When one uses a valve in a position requiring power rather than voltage gain it should ideally be operated in accordance with its manufacturers specifications. That is to say the correct voltages should be used and most importantly, correct biasing should be employed. In the case of a radio frequency power amplifier the valve must be operated in the correct class. The word "class" refers to the combination of anode current and grid voltage being employed, over the expected range of inputs and outputs; the combination dictates the way power is delivered to a load and with what efficiency and distortion.
An ordinary audio power amplifier stage that drives a loudspeaker in a radio is usually operated in Class A. The valve is biased so that it works on the linear portion of its input-output characteristics. In other words the output voltage will be exactly N times the amplitude of the input voltage. Put in half a volt of 1000Hz sine wave on the grid and you get say 10 volts of 1000Hz sine wave at the anode. Put in double the voltage and you should get 20 volts at the anode. Unfortunately, although class A can be really linear, it is poor in respect of efficiency because lots of the current through the valve just gets dissipated as waste heat. What if the valve is working in a non-linear fashion? Well, not only will the output fall off from the expected voltage (or increase for that matter) as the input increases (or decreases) but the non-linearity will produce strange results.
Compared with a DC input, a sine wave input signal will never produce a constant instantaneous grid voltage so to amplify a sine wave (or any alternating signal) a linear circuit in which the gain of the valve will be identical at every instant is important, so the sine wave will be undistorted when it's reached the anode circuit.
If the transfer characteristic is not linear the gain of the valve, driven by an alternating voltage, will be changing continuously with input voltage and probably following some repeatable pattern. At any particular alternating input voltage the output will be either higher or lower than than that at DC (other than the gain corresponding to the DC voltage) at every instant so a sine wave will be distorted when it's reached the anode circuit.
I like looking at the extreme limits to get an idea of what will happen. If the HT supply to the valve is 100 volts and its gain is 10 and we input 20 volts at its grid what will happen? We would have expected the output at the anode to rise to 200 volts but as the HT is only 100 volts this will be impossible and the output will be clipped. Clipping in mathematical terms equates with the generation of harmonics and our sine wave of 1000Hz will still be present (if one examines the spectrum) but accompanied by (typically) 3000Hz, 5000Hz and other undesired signals. In fact, non-linearity when looked at mathematically will produce umpteen different signals in the anode circuit not just odd harmonics and certainly not a good idea when working with the power amplification of RF signals, although great for mixing and other requirements, such as frequency multipliers.
Pausing at this point to consider frequency multipliers. If the anode circuit of our amplifier includes a tuned circuit, set to say the third harmonic of the input signal, a picture on an oscilloscope will reveal (all things being realistic) what appears to be an undistorted sine wave of three times the input frequency. This is because, although there are a large number of RF signals present, the anode circuit impedance is so high at 3xf compared with other frequencies the third harmonic will seem to be the only signal present. For example, if the input frequency is 10MHz and the anode tuned circuit has an impedance at 30MHz of 10,000ohms and signals of various frequencies are attenuated in terms of power at the same general rate, as their frequency increases (say for example the 10MHz signal represents 50% of the total power and the 30MHz signal say 25%) then the 30MHz signal would be 50 times the amplitude of the 10MHz signal at the anode.
I calculated this as follows.. say the RF current component of total anode current is 10mA. The 10MHz component is 50% of the total = 5mA and the 30MHz component is 25% = 2.5mA. The tuned circuit has a Q great enough to present 10,000 ohms impedance at 30MHz so our 2.5mA will produce 25 volts. The tuned circuit has an impedance of 100 ohms at 10MHz so the 5mA will produce only 0.5volts. Of course these figures will be different if the tuned circuit is not so sharply resonant.
One can use Class A for RF power amplification, but only for relatively low powers as efficiencies are awful. Class A audio amplifiers get very hot and only good for up to say 5 watts. If greater output power is required, a different class of operation is best. After Class A is Class B then Class C (I'll ignore Class D). You can also specify Class AB (even AB1 and AB2). These modes of operation arrange that the power amplifier valve only turns fully on when output is needed. To arrange this we must apply a negative voltage to its grid. The higher this negative voltage the more you push towards Class C (and are forced towards non-linearity). Some modes are best used when two amplifier valves are operated in what's called "push-pull" operation. This is when one valve works hard amplifying the upper part of the sine wave and the other the bottom half, the two outputs being combined in a transformer.
To decide on the best mode or class of operation one has to consider the signal being amplified. Because the 19 set employs grid modulation the 807 has to operate in a linear fashion. A 19 set MkII uses automatic bias to set the mode of operation. Automatic bias is obtained two ways, one by inserting a resistor in the cathode lead. Current through this resistor makes the cathode positive with respect to ground (the earthy end of the cathode resistor) and as the 807 grid is biased to ground it will effectively be negative with respect to its cathode. The specific negative voltage dictates the anode current, hence the cathode current hence the bias voltage (explaining the term "automatic"). The second auto-bias voltage comes from RF current flowing through the 807 grid leak. This arrangement could be problematical but the designers solved the problem by adding a regulator using a double-diode valve. The designers looked at the 807 characteristic curves and worked out the correct values for the cathode resistor and regulator components as well as figuring out the best RF and audio drive levels, including preset adjusters to deal with manufacturing differences.
If we wish to conserve battery power and increase the power output somewhat (remember the 19 set uses a 12 volt battery in its operational life) then we can back off the 807 anode current by increasing the negative grid voltage. If, ordinarily the 807 draws 100mA at 500 volts with minus 20 volts on the grid supplied through its auto-bias cathode resistor then if do away with the latter and arrange minus 20 volts to be supplied to the grid the anode current will hopefully remain the same, however we've gained 20 volts by removing the cathode resistor.
The 19 set power supply has a 500 volt output which originally connected its negative side to ground. Clearly, when the MkII was redesigned and the MkIII specified battery economy was one of the drivers. The best modification was to add a switch so that the 500 volt supply was only activated when transmit was required. The amount of power involved is not particularly great as it's only the inefficiency part of the drain from the dynamotor that's saved. In the MkIII version (with its larger plug) the negative output goes directly to the grid circuit of the 807 where it grounds through a resistor. Basically we've lost the 20 volts we gained as the 500 volt supply is now fed via this resistor leaving only 480 volts for the anode, which is exactly the same as in the MkII, so what's changed? Maybe an inspection of the various resistors used and a view of what the various control switching does is needed...
First there's a difference in the R/T and CW arrangements. In the MkIII The R/T circuit has the negative bias as described above but in CW mode the bias voltage is grounded. Shorting out the grid bias increases the HT to the full 500 volts and additionally places the operating point of the 807 further up the anode current versus grid voltage curve. The extra current increases the 807 power input and of course its RF output. The valve is no longer working in class A, closer to class C which is more efficient but no longer linear. This non-linear aspect requires more careful tuning to avoid unwanted RF output signals.
If a resistor is inserted in the 500 volt supply ground return you get the extra negative bias for the 807 by connecting the high voltage zero to the 807 grid (in the MkIII set) instead of ground (MkI and MkII sets).