A Class A Project
Audio Meltdown: Cooling the Core
Recently I told a friend I was building a one off 60W Class A amplifier.
He remarked that Class A amp's got pretty hot; how much heat did it give
off? I replied that two channels produced about 600W, and that I
was planning to water-cool the beast. He looked at me strangely,
seemed to gather his wits, and asked how the hell I was going to do that.
Carefully I explained in some detail, (mildly surprised at his strange,
nervous tic), and he went away shaking his head.
A few days later he rang, telling me it had taken some time for all this
to sink in. Why didn't I write an article telling others why I decided
on this crazy project, with a discussion of the engineering problems,
any 'artistic' considerations, and how I did it? The following is
my attempt to do just that. It is a strange story, involving quite
a bit of engineering, a strange audio vision, (work that out!), and a
quiet, grim expression!
For some years I have read every bit of hifi/audio I could lay my hands
on. I built my first amp at age 12 - a 12W push-pull solid-state
amplifier using Fairchild semiconductors. At the time I thought
it was pretty good, but now I realise it sounded dreadful. In 1963
transistors were new technology, and certainly no-one my age would be
building a valve amp. They were old technology, passé and
uncool. Hmmm. Strange how times change...
As a child I learned piano, and later the pipe organ, and nothing sounded
as good as the real thing. But after some time, I forgot about hifi,
and then after a career in the military, I decided to pursue a childhood
dream and again took it up.
This time I read, tinkered, built, and listened. In particular, I listened.
I went to University libraries and consulted all sorts of old and new
engineering texts. I even bought some impressive books of my own.
I met the frenetic, live concert guru, and listened to his informed tirades
about professional audio. I watched the sorcerer as he muttered
his incantations, and marveled at his 100,000 Watt compositions, even
if he sometimes vanished from view in the smoke! With his advice
ringing in my ears, I spent time building strange solid-state push-pull
amp's with many output devices, and learned the basic engineering of transistor
technology in powerful amplifiers. I stumbled through the silicon
cemetery until I came to recognise the tombstones. I asked others
with better hearing to listen for me, and slowly, (thanks Tony, Gary,
and Danny!), I too began to hear the sorts of things they listened for.
I listened to Peter H's single-ended Cary and Peter A's single-ended EAR
859 and was deeply impressed. Slowly the basic truths of the engineering
compromise grew clearer - everything lies somewhere along an engineering
line, with more of this giving less of that. Eventually, I began
to see how amplifiers were designed, and realising that guitars with valve
amp's sounded better than those with solid-state amp's, I saw that the
human ear seemed to favour the sound of valves. And in all this,
I cautiously wandered from the solid-state camp towards the valve camp,
although I could see that each had its place, yea, even in audio.
And what was this special favour? Somehow the valve amp was sweeter, and
less mechanical, than a solid-state amp. At full pelt it sounded
better, too - ask any musician. Voices were much more natural than with
solid-state amp's, even good ones, and on fabled single ended valve amplifiers
the voice seemed to issue from well in front of the instruments - right
where it should be - giving a sensation of depth and 'being there'.
This seemed absent with solid-state amp's. In short, the solid-state amp
stripped away the emotion of the performance. If anything, the valve amplifier
seemed to enhance it.
From the design standpoint I came to see that valves were magnificent
voltage amplifiers, and transistors were marvelous current amplifiers.
Audio amplifiers require both types; so, if it were possible to keep the
best qualities of the valve, why not combine them? Thus was born
the Glass Harmony, a unique hybrid amplifier.
The Glass Harmony
I spent about three years getting the Glass Harmony right. I started with
Nelson Pass' Zen amp but soon turned the design on its 'ear'. I
joined a valve audio email discussion group, and began friendships with
enthusiasts from different countries. This strategy added to my
knowledge and informed me of technical developments overseas. My
first attempts, as associates will remember, were less than memorable.
(Fortunately I realised the journey was more important than the destination!)
The Glass Harmony now uses a simple triode to amplify the voltage some
43 times (32.6dB) and a single ended MOSFET output stage to amplify the
current some 3000 times. Just getting the valve to talk to the transistors
was very difficult, and took about 18 months. But when I listened
carefully, I had the feeling that with a little more tweaking the sound
would be sensational.
