Class E PAs - FET Evaluations at 40 Meters

[Tom WA3KLR]

(Forgive the long introduction, but is good information I think.)

I have kept my eye on the n-channel power MOSFETs the last few years and singled out ones that look more promising than others.  Attention has been given to find ones that appear to be easier to drive for the same drain characteristics.  I picked FETs that have an Rds(on) in the region of 0.2 to 0.4 Ohms for a design that would use 3 - 4 FETs in parallel in a single-ended PA.  Although the lower the Rds(on), the better the efficiency in one respect; going to FETs lower than 0.2 Ohms means a large device that probably will be difficult to drive.

It is hard for me to guess the best inexpensive switch-mode power supply FETs for the Class E transmitters on the low bands.  Comparing data sheets is sometimes like comparing apples to oranges since the series gate drive resistor chosen by the manufacturer for the device testing is not the same value.  I don’t claim to know of every FET that is out there.  My selections are not the authoritative best of the best, but are my candidates from what research I managed to do. 

I have evaluated 10 power FETs in Class E operation at 7.3725 MHz. CW at up to 186 Watts output.  (More output can be obtained with bigger power supply.)  The near goal is not to build a 40 meter deck now, but to be able to see differences in the FETs.  My intuition tells me that anything will work o.k. at 160 meters and with some differentiation at 75 meters.  I made a single-ended breadboard that accepts one FET.  The PA designed for the testing was based on my triple-output variable lab supply which will do 62 Volts maximum at 3.2 Amperes maximum.  This implies a load impedance of 19.3 Ohms to squeeze the most power out of the supply in the test.  This equates to a Class E PA Design Resistance of about 9.6 Ohms and 198 Watts maximum d.c. input power.  One rationalization here is that due to the range of FET Rds(on)’s being 0.15 to 0.96 Ohms, the 9.6 Ohm PA resistance will minimize the variation of FET Rds(on) as a factor in differences in output performance.  Obviously a lower Rds(on) is less loss, but I am concerned about turn-on/turn-off time losses.  The PA designs in actual transmitters on the air now, I believe, center on 3 to 4 Ohms design R.

The circuit design Q was 5.  The series tank inductor I wound turned out to be about 18 % higher inductance than intended, but this just means that the circuit Q is about 6 instead of 5.  The inductor was wound with about 4 feet of #6 AWG solid copper wire (89 cents per foot at Lowe’s these days).

The breadboard is a large sheet of 0.063” aluminum which is from a rack cabinet top cover.  I started with an air variable shunt capacitor, but the series inductance proved to put much ringing on the drain waveform.  I tested with 2 small pieces of solid copper printed circuit board instead for the shunt capacitor, which worked out very well (photos attached).

The load match capacitor and series tank capacitor were preset to the calculated values. At first, good Class E operation was not achieved but appeared to be in the ballpark.  I had not built any Class E PA’s before.  The tuning was not near the theoretical values.  One problem was the shunt capacitor inductance mentioned above which was cured by the use of printed circuit board as low-inductance capacitor.  Optimum tuning of the series tank capacitor was resulting in a much higher value of capacitance than calculated and the drain waveform never showed the waveform for a tank capacitor going to higher capacitance than the optimum point. 

I knew that the FETs, by the data sheet information, switch on faster than they switch off, which is unfortunate for several reasons for class E.  (For one, this causes the drain duty cycle to be biased to higher than 50/50 conduction with a 50/50 drive waveform, which is bad for efficiency.  Second, in class E operation, when the FET first conducts, the current is zero and ramps up through this half of the r.f. cycle – a non-critical turn-on condition.  However when the FET cuts off, it is conducting current at about 80 % of the peak value reached a little earlier.  Ideally it should cut off very fast.  It would be nice for Class E PA’s that the FETs cut off faster than they turn on!)  As suspected, simulations of a class E PA with an ideal switch running more than a 50/50 duty cycle exhibited the drain waveform characteristics I was seeing.  Close examination of the output of the exciter i.c. – a mini-dip i.c. with a build-in crystal oscillator and divider was providing a 52/48 duty cycle.  The FET driver i.c. – a DEIC420 did an admiral job of maintaining an unchanged duty cycle input-to-output.  The drain waveform did appear to show a shortened off time.  The simulation of ideal operation at 7.3725 MHz shows that the width of the off-time drain voltage swing should be about 39 - 40 nanoseconds measured relative to ˝ of the peak voltage.  I made a circuit modification between the mini-dip output and the DEIC420 driver input with a small pot, diode and 62 pf. capacitor which allows me to vary the duty cycle down to very low values.  To get the initial FETs tested to run approximately 50/50 operation, I had to provide a 38 % duty cycle to the DEIC420 input.  This setting was used for most of the testing.

