This is an update to an earlier blog post describing a moderate power 2200m class E amplifier with very low drive requirement. The design had been evolving for some time but is now in a finished state as far as I am concerned. In its current configuration I have many hours on this little amplifier running 250 to 275 watts RF output, including numerous nights running 80% or higher duty cycle for hours at a time. It has proven to be very reliable. The basic design requires just 0 dBm drive (one milliwatt), but I have included a built in 20 dB attenuator in mine to accommodate the +20 to +24 dBm drive provided by my various exciters.

Use caution when selecting capacitors for the output circuit, namely C1 through C4. It may be very tempting to use a single capacitor of the specified value, but doing so will likely mean operating the capacitor beyond its voltage ratings if it is a film capacitor. Film capacitors must have their voltage derated as frequency increases. Capacitor data sheets usually have curves for this derating. I had capacitors in a 630 meter amplifier fail because I had not taken that into account. Generally smaller value capacitors can handle more voltage at a given frequency than higher value ones, which is why I use several low value capacitors in parallel to reach the desired capacitance.

When selecting a FET, choose something rated 200 volts or more if you plan to run this amplifier at full power. Voltage peaks at the FET drain are about 3.5 times the applied DC voltage. So with 40 VDC on it, the FET is going to see a peak voltage around 140 volts on every RF cycle.

The FET requires good heat sinking. I prefer either directly mounting the FET with a bit of thermal grease to a heat sink isolated from ground (note the heatsink will have drain voltage and RF on it, so be careful what might come into contact with it) or a mica insulator with thermal grease for a normal grounded heat sink. I do not recommend using greaseless Sil-pad thermal pads as they may be unable to provide adequate cooling efficiency. The heat sink on my amplifier is about 5 x 3 x 1.5 inches. A fan on the heatsink is not required for low duty cycle such as two minute WSPR transmissions at 33% or lower duty cycle. For long T/R period modes or frequent transmission resulting in high duty cycle, you will need either a larger heatsink or a fan. I also have a fan on the bottom pushing air into the amplifier and air exhaust vents on the other end. Again, this is not needed for short transmissions of low duty cycle but if you are going to run 15 or 30 minute modes or very frequent transmissions, it will be necessary to supply cooling air to L1 and L2. I also have an internal fan assuring high volume air flow across those inductors, though that is probably not needed. It was there to move air across the inductors before I added the bottom cover and intake fan, and I didn’t bother removing it.

As noted in the previous post, this amplifier was constructed by making “islands” in a solid copper plated PC board using a Dremel rotary tool. Other methods will surely work as well.

The only future change I might make is to replace the little TO-220 size 34N20 FET with a FDA59N30 in the larger TO-3P package. I have not experienced any FET failures with the present configuration but I like the larger FETs for better cooling efficiency.

The story of this amplifier starts back in 2017 when I held a FCC Part 5 experimental license (WI2XTC). This was prior to FCC granting amateur privileges on 2200 and 630 meters. I was looking for a kilowatt-class amplifier that seemed reasonably easy and inexpensive to reproduce. I settled on the W1VD kilowatt class D design.

After building the amplifier I had a lot of problems with blown FETs. After many months of testing, troubleshooting and trying various things, I got that problem under control for the most part. It turns out this was not a problem with the design or my construction, but simply that modern modes and operating practices are not consistent with the intent of the design. The amplifier was designed for a steady RF signal at its input, with transmission to start and stop by using one of the keying inputs to enable and disable the FET driver IC. That just isn’t how modern computer generated modes work. The software usually provides for PTT (amplifier keying), but it is the opposite of what would be needed to keep the amplifier happy. PTT is asserted before RF generation starts and held until after RF stops. The amplifier needed the opposite to be safe: PTT asserted after the start of RF generation and released before RF stops.

Initially I was experiencing frequent FET failures with any exciter I used, but they were far more common, in fact almost guaranteed using an exciter capable of amplitude shaping the start and/or end of the RF envelope. I don’t have a storage oscilloscope, but after seeing some FET drain waveforms provided by other users of the same amplifier it was apparent there were (or could be) voltage spikes exceeding the 200 volt rating of the FQP34N20 FETs at the start and especially end of a transmission. Additionally there appeared to be extended times of zero voltage on one pair of FETs or the other, possibly indicating a longer than normal on time. One might wonder if there were current surges occurring at those times. After a lengthy search for replacement FETs rated for higher voltage and current but otherwise having similar ratings to the FQP34N20, I tried the FDA59N30. That eliminated any blowing of FETs with exciters or modes that are not capable of RF envelope shaping, such as the QRP Labs Ultimate 3S which I use extensively. I had just one FET failure in more than a year of operation. It appeared that may have been due to overheating. I found the mounting screw on the failed FET was not tight. Both the mounting surface of the FET and the Sil-Pad underneath were discolored in a way that suggested excessive heating. The FDA59N30 is a current production part while the FQP34N20 is long discontinued and becoming very hard to find except from some overseas sources which are selling counterfeit devices.

