I have a medium term project in mind involving a portable HF digital mode transceiver. To get there, I need to build a lot of individual components that will come together in the final system. The most challenging so far has been developing an acceptable output power amplifier.

For a little bit of background, this type of hardware will enable me to communicate using digital communications modes like RTTY, PSK31, or (my favorite) Hellschreiber. One important key is to get enough power out that the recipient can actually hear the transmission, and that requires an RF power amplifier.

There are lots of published designs, but my goal is ultimately self educational here. So I don’t want to just use a kit amplifier, or simply replicate a circuit from a project book. I want to slog through the process of design, understanding each piece that goes into it, so that I know I’ve learned these things. My tentative benchmark is that my 6 year-old should be able to point to any part, and I should be able to explain what it is and why it’s there!

I have built a small amplifier previously that works up to about 750mW based on the 2N2222 transistor:

But to go much further than that you are looking at using beefier transistors designed for power or RF applications, or there is a deep technical well to fall into on highly efficient designs (e.g., class E amplifiers). I decided that the former makes more sense for a “first project”, and I’ll save the complexity and extra challenging theory of the latter for when my feet are thoroughly wet!

Suitable Output Device

A lot of classic homebrew amplifier designs use devices that were ubiquitous in the last century, but are getting harder to find. If you pop open a book on amateur radio amplifiers, or read well-received articles from the past, the output transistors are often simply unattainable. There was a burst of production of low- to mid-power BJTs for HF use in the 70s through the 90s (maybe the CB boom was part of that), but the semiconductor industry has moved on, and inexpensive parts in convenient packages are harder to find.

I want to play with some of the cool parts (like the 2SK3476 from Toshiba, or the AFT09 series from NXP), etc., but to get my teeth cut I settled for the very popular IRF510 that you see in lots of ham radio circles. In discussions with my mentors, I found that it is somewhat challenging to use, but straightforward, and the ability to get them for under a dollar apiece makes them attractive.

Synthesizing from a number of other amateur radio homebrew circuits I have reviewed, I settled on a simple starting point. This is about the simplest possible IRF510-based amplifier I’ve found, and it’s from Experimental Methods in RF Design (EMRFD from now on), figure 2.98. The book says it’s designed for class AB operation (and I’ve confirmed that it can’t quite be pushed to full class A in my case), but can also be biased for class B or C.

The challenge with the IRF510 is its high gate capacitance of 180pF. It means you need to dump a lot of power into the gate to keep the voltage up and keep conducting at the drain. One of my mentors suggested a full watt of input is required to realize the IRF510’s potential!

Driver Stage Design

So I needed to build a driver stage to take a small input signal (I assumed a 1mW drive as my source). I read a lot of articles on power amplifier design, and ended up rather liking a push-pull design combining a PNP and NPN transistor pair. I kind of liked the idea of a transformerless design, and thought I’d try input and output matching L networks. Something like this:

I built this one, and immediately ran into problems. There was no real amplification, and I wasn’t sure the matching networks were doing anything for me. So I posted on the RF Electronics subreddit, and received a lot of different (and useful) advice. I am really thankful for the help I got from those guys!

This push-pull arrangement is basically two current-gain stages back to back, which I had not properly understood, and with a small drive I wasn’t going to see any effective power gain into a 50 ohm load.

One of the suggestions was to use an initial stage of voltage gain, and then use my existing circuit for current gain. Another thought (which I have yet to verify) was that the load may affect my output network. I don’t know if there is any subtlety there to worry about, so I decided I’d simplify input and output too, just in case, and save complicated things for later. This is the modified circuit:

I cut the board and put it together, coming out as shown in the following image:

And after all this work, I found that it still didn’t amplify worth anything. The signal would pass right through the amplifier, and actually with some attenuation! Frustrated, I asked some more questions on Reddit. Someone suggested that my transistors might be backwards, but I was confident that wasn’t it because I’d double-checked the pinout with the datasheet.

