To All from PA0SE
The following subject may have been discussed on the reflector before but I
can't remember it.
Class D and E final amplifiers have high efficiency because they produce square
waves. When the voltage between source and drain of the FETs is
high, current is zero; when current flows voltage is almost zero…
Dear Dick, LF Group,
The first thing is that class E refers to
a very specific circuit configuration – see Alan G3NYK’s web pages
at http://www.alan.melia.btinternet.co.uk/classepa.htm
for a discussion of the concept.
The basic Class D amplifier configuration
comes in two flavours; current-fed and voltage-fed. The voltage fed type drives
a load resistor with a square-wave voltage, through a series-tuned tank circuit
(ideally with infinite loaded Q), resulting in a sinusoidal current in the load.
The current-fed type drives a load resistor shunted with a parallel tuned tank
circuit (again, ideally with infinite loaded Q) with a square wave current,
resulting in a sinusoidal voltage across the load. In both cases, the switching
devices produce a square wave by alternately reversing the direction of the DC
supply voltage or current at the desired output frequency. The voltage-fed type
uses a conventional DC supply, decoupled to ground in the normal way with low
impedance capacitors. A constant-current supply is not a convenient thing to
produce, so in practice the current-fed type uses a conventional low impedance
DC supply with a high-impedance series choke, which is a satisfactory
approximation.
As Dick points out, one could not apply a
square-wave voltage to a tank circuit with a shunt capacitor without having
huge (theoretically infinite for ideal components) peak currents and power
losses, as you tried to charge and discharge the capacitor instantaneously (I =
C.dv/dt). Likewise, if you tried to apply a square-wave current to a series-tuned
tank circuit, you would have huge voltage spikes from trying to reverse the
tank inductor current instantaneously (V=L.di/dt), with similar disastrous
results. So one cannot pick and choose – a voltage switching circuit must
drive a load with series inductance, to keep the current to finite levels,
whilst the current switching version must have shunt capacitance to avoid the
voltage spikes.
Note that the function of the resonant
tank circuit is NOT primarily to “filter out harmonics”, but to act
as an energy storage “flywheel”, which allows the conversion of the
square wave drive into a sinusoidal output without loss of power. It also
achieves desirable switching conditions for the power transistors – in the
voltage fed case, the load current is zero when the voltage reversal occurs,
whilst with the current fed circuit, the voltage across the load is zero when
current switching occurs. This means that, even with fairly slow switching of
the transistors, transient power dissipation is low.
The “Decca” circuit is a clear
example of a voltage-fed class D amplifier with a series L and C tank circuit.
But with the push-pull designs typified by the G3YXM and G0MRF circuits, things
are not so clear. For a start, neither has a resonant tank circuit as such.
However, the output low pass filter is “Quasi-parallel-resonant”,
in that it has resistive impedance at the output frequency (or almost so in the
G0MRF case), and a capacitive impedance at higher frequencies, so it acts in a
somewhat similar way to a very low Q parallel tuned tank circuit. This suggests
a current-fed configuration, and indeed the ‘YXM design has the DC supply
fed to the PA transformer primary via a choke, as one would expect. However,
the ‘MRF has the transformer centre tap decoupled to ground via a (fairly)
low impedance capacitor. So at first sight, it would appear that each time
switching occurs, one MOSFET or the other must try to transfer charge instantaneously
from this capacitor to the filter input capacitor through the transformer,
resulting in a large transient current spike. However, in reality, life is more
complicated. The transformer has substantial leakage inductance, which prevents
a very rapid rise in current – so the circuit is somewhat in a grey area
between voltage- and current-fed class D. The trouble is that the leakage
inductance resonates with the various capacitors, and the other inductors in
the circuit, to produce several different resonant frequencies in the output
network, all of which are driven by the switching transients. This results in a
very complicated waveform with high amplitude ringing at high frequencies. Although
the circuit “works” in the sense that it produces substantial 136kHz
output, it results in high power losses in the PA components, if you are
unlucky. I suspect that the reason some people have had problems with the MRF
design, while for others it works well, is that the overall circuit behaviour
is quite sensitive to the exact amount of leakage inductance, and so on the exact
physical construction of the transformer. When the ringing occurs at just the
right frequency, the circuit could act as a class ”F” amplifier,
but that is another story…
The mods I came up with for the ‘MRF
circuit aim to convert it into something nearer the “textbook”
current fed configuration. Particularly, I tried to reduce the leakage
inductance of the transformer – the original design had a leakage
inductance of about 2.5uH, which has a reactance of the same order as the load
impedance of a few ohms on the primary side of the transformer, so is bound to
substantially affect circuit behaviour. The new transformer I described reduced
this to about 0.25uH. Eliminating the “decoupling” capacitor and
increasing the feed choke inductance makes the drain current a fair
approximation to a square wave as desired. The reduced leakage inductance and
the introduction of the “snubber” RC networks get rid of a lot of
the HF ringing, and “simplify” the circuit behaviour. The low-pass,
Low-Q output filter network means that the PA voltage waveform is not the ideal
sinusoidal form, but seems near enough not to make a lot of difference. I have
experimented with higher-Q tank circuits, and also with voltage-fed
arrangements with a series LC tank, which do improve the waveform. However,
since the circuit is about 90% efficient anyway, there is not a great deal of
improvement to be had, plus the higher loaded Q circuits have the disadvantage
of needing bigger, lower loss, output Ls and Cs which must be more accurately
adjusted in value. Also, the low-pass configuration gives better reduction of
harmonics; even if a higher Q tank circuit is used, an additional LPF will
probably still be needed. Although the antenna itself will give substantial
reduction of lower order harmonics, the switching transients inevitably mean
that significant harmonics are generated throughout the MF and HF range at 136kHz
intervals. It is unlikely that the antenna-plus-loading coil will give good
rejection throughout that range.
Cheers, Jim Moritz
73 de M0BMU