Once I felt that the circuit was performing well enough, I tried to use the same setting for all FETs.  The tank and load air variable capacitors had graduated skirt knobs so that the settings could be recorded and repeated if necessary.  Also the capacitors positions versus capacitance were characterized earlier so that a setting had a known capacitance that could be read from an Excel graph.

 The wattmeter used was a Bird 43 with a 1000 Watts HF slug.  The r.f. voltage across the power attenuator input following the wattmeter was measured with a 100 MHz. d.c. coupled scope and then the peak to peak voltage re-created with a high voltage supply and read with a digital voltmeter.  This was my means of coming up with a correction factor for the wattmeter in the range of the power output being measured.  Digital multimeters were used for the driver and final drain current, and final drain supply voltage.

The driver i.c. current is a measure of the work required to drive the FET gate.  If a load capacitance has series inductance associated with it, this causes the load to be greater.  One FET gate loaded the driver enough to cause the waveform to be somewhat sinusoidal along with nasty ringing.

Rds(on) was estimated very crudely by eyeballing the FETs drain voltage at approximately 70 % through the ON period.  Here the drain current should be at a peak, 9 Amperes assumed from a simulation.  Ringing present on this low voltage added to the difficulty of making an accurate reading.  The scope probe was never used with a ground lead.  The ground collar at the tip was directly touching the ground sheet or a ground stud.


Datas:

In no particular order:

                                      Specifications                                          
                                                      Typ.                             Est.
                                                      Rds RF **P.S.**   Driver  Rds
FET type    Mfr.     Pkg.        BVdss Id (on) out  Vdd   Id   current (on)  Eff.   Notes

IRFP450B Fairchild TO-3P       500   14  0.39 146 62.7 2.64 1.10   0.5  88 %     
IRF740 Int’l Rect. TO-220AB   400   10 0.55 138  62.7 2.73 0.58  1.3  81 %
IXFH28N50F IXYS TO-247AD   500   28 0.19 138 62.7 2.39 1.12   0.3  92 %      4.
                                        custom-tuned 186 62.6 3.12   “       “    95 %    1., 4.
FCA20N60 Fairchild TO-3P      600    20 0.15 144 62.7 2.63 1.02  0.3  87 %
                                         Custom-tuned 173 62.6 3.08 1.00   “    89.5 %     1.
FCP11N60F Fairchild TO-220   600   11 0.32 150  62.7 2.67 0.70 0.45  89 %
FDH15N50 Fairchild TO-247AD 500   15 0.38 153  62.7 2.68 0.60 0.67  91 %
FQA11N90 Fairchild TO-3P      900 11.4 0.96 115 62.7 2.58 0.92  1.3   71 %         2.
                                          Custom-tuned 94 62.7 1.89    “      “   79.5 %     1., 2.
                                              "         "  sample #2                      77 %       1., 2.
FQA18N50V2 Fairchild TO-3P   500   20 0.27 150 62.7 2.63 0.68   0.39   91 %
FQP18N50V2 Fairchild TO-220  500   18 0.27 150 62.7 2.65 0.68   0.4     90 %
STW20NK50Z ST TO-247        500   17 0.23 145 62.8 2.61 1.07   0.4     88 %
                                         Custom-tuned 173 62.7 3.01 1.04  0.4     91.5 %  1., 3.

