April 30, 2002
Part 4: Power Amplifier
DogZilla uses a hextet of venerable 807's as output valves (V5-V10). This tube has
been described as "the valve that won the war," because of its widespread use by the
Allied forces during WWII. Its reliability and ability to take a beating are almost legendary, and
amateur radio operators have been know to squeeze out hundreds of watts by running them in oil baths.
I had originally intended to push them to the max, in Class B with HT around 800 volts, for maximum
output of over 300 watts from the sextet. However, some experimentation convinced me that the
sonic advantage of Class AB2 far outweighed the extra 1.8 dB of output power, and the
present design has been finalised at the more reasonable 200 watt level.
Starting at the back-end and working forward, we have the transformer (Hammond 1650R). Originally I
had two of them, wired in parallel, both primary and secondary. However, I found that these are so
conservatively rated (for hi-fi use) that a single unit works just fine at over double the output
power in an instrument amp application.
The secondaries are wired through an impedance selector
switch, such that either the "16-ohm" or "8-ohm" configurations
can be used. However, the actual recommended speaker impedances are 8 ohms and 4 ohms, respectively.
This means that 1/2 of the nominal impedance is reflected back to the combined primary, or about
2500 ohms plate-to-plate. Considering that the recommended load impedance for a single pair of
807's in Class AB2 is around 7300 ohms, this ends up providing an excellent match for
these devices under these operating conditions.
As mentioned in the control supply section, the output transformer
secondaries are connected to a dummy load resistor before the HT section is engaged, and for
about 1/2 second afterwards, to prevent speaker pops on power-up. Not shown on the schematic is
a small 4-ohm monitor speaker included on the prototype, in series with a small toggle switch and
a 125 ohm, 20w resistor to limit maximum power to about 200 mW (or 1/1000th of the
full output power!).
Schematic, Power Amplifier
(Note: In this and subsequent signal-path schematics, the voltage or power shown in
blue at the output is the maximum peak level available at that stage; it therefore gives an
indication of headroom. The voltage shown in red at the input is the peak level required to
obtain maximum undistorted output, and therefore indicates the gain or insertion loss of that
section. Level controls, if applicable, are assumed to be in the maximum position.)
The output circuit is a completely traditional push-pull topology, in true pentode mode (fixed
screen voltage). There are a few caveats to be observed, however. Remember that the 807 is quite
content to operate at RF frequencies, and it's quite eager to go into RF oscillation at the
slightest provocation. The following points were found (mostly empirically) to assure stability:
- Do not simply tie the screen grids together. This is a guaranteed recipe for glowing plates and
enough RF hash to blank AM radios for blocks around. Each screen should have its own
"grid-stopper" resistor, right at the socket, with no capacitance to ground at the
screen terminals. I found that 240-ohm 2W metal film resistors effectively suppressed the
tendency to become an autodyne oscillator. Each bank of grid-stoppers connects to a bus rail, which
in turn is wired to the output of the screen regulator.
- A single small bypass capacitor from the junction of the bus rails directly to chassis
ground (as indicated by the chassis ground symbol) further improves stability. (Note that
this is one of the few cases where the usually recommended "star-grounding" approach is
not the way to go.)
- Similarly, the control grids should each have their own grid-stopper. At present, 1.2k is in
use, and results in unconditionally stable response. Once the amplifier is completed, I may
experiment with lower values to improve linearity when the grids start drawing current.
- Small 10-ohm resistors in each cathode lead are used for individual current metering. You
wouldn't think that these resistors would have much to do with stability - or at least I didn't
think so. Yet there were persistent RF artifacts on the leading edge of the waveforms, just about
at the Class AB1 switching point. These came and went at whim, were sensitive to
lead dressing, and just generally made a nuisance of themselves. The culprit ended up being the
miniscule inductance of the 10-ohm metal-film sense resistors, combined with the wire runs to the
meter box. There are two possible solutions to this electronic gadfly:
I opted for the second solution, since I'm quite leery about the resistance stability of
- Replace the cathode current-sense resistors with carbon-comp types.
- Bypass the metal film resistors with small (.001 uF) mylar capacitors
- Keep leads to grids and screens dressed out of the way, as much as possible. Complimentary
signals (i.e. plate transformer primaries, filament lines) should be run together as twisted pairs.
I even went so far as to strap together each of the three pairs of HT wire running to the anode
plate-caps on the tubes.
- Run the cathode grounds to a single common point, and from there to the central "star ground"
with a heavier-than-usual wire (e.g. 12 gauge).
- Be aware of maximum grid-circuit resistance on the output valves. This is given as 30k for the
807 in Class AB2 mode. Or, for three in parallel, this would work out to 10k. On the
surface, it would appear that I'm violating this maximum by a factor of 2:1, since R47 and
R48 are each 20k. However, the effective dynamic resistance of the voltage follower tubes
themselves will be on the same order of magnitude (under quiescent conditions), and will be
effectively in parallel with the cathode load resistors. In that sense, we're running right at
the recommended maximum effective grid resistance. Testing has shown no sign of difficulty in
this regard, but if you're queasy about this you can use paralleled 6SN7s (or a larger triode such
as 6AH4) and decrease the cathode loads appropriately.
Voltage Followers and Biasing
The control grids of the output stage are fed by direct coupling to a pair of voltage follower
stages consisting of the two halves of V11, a 6SN7GTB. To optimise voltage swing and output
current capability whilst staying well within the recommended maximum dissipation, positive supply
is taken from the +220V auxiliary supply. To allow the necessary substantial negative excursions
of the cathodes, the cathode return is taken to the -220 volt auxiliary supply. This also insures
that the grid voltages on the finals is well-established long before the main B+ supply kicks in.
Resistor R49 and capacitor C17 decouple the plate supply to the voltage followers.
Two identical biasing networks referenced to the regulated -150 volt supply provide fixed grid
bias, adjustable via trimmers R53 and R56. With reasonably matched sets of 807 tubes,
the bias point for each valve (nominal 30 mA quiescent, and about 90 mA under full power) will
normally only vary by a few milliamps in either direction. Burn testing and experimentation with
different effective line voltages have shown the final result to be eminently stable and reliable.