Since a zener diode has slightly variable voltage characteristics depending on the current and this curve is not linear, an unwanted nonlinearity will be produced in the zener-biased amplifier stage as the tube nears and reaches cutoff and the current through the diode decreases to a level below the specified minimum current (this may not even be specified in some datasheets).
The zener diode should be bypassed by a capacitor having a low reactance at the lowest frequency of interest. The reactance should be no more than 10% (and preferrably as low as 1%) of the of the product of the average zener current multiplied by the zener voltage.
For example, a plate modulated 807(1) under carrier conditions of 475V (all voltages ref GND.) on the plate, 325V on the screen derived from a 25K resistor connected to modulated B+, 0V on the grid, and +75V on the cathode (due to the zener diode) running 80mA plate current, 6mA screen current, and 3.5mA grid current (89.5mA cathode current) operating on a frequency of 1.75MHz has an impedance from cathode to ground of 75/.0895 or 838 ohms and would require a capacitor from the cathode to ground with a reactance of 1% of 838 ohms, or about 8.4 ohms. At 1.75MHz this is close to 0.01uF (Xc=9.09 ohms).
Because an 0.01uF capacitor may have some unwanted inductive reactance or series resistance at this frequency, adding a 0.001uF cap in parallel will help reduce distortion due to these effects. The effects become worse and change unpredictably as the frequency rises. In order to avoid the effect in inexpensive capacitors wherein the capacitance changes with the instantaneous voltage, a high quality plastic capacitor should be used wherever possible.
Under modulated conditions where modulation may occasionally (try to) become more than 100% in the negative direction and the tube may be (suddenly) cut off, and assuming a lowest modulation frequency of 175Hz (to make it simple), the capacitance required to equal 84 ohms (1% of the 838 ohm impedance) becomes 100uF.
An electrolytic may be used as long as its similar and additional undesirable properties are also taken into account. The 100uF electrolytic capacitor will have unwanted properties at most audio freqencies, so something smaller like a 10uF high quality plastic capacitor should be placed across the 100uF capacitor as well. So the 0.01uF and 0.001uF capacitors may be placed at the cathode pin, and the larger 100uF and 10uF capacitors for audio frequencies may be placed more conveniently near the zener diode.
This may seem like gross overkill and extra trouble but it is proven in the field of broadband wired communications that capacitor quality and construction directly affects the amount of distortion of a complex waveform. In a sensitive system using up to 128 or 256 carriers and DMT+16QAM a cheap capacitor substituted by a careless contract manufacturer will destroy the performance of the system by introducing phase and amplitude variations. The variations, if severe enough, look like indeterminate points in the QAM constellation. The problem is exacerbated when the signal is subjected to clipping. For reference, the impedances in typical ADSL2+ circuits is about 100 ohms and the frequency range is from about 40KHz to 2.2MHz with peak amplitudes of 10-20V, so the case for decoupling and quality components is relevant to the Zener-based cathode bias discussion. In the RF amplifier, the performance will not be destroyed, but the quality of the obtainable performance will be diminished as the capacitor introduces the distortion in the cathode circuit.
Certainly the cited conditions for the 807 have their own issues, such as deriving the screen voltage by a resistor from the modulated B+ supply. These are not addressed herein but are assumed to be corrected by the devout enthusiast to gain, by means of proper design, maximum linearity from the stage.
Why has this not been expounded upon in the ARRL handbook or other volumes? Probably because the topic of using Zener bias and the study of the artifacts produced by it in its usual embodiment is so far removed from the topic of what has only recently been learned in the laboratories of wired broadband where distortion is minimized in order to reach ever-increasing demands for performance at greater distances, that those writing the articles are not aware of these matters.
Bypassing the zener diode and selecting the better components for the job can greatly reduce the total distortion caused by the nonlinear voltage characteristic of the zener diode (2-3%) for straight RF. The same applies to the audio frequency current variations and thus the two sets of capacitors.
So far, the discussion of the impedance of the zener diode bias circuit has been referenced to carrier conditions. Under 100% sinewave modulation, the cathode current increases, therefore, capacitor values could also be increased by the same factor to accomodate the reduction in impedance.
It has been stated that people have been using unbypassed Zener diodes as cathode biasing devices for years and never received a complaint. This is entirely believable considering that the signal path, receiver nonlinearity, and the 3-6% distortion inherent in even a good speaker would make it impossible to hear this at the far end. It can be measured at the near end, and that is the point of this article -to suggest one way to reduce distortion in the RF power amplifier which uses Zener bias.
I may have some error in here but I have done my best to explain this point.
(Perhaps I should submit this to QST for April 1 for non-believers.. *-P
Patrick Jankowiak KD5OEI