Designing and Building CMCD Amps - Part 1 - Theory and Overview

I would count this as a work-in-progress post. I want to write down what I've learned primarily, and I will probably update the information here over time. I've been working on CMCD amps with my Dad for a year or two now, starting with a foray into class E amps for their high efficiency. CMCD amps are more appealing to us because of their typically higher output power. The saying is that CMCD has a higher utilization of the switches than class E, meaning you get more power even using the same FETs.

We have settled on using GaN FETs for a few reasons I'll probably go into later, long story short is they're easy, fairly cheap, and fun.

Quick overview of CMCD amplifiers:

A Current-Mode Class D amplifier is the opposite of a voltage-mode Class D amplifier. And a class D amplifier according to Wikipedia is:

"A class-D amplifier or switching amplifier is an electronic amplifier in which the amplifying devices (transistors, usually MOSFETs) operate as electronic switches, and not as linear gain devices as in other amplifiers. They operate by rapidly switching back and forth between the supply rails-"

A CMCD amplifier uses RF chokes to create 'virtual' constant current sources which feeds the actual FET switches. The switches spend as little time as possible in their linear region, being switched rapidly ON/OFF instead. This would ordinarily create a square-ish output of the circuit, but instead a resonant circuit is used to create a sinusoidal output. The primary purpose of the resonant circuit is actually to create Zero Voltage Switching (ZVS), returning the voltage across the switches to 0v right before they change state. ZVS is a common scheme for eliminating much of the switching losses incurred when switching FETs at high power and high frequency. With ZVS, it's possible to create very high efficiency amplifiers at very high frequencies.

Figure 1: Scientific Figure on ResearchGate. Available from: https://www.researchgate.net/figure/Schematics-of-a-conventional-CMCD-PA_fig2_341591078 [accessed 5 Aug, 2022]

Figure 1 shows a common representation of CMCD amps. In this case, the center of the inductor is fed with the RF choke

Problem 1 - Switching

The first thing you have to solve is driving the FETs. They must be driven 180 degrees out-of-phase (with opposite signals, essentially). This can be done many different ways of course, but whatever solution you settle on has to drive the FETs quickly into saturation. The other typical challenges of driving FETs apply, ringing/overshoot/undershoot on the gates and high drive power required at high frequencies being common issues.

Having a duty-cycle adjust circuit is advantageous, for a few reasons. First off, having exactly a 50% duty cycle can cause issues with both FETs turning on nearly or actually at the same time. It seems to me that when the turn on/off time is a significant fraction of your signal period, that lowering the duty cycle can help align the switching events to the zero-voltage crossing points of the tank circuit. The other reason to have a duty-cycle adjust circuit, even if you want exactly 50% duty cycle, is that devices like FET drivers or other digital-input devices often have some variation in their trigger voltage. You can end up with a situation where one FET driver turns on a lot sooner than another, just due to it reading a HIGH input sooner on the input signal as it rises.

Problem 2 - Tank Circuit (LC/Resonant Circuit)

There is a surprising level of complexity with the tank circuit. Especially when targeting higher power with high currents and high voltages. This boils down to the 'circulating' current in the tank, and the quality of the components you choose. The Q value of the tank is very important, but it doesn't seem to me that you can target a single value for all designs. You have to balance the heating of the inductor and the capacitors carefully, but lower the Q too much and your ZVS breaks down or your output signal has very high harmonic content.

We've primarily been targeting 14MHz, and one of the biggest issues we've had is getting high quality capacitors. Often times manufacturers don't list the ESR or the Q of the capacitor at that frequency, and different manufacturers count wildly different values as things like "high Q" or "high frequency". Another critical value that's often unlisted is the thermal resistance of the capacitor. They tend to need to dissipate quite a bit of power, and it seems like the SMD packages are often quite bad at that. The voltage is also multiplied in the tank circuit, so you have to be careful not to over-volt your capacitors. Another thing we've ran into a few times is the capacitance changing drastically with temperature, meaning the amplifiers often have a different frequency "sweet spot" as they warm up. Essentially, you just have to be extremely careful when choosing a capacitor.

We've settled on using air-core inductors for the tank inductance. The inductors we've made are typically thick copper wire (or, when we can, copper tubing). Being copper and quite large compared to the capacitors, often you can afford to burn more power in the inductor, especially if there is airflow over it. Inductors are a strange tradeoff of ESR and inductance. A larger inductor has higher inductance and thus lower tank current, but also has higher ESR and lower Q. A wider diameter inductor is preferable to a longer one since diameter scales inductance faster than length.

Problem 3 - Signal Output

The current design uses a balun as the output. That's from this paper I believe. Actually we've gone through a lot of papers to understand and build transmission line based transformers. The basic idea is that you have a non-ground referenced signal across your tank circuit, it's balanced and symmetrical. You then usually want to take that circuit and send it via coax to your load. If you simply hookup your coax across the tank circuit, you permanently ground half your circuit if the load is grounded, and if not you are at least hooking up an unbalanced transmission line to a balanced source.

I'm not entirely sure yet how this sort of output behaves when driving a heavily reactive load. We used fairly low quality dummy loads for a while which were either slightly capacitive or slightly inductive. Those loads worked fine, more or less. We did have issues with them getting more reactive as they got hotter, and mis-tuning the amplifier. So, this amplifier may be more susceptible to reactive loads than other kinds without a tank circuit. Something like an L match or matching circuit should be all that's necessary to negate this issue, and we've used a manual tuner with no issue.