During the summer of 2020 I was invited to join the early testing team for the new FST4 and FST4W modes being developed for use at LF and MF. It was one of the better things to happen in 2020! Initially I was able to run these modes using the phasing exciter but I noticed some peculiar glitches on the scopematch (sophisticated RF power and SWR monitor using an oscilloscope) at the start and particularly at the end of transmissions. I also had some intermittent problems with amplitude and phase fluctuations during FST4/W transmissions. Eventually while looking for the source of that problem, I discovered the IR2110 FET driver was not entirely healthy. One side was OK but the other was providing only weak gate drive to the FETs. I replaced the driver and that was the end of being able to transmit using the new modes! FETs were constantly meeting their demise with the new modes, while other modes were OK. My pile of dead FETs was again growing rapidly! At about the same time I learned something about the new modes that I had not previously known. They were intentionally using envelope shaping at the beginning and end of transmissions! (Note: with the general availability release of WSJT-X 2.3.0, the FST4/W envelope shaping can be disabled.) Sure enough, when I tried manually enabling the FET driver after the envelope shaping at the start and disabling it before RF shaping commenced at the end of transmissions, FETs did not fail.

It has never been clear to me exactly where in the amplifier problems start with non-constant amplitude drive, but clearly bad things were happening somewhere. Was it strictly in the output circuit, or was something going wrong in the driver or the pre-driver logic? It would be fair to say I was never entirely happy with the configuration of the amplifier anyway. Since it used a flip-flop to clock the IR2110 FET driver, it required the RF input signal to be at twice the operating frequency. For 137.5 kHz operation, it needed RF drive at 275 kHz. In order to achieve that with common exciters and relatively minimum hassle, I used a frequency doubler circuit before the driver. That always seemed like unnecessary complexity to me, but at the time of construction there were few, if any alternate driver designs for class D amplifiers that didn’t use a flip-flop, therefore requiring drive at twice the operating frequency. The doubler also caused some problems running EbNaut, which uses 180 degree phase shift keying.

While struggling to think of a solution to the FST4/W envelope shaping killing FETs en masse, it came to my attention that others were now using drivers for class D amplifiers that did not use the flip-flop and worked with “normal” drive at the operating frequency. I decided to try an experiment. Melding elements from three different designs, I came up with a driver that provided all of the control inputs of the original, required no doubler and allowed for some adjustment of the duty cycle. The circuit uses a 1:9 impedance step up transformer driving a pair of LM311 comparators. The comparator outputs control the IR2110 FET driver. One obvious advantage is that this does not require drive at twice the operating frequency. Another is that, unlike the original driver configuration this one allows for some adjustment of the length of the drive pulses to the FETs. This made it possible to get cleaner drain waveforms with less high frequency ringing.

With the original driver there was always some high frequency drain ringing. With this driver it can be almost entirely eliminated by adjusting the 20 ohm trimmer to vary the duty cycle or length of drive pulses. There is a tradeoff between the circuits. The original amplifier input consisted of a frequency doubler and the flip-flop preceding the IR2110. With that configuration there was little to no change in drain waveforms over a 15 dB drive power range. With the new circuit, drain waveforms change with drive level. The change is minimal over about a 6 dB range but increases outside those limits. The range of acceptable input can be pushed to 15 dB before things start looking really alarming. This worked fine for all but FST4/W modes (prior to the 2.3.0 GA release with envelope shaping disabled). With FST4/W slowly rising from zero to full power at the start of a transmission and slowly decreasing from full power to zero at the end, the drain waveforms went through some ugly periods. I was still occasionally losing FETs.

