Mid-Power HF Amplifiers

Updated 20191014
New thoughts on HF amplifiers
When out on mobile radio trips I dream about a big signal to shout back at the DX I can hear (SSB). My Elecraft KXPA100 is wasteful, like all class-AB designs. It also needs a healthy 13V+ supply to get a full 100W output, whereas a good car battery drops to <12.5V when off charge. Some amateur class-AB amplifiers are still in the 1980s with obsolete IRF510 or Motorola RF parts.

This page looks at 2 designs of HF amplifier for mobile/portable operation.

Zero-Z (zero output impedance) HF Amplifier (2018-)
The high efficiency HF amplifier is now called Zero-Z - snappy and describes what it does.

Some HF amplifiers are designed around 50Ω output impedance. That gives them a maximum efficiency of 50% with a 50Ω antenna. Clearly it’s not good to transmit 100W and waste 100W, and at lower output they are much worse!

So far, I did some practical tests but a lot more simulation. Here’s a simulated frequency spectrum in LTSPICE XVII with 3kHz AM modulation. Peak power 120W from an 18V power supply at 92% efficiency.

40% modulation Blackmann window

Though the result looks spiky, the spurious is <50dBc which is the required limit for ham radio equipment. This is only an AM modulated test, the next plot shows a 2-tone test (SSB) signal simulated in LTSPICE. The frequency is 14.2MHz.
The FFT resolution is a coarse 4kHz on 10kHz spacing, but shows the two peaks IMD at -30dBc. With an audio tone spacing of 2kHz this is below -40dBc. The 2-tone waveforms in the time domain also show quite faithful reproduction between input and output. I know how annoying the splatter is on 20m, don’t want to make it worse!

This is the first test PCB, which didn’t work well but led the way to correct many problems!

Class-E PCB

Initial tests confirm high efficiency from the power stage, with 65W produced from 13.8V, 5.3A = 93%. Modulating this raw power yet keeping safety margins on the FETs is a difficult challenge.

Much of the technology is new or not widely given in books anywhere. I found some approaches to circuit design which may overcome the failure of previous experimenters with this type of amplifier. As said above, I’m designing a battery powered amplifier in the 100W class for field trips. There’s little point in having a fixed station amplifier with mains power, because the efficiency and size saved by switching type amplifiers is not really needed. A mobile amplifier with 100W+ powered from an easily available battery would be a more valuable invention.

The bands above 20MHz will remain closed for many years. The only viable bands for mobile SSB operation are then 80m,40m,20m,17m. So, I’m designing for this frequency range to make things easier.

If anyone would like to collaborate on this project I can share far more information privately than given here? Unlike the class-AB projects the Zero-Z work is closed source, and no further details can be shared.

Realisable Efficiency
A practical amplifier needs a drive source, transformers, filters, ancillary circuits. Looking at power loss in a typical class-AB amplifier in terms of dB and percentage:
  • 1dB = 25% loss in output transformer
  • 0.5dB = 12% loss in output filters
  • 0.4dB = 10% loss in switching and connectors/cables/PCB tracks
  • The drain efficiency will be 60% at best, 40% lost

So for 100W, dc input power will be >250W and a 12V amplifier takes 20A of current. For lower power outputs efficiency drops substantially, down to perhaps 30%. SSB duty cycle is generally only 25%, so amplifier efficiency will be way below 50%.

It’s tempting to think zero-impedance amplifiers provide >90% efficiency, but other losses become important. Designing a system with combined efficiency above 75% maybe impossible given the constraints of all the passive circuit elements.

For the example of a 100W amplifier, it will pull 133W, or 7.4A from an 18V power supply. Much better than class-AB, but there will always be substantial waste. Wastage is dependant on the power loss of every power handling component in the system. A zero-impedance transmitter stage is not realisable, but something close is.

100W Class-AB (2018-)
The new MRF101 LDMOS from NXP provide many advantages over the Mitsubishi RDxx series found in most HF power amplifiers.
  • Lower cost
  • Higher gain
  • Easy drive
  • Easy heatsinking
One major disadvantage - they cannot work from a 12V power supply. As concluded previously, it’s practically impossible to persuade DMOS switching FETs to give reliable 1-30MHz RF output >50W. With the MRF101 easily available at about 10x the cost of DMOS FETs but half the price of the RD100xxx devices, they look a superior solution.