Well, turns out I was using the ON Semiconductor datasheet for the 2N2222. But my pile of 2N2222s came from a bulk pack I’d ordered that apparently uses a different pinout – the European standard pinout (EBC instead of CBE, ack!). So it was an innocent mistake, but embarrassing nonetheless! After I flipped the transistors, I found that I got a solid 180-250mW of power output with a 1mW drive, depending on the setting of the feedback trimmer and which HF band I was using.

Putting the Stages Together

I hoped this driver stage should be able to drive an IRF510, so I proceeded with the three-stage design. I made a couple changes, but nothing too significant because the EMRFD figure 2.98 amplifier is about as simple as you can get for the IRF510! Everything seemed to check out in an LTSpice simulation, and the final circuit looks like this:

Note, I used 1N914 diodes, not 1N4148 as shown in the schematic. The 1:4 autotransformer at the output presents the 50 ohm output impedance as 12.5 ohms to the drain of the IRF510, so it should be able to push some real power out. I was hoping for a few watts.

Fabrication and Assembly

I’ve recently begun to mill my own PCBs, which is the single most significant improvement I’ve made that speeds up my learning curve. But this would be my most complicated milled PCB yet – with conservative feed speeds, it took a bit over an hour to mill:

After I cleaned up the burrs and did some continuity testing, I assembled the parts and soldered it up. It came out looking like this:

And firing it up for the first time, I set the bias for about 100mA of quiescent current and fed it a signal in the 40m band, passing the output through a low pass filter. The result was quite stunning, at 3.92 watts!

Managing Heat

The first problem I noticed was heat. I had put a heatsink on the IRF510, but it was clearly inadequate. Running at 4W output for a little while, the heatsink reached about 140 degrees F! This setup would be good for testing in short bursts, but I will need to find a much bigger heatsink for the future.

At least I could experiment with a bunch of parameters. One important thing I learned about this NPN/PNP push-pull configuration is that if you can’t run the output close to class A, there will be an asymmetric dissipation in the second stage transistors. E.g., if I set the bias for class B or C (i.e., no quiescent current), then only half the input signal can turn on the transistor. This means the 2N2907 does all the work, and the 2N2222 does almost nothing. I even managed to blow the 2N2907 during this testing by loading it very highly!

So I learned that I really have to run this thing in at least class AB if I’m going to push for higher power to make sure that the load on the second stage is distributed between the two transistors. I also slapped a little heatsink on the 2N2907, just in case.

Then I realized I had a spare 12V ducted fan from my 3D printing adventures lying around. And I’d included a second set of power headers on the board so I could chain it to another 12V device… it was perfect! With a ducted fan directed at the heatsink, I could ramp the bias current to 250mA, and even raise the supply voltage to 14V and beyond, keeping the heatsink temperature below 100 degrees F!

At this point, with the fan, I’ve run this thing up to about 12.5W output (this was at a higher supply voltage, BTW), and not blown another transistor. So for all the annoyance of the noisy fan, it sure makes a big difference to get some convective airflow!

Measuring Performance

For my main purposes, I don’t really need a linear amplifier – non-linear behavior is fine for CW and FSK modes, after all. But I was curious just what sort of monster I had created; so I wanted to measure it and understand where it fits in the world of amplifiers.

The first test was pretty straightforward – measure input and output power across a large range, and see where gain compression comes in (and how linear the gain is). I did a bunch of measurements with my signal generator and oscilloscope, yielding the following graph:

This was far better than I expected. I then went on to try a two-tone measurement to see how the amplifier performs in terms of intermodulation distortion. This involved a lot of learning – I hadn’t really ever thought about what IMD is; and I didn’t even understand how the measurements are done!

I knew I would need two signals, close together, and an RF combiner to feed them both into the amplifier. My first measurements were confusing (and seemed shockingly bad!). Worse, when I just tested the signal direct to the scope without the amplifier, even that looked terrible. I started wondering if my (super cheap Koolertron) signal generator had too much IMD to be a usable source. This is when I asked the experts on the QRZ homebrew forum a few questions.

