Class E Notes

the result of experiments with high efficiency class E amplifiers

Posted By: Robert Nickels (ranickels)

There is a lot of misunderstanding about how a Class E amplifier works.     As the result of studying the literature and experimenting, I thought I'd share what I have learned over the past several years.   

Below is an example of a test amplifier I used to optimize my 2 watt wspr transmitter boards.   It can be visualized as two circuits - the amplifier itself and an impedance matching section.    The amplifier consists of the swtich (NMOS FET), a parallel capacitor, and a series tuned circut.    The shunt (parallel) capacitor adds to the Coss capacitance of the FET as part of the tuned circuit, but also acts to limit the voltage across the FET when it is turned off, and to widen the off-time waveform.   The value is not super-critical but it must be capable of withstanding at least 3.5X the drain voltage and should be mounted as close as physically possible to the FET (ideally right across the leads) because inductance here will add harmonics to the output.   The fundamental issue with all class E designs is that while NMOS FETs make nearly ideal switches, the gate structure is basically a capacitor, and the higher the power capability, the larger the gate capacitance becomes, and as the nice square edges of the drive signal are rounded off, the less efficient the switch becomes.     Fortunately this issue is manageable with the devices used at QRP power levels where digital logic ICs (especially the 74AC family) can do the job.   There are many low-side FET driver ICs on the market that work great for power supplies, but sadly most of them are not fast enough for RF.

Contrary to intuition, the series network is not made resonant at the operating frequency, and is slightly reactive (resonates lower in frequency).   This is because the goal when optimizing a class E amplifier is to minimize the time when the voltage and current across the switch (FET) are both high.  Graphically, an ideal class E drain waveform has several key characteristics that are described in the inventor's paper, but which look like the attached "Ideal waveform" photo.    In my experience (and verified by more seasoned class E designers), formulas and design aids get you in the ballpark, but simulation (I use ltspice) and actual lab testing are required to arrive at an efficient optimized design where efficiency even greater than 90% are achieved.

As proof of this, the experimenter will soon discover that the answers provided by various online and downloadable class E design aids do not always agree.  However I will list several here with the usual advice that "YMMV":

VK1SV online calculator

WAØITP downloadable class E spreadsheet

Tonne Software downloadable ClassE design aid

VK2ZAY online tool

NU2B downloadable design tools (DOS and xls)

And for those who wish to go straight to the source, the design equations can be found in the inventor, Nathan Sokal's article "Class E Power Amplifiers" in QEX for Jan-Feb 2001.     Further information can be found in research papers and in US Patent No. 3,919,656 that he and his son obtained in 1975.     This article is copyright by ARRL so is not included here but may be found online  LINK

The WAØITP spreadsheet has always produced "good ballpark" efficiency for me by using nearest standard values to those specified by the spreadsheet and is my recommended best choice of design tools.    This tool calculates the values fo Dave Cripe's design which has been proven in many applications.

However you get there, the bottom line is:  if you don't have this waveform, you don't have class E.    The design process involves tweaking and adjusting to optimize the drain waveform, because that is the heart of the efficiency of  the class E design.   Ohms Law tells us, P=E*I, so minimizing E and I means the amount of power dissipated in the transistor is minimized.   A typical optimized class E amplifier will be able to convert at least 80% of input power into RF, i.e.. 80% efficiency.   That is compared with 50-60% for a class C amplifier, and as high as 90% is possible.

The problem is, the output impedance of most class E amplifiers is not 50 ohms, and because of the square wave switching, harmonics are present - the 2nd harmonic is usually the strongest by far.    Thus an impedance matching stage is required, and a  harmonic filter.    The circuit below accomplished both with the fewest parts:

On the left hand side of the red vertical line, the MOSFET is switched on/off by a squarewave at the desired frequency, so it is on (saturated) half the time and off the other half.   Visualize what happens if you switch a pulse into a series resonant circuit (band pass) - you get a big increase of current because the impedance is low, but when the switch goes off,  you'd get nothing.   But the resonant tuned circut keeps current flowing into the load and thru the shunt capacitor when the FET is off, which creates a continuous RF output.   Essentially the FET "pumps" a series of pulses at the RF frequency into the load in the same way that a switching power supply does to generate DC.    The mark of a proper working class E PA is the characteristic waveform at the drain which results in highest efficiency, i.e. that switching is being done when the voltage across the FET is as close to zero as possible.

