the potential plate, which from previous experiments should lower the resonance peak by less than 2%.  The plan for the other end is a plug of some type to support the AFE.  And this larger diameter form could likely benefit from an internal stiffening rib or two placed near the middle.  Printing these elements in any quantity would likely benefit from a printer nozzle diameter increase from 0.4mm, perhaps 0.8?

The kit coils are 1mH for pitch and 2mH for volume.  My reasoning here was that when folks placed their hand on the volume loop, the lowered volume field frequency then couldn't overlap and potentially disrupt the pitch field.  With the use of an insulated  volume plate this is much less likely, so I'm strongly considering flipping this, i.e. using 8mH for pitch and 4mH for volume, which would at least get the pitch field out of the AM band.

Now to print a 4mH 70mm/AWG30 volume coil.  It's rather tedious work, but I really don't mind putting this extra effort into the coils, they're the beating heart of the Theremin after all.

[EDIT] Just wound the 4mH coil:

Above: 4mH coil (left) 8mH coil (right).

The 4mH coil Q is 213, which is a new record for all my coil winding!

Next step ? 

Engraving depicting John Henry Pepper demonstrating the giant induction coil at the Polytechnic Institute, London, before Princesses Louise and Beatrice.
John Henry Pepper (1821-1900) a British scientist and inventor who toured the English-speaking world with his scientific demonstrations.

"Next step ? "  - André

Ha ha!  Too bad they didn't have ferrite back then, Pepper probably could have fit that coil in a suitcase rather than on a flatcar.  Sort of related: I've had folks ask me about making large pieces of metal into Theremin antennas, and to do it right requires very large coil values (the inductance needs to be a lot bigger than the capacitance for high Q).

Just ordered a few items off eBay for D-Lev experimentation and got acid flashbacks of the Covid supply chain collapse.  The stupid Trump tariffs are jacking everything up to like double what I paid in the past.  

AFE Candidate

If we filter the antenna signal in a way that provides very little phase shift from the antenna to the sense terminal, we can perhaps dispense with identical filtering of the reference signal in order to maintain quadrature at the LC_DPLL phase detector.  One way to do this would be to perform both high pass and low pass filtering of the antenna signal, so that the phase lead/lag more or less cancels out around the LC resonance frequency.  Because the vast majority of terrestrial RF transmission exists above LC resonance, we're mainly concerned with attenuating any RF there.  But filtering the RF below resonance probably doesn't hurt, and could perhaps help with mains hum, though that's more of an FM or intermodulation phenomenon which is most effectively dealt with via digital filtering.

The above circuit candidate uses an RC network to both filter and attenuate the antenna signal voltage.  C1/R1 form a high pass filter, as do C2/R2.  C3/R2 and C4/R3 form low pass filters.  So the whole thing is a loose 4th order band pass filter.  It employs no active components to accomplish the filtering, so it should effectively function up to fairly high frequencies.  But this also means that the filter components are highly coupled together, making adjustment a somewhat tedious process, and the filter response a rather droopy affair.  Note in the graph above that the phase error is essentially zero at LC resonance, and isn't too bad some distance from resonance, which is important because the operating frequency of a Theremin can vary rather substantially during operation.  Setting minimum phase error for the far field is probably the best.  Note also that the graph is of Vsense/Vant, which plots the transfer function of both signal and noise from the antenna to the sensing point.

Remember that we're also concerned with loading of the coil in order to minimize loading in the form of capacitance (so as not to overly diminish sensitivity) and loading in the form of resistance (so as not to overly diminish Q), where the goal is to get sufficient attenuation of the antenna signal in order to safely feed the sense signal to 3.3V logic without clipping.  Rbias here is the usual inverter feedback resistor which establishes the network bias at the threshold voltage of the inverter.  Since C2, C3, and C4 isolate the inverter input from any DC sources, the whole thing is able to float.  

The main downside to all the coupling interaction is there is no simple and convenient way to adjust the amplitude of the Vsense signal, so here we have to fall back on adjusting C1, much like the current AFE.  Having a standard plate bolted to the coil should help to control things here.

I've had the above circuit breadboarded and running on the bench for a couple of days.  From what I've seen I think it's an improvement, though perhaps not earth shattering.  Outside of gross behavior, it can be quite difficult to suss out marginal improvement / degradation with this sort of testing, as it relies on quite variable environmental conditions.  There isn't too much danger of incorporating this circuit on the PCB, the pads can be left vacant or jumpered with 0 Ohm resistors to return to the simple C divider AFE.  I'm trying very hard to not degrade the D-Lev in its current form.

The Unvarnished Truth

For some reason I've never done any explicit experiments re. coil doping.  From the very start of this project I've been using clear nail polish to secure the end turns, and PVC heatshrink tubing to secure and protect the entirety of the windings against physical damage.  Someone else who built a couple of D-Levs varnished their coils.  Do any of these coatings negatively impact coil Q?

I took the "losing" 8mH 34AWG coil from my recent experiments and applied two coats of "Varathane Ultimate Polyurethane Interior Crystal Clear Satin" water-based varnish.  Testing on my coil jig before and after revealed a rather large drop in Q, though once the varnish was given 12 hours to more fully dry the Q was more or less back to the original 152 or so.  So I'm guessing water is a big Q quencher.

I also took a 1.6mH coil from past experiments which was naked except for the end nail polish.  I applied the customary heatshrink tubing and again the Q was more or less back to the original 165 or so.  I then cut the heatshrink tubing off and applied a coat of nail polish to the entire coil, after which the Q was unaffected.

For these experiments I also looked at the jig and self resonance frequencies, which were not meaningfully affected.

So below are the proposed 4mH & 8mH kit coils with two coats of satin varnish - I think they look quite nice!  I plan on also covering them with PVC heatshrink, which I should probably order soon.