Updated October 2011.
Download the new page to replace the page from 2006 because some links
won't work and there is some new information.

8585 Amplifier in 2004 with 8 x GE 6550A.


Contents of this page....

Picture of 8585 on a bench.

Fig 1,
Schematic of one channel of 8585 amp, 2006, input and driver stages.
Fig 2,
Schematic of one channel of 8585 amp, 2006, output stage.
8585 General description,
Tube choices,
Local output stage cathode FB and B+ regulation,
Global and Local NFB,
Power output,
Max Vrms output for various RL,
THD, Frequency response,
Speaker load value and Amp Output Resistance,
Inbuilt line level preamp,
Tubr layout,
Fuse replacement, Fuses within amp,
Fig 3, Schematic of 8585 PSU,
Fuse locations,
Turn on delay,
Protection circuit function for owners,
LEDS at front panel of amplifier,
Power consumption,
Amplifier weight,
Fig 4, Graph for power output vs load for 4 x KT90.
Fig 5, Graph of harmonic content for middle setting of volume control.
Fig 6,
Graph of harmonic content for maximum volume control setting.
Fig 7, Graph of 3 THD curves of 8585 THD for comparisons.
Fig 8, Graph of harmonic content for 12AU7 input preamp stage.
Fig 9, Schematic of 8585 amp as supplied in 2004.
Fig 10, Graph of  THD for 2004 version of 8585 with 4, 6 and 8 ohm loads.
Fig 11, Schematic of protection circuit and explanation.
Fig 12, Schematic of 8585 in block diagram form.
Many notes on NFB and PP basic operation,
Fig 13, Graph for loadline analysis for 4 x KT90 used with 12.5% CFB.
How the loadlines are drawn, Step 1 to Step 16.
Fig 14, Picture of 8585 under chassis. Following notes.


Use of the 8585 amp without reading all these notes is possible but



Fig 1,

Fig 2,

8585 General Description.
The basic design of the push-pull 8585 was conceived in 1995, following
several years of trials with various PP topologies to get the most accurate,
dynamic, and subjectively pleasing sound, along with the best measurements
possible with the least use of corrective circuitry known as global NFB.
The fundamental output stage circuit design of the 8585 is based on the
use of beam tetrodes or pentodes with renowned Acoustical Connection,
famously exploited
Peter Walker in the 1950s in Quad II amplifiers,
and continued in the much later Quad-II-40 amps with the same circuit
by Andy Grove, and in Quad 80 amps.

The performance of my 8585 output transformers surpasses anything
built for Quad amplifiers.
The October 2006 version 8585 is an integrated amplifier, with 6 switched
line level inputs, balance control, line level SET preamp amp, and volume

Incremental improvements since 1995 evolved the present fine sounding amplifier.

The only change to the 8585 schematic since 2006 has been in the PSU
where the shunt regulator using 4 x 75V zener diodes in series has been
by the regulator shown in the revised PSU schematic below.

The string of 4 seriesed zeners were found to change their zener voltage
to less than the 75V nominal value, and probably from repeated heat cycles.

Tube Choices
There are presently 4 x KT90 output tubes per channel, but the circuit allows
the use of all the main octal based output tubes such as 6V6, 6L6, 5881, EL34,
6CA7, KT66, KT88, 6550, KT90, and now KT120, in 2011.
However, only the 6L6GC, 5881, KT66, KT88, 6550, KT90 and KT120
are able to be plugged in and biased to suit the existing circuit.
The use of EL34 and 6CA7 are usable, but require minor alteration to
the circuit to ensure the correct grid bias voltage these 2 tube types.
For 6V6 the anode and screen supply must be re-wired to provide an
anode supply of no more than +350V and a screen supply of about +300V.
There is a tap on the 180Vrms A-B HT winding which gives a lower B+.

Local output stage cathode feedback and B+ Regulation.
The output stage for each channel has 4 output tubes, two for each side of the
push-pull circuit, and these are configured to allow 12.5% of the total anode to
cathode signal voltage to be fed back locally to the output tube cathodes from
a tertiary winding on the OPT.
With a 5 ohm load, the 12.5% of local NFB equates to about 7 dB of local NFB.