About a year later I had it tweaked properly and ready for long listening
tests. But it was rather costly to build, and the heat problem forced
me to use four heatsinks which in turn meant two boxes, a 'monoblock'
for each channel. Although each channel dissipated 150 Watts of
heat, the audio output was a meagre 28 Watts. This is a horrific
18% efficiency - less than an automobile engine - and it was an affront
to any engineer. But the sound was ethereal, with shattering resolution
and absolutely no trace of intermodulation distortion. Progress at last!
The amplifier could drive almost any loudspeaker, and its sonics set me
thinking. It had promise. I had seen brochures on the $US50,000 WAVAC
Japanese amplifier, which used an 833 triode and reputedly gave a luscious
100W per channel. 28 Watts was pretty small, so why not go bigger???
So I fleshed out the schematic and did the sums. Instead of four
output devices I could use six. The driver stage could use a 30%
boosted current, to cope with the extra drive requirement. The voltage
amplifier need not change at all. The low voltage supply would need
to rise from 50Vdc to 70Vdc, and idle current would rise from 3A to 4.4A.
Dissipation per channel would then be around 310W - and power would be
about 62Wrms into 8 ohms. All that energy wasted was certainly extravagant,
but a running cost of 7 cents an hour was still affordable. For reasons
unexplained, the Glass Harmony's 28W sounded like a 150W MOSFET push-pull
amp, so this should sound like a Baldwin Steam Locomotive! The big
problem was the heat.
The Power Supply
The power supply needed careful thought. The valve B+ needed 325Vac, the
filaments 6.3Vac reduced by dropper resistors, and the output stages required
8.8A at 70Vdc.
Much experimentation revealed that line level, indirectly heated triodes
sound best with AC on the filament, providing the hum can be banished.
Fortunately this is not too difficult, and a couple of 100R resistors
from each side of the filament to ground usually do the trick. Since
the cold resistance of a valve filament is about ten times less than when
hot, a huge current surge rips through the filament at switch-on, shortening
its life. It is a good idea to power the filament via a resistor
dimensioned to 50% of the hot filament resistance. This reduces
cold surge current to a factor of about three, thus prolonging filament
The high tension valve supply has been the subject of much experimentation.
The Glass Harmony uses a R-C-R-C-R-C filter configuration, (resistor input,
double pi filter), to very good effect. The valve draws 3mA; a 2mA
bleeder is used. However, the optimum is a shunt supply driven from
a series current source. I admit, however, that this solution is
quite complex, and for this project I decided to replicate a supply which
works very well in my convection cooled Glass Harmony. The one refinement
is the final capacitor; 33uF of SCR metallised polypropylene, which gives
a uniform, low output impedance across the frequency range and very effective
AC grounding. The two other capacitors are quality Nichicon 100mF
electrolytics, and the resistors, in order from the rectifier, are 2K2,
4K7, and 10K. The ripple content at the top of the plate resistor
The low tension, 70Vdc supply calls for some heavy industry. The
power transformer, a 625VA unit, is very heavy, and rated at 9A continuous
56Vac. With full wave, 25A bridge rectification into 2 x 22,000mF
capacitors, two separate 25H chokes, and two further 22,000mF capacitors,
we finish with 70Vdc at 8.8A continuous rating. The chokes have a DC resistance
0.33W, drop 1.5V each, and run quite hot. They are vitally important
for reducing ripple, which measures less than 10mVrms, and has negligible
bad effect on the sonics of the amplifier.
Cooling the Core
In the late eighties I built a Crescendo MOSFET amplifier out of the European
publication Elektor. To improve the sound I set the idle current at 150mA,
(10 Watts), per device, and it ran very hot. After trying fan cooling
and finding it noisy, I chose water cooling, using the purge pump from
a dishwasher and 20 feet of buried pipe as a heatsink. I mounted
a level sensor on the water reservoir, and fitted a solenoid tap to top
up any lost water from the house water supply. It was all automatic
and worked flawlessly with complete reliability. The semiconductors ran
at below room temperature - about 150 a 400 day - and no wonder, flow
was above 200 litres per hour! These were impressive figures, and the
beast lives to this day. I'd hit on an idea for cooling my new Class A
The 60W Glass Harmony would dissipate about 610 Watts continuously - more
than five times that of the Crescendo. Water could easily do the
job, and now I had ideas of doing it better this time. I needed
to improve the efficiency of the cooling, with a reduced flow. This
would be suitable for a total loss system which could be made reasonably
portable and much simpler.