To the right of the vertical line, two things must happen because the impedance is much lower than 50 ohms, and harmonics are present (especially the 2nd) as a result of the digital switching.     First,  a pi-net is used to transform the impedance from typically 10 ohms or so for QRP power levels up to 50 ohms.  Transformers are also used but the pi-net is simple and cheap and only takes three parts.   It also provides a free bonus because by adding one more capacitor across the inductor, we can tune it to the second harmonic where it will (typically provide 45 to 50dB of second harmonic attenuation.     And we're done!

This circuit was described by Nathan Sokal in 2001 and have been used by many designers since, up to and including large broadcast transmitters and the beauties built by WA1QIX (check out his website:  classeradio.com to see how this concept is scalable)       As you can see there are few components but they are specific to a band, so some consideration must be given to how a class E transmitter is made to cover multiple bands.   It's really no different than a linear PA which requires a bank of low-pass filters that must be switched in/out for each band.   The cost and space required by the output filter is a sizeable part of the cost of any radio, whether SDR or conventional, and switching the filters for a class E rig would be done the same way, however the losses associated with switching and instability resulting from added inductance and capacitance in the switching elements are the main reasons that class E is best suited for single-band operation, or where the entire frequency-dependent PA section is swapped when the operating frequency is  changed.   The ill-fated SGC "Mini-Lini" was a class E amplifier that was not successful but which used this concept (photo below).

One final thing about the power supply - the RF choke (L1) is non critical, it just needs to pass full DC to the MOSFETs AND have sufficient impedance to keep the power supply from loading down the RF signal.   The recommended value is 10X the value of L2, and most design guides recommend a value that is 5-15X the inductance of the series inductor.

Noted class E designer David Cripe NMØS (note the appropriateness of his callsign!) has created numerous class E rigs for the 4 State QRP group and often gives presentations about class E which can be found online and are highly recommended.   Full information can be found at http://www.4sqrp.com/index.php but for convenience I will post the schematics of two below.    The Bayou Jumper is a retired kit that featured a class E PA and impedance matcher as shown above with a single IRF-510 FET.   The Cricket is a tiny transceiver that also uses the FET as a mixer and a single 2N7000 FET, but as can be seen, the same circuitry.   Another good example of the same class E PA with the addition of a bootstrapped class G AM modulator can be found in Dave's latest design, the Nouveau 75 AM transceiver that I discuss here.

My own design of a 2 watt class E transmitter that is driven by the programmable clock output of a Raspberry Pi uses a single BS-170 at 2 watts output.   A separate optimized class E board is used for each band with provisions to select from up to 8 bands  using GPIO signals for automated multi-band operation (the stacking configuration has seen been redesigned due to the poor quality of stackable pin header connectors).   With modest omnidirectional antennas on 80, 40, and 30 meters this wspr station is routinely spotted by stations all over the world.    Class E is ideal for digital modes.

In sum, class E is a highly efficient amplifier design, both in terms of power and cost, and unquestionably give the most "RF bang for the buck".  That's why Dave's class E amplifier was chosen as a co-winner for the "Lowest Cost Entry" in the ARRL Homebrew Challenge (Oct. 2010 QST).  But it is inherently a tuned device and requires care in component selection, PCB layout, and in tuning/optimization as the power level is increased.   Most of these concerns are easily managed at QRP power levels, and that's why class E is ideal for experimenters and QRP/minimalist fans.

Update:

After receiving some new PC boards for my implementation of the uSDX micro-SDR I did a quick test using the textbook class E series-resonant circuit above and components selected using the WA0ITP spreadsheet.   Excitation was provided by an Si5351-based VFO with a single 74AC04 inverter as a gate driver.  Here are the test results for the 20 meter band:

Input power:   6.2 watts

Output power:   5.5 watts    (47 volts P-P across a 50 ohm load)

Efficiency = 5.5/6/2 = 89%