The reason for applying two lots of negative feedback, with one lot in the form
of local cathode FB  in the output stage and the other globally is because the
local FB in the output stage has a very favorable effect on the spectral content
of distortion harmonics which are reduced to lower levels than triode or ultralinear
connection but without paying a penalty of requiring too high a drive voltage to
the output stage tube grids. This local NFB around as short a path as possible
results in minimal secondary IMD distortion products being produced. 
Therefore less global NFB is then required to make up a total of less than 20dB
and it is thus easier to achieve unconditional stability without requiring such
critical stabilizing techniques.
If the output tubes were connected in pure beam tetrode mode the harmonic
spectra would remain more complex and the amplifier would be more difficult
to stabilise, even though 20dB of global NFB would reduce all these harmonics.
The local CFB prevents considerable production of unwanted harmonics before
needing global NFB to complete the work. To accommodate the lesser amount of
global NFB, the input driver stage has been designed to give about 1/3 of the
usual amount of THD for the wanted drive voltages.
The anode supply is +500Vdc nominally using an 800 VA toroidal power
transformer, silicon diodes, and large value capacitors, with one filter choke.
The output stage screen supply is +330Vdc, and actively shunt regulated.
The output tubes have fixed grid bias voltage applied, once the level of
cathode current is adjusted, see below. 

The output stage is driven by a "long tail pair" differential amp using EL84/6BQ5
strapped as triodes and supplied with DC via a centre tapped choke and with 8.2k
from each end of the choke to the EL84 anodes.
(A choke is a winding of wire around special grain oriented silicon steel core aka
GOSS and it has high impedance to signal currents, but has low resistance to allow
a high direct current anode supply easily ).
The novel and little used choke loading method in all my best PP amps reduces
dynamic signal current change in the EL84 to about 1/3 the usual level compared
to use of a pure resistance supplying Idc to the tube.
The EL84 anode voltage swing is increased with the choke and the choke avoids
the EL84 having to waste its efforts to make signal power where it is not wanted
or needed in low ohm value resistors bringing DC to the anode.
EL84 anodes are loaded mainly by the pair of 100k bias resistors to each side of
the PP circuit. Thus each EL84 anode has a capictor coupled load of 50k plus
the choke impedance which is many times higher than 50k at most audio frequencies.
In this case, the use of the choke was found to reduce THD to about 1/3, or -10 dB,
compared to conventional circuits using cheaper dc carrying resistance loads.

Global and Local NFB
The input stage of the power amp is a 12AU7 with both halves paralleled.
The dc supply to this input stage is via a pair of 100k resistors in parallel
which means that although the anode signal voltage is up to about 7Vrms, there is
not a huge anode current change in the parallel triodes.
The 12AU7 is set up as a single ended triode and there is 12 dB of "global"
NFB from the output transformer secondary applied to the cathode via a
low resistance network.
At clipping the amp requires about 2Vrms applied to the 12AU7 input grid
and there is 1.5Vrms of fed back voltage applied at the cathode and the
difference between the input voltage and fed back voltage is 0.5Vrms and this
is amplified to make up to about 7Vrms to power the following Long Tail Pair
driver stage ahead of the output stage.
This basic method of globally applied NFB has been used for 60 years at least,
and there is a total of 19dB of NFB applied consisting of 7dB of local output
stage NFB which reduces the output impedance of the output tubes to about
2 ohms.
The 12dB of global NFB reduces the 2 ohms down to about 0.4 ohms
which results in a damping factor of greater than 10 for a 5 ohm load.
The distortions in pure beam tetrode connected tubes is reduced by the
output stage NFB by about 12dB and then the global NFB further reduces
the distortions by another 12 dB.

Without using any NFB the amplifier would be useless because the output
resistance would be many times the speaker impedances and the distortions
would be both audible and objectionable.