Notes
1.   If I felt a FET was different and would benefit, I re-tuned some.
2.   FET ran very hot.
3.   FET appeared faster than the others, adjusted drive duty cycle to 45/55.
4.   FET overloaded the FET driver i.c. output.

Conclusions:

The lowest efficiency was with the Fairchild FQA11N90.  Two FETs were tried here.

The highest efficiency was obtained with the IXYS IXFH28N50F.  This is a low Rds(on) FET with metal gate.  It overloaded the driver i.c. and drew the most current from the driver.  Great drain performance, but at a cost of the most driver power.  $11 in 2004 when I bought them.

The second highest in efficiency is the STW20NK50Z.  Again not easy to drive.  I think that this FET was not picked by me but was a substitute recommended by the Mouser website.  I had delegated it earlier to PWM use and it almost didn’t make it in the evaluation, but I included it and it did good.

                     Efficiency  Drive power
IRFP450B             8          8   “tough drive”
IRF740               9          1  “easy to drive”
IXFH28N50F        1          9   “tough drive”
FCA20N60           6          6
FCP11N60F         7          4  “easy to drive”
FDH15N50           3          2  “easy to drive”
FQA11N90          10         5
FQA18N50V2      4          3  “easy to drive”
FQP18N50V2      5          3  “easy to drive”
STW20NK50Z     2          7

I was surprised how many actually did work at 40 meters.  The waveforms look good by eye, but the efficiency numbers tell the story.  I would say that in efficiency, # 1 – 7 can be used at 40 meters, if you have them on hand.

The FDH15N50 seems to be the overall choice for the one that is the most efficient AND easy to drive.  Close behind this is the FQA/FQP18N50V2 which always looked promising from when I first saw the data sheets 3 years ago.  FCP11N60F is a good compromise device here also.

A spectrum analyzer was used sometimes to monitor the r.f. output.  Generally, the second harmonic was down about 29 – 31 dB., the third harmonic down about 49 – 52 dB., and the fourth harmonic down about 62 dB. or greater, which was the grass level.   Occasionally the 15th and/or 16th harmonic popped up above the noise floor (110.6 and 118 MHz.) at about -57 dB.   No other spurs were seen in the HF or VHF region.  I looked up to 200 MHz.  (A photo is attached of this.)

No FETs were damaged.  A second FQA11N90 was tested to confirm low efficiency results there.

Peak drain voltages range from +180 to +216 Volts d.c.  In general it was very close to the simulation.  The ideal case theory is that a perfectly tuned final drain will swing to 3.56 times the applied voltage. (62 V. x 3.56 = 221 V.)

I hope this experiment is a help to the science and implementation of Class E power amplifiers.

Photos below:
1. breadboard top view
2. breadboard side view
3. FET-under-test close-up

[Tom WA3KLR]

Two more photos:

4. Scope view of FCP11N60F drain voltage - 50 V/division vertically and 20 nanoseconds/division horizontally.

5. Spectrum analyzer on FCP11N60F r.f. output, 1 to 200 MHz. 
     10 dB. per division vertically.

[nu2b]

Thanks Tom,
Nice work and a very solid presentation.
I wonder if you ever tried custom tuning the FDH15N50?
Regards,
BobbyT

[Tom WA3KLR]

Hi Bobby,

Thanks for the nice comments.  I have spent a week of elapsed time working on the evaluation and will take a break but I am interested in doing a little more lab work.

- - -

The circuit details:

The oscillator i.c. is an ECS 14.7456 MHz. unit, 8-pin mini-dip, Mouser p/n 520-DC01474.  I have a selection of these at various oscillator frequencies.  Has buffered output at oscillator frequency, and divided output selectable for divide by 2,4,8,16,32,64, or 128.  Runs off of + 5 V.  Handy devices.

The driver i.c. is a DEIC420 CMOS i.c. obtained through IXYS.  I bought it 3 years ago at a cost of $30, ouch.  I first ran the part at 8.25 Volts.  At this voltage the current drain was 378 mA. with no load.  The part drew over 2 1/2 Amps driving a 0.01 uf. film cap. with leads.

