This is a continuation of Designing a 20/40 band CW rig – Part 2.
Too many harmonics
After the last post, that looked at the TX circuitry and filters, I went back and looked at the TX efficiency. I was also a little unhappy that there was quite a bit of power in the harmonics, which were being shunted to ground. One effect of all of this was the BS170s were presenting a variable input impedance to the previous BJT driver stage. Now both MOSFETS and BJTs are current sources, the former voltage controlled and the latter current. The voltage that is present on the output of an active device is not generated by that device directly, rather the device generates current. The voltage is a function of that current and the load. Now if the load is variable, then the voltage is going to change, even if the current output from a device is a nice clean sine wave, the voltage will not be so if the load is changing. We could kind of get away with this if the load device is a BJT as it is current controlled, but MOSFETS are voltage controlled.
The answer is not more filtering, the answer was to go back and look at the design. I first tried to lower the resistances in the BS170 bias network to swamp out the impedance changes. This worked to some degree, but it was not a final solution:
- More current in the bias network is overall power efficiency loss. We want to keep bias currents lean.
- My driving device, a MMBR941 has a maximum mean collector current of 50mA and I was within 20% of this. In the words of a well known Star Trek character “I cannot give her anymore power capt’n”.
Replace the MMBR941 with another device
It takes a long time to search around for suitable devices, and they also need to not be expensive. Instead of a long search, BS170 devices are nice and cheap, so I began with using another one of these to drive the 3 BS170 finals. It was easy enough, I already have a bias divider network to set the bias, and by moving this up and down, I can control the drain current. All of the stuff on the collector and emittor of the BJT can go, except for what is now the drain inductor.
This approach gave me more drive, and allowed greater 2nd network bias currents, but the additional losses here about matched the gains post BS170 finals. Not too good.
True Class C design
The general idea of Class C is that the active device is operating for less than 50% of the cycle. This is true, but a tank circuit is needed as well. My initial design had one of these, but on the output. The literature that I saw has the tank network between the voltage supply rail and the drain. So I needed to change what is basically a resonant low pass filter on the output to a tank at the traditional spot, between supply and the BS170 drains.
This approach certainly worked, and the output was much cleaner, but not near perfect. One thing that is needed is that the resonant tank circuit needs very low values of L and high values of C. Being a shunt, it is still presenting a very low impedance to the finals on harmonic frequencies. Because the fundamental load frequency also has a lowish impedance of around 10 to 12 ohms, the shunt network needs to have a net reactance well below this on the harmonics. A bit tough to do, and still we have the problem of variable impedance being presented to the driver stage.
Class E design
I decided to go for an alternative approach, a Class E design. The circuit I showed in the previous post has a bit of a change:
As can be seen on the circuit, a BS170 is being used as a driver. I have also gone from three BS170 finals to two.
A Class E amplifier uses a shunt capacitor across the transistor to complete the waveform, along with a series resonant circuit. This presents high impedance to harmonics, while passing the desired frequency. Actually, one of the tricks with Class E is that this resonant circuit is not centred on the frequency of interest, but a little below. A calculator is available here, thanks to Alan G3NYK. That site has a LF orientation, but the principle certainly works on HF or higher.
Class E amplifiers also have a reputation for higher efficiency than Class C, and I am finding this to be the case. I was almost able to get 5 watts RF output from a single BS170 on 20m, and could do it easily on 40m. We want 5W on both bands, with a bit of margin on the devices, so I am still using 2. By designing for a specific Q on the resonant circuit, and the selection of the shunt capacitor, various power levels are achievable for a given input to the transistors.
With the change of circuit design, I also had to review how I could dampen the oscillator input when not TXing. Unfortunately, I have found that the leakage currents on the various MOSFET devices I have used are too great to effectively shut it out. I also tried using a NDT2955 device to cut the power to the finals when not TXing. This device has a very low voltage drop across it when used as a switch, even at several amps. Unfortunately it sets up some strange oscillation in the Class E circuit, so no success there. It really didn’t make much difference, because the BS170 will act the same way on a small signal, allowing it to leak through, regardless of the gate bias set well below cutoff, or the power being cut from the drain. In the end, it was back to a simple 4401 to sink away the unwanted signal when not TXing, and this had the best effect between the driver and the finals.
What was the effect of all of this: I achieved a efficiency on the finals of 81% on 20m and 85% on 40m. I had to tone down the circuit on 40m, because the circuit adapted directly from 20m at 5 watts produced 7.5 watts on 40m. I also would not want to change the biasing networks based on the band, so the flexibility of being able to control power gain in the components of the Class E amplifier itself was nice.
I also was able to significantly reduce the current on the final bias network. The load impedance presented to the driver from the finals is still quite variable, but with Class E, it does not matter as the output is remarkably clean. Some Class E approaches are fed directly with a square wave and this can allow for even greater efficiency. I could look into this, but I think I have captured all the low hanging fruit. I still had to allow a moderate level of current on the first bias network, but it is only about 2mA. There is about 40mA spent on the driver and less than 1mA on the second bias network. This is a total of less than 45mA, for around 550mW. In total, including the driver and bias networks, we should be spending less than 7W to drive 5W of RF on 20m and less than 6.5W on 40m. These levels are well below half of a FT-817 (not counting the FT-817’s draw for other things like the screen, activating the coil on the rear connector, etc).
The RX circuit needed some adjustments to deal with with the change in the output signal. I adjusted the bias on what is now Q12 up to about 11V, as the peak to peak voltage of the TX output here. I also need the base of the signal at least around 0V, otherwise the body diode of Q12 will begin to conduct. We will still be well within the breakdown limits of the device. Both the BS170 and 2N7002 have 60V drain source breakdown
Side note: What is this body diode on the 2N7002? It is what comes with the territory with MOSFETS, it is part of their nature, and for a N channel, it means that the drain will get passed through to the source if the drain is below the source, like a forward biased diode. Also, when reversed biased – the normal usage of a MOSFET, the diode acts like a zener diode, with the drain to source breakdown voltage being the zener voltage.
Speaking of zener diodes, I will be putting one across the BS170 finals, at somewhere in the high 40V, I’ll just need to select the device. Of course we never want to see the MOSFET put to its breakdown voltage. One impact of Class E design is that there are higher voltages than what would be possible with Class A or even Class C.
The Class E matching reactive components are of course band specific, so the location of the relay will need to change, with it being right up on the BS170’s. The peak current across the relay is about 1A, or 700mA RMS. This is well within the limits of the G6SK-2F unit I plan to use.
One other thing I should mention is that I am using two coil latching relays. These will only use power when I need to change the relay orientation. I’ll use 5V relays with the supply fed by a 5V regulator.
Next time, I’ll look at the audio part of the circuit.
73 de Wayne VK3WAM