Saturday, February 8, 2014

A Treatise on Audio Modulation of Solid State Tesla Coils

The question of audio modulation often comes up in the discussion of Tesla Coils. In the solid state world using conventional topologies for Tesla Coils there are a few methods of achieving audio modulation. It is important to explain the most common ones and contrast each in order to answer the question accurately. The short answer is YES IT CAN BE DONE.

1. Direct feedback signal modulation; FM: This is the method used in all VCO based designs such as the popular 4046PLL type. The frequency of the drive signal is varied around a set point, thus frequency modulating the drive signal. FM modulation of the drive signal relies upon the bandwidth of the resonator to perform slope detection to impose amplitude modulation on the output arc as the drive frequency moves away from the resonance point.
a) closed loop: Here the driver auto-tracks the exact resonant frequency of the secondary coil and uses this as the set point that the drive signal is FM modulated around. As the coil is always run at resonance, this has the best output power achievable. I’ve tested this but not published the results as they are unsatisfactory to my standards. The issue is that at exact resonance the Q of the resonator is so high that you don’t have adequate bandwidth for the audio signal. The result is that you get a highly garbled audio output as there is a different phase shift between the positive and negative half cycles of the audio. I found that by artificially lowering the Q of the resonator you can increase the bandwidth of the resonator to the point that clear audio is produced. By artificially lowering the Q I mean place a grounded object in close proximity to the toroid. Unfortunately my main TC that I used for this kind of experiment is still awaiting repairs after a primary overheat while running at 3200Watts. I’ve done some -almost equivalent- testing on my small coil which runs open loop, but since it is open loop I can’t tell if the bandwidth increase is completely from the lowered Q from the increased loading or if it is because the increased capacitance seen at the topload which consequently drops the resonance point, in effect raising the drive point above resonance.

b)open loop: Here the driver is basically just a VCO where the user adjusts the set point manually. This drive method is the most reliable and easiest to reproduce. This is the method used for my 4046PLLMod modulator. It has the inherent benefit that if something were to come in close proximity to the toroid that the resonance point will shift further down from the set point reducing output power further. This is a user safety benefit, not a user enjoyment benefit. Due to the way the phase shift between voltage and current occurs in multiple switch topologies used in the inverter (half or full bridge) the set point must never be allowed to be set below resonance. This results in an overly capacitive load which causes massive current commutation through the body and/or isolation diodes on the switches.
  •  Build consequences: The inverter must thusly be built to withstand occasional operation under these conditions. This requires the use of two high current ultrafast schottky diodes per mosfet. One diode is in series, forward biased, with the diode’s cathode connected to the fet’s Drain. The other diode is antiparallel with the series diode/fet combination; anode to the fet’s Source and cathode to the first diode’s anode. This blocks off the internal diode and reroutes reverse polarity current literally around the fet. If this is not implemented running the inverter in a capacitive environment will cause a quick and spectacular death of all the silicon involved.

Running the coil where set point equals exact resonance has the aforementioned affect of inadequate bandwidth for audio reproduction. As it is open loop attempting to manually lower the Q and increase bandwidth also lowers the resonant frequency, so one can’t be sure which and how much of which effect is actually yielding the increased bandwidth. In my own tests a very minimal frequency shift was seen and the equivalent shift is not enough to produce the increase in bandwidth seen. This confirms the theory that adding loading will reduce the Q of the resonator, and that such a drop in Q increases bandwidth. The one benefit that suggests further developments are required here is the same for the closed-loop method; the highest output power of any modulation scheme is seen.

Running the coil where the set point is a set frequency above the resonance point drops the Q of the resonator (at the actual drive frequency, the Q of the resonator to the resonance frequency is unchanged) to yield enough bandwidth for adequate audio reproduction quality. The distance above resonance needed is related to the slope of the frequency response of the resonator, and is thus a function of the resonator Q. Testing on my own coils have yielded that approximately 6.5KHz above resonance yields the best results. If you adjust the set point too far above resonance the slope appears more and more nonlinear, like adjusting a grid/gate/base bias down into cutoff. This has multiple results, primarily there is reduced power transferred to the secondary so there is a smaller output corona which is less able to reproduce the lower frequencies of the audio. The user will primarily notice a low frequency roll off as the set point is set higher. Secondly due to the decreasing linearity of the slope response of the resonator there will be reduced change in amplitude for a given change in frequency; less and less modulation is seen in the output corona. It is thus that the user should determine the resonance point by running the coil in closed loop mode with a stable, reliable, repeatable feedback network and record the resonant frequency. Then a transition should be made to the open loop FM modulated drive signal and the user should adjust the set point to where the audio quality sounds its best. It is best to use a high bandwidth selection of audio/music for this process. The user must take care to observe proper high voltage high frequency/RF isolation practices if the adjustments are to be made while the coil is in operation. Moving the set point into the capacitive region below resonance can be identified by a harsh tone in the output corona when no modulation is present. An at-resonance or above-resonance (inductive) environment can be identified by a near-silent hissing coronal output. Good DC bus filtering of ripple is thus necessary to eliminate any hum from the bus input rectification. As the set point is adjusted up from resonance the bass frequencies will strengthen and begin to reduce in distortion. The optimal set point can be identified as the point where bass response is still strong yet there is little to no distortion. If the output is observed on an oscilloscope one can set the time base to where the RF envelope is displayed and bass frequencies will stop appearing to be clipped (flat-topping). At the optimal set point there will still be a noticeable drop in output power compared to driving at exact resonance, but this is not always a detriment, as it allows for longer run times. After all, what good is perfect audio modulation, if you can’t even play a whole song on it!?