During the PA testing it seemed that the FETs performed a little better with more gate swing.  I increased the supply voltage to 10.3 V.  It is not worthwhile going higher than this to drive the FETs as the driver i.c. just draws more power and the gate's temperature rises more?  Nothing else is gained.

The shunt capacitance is from 2 pieces of p.c.b. totaling 427 pf.  With the capacitance around the FET itself, the total is 450 pf. for a mounted TO-220 and 470 pf. for a mounted TO-3/247.

The tank capacitance was set to 560 - 684 pf.  This was total of the air variable and a 0.090" double-sided p.c.b. piece equal to 190 pf.

The tank inductor is 1.7 uH. #6 AWG solid copper wire, about 2" i.d., 5 1/2 turns spaced out.

The loading capacitor was set in the range of 1125 - 1240 pf.

A 62 uH choke was used for the feed choke.

The spreadsheet calculated values were:
466 pf. for the shunt capacitor.
470 pf. for the tank capacitor.
1.67 uH for the tank inductor.
884 pf. for the loading/ 50 Ohm matching capacitor.
The feed choke was 62 uH.

The FET contributes a parametric and net effective shunt capacitance also.  I tried the shunt capacitance of just one pcb (233 pf.) and the efficiency and harmonic output were somewhat worse.

The main difference is that the tank capacitance and loading capacitance is larger than predicted.  It then appears that the circuit is operating at an r.f impedance that is lower than expected.  My guess is that this is all due to the slow turn-on and turn-off of the FET versus ideal switch operation which the Sokal equations must be based on.

[steve_qix]

Hi Tom,

Nice experiment !!  Very nice !!!!! I've found - for a given power level - the lower the R D-S On, in general, the better the efficiency - for any given voltage/current combination.  Your results definitely confirm this in a farily concrete manner. 

The IXYS MOSFETs DO perform VERY well. I have a transmitter with a bunch of IXFH12N100 (1000 volt, 12 Amp devices), and it is a very good transmitter - 6 years old and never one problem from the day I built it.  2 bands, 160 and 75 meters.

I find the the higher voltage MOSFETs like the FQA11N90 and the IXFH12N100, which are both relatively high voltage, not too low R D-S on devices (about 1 ohm, which is pretty high) will not perform well at high current.  My personal recommendation for these devices is about 1.2 amperes maximum, at-carrier current, per device.  With higher currents, the efficiency will deteriorate.  These 2 MOSFETs are generally good for 50 to 60 watts carrier output per device.  They work well in high voltage (40 to 55 volts), lower current applications.

Another one to look at is the FDH45N50 or the FDH50N50.  These are what I like to call "high yield" devices.  They will yield high output for a single device, without having to resort to exotic circuitry.  This MOSFET will deliver between 100 and 150 watts output per device.  These work well in medium voltage applications (25 volts at carrier @ around 5.5 amperes).

In my work over the years with class E, I have found that all MOSFETs (unless they are VERY slow with bad gate structures) will yield very good efficiency (in the vicinity of 90%) if operated such that the R D-S On  is low as compared to the load resistance.  In all cases that I've ever tried, the higher the current, the lower the efficiency, once a nominal power output level is reached.

Very respectable piece of work!  ;-)

Regards,

Steve

[WA1GFZ]

I agree with Steve's numbers on the 11N90 at an amp or so at carrier. Above that current will work but efficiency goes down
I use IRF840s in the 160 meter rig and monitor package temperature of 1 FET.
I run them at .5 amps steady state and the package hits about 100 degrees F after an old buzzard. During the summer about 10 to 15 degrees hotter. They are also 1 ohm class FETS but only rated for 8 amps in a TO220 package rated for 75 watts dissipation..
The lower RDS on FETs look better as the drivers get better

[Tom WA3KLR]

Bobby,

I did some more work with the 15N50.  I obtained a bunch of old 33 and 47 Ohm 1 – 2? Watt carbon composition resistors and used them all in parallel as a drain pull-up on the 15N50 for setting the drive duty cycle for 50/50 drain output.  The drive duty cycle wound up to be 42 %.  Then, I was able to get 179 Watts out with efficiencies in the range of 93 – 94 % with capacitor tweaking.  Probably most efficiencies reported by me can be raised by more fiddling.  This would put many in the same efficiency range.  This gives the 15N50 a best of both worlds rating.  I bought them about 2 ˝ years ago from Mouser.  Yesterday their site reported not available.