I wondered if an RF sensing circuit could solve that problem. It should be possible to sample the incoming RF drive, rectify it and use the resulting DC voltage to control a comparator which would enable the IR2110 only after drive had reached a safe amplitude. The question I had was would it be fast enough to disable the FET driver at the end while the envelope was decaying. If it was too slow, it might not disable the driver before the amplitude reached a low enough level to cause problems. Never blindly trust my math or circuit design skills but by my reasoning it looked possible. The envelope shaping occurs over approximately 2.5 seconds for a FST4/W 1800 second transmission. I believe it scales linearly with the T/R period, so for the 15 second transmission it should be about .02 second. There should be plenty of time to shut things down if I used an RC time constant of about .0002 second. Instead of rambling though all of my rough calculations let me just say I tested the circuit as built with several hours of FST4-15 transmissions, which would require the fastest timing. There were no glitches evident and no FETs were harmed during the test.

I built the new driver circuit on a board which was the same size as the original and used the same connectors, so it was a drop in replacement for the amplifier. Similarly the RF sensing and control circuit is a drop in replacement for the no longer needed frequency doubler.

The remainder of the amplifier remains mostly unchanged from the W1VD circuit except for the substitution of FETs as discussed earlier. It should be noted that for power levels above 400 to 500 watts at high duty cycle, a small fan cooling the output transformer is a good idea.

The power supply is about as simple (and efficient) as you can get, a few luxuries notwithstanding. It is unregulated, consisting of a variac, transformer, two bridge rectifiers and two large filter capacitors. It can provide 0 to 50 volts for the FET drains. The power transformer has two secondaries. Each has its own rectifier and filter capacitor. The two are combined at the output terminals of the supply. A small fan is used to blow air across the bridge rectifiers to aid in cooling. Because the filter capacitors are large and the transformer resistance is low, a soft start circuit is used to prevent inrush current problems. There is a separate transformer with a similar configuration to supply 12 volts to the amplifier driver circuits. The 12 volt supply also controls the soft start by means of a comparator which closes a relay to short out a resistor in the AC input to the variac after a short (adjustable) delay.

In early 2020 I began phasing out much of the first generation LF equipment and building replacements. My LF operating interests focus largely on DX. As I have learned more about all of this, it became obvious I needed some upgrades. This is the second in a series of posts about new equipment for our lowest frequency amateur radio allocation.

Like the first generation receiver, the transmitting downconverter did not have adequate frequency stability for slow modes on LF. I also wanted something that didn’t tie up my only HF rig when operating on 2200 meters. After reviewing several designs for phasing exciters I settled on a design by W1VD. I built mine Manhattan style using MEPads and MESquares from QRPme.

The MPS6650 and MPS6652 transistors used by W1VD are no longer available. I successfully substituted BC33716BU and BC32716BU devices but I have not been able to achieve the stated +20 dBm output. Mine will only make +16 dBm before the output waveform becomes distorted. This works OK with my amplifier but is a subject I would like to revisit at a later date.

Initially I encountered some difficulty getting good carrier and opposite sideband suppression. I traced the problem to the LO signal to the two mixers not being 90 degrees out of phase. I built several variants of the quadrature hybrid but I could not get accurate 90 degree phase shift or equal amplitude. Trying some alternate approaches, I achieved success using a Wilkinson divider and phase shift network. Some cut and try adjustment of two capacitor values was needed but in the end I achieved accurate 90 degree phase shift with similar amplitude on both ports. I used 6 dB resistive attenuators on the two LO signals before feeding the mixers. The two outputs from this circuit go directly to pin 8 on the two SBL-3 mixers in the exciter. The 6 dB pad, C1, C2, T1, C3, C4 and the associated 49.9 ohm resistor shown in the W1VD exciter schematic were omitted. With this arrangement I was able to achieve better than 55 dB carrier and opposite sideband rejection after careful adjustment of the level and phase balance trimmers in the exciter. If you build this and find it is operating on the wrong sideband, reverse the LO inputs to the mixers. If you look closely at the blue and orange wires coming off the LO divider and phase shift board, you will see they cross over each other on the way to the mixers on the main board below. Mine had ended up being on lower sideband the first time around! One other change should be made to the phasing exciter if you will be operating it into a 50 ohm load: omit the 49.9 ohm resistor in series with the output. The 1 uF capacitor should connect directly to the junction of the two 5.1 ohm resistors.

I am using the same Leo Bodnar GPS Clock that supplies 408000 Hz to the new receiver. It supplies 136000 Hz square wave to the exciter, which I low pass filter before the divider.

I have many hours of operation with this exciter in various modes. It has performed well. One thing this exciter does not like is magnetic fields which can couple 60 Hz energy to the audio circuits. Don’t put it too close to a linear power supply!