It turns out the RF combiner I had made wasn’t working. I’d copied the schematic from the ARRL handbook amplifier chapter. But it was a hybrid transformer design, which features very poor isolation between the two input channels. What was happening was that IMD was occurring in the signal generator because the two channels were feeding into each other.

The wizards on QRZ suggested a resistive or attenuating combiner, and suggested a schematic. I poked around in LTSpice a little bit, and came up with this, which seemed quite promising. This is also another situation where being able to mill a circuit board is a godsend – iterative development is wonderful, and it took about 5 minutes to mill this board (after about 20 minutes to do schematic capture and PCB layout).

Armed with this, I did another IMD test with the signal generator, and found that it was absolutely beautiful. I couldn’t see any IMD in the 60dB between signal and the noise floor. Excellent, now I can trust my measurements, and characterize the amplifier!

And here’s another interesting learning moment. I had read some simple definitions of adequate IMD performance, but had critically misunderstood some things. I thought I should apply signal to the maximum power output for the amplifier, and measure the IMD products; acceptable linear performance is any power up to where the IMD products are >=30 dB below the fundamental.

But that turned out to be missing a couple details. The QRZ folks showed me an example of an amplifier test, and I realized what was going on. By definition, the IMD testing parameters work like this:

  • Choose a peak envelope power for the amplifier
  • Apply two signals 1-10khz apart at PEP - 6dB (one quarter total power)
  • Measure IMD products

The test setup is shown in the following image. The signal generator sends two signals, one at 7.100 MHz and the other at 7.101 MHz. These are fed to the two input ports to the resistive RF combiner, whose output goes to the amplifier input. Its output goes through a low pass filter for the 40m band (5th order Chebyshev), and then into the oscilloscope for an FFT measurement centered at 7.1005 MHz.

And this gives the IMD measurement at the tested PEP. The following shows that the amplifier is probably good through 1.5W, and marginal a little over 2W:

And, for completeness, I used this online calculator to take my measurements and calculate that the third order intercept point is about 42 dBm. So really, this isn’t all that bad, considering it’s my first three-stage amplifier project!

Conclusions

The first conclusion is that seeing physics work and building things with your hands is just plain fun. It is really satisfying to go through the iterative development process, run into challenges, learn new things, and solve problems… and ultimately achieve the objective!

I’ve found that with adequate cooling, I can safely run this amplifier non-linearly to about 7W output with 1mW input. This is a fantastic result, and far more than I had tentatively hoped when I started work on this.

If I want to run it with a linear mode, it probably can’t be responsibly operated above 2W PEP because of the intermodulation distortion. But then, my previous amplifier project that did 750mW in non-linear operation can’t come close, so I did at least dramatically increase my homebrew capability.

But really, one of the most important things I learned is that a NPN/PNP push-pull driver doesn’t make sense unless the next stage runs in full class A. Amplifying in less than 360 degrees of the cycle means one of the driver transistors does more work than the other, and that’s distasteful from an engineering perspective. I will probably switch to a different driver architecture.

I learned a LOT about the definition of linearity, quality of linear amplifiers, measurement, and a few practical gotchas with amplifier construction. I will definitely make more amplifiers, but this one has really helped cement a lot of the principles and lessons to be learned from this corner of the RF world.

Next Steps

I have several directions I want to take things. First, I’d like to work out keyshaping and effective keying strategies for an amplifier roughly like this.

And I also recently came into possession of a nice stack of 2SC2078 transistors, which might be a lot easier to drive than the IRF510. I also have a few of the fancy VHF transistors that are out these days (2SK3476 and AFT09MS015) that would be good targets for the next amplifier. I’ve learned a few useful things, so I think I can make some useful stuff with those now.

But really, I have achieved enough to proceed with my mid-term project – a digital mode transceiver. This amplifier, or one very much like it, would satisfy my requirements to complete that project, and I should probably try to stay focused on that goal!