Steve,

Yes, to be fair to the 11N90, it is the highest impedance device of the 10 FETs.  The IXFH28N50F which was #1 in efficiency has the highest current rating and one of the lowest Rds(on)s of the group.

My interest in doing the testing was to see what FETs were fast enough at 40 meters and relative drive requirements.  This tables repeats some of my ranking data
(and let’s try an (Id/Rds(on)) quotient figure-of-merit and see what happens):

Efficiency  Id rating     Rds(on)  Drive Ease        Id/Rds(on)
1                  28 A.        0.19        9                  147
2                  17            0.23        7                   74
3                  15     *     0.38        2   0.604 A.     39
4                  20            0.27        3                   74
5                  18            0.27        3                   67
6                  20            0.15        6                 133
7                  11            0.32        4                  34
8                  14     *     0.39        8  1.104 A.     36
9                  10            0.55        1                  18
10                11.4          0.96        5                  12

One interesting combination that jumps out at me is a pair of FETs with almost identical current and Rds(on) ratings; #3 and #8 in the efficiency rating.  #3 was 2 in the drive ranking and #8 was 8 in drive; same impedance but different in drive.  This is exactly the type of thing I was looking to identify.

#3 is the FDH15N50 and #8 is the IRFP450B. The FDH15N50 benefits from being several generations newer power FET design which is focusing on drive efficiency.

The efficiency results roughly follow the Id/Rds quotient!  The 133 number is out of place though.  This is the FCA20N60.  Since it does not follow efficiency by beefiness, perhaps it's out of place efficiency is due to being the slow one of the group.

Let’s try a table ranked by just this new quotient and compare with the Drive Ranking and see what happens:

Id/Rds(on)  Drive Ease
147              9                IXFH28N50F
133              6
74                7
74                3                FQA18N50V2
67                3                FQP18N50V2
39                2                FDH15N50
36                8
34                4
18                1
12                5                FQA11N90

The Id/Rds(on) is a measure of impedance beefiness.  One expects the drive requirements to correlate to this ranking.  A nice FET would be one that has a lower drive number in the series as you read down in the DriveEase column.  What looks attractive here are the FQx18N50V2 FETs and the FDH15N50 which we already identified as being attractive.

Does this numerology seem meaningful? 

Well, enough for now.  I’m going to go vote for the next American Idol.

[Steve - WB3HUZ]

You are approaching this like a good engineer Tom. Rather than just getting caught up with one spec, like efficiency (not that important in amateur radio service, most especially past some point like 90%), you are looking at other factors that may serve as useful trade-offs in the big picture (drive, 40 meter capability, etc). Thanks for posting the numbers with various factors considered. Makes it easy to understand, even for a dummy like me. Excellent work!

[Tom WA3KLR]

Thanks Steve.  Hopefully we will work each other with Class E rigs sometime in the next year or so.

[Tom WA3KLR]

More on the FET driver i.c.:

I characterized the CMOS DEIC420 driver i.c. before running the FETs.  I used 2000 pf. and 5100 pf. silver mica capacitors for the loads.  I bent the capacitor leads to make connection to the i.c. output and ground plane with minimal lead length.  Then I used the full lead length of the capacitors.  The Excel graph of the resulting curves is posted below.  How much the i.c. draws depends on the amount of capacitance and the series inductance of that capacitor.

I will assume that the FET gate is an impedance that is half-way between both curves.  From the curves and driver currents recorded, it appears that the FET gates effective capacitance range from about 5500 to 8500 pf.  This is a reactance of 3.9 Ohms down to 2.5 Ohms.