A Shower an Hour
610 Watts is around 150 calories of heat each second. For a 150
temperature rise, we need a flow of 10 cc's per second, or 36 litres per
hour. Using only 5mm internal diameter tubing, this is a flow speed
of about 1.9km/hr; quite slow. You'd use 36 litres of water in a
10 minute shower, so although wasteful, a total loss cooling system is
not such an outrageous extravagance. And it's simple; this system
would be 'portable', as all I would require in a strange home would be
a kitchen sink, some rather long hoses, and a lockable gizmo for the tap
to prevent people turning it off!
The heatsink design would necessarily be more refined than my Crescendo
system, where the rush of cooling water was just audible. Later
on I could always close the cooling loop with the use of a small pump
and buried pipes. Doubtless the garden would be quite grateful for
any extra water!
Next I designed the circuit board. I used the Glass Harmony design
as a start point, and over many tens of hours within a month I had developed
a simple, elegant board with just one link. PCB design is not easy,
particularly with single sided boards, and takes countless iterations
to achieve any kind of elegance. The best boards I have seen are
found in HP and Tektronix equipment - they are excellent models.
Each design iteration must be printed and carefully studied. Eventually
improved ways of doing the same thing appear in the mind's eye.
I have found that a well designed, elegant board is easier to manufacture
and assemble and usually pays dividends in circuit stability and ease
The tube, B+ and AC filament supplies, and all signal processing are included
on this board. All six devices, flat pack MOSFET's, are mounted
in line on a thick aluminium strip about 17cm long. Each MOSFET idles
at 50 Watts dissipation, between 33% and 40% of their rated maxima at
250 Celsius. I welded the 4mm aluminium strip perpendicularly to the middle
of a 100mm x 40mm aluminium box section. Capped at each end, and
with an identical strip on the other side for the second channel, the
box section would look like a submarine conning tower complete with stabilisers
- though with the water running through it, not around it. Now the
problem was to equalise the water flow for all 12 devices to achieve even
cooling across all twelve devices.
I hit on a simple idea. A 10mm aluminium tube lies at the bottom of the
box section. Another is supported at the top. Both are sealed
at one end; water enters the bottom tube at the other end, and exits from
the top tube. Water jets from six tiny holes drilled vertically
along the length of the bottom tube to coincide with the positions of
the MOSFET's. These holes are dimensioned so that each passes the
same quantity of water for the chosen flow rate of 10 cc's per second.
The top tube is identically drilled. Thus water flows into the bottom
tube and issues equally from each hole, moving vertically up the box section
to the upper tube. In this way a small flow of water is evenly warmed
along the full length of the heatsink, keeping all devices at the same
temperature. Any increase in water flow above 10 cc's per second
may well unbalance the flow, but no one device heats more than 150 above
the incoming water temperature. A temperature-sensitive trip switch bonded
to the sink quickly shuts off the power should the beast take thermal
The two tubes are inserted into the top and bottom of the box section
with O rings and secured with a locating clip. This enables quick
removal for cleaning if necessary. The cap at one end of the box
section is 5mm thick to accommodate an O-ring seal around each tube.
This is simple, effective construction with a minimum of aluminium welding.
By mounting the box section so that the tubes are at the rear of the amplifier
cabinet, any water leaks remain external to the cabinetry, preventing
unthinkable accidents with mains power and water.
This covers the major design issues associated with the 60W Glass Harmony.
Once built, the amplifier proved itself extraordinarily refined, and yet
thunderous it sounds like the audio equivalent of the steam traction
Do we have any steam buffs . . .?
© Copyright Hugh R. Dean 1999
All rights reserved