In early 2020 I began phasing out much of the first generation LF equipment and building replacements. My LF operating interests focus largely on DX. As I have learned more about all of this, it became obvious I needed some upgrades. This is the first in a series of posts about new equipment for our lowest frequency amateur radio allocation.

After using the original modified SoftRock Lite II receiver for three years, it was time to move on. That first receiver served me very well. With it I was able to make three trans-Atlantic QSOs, and heard a lot of DX on various modes. In the end, however, I wasn’t satisfied with the frequency stability of the crystal oscillator, which was about 1 ppm, or a little less than 0.15 Hz drift on 2200 meters. That may seem completely insignificant to the HF, VHF or microwave operator but for the most serious DX pursuits on LF it not sufficient. With the one watt EIRP legal power limit, propagation and high noise levels at 137 kHz we need very slow modes to succeed over great distances. As a general concept, the slower the mode the greater the frequency stability needed. Legacy modes include QRSS (extremely slow CW meant to be read visually from a waterfall) and its derivatives like DFCW. Readers may recall my first DX QSO with 2E0ILY used DFCW60, meaning that each “dit” or “dah” takes 60 seconds to send! Drift of 0.15 Hz is clearly visible at that speed and can lead to difficulty “reading” signals at even slower speeds. Today we have various slow digital modes for beaconing and QSOs. At the extreme, EbNaut requires transmitter and receiver drift be no more than a few tens of microHertz! Others are more tolerant but current evolution suggests one should strive to stay within 0.01 Hz or better during any 30 minute period if DX is of prime interest.

During those first three years I had tried various receiver, filter and preamp configurations. I now know what is needed with the SoftRock and my available antennas. I wanted to combine the filter, preamp and receiver into one box but I wanted to use a GPS referenced local oscillator for stability. In the end I settled on a design which puts all but the local oscillator into one box. The LO is a separate Leo Bodnar GPS Clock which supplies 408 kHz for the receiver (divided by four in the SoftRock quadrature LO generator) and 136 kHz for a 2200m phasing exciter.

The major building block for the receiver is a SoftRock Lite II kit from Five Dash. A few modifications need to be made for 2200m operation. The schematic shows the values for parts that need to change for operation on this band (C3, C4, C10, C11, C12, L1, T1, R5, R6, R16), as well as the removal of the crystal and external LO connections in its former place. The capacitors can be ceramic. I recommend mounting the SoftRock Lite II board with the insulating hardware that comes with it. Ideally one wants everything isolated from the metal box except for the shield of the audio cable connector. To maintain that one ground point I run the receiver either on a battery or an isolated wall wart.

For the front end stages I have married a filter design by YU1LM and a preamp design by W1VD. The filter provides a bandpass response to keep out of band signals from overwhelming the receiver, while the preamp provides about 20 dB gain which is needed with many small receiving antennas on LF. You want enough gain in the front end and receiver so that the noise floor comes up at least 10 dB when you connect the antenna. If this seems a little different from conventional advice, consider that we are dealing with extremely weak signals where even fractions of a dB can make a difference. If we want to keep the signal to noise ratio from being degraded a meaningful amount, we need that much gain to be sure the SoftRock and sound card noise floor don’t degrade S/N of the system. With the exception of the 10 uF electrolytic, all capacitors are ceramic types.

Next I needed a suitably stable local oscillator. We need a final LO frequency that is close enough to the 2200 meter band to allow tuning it with whatever sound card will be used. If the sound card sample rate is 96 kHz, we need to be within 48 kHz of the receiving frequency. I recommend staying a few kHz less than that due to the way anti-aliasing filters in sound cards work. This means we want our LO to be between about 96 kHz and 178 kHz in practice, preferably avoiding putting it “in band”. The LO frequency is divided by four in the SoftRock quadrature generator circuit. This means we need to inject a frequency four times higher into the receiver. Anything between 384 and 712 kHz will work. I was already using one of the two outputs from the GPS Clock to provide 136000 Hz LO to my phasing exciter. Available frequencies for the second output are somewhat limited and tied to the first frequency but in this case 408000 Hz is one of the options, and it is perfect. That puts our final LO at 102 kHz, comfortably within range, yet far enough removed from the band of interest to put the image frequency around 67 kHz, well down the slope of the receiver front end bandpass filter. Perfect!