With an 8500 pf. capacitor, if there is a total of 55 nH. of lead inductance, the load will be series resonant at the driving frequency of 7.3725 Mhz., which greatly reduces the load impedance and increases the i.c. current.  If the lead inductance is even half of this, just 28 nH. which is a small amount, the load reactance is cut in half.  So with the square wave voltage driver i.c., minimizing the gate circuit lead inductance is very important.

Driver i.c. curve graph below:

[AB1GX]

I was wondering if anyone has tried to put an inductor in parallel with the gate capacitance so the input circuit resonates.  The inductor can be biased up to +5V to match the FETs on voltage and swing between +20V and -10V peak to peak.  That might really reduce the drive requirements.

Tom, AX1GX

[WA1GFZ]

Tom,
I put two turns spread #14 1/2 inch I.D. phase to phase on my 75 meter final to dial in the VSWR. It runs a bit warm.

[KF1Z]

I was wondering if anyone has tried to put an inductor in parallel with the gate capacitance so the input circuit resonates.  The inductor can be biased up to +5V to match the FETs on voltage and swing between +20V and -10V peak to peak.  That might really reduce the drive requirements.

Tom, AX1GX

Tom,
You may find this interesting...
http://classe.monkeypuppet.com/viewtopic.php?t=510

[WA1GFZ]

Tom,
Just for the heck of it I looked at the Motorola MRF 15X family and find the input to reverse transfer C ranging from .042 (MRF157) to .143 (MRF150)  Crss/Ciss gfz

[Tom WA3KLR]

Drive science achieved?

Well, after some more thinking I hope I have come with the final analysis.  In the last installment I came up with a drive ease ranking based on relative drive power data.  I also published a graph of the characterization of current drawn by the CMOS driver i.c. at 8.25 Volts.  This FET testing was completed at driver supply voltage of 10.32 V. however.  This morning I re-checked the current the driver draws at 10.32 Volts no load.  This allows me to come up with a Drive Power number; Pdvr = 10.32 x (driver current – 0.46 A.) 

Having this “absolute” number allows me to now come up with a “Drive Efficiency” number using the “FET beefiness” quotient (25 C. continuous Id rating/Rds(on)). 
The Drive Efficiency is (FET beefiness/drive power).

Just comparing the drive power I measured on the lab bench to the data sheet’s Qg (total gate charge) shows a very good correlation, no surprise.  The data sheet value for the Qg of the IRF740 seems off however.

FET type         Drive Power (W.)   Qg (nC.)
IXFH28N50F       6.81                    95
IRFP450B           6.65                    87
STW20NK50Z     6.32                     85
FCA20N60          5.73                    75
FQA11N90          4.75                    72
FCP11N60F         2.46                    40
FQA18N50V2       2.27                    42
FQP18N50V2       2.24                     42
FDH15N50          1.49                     33 
IRF740              1.27                     63 ?

Now for the ranking via the new Drive Efficiency number (the higher the better):
FET type       FET Beefiness  Drive Power   Drive Efficiency

FQA18N50V2      75                 2.27             33.2
FQP18N50V2      68                 2.24             30.3     
FDH15N50         39                  1.49             26.6     
FCA20N60         133                 5.73            23.3
IXFH28N50F      147                 6.81             21.6
IRF740              18                  1.27             14.3
FCP11N60F        34                  2.46             14.0                   
STW20NK50Z     74                  6.32             11.7     
IRFP450B          36                  6.65               5.4
FQA11N90         12                  4.75               2.5     
                   
So it appears that a FET can be sized up for drive efficiency by looking at the data sheet and pulling out these 3 numbers:
Id @ 25 C. continuous
Rds(on)
Qg

The rule of thumb formula for Drive Efficiency (based on the 7.37 MHz. drive data) becomes:
                                                        Id
Drive Efficiency  =   ------------------------------------
                                  Rds(on) x ((0.09 x Qg ) – 1.4)

Comparing the lab data results against the data sheet rule of thumb for Drive Efficiency numbers:

FET type     Lab Drive Efficiency    Data Sheet-ROT formula

FQA18N50V2        33.2                  31.7
FQP18N50V2        30.3                  28.5
FDH15N50           26.6                   25.1
FCA20N60           23.3                   24.9
IXFH28N50F         21.6                  20.6
IRF740                14.3                   4.6 * Flyer, mfr. error? Data ignored for formula.
FCP11N60F           14.0                  15.6
STW20NK50Z        11.7                  11.8
IRFP450B               5.4                    5.6
FQA11N90              2.5                    2.3

Now you can find the next great FET-du-jour.