A lot of ground-work was already done here in terms of auto-band switching; SWR protection and monitoring; boost power supply; Low pass filters; output transformers, it was decided to have a last push and productionise the best technology combined with the latest FETs. This project should provide a minimal parts count 100W (PEP) amplifier which has several advantages over products like the Elecraft KXPA-100.

A prototype PCB and 160x100mm enclosure are near completion as shown here.


As with previous HF amplifier projects, the hardware is open source and software is closed source. I refuse to be ripped off by Chinese manufacturers. As of late 2019, there’s much work to do on the software, but the circuit diagram and Gerber files are available from these links.
The items list (BOM) will follow when the software is more complete and the design more proven.

Hardware Description
Connections are:
Input power, 20A maximum, 15A average
RF in BNC, 2W maximum
RF to antenna BNC, SWR 2.5:1 maximum
PTT phono socket, short to ground on transmit as provided on many radios

The amplifier needs a 27V power supply, which can be made by “600W” boost modules available on eBay. Search for “600W boost converter” on eBay or Amazon. Internally there are positions for terminal blocks to accept wires from the boost module. Note polarity is shown on PCB silkscreen.

Power Stage
This section gives secrets of HF PA design not often written in textbooks. The power stage is based on NXP MRF101A/B LDMOS transistors. A mirrored pair is used to greatly improve the PCB layout, as the gates and drains are adjacent. The FETs then sit ideally placed to be thermally attached to a large heatsink. Balanced layout is a given with this type of design, it spreads the power dissipation across 2 semiconductors, and reduces 2nd harmonics. At the expense of a small increase in complexity.

The output impedance of a linear amplifier stage is:
Where Vs = supply voltage, Vsat = saturation voltage (typically <0.5V).

The stage is designed for a matching impedance of 12.5Ω. For 110W out at 27V supply, the impedance is about 12.5Ω assuming saturation voltage of 0.5V. This means the output filter must step up by 4x. That ratio enables a 1:4 transmission line transformer, far lower loss and lower cost than the dreadful tube-in-bead design.

At the input there is a min-loss-pad of 2dB which reduces the input impedance to 25Ω. That reduces the chance of overload at the input and ensures a reasonable match to the driver transmitter. 25Ω also helps broadband matching of the capacitive FET input. A balancing choke T3 ensures the 2 FETs are driven 180 degrees out of phase. The transformer is a ready wound part to cut costs. The input bias to the FETs comes from an MCP47FEB 10-bit DAC. The DAC is controlled by the MSP430 to give correct bias without manual adjustment. The bias is dropped down slightly at high temperatures.

The FETs drains are fed by a balancing choke, T8 which has 2 purposes. It ensures that the dc and ac supplied to the FETs is equal and opposite - i.e. balanced. A small amount of negative feedback is tapped off by a single turn thru T8 and connected to the FET gates. The feedback reduces distortion and gain. Gain is plenty high enough so a few dB reduction is actually better.

Input and output of the power stage is switched with Axicom IMC miniature SPDT relays, offering the highest current handling in such a small package.

Current Monitor and dc Power Input
A Texas Instruments INA138 high-side current monitor gives a voltage output for the MSP430. It needs to measure the current for setting the bias and for overload protection. The current is monitored by voltage drop across a 20mΩ resistor. A large capacitor helps absorb the high current pulses of the power stage. A large ferrite bead U8 helps filter the power stage.

MSP430 Microcontroller and LCD
The concept is to provide advanced features whilst removing expensive parts. For instance the MSP430 capture unit acts as a frequency counter for band selection, avoiding expensive mechanical switches on the front panel. Automatic bias adjustment avoids expensive manual potentiometers. The MSP430 has just 28 pins but each is heavily utilised, such as power save which turns off band selection relays and the LCD backlight. The only external logic chip is a 74AC238 which saves some pins and adds that nice power saving feature.

A 16x2 LCD is cheaper than a large number of LEDs, once the cost of an LED driver is factored in. These modules are made in China at very low cost and high volume. The display enables the secondary feature of copy protection too. The trimpot for contrast adjustment is more expensive than a DAC for contrast adjustment, perhaps in future I should do it that way. Obviously the LCD is used in 4-wire mode to save pins.

The MSP430 does all monitoring and control, in particular Rx/Tx sequencing and fault protection. The main defence against faults is simply switching back to receive. The main risk with that is welding the output relay if it switches at a peak of power. So the MSP430 does a sequence of reducing gate bias to zero and cuts off the boost power module, then waits for 10ms before switching the Rx/Tx relays.