First I tried injecting the 408 kHz square wave directly into the SoftRock. It worked but I didn’t have a good feeling about it. For one thing, that meant that the SoftRock and GPS clock grounds were connected, a situation which I was trying to avoid in case of ground loops and noise getting into the system. The GPS Clock also didn’t like the impedance, causing it to put out not only the harmonic rich square wave but also a significant amount of HF energy as ringing due to impedance mismatch. I tried using a transformer (for ground isolation) and low pass filter to clean up harmonics but this made the GPS Clock even less happy with a lot of ringing due to reflections. Since I had signal to spare I solved this, albeit somewhat crudely, but inserting a 10 dB attenuator between the GPS Clock and transformer. This gave a nice clean sine wave at sufficient level into the SoftRock LO circuit. I don’t claim this design to be elegant or perfect, but I do claim it works well for me. I used film capacitors in the filter because I had them on hand, but ceramic should be quite acceptable.

This new receiver has been in operation for several months. Sensitivity and gain is more than adequate for use with my LNV antenna. Frequency stability is now determined almost entirely by sound card sample rate drift and is on the order of 0.01 Hz over several hours. This is sufficient for all but EbNaut, where the sound card sample rate requires continuous monitoring and correction. I have not conquered that yet.

On the morning of January 15 I was nearing the end of a 72 hour test of the JT9 submodes (JT9-10, JT9-5, JT9-2, JT9-1) on 136.395 kHz. The transmitter had been running 87% duty cycle for two days and as far as I knew all had been well. On this morning I checked in on things when I got up just before sunrise. It was running as expected with the waveforms on the ScopeMatch looking normal. I went about some morning chores and came back about 20 minutes later to check again. The transmitter was still running but the antenna was far off resonance. Minor changes are common but this was more than a minor change. I knew something was very wrong.

I quickly shut down the transmitter, grabbed my binoculars and went to the window to inspect the antenna. All wires were up and intact. I then hastily bundled up and went outside to check the loading coil / variometer. It didn’t take long to realize where the trouble was. When I removed the cover from the assembly housing, acrid smoke came billowing out and I could feel heat radiating from somewhere inside. This was not good! Since the smoke was so thick and presumably toxic, I could not do a full inspection until things had aired out a while.

Upon subsequent inspection I found the bottom of the moving inner coil badly damaged. I can only guess as to what happened. Careful inspection of the following pictures will reveal something of the construction. There was a wire (12 AWG solid, insulated) running down the length of the form on the inside. This provides connection from the bottom of the inner coil to a a terminal at the top of the coil form which is jumpered to the top of the large outer coil. At both ends, the method of feeding through the form was a 18-8 stainless machine screw with washers and nuts as needed. On the inside the ring lug on the wire was between the head of the machine screw and the coil form. Stainless hardware may not have been an optimal choice. It stays clean practically forever but it has poor electrical properties. I had assumed it would be fine with the expected 2 amps or so of low frequency RF current.

What I suspect happened is that over time, probably aided by thermal expansion and contraction cycles of the PVC form, the hardware became loose on that bottom connection. As it began to loosen slightly, resistance of the connections may have increased somewhat, leading to more heat being generated. This may in turn have led to some slight softening of the PVC, allowing pressure on the connections to relax even more. I believe eventually it became so loose there was arcing which produced extreme heat in a localized area, eventually leading to the damage.

In hindsight, there may have been two warning signs that something was not right. If these were signs of failure in progress, things had been going south for some time. About two or three weeks prior to this incident I had noticed that when I was transmitting I would sometimes see “fuzz” appearing on both sides of my signal when viewed on the waterfall of my SDR receiver. It usually lasted only for several seconds, then cleared up. I did wonder about arcing, but the ScopeMatch looked perfectly normal. I put it down to just another artifact of severe receiver overload. It’s not as though my signal ever looked clean in the local receiver! There was always plenty of junk, no doubt worsened by the use of back to back diodes across the receiver front end to prevent damage from my own transmissions. But this particular “fuzz” phenomenon was something I hadn’t recalled seeing previously.

The second possible warning sign came 24 hours prior to discovery of the failure. On that morning resonance suddenly “jumped” higher in frequency. It wasn’t a big change, but was something I hadn’t seen before in benign weather conditions. Re-resonating took care of it but about an hour later it “jumped” back to the original resonance condition and needed to be adjusted again. This unexplained behavior should have been a warning that something was not right.

Much of what I think I know about this failure is speculation based on inspection after the fact. My theory seems further supported by the fact that the other stainless machine screws passing through this form had all loosened considerably. I know they were tight when it was built, but I was able to remove them using just my fingers. I will never know for sure exactly what happened, but the new inner coil will be designed to avoid the suspected failure scenario. If it fails again, I will have to reexamine my theories!