[WA1GFZ]

Tom,
Interesting you find lower voltage FETs easier to work with. The only problem is your forced to run lower drain voltages to get the same safety factor as the 11n90.
When you run lower voltages you need to run higher currents and lower operating Z. I would not run more than 85 volts peak on a 500 volt FET.
I would rate the IXFH28N50F the highest single part input at carrier 30 volts 3 amps
or 90 watts (just looking at the numbersSWAG) The rest of them even or lower power than the 11N90 at 50 watts.
my rating system is about 30 volts carrier on a 500 volt part running about 1/10 current rating at carrier.  I find you can push the current higher but the RDS on
eats into efficiency.  I'm sure you could run higher operating voltages as long as the load never bad.
The last time the 160 meter rig blew a FET was when a Tree took down the antenna and I keyed the rig about 10 times without glancing over at the reflected power. My modulator modulator was in current limit but Drain voltage was high meaning some high Z was reflected back at the rig. Normally I see a bit over 300 volts on the drains during modulation peaks IRF840s. I consider this safety factor marginal. gfz 

[AB1GX]

Yes, I agree that lower voltage requires lower Z, and I think that's the way to go.  The inductor gets small (.35uH).  Efficiency isn't an issue with FETs that have Rsd of under .04 Ohm @ 50A. 

I think 200V peak on the drain is doable at 1.5KW into 50 Ohm.    That's 50V @ 35A input.  300V FETs might be the way to go.

[WA1GFZ]

You are forgetting the peak drain voltage class e is about 3.5 times drain DC input.
At 130 volts that is about 450 volts peak. A 900 volt FET provides 100% derating at that voltage so if your antenna dies making VSWR crazy.  Sure you can use a lower voltage FETs but at higher risk of failure.
Lower Z allows smaler inductors but you will need bigger caps that can handle high circulating current.
I had no problem coning up with isolation transformers to build my 75 meter final but I had a heck of a time coming up with lower voltages at high current. gfz

[AB1GX]

Low voltage and high current is key as you say.  My inductor is a sheet of copper rolled into a cylinder that's ~7" in dia and 8" long.  The ends bend down to form the conductors (8" x 2 1/2") of the homebrew kapton caps.  The output of this pi network then goes into a conventional series LC tank (1/4" copper tubing at 2 1/2" dia with a couple of hundred pf air variable) and then to a 50 Ohm load.

Low Z is hard, but not totally impossible. (And yeah, I have a pile of FETs I've nuked)  As you know the FET killer is screwing up the gate drive!

During testing, I'm only using 10V @ 10A DC input with very low Rsd FETs (only 60V so they're cheap) and testing with/without 50 Ohm load to check the voltage rise on the drain.  What's interesting is with no load the drain voltage doesn't rise that much, the circulation current remains huge, and the input current falls to almost nothing.

But 1.5KW with this low Z configuration may be a different story...

[Tom WA3KLR]

Some notes on drive at 80 and 160 meters relative to my findings at 40 meters:

I have tested driving the IXFH28N50F at 80 and 160 meters.  This was the FET with the heaviest gate load.  The lab measurements agree with what the theory says that drive power will drop in proportion to frequency.  Since you do have the Miller effect – feedback to the gate, the drain should be energized so that you have a swinging load for the test.  This does cause the drive power to increase.  It’s interesting that I was able to run the FETs at 80 and 160 meters drive with the test bed tank circuit still set to 40 meters.

IXFH28N50F results – 
                    Drive power used
@  40 meters     6.81 Watts
@  80 meters     3.08 Watts
@ 160 meters    1.71 Watts

Although I felt that the DEIC420 was not quite driving the FET well at 40 meters, my guess is that you should be able to drive at least 3 of the IXFH28N50F at 160 meters, no sweat.