2. Drive signal gating - a.k.a. PWM Modulation: This is the method most often employed as it requires the least difficulty in implementation, and the coil is inherently run at a lower duty cycle than CW. This method requires the use of gate drive chips which have an enable input. As the feedback signal is effectively interrupted to produce the amplitude modulation in the output there are times where feedback is lost. This requires the use of a self-starting or self-pinging driver. This is why discrete gate drive cannot be used unless a complex network of logic gating is used to detect when no feedback is present and then ping the discrete gate drive to reinitialize oscillation and feedback. Thankfully Texas Instruments offers complementary gate drive chips which will operate at very high frequencies and very high peak currents that have enable inputs; the UCC37321 and UCC37322. The PWM modulation can be implemented by a number of methods, but the TL494 and SG3525/KA3525 chips are the most commonly used. I’ve successfully used the KA3535 chips in my own tests. The PWM carrier frequency must be chosen to be sufficiently high enough as to render it inaudible to human ears and also be high enough that accurate reproduction of the audio signal’s harmonic content is achieved. If the frequency is too low the effective bandwidth is insufficient for quality audio. This in practice means that the carrier frequency must be above 48KHz or so. Now, as we are interrupting the CW operation of the SSTC we will see a greatly reduced output corona, and in fact if one were to try such a "low" carrier frequency like 48KHz there would be little to no human-detectable output. As the carrier frequency approaches reciprocals of the resonant drive frequency the output of the SSTC will begin to increase. For example, for a 340KHz resonant drive frequency an output maxima will be observed around 113KHz PWM carrier frequency. The exact carrier frequency will need to be adjusted manually to shift IF (intermediate frequency) and BF (beat frequency) harmonics out of the range of human hearing, or loud squealing will be heard in the output. The trick is thus finding a carrier frequency that has enough bandwidth to reproduce the audio with sufficient quality, while having harmonic content of unwanted frequencies outside the range of human hearing, and simultaneously producing a strong enough amplitude corona output to get faithful and audible reproduction of the modulation signal at the output. In my own tests with a 340KHz resonant SSTC and a PWM carrier frequency range of 48KHz-250KHz there were several maxima nodes between 80KHz and 180KHz (which correspond with reciprocals of the drive frequency), and at the "best sounding" of them the audio reproduction quality was significantly less than that achieved by FM modulating the drive signal. Likewise, even when at the strongest power output maxima in that range, the actual output power was less than seen by the FM modulation scheme. More experimentation and developments are required in determining the effect and optimal pulse width set point that the PWM modulation uses. The optimal set point for pulse width depends on how much of a pulse width increase a full scale deflection input signal produces, but the biasing of the set point would be similar to the biasing in a class-A amplifier, where you want to use the highest set point that keeps the signal within the linear region. Given the relationship between secondary resonance frequency and PWM carrier frequency one can postulate that higher frequency SSTCs would benefit more from this topology, except in the case in which the PWM carrier frequency can be made to be the resonant drive frequency, but then we are no longer performing feedback/drive signal gating! In that circumstance it would be best to use a 50% duty cycle PWM set point and keep the modulation to acceptable limits so that the inverter is not driven far from 50% duty cycle pulses.

3. High Level Bus Modulation - Class-H: In this methodology we AM modulate the supply rail via a modulation transformer, as was done in the old days of Tube AM transmitters. The trick here, is to find, or make, a modulation transformer that has the required low impedance to current at the resonant drive frequency of the secondary, yet can handle the large DC current draw of the primary. This is NOT an easy task. I have yet to succeed in this method because of the difficulty in locating a sufficient transformer. Also, because of the large voltage swing needed on the supply rail to produce a sufficient modulation of the output corona a relatively high power amplifier must be used to drive the modulation transformer. Unlike with thermionic valve technology where the vacuum tubes were inherently high impedance devices, the modern SSTC is a low impedance device, so transformer parameters used in the olden days will not suffice, even though the concept of the overall topology is the same.

In the end all methods of audio modulating the tesla coil are all endeavors of AM modulating the output, either directly, or by utilization of slope detection by the resonator. More complex and less common methods are often employed in such topologies as the Single-Ended Class-E or Class-D/E Half-Bridge, but their operation is too complex to discuss here, and is often considered esoteric by most tesla coilers. I’ve even read about an experiment where audio modulation of the output was produced by placing a variable resistive load (class-A operation transistor) in parallel to the primary coil. Such methods always remind me of the quote by Einstein in reference to nuclear power technology: "Hell of a way to boil water!". Though the subjects are far removed, the idea is the same; extreme lengths are taken to do simple things, with efficiency often overlooked.