It would be better to have a solid state (PIN diode) input switch, that changes over much faster than a mechanical relay could. It would cut the input power much faster. That is for any future iterations and definitely for the Zero-Z design. The system will not go into transmit until the input magnitude is fairly low. This avoids switching the relays to transmit under high power.

As mentioned, the MSP430 capture unit acts as a frequency counter for automatic band selection. If operation on a different band is detected while the amplifier is in transmit, it returns to receive mode to protect the band switching relays. Changing bands while transmitting is locked out on most radios anyway. The MSP430 has a maximum logic speed of 16MHz, so a divide-by-2 gate is used to bring down the possible 29.7MHz input to below 15MHz where the capture/compare is happy with it. Protective diode clipper is ahead of the frequency divider chip.

The software reads the supply current, forward/reverse power, input signal magnitude and supply voltage sequentially every 10ms. It also sends characters to the LCD at that rate and checks for encoder movement. Fault conditions have to be seen on 2 adjacent cycles to register as a fault. Every 50ms the auto band selection frequency and temperature are checked. Until the band selection count has been registered twice, band switching is not made.

The MSP430 runs at 8MHz but enters sleep mode once its processing cycle is complete. During sleep the ADC unit is sequentially reading the ADC channels. In receive mode, software doesn’t run the processing on forward and reverse power ADC results, nor the frequency count. So, it enters sleep mode more often in receive to save a small amount of power.

Input signal magnitude is measured to avoid overdriving the input. LDMOS FETs are sensitive to being overdriven. If the input power is excessive, the unit does the return to receive mode sequence described above. The magnitude is carefully split from the drive to the frequency counter.

A second PCB is in the amplifier to boost the 12-14V car supply to about 27V. There’s a facility included to shut off the boost power supply during receive, to prevent noise. The Chinese boost modules use an old TL494 switching controller. Applying 3.3V to pin 4 of that chip disables it.

Temperature Monitor
An NXP PCT2075GV temperature sensor is fitted close to the power FETs. The temperature is read via I2C bus, and displayed on the LCD. The MSP430 uses the temperature reading to slightly reduce bias when the FETs are hot. It also returns the amplifier to receive if the reading is above 90C.

Lowpass Filters
A 6 section low pass filter covering 160m to 10m is onboard. The bands are selected thru a 74AC238 3-8 line decoder and 2N7002 FETs as pull downs. The FETs have body diodes, avoiding discrete diodes for discharging relay magnetic fields. The filters are uprated versions of the designs used on previous amplifier projects.

SWR Bridge
A standard 2-toroid SWR bridge. Just a few turns are enough to get the right voltage for the ADC inputs. Right on the output, a spark gap is included in case of lightning strikes.

Software Description
The software is written in Code Composer Studio (CCS) for an MSP430G2553 target. Eventually the object code will be available in MSP4300 .txt format. The file can be programmed into the board using any MSP430 programmer supporting Spy-BiWire 4 pin programming. MSP-FET or many of the Launchpad Kits (with 4 wire cable) are fine.

Once programmed and running the software asks for a key code. I can supply a key code which is specific to each board made. This prevents copying of the software and the whole project. Before the key code is entered, the unit is in calibration mode, which sets the standing bias of the output stage.

The LCD is 16x2 line alphanumeric and controlled by the rotary encoder/push button. The top line shows the amplifier status: TRANSMIT/RECEIVE/BYPASS

TRANSMIT (transmit mode with no faults)
RECEIVE (either receive mode or return from transmit to receive due to a fault)
BYPASS (the PTT line is disabled and transmit power is passed straight through)

Rotating the encoder knob scrolls the bottom line through these sets of information.
For transmit:
PEP SWR (peak power output and SWR)
VLT CUR (supply voltage and current)
BND TMP (auto-band selection and temperature)

For receive:
FLT (cause of fault: high SWR, over current, over temperature, over voltage, high input)
VLT CUR (supply voltage and current)
BND TMP (auto-band selection and temperature)

Pressing the encoder knob at any time sends the unit into bypass mode. The LCD backlight goes off after about a minute in receive mode if the encoder is not touched. Touching the encoder or going into transmit makes the LCD backlight come on. This is done to save the 20mA of backlight current in long receive standby periods. A fault condition latches in for 30 seconds and switches to receive. Keying to transmit is not possible during this timeout.