I’ll be interested in evaluating one of the new IXYS ‘515 driver i.c.s.

[Tom WA3KLR]

Modulation Linearity of the 40 meter Class E Test PA:

Just a side piece of data here.  I took measurements of the Class E PA r.f. output power versus the drain supply voltage.  I used a HP8901A Modulation Analyzer for reading the power level, which has a digital readout.

The FET installed at the time was the IXYS IXFH28N50F.  The modulation linearity may be able to be improved with tuning/linearity optimization.  This plot was based on what the circuit tuning just happened to be.  The PA was running about 170 Watts out CW at full voltage, 62.5 Volts d.c. 

The nice thing is that the power is not starting to droop at all.  In fact, with the slight non-linearity that there is, the output power is rising a bit as maximum power is approached.

Excel graph .pdf file below:

[WA1GFZ]

Tom,
I think you could run a bit higher voltage since carrier is under 50 watts. Also I bet you could run about 2.5 amps at carrier.
looks good though. fc

[steve_qix]

This is a very interesting discussion.

I have found, over a number of years doing this sort of thing that the higher the impedance (and therefore, the higher the voltage), the better. 

Even though the high voltage MOSFETs (such as the IXFH12N100 and the FQA11N90) have larger die structures, and therefore require more drive than lower voltage MOSFETs, the overall results with high impedance are better.  I, and many others, regularly achieve drain efficiencies of over 90% with high voltage MOSFETs, so that is not really the issue.  All of the MOSFETs are capable of this level of efficiency if they are operated correctly.

For me, the important consideration then becomes cost and complexity of construction.

As the impedance drops (lower voltage, higher current), the requirements for the shunt capacitor increase significantly.  Because more current is involved, the shunt capacitor has to be a very good capacitor with extremely low internal resistance (inductance).  And, the capacitor will have to be of a higher value, as the value of the shunt capacitor is inversely proportional to the impedance.  This pushes the cost of the shunt capacitor up dramatically.

The 2nd problem of a lower impedance RF amplifier involves the layout and construction.  As the impedance drops, lengths of leads, thickness of material, wire size and layout all become much more important.

And lastly, there is the effect dropping the impedance has on modulator, power supply and interconnects.   Let's say you are building an "AM Killowatt" transmitter.  At 45 volts, that's around 24 amperes total current (using 900 volt or better MOSFETs) - an impedance of around 1.9 ohms.  This is about as low as one would want to go with a single RF amplifier/modulator combination.  If one were to operate the same power level at 22.5 volts (using 450 volt or better MOSFETs and maintaining the same safety factor as above), the current would then be around 45 amperes, and the impedance would be .5 ohms.  I don't want to think about going lower than this !!

It is much easier to build a modulator to work into 1.9 ohms than it is for .5 ohms !!!  Consider the peak current can rise to 3 times the average (200% positive modulation) - at .5 ohms (at a KW), the peak current will be 135 amperes !!!!!! 

Anyway - that's been my experience with all of this.  Keeping the impedance as high as possible reduces the construction cost and complexity as you get into high power.  Sure, I designed a class E RF amp that ran off of 5 volts (for a piece of medical equipment), but the total power output was only a few watts 

Good stuff !

Regards,

Steve

[WA1GFZ]

I agree Many guys building this stuff have problems blowing FETs due to layout problems. You drop the Z by a factor of 4 ane it gets much harder. Shunt caps and output tank caps become a big problem.
Also I have found running the output transformer higher then 1:2 hurts efficiency a bit due to leakage inductance of the secondary winding.
Almost any FET will work but the nice idea of 130 volts peak is isolation transformers are easy to come by. gfz

[AB1GX]

Yes, the switching circuit must be very low Z to use low-voltage, low-resistance FETs.  Smaller copper inductors and larger copper caps are required, but a transformer is not needed.  Operation below 50V also makes the switching modulator much easier to design.

I just prefer 250V .01 Ohm 100A FETs over 900V 1 Ohm FETs.  Lower drive is another advantage.