Power Output
The output power is 84 watts into 5 ohms at less than 0.3% thd which is at
the onset of overload or clipping.
Power is class AB1 with an initial 20 watts of pure class A.
There can be 57 watts watts into 8 ohms at less than 0.2% thd, class AB1,
with the first 40 watts being pure class A.
Refer to graph BELOW for power output vs load to examine the maximum
power output levels possible.
The use of 4 ohm loads and KT90 gives 100Watts.

Maximum Vrms signal levels into loads with the 3 output terminals
strapped together are as follows:-
33 ohms, 18W, 24.4Vrms, 34.5 peak volts,
15 ohms, 32W, 22Vrms, 31.1Vpk,
8 ohms, 57W, 21.3Vrms, 30.2Vpk,
5 ohms, 84W, 20.5Vrms, 28.9Vpk,
2 ohms, 112W, 15Vrms, 21.21vpk,

Three pairs of Quad ESL will have an impedance varying from
11ohms at 60Hz to 0.6ohms at 18kHz.

This may seem like an impossible speaker impedance to drive but 90% of
the power in music is within the 60Hz to 3kHz band where Z is between
11 and 3 ohms, and only a tiny amount of power is needed to produce
frequencies above 3 kHz, and since the 8585 has a power ability equal to
4 Quad-II amplifiers in parallel and considering that stacking the Quads will
increase their power sensitivity approximately 3dB, then the 8585 will not
have more difficulty driving 3 pairs of ESL57 than a single pair of Quad-II amps
will have driving one pair of ESL57 which are regarded as a compatible match
of amp and speaker since both were designed in the same era by the brilliant
Peter Walker.

Even with just a 0.6 ohm load there is 60 watts of power available from the
8585 which means that the output current ability without clipping is about
10Arms, or 14 amps peak.

At any load above 3 ohms,
1 watt of power, thd < 0.02%,
4 watts, thd < 0.03%,
16 watts, thd < 0.1%.
80 watts, 5 ohms, thd < 0.3%,
See the Graphs BELOW with following notes for Harmonic Products
to view levels of THD with 5 ohms.

Frequency Response.
The frequency response for 1 kHz and 5 ohms is from 14 Hz to 65 kHz,
at 80 watts, limited by saturation of the OPT at LF, and bandwidth limiting at HF.
The response widens from about 5 Hz to 68 kHz at ordinary loud listening levels.
With a test load comprising and RC series network of 0.5ohms in series with 6 uF
and with 5 ohms shunting the RC network there is no peaking in the response.
Pure capacitance loads of any value between 6uF and 0.1 uF may be connected
across the output terminals with the HF response showing less than 1dB of peaking
at 20kHz, and not more than 6dB of peaking between 20kHz and 200kHz,
so the amplifier is stable with any value of C load.
Tests were done on C loads at low output voltage levels of 1Vrms output to ensure
that the diminishing impedance of C loads at HF did not cause the active protection
circuit to activate because of excessive dc anode current draw from the power supply.
For example a 2uF capacitor has 2.48 ohms of purely reactive impedance at 32kHz
and if the output voltage level was raised to equal that with 2.5 ohms of pure
resistance at clipping at 1 kHz, the amp will shut down within a couple of seconds. 

Speaker and Amplifier Output Impedance.
Any type of load is permissible, including dynamic, ribbon or ES speakers,
and the amp will drive any load above 3 ohms for an average power level
of 5 watts, which allows for peaks in the music to be 50 watts or more.
There are 3 pairs of output terminals arranged so there is 0.6 ohms series
resistance from each output to the common internal connection at the amp
so that if 3 pairs of Quad ESL are connected with each pair taken to each
pair of output terminals then each pair of speakers is fed via its own separate
0.6 ohm resistance.
The output source resistance in series with the ESL is thus the amplifier Rout
of 0.4 ohms plus the 0.6 ohms giving a total of 1 ohm which is recommended
for Quad ESL57 in order to get a flat response to 20kHz.
The 8585 will thus mimic the action of the Quad II amp but give greater
stability and lower distortions.
For normal speakers where the lowest amplifier resistance may be desired,
all three active terminals to each channel may be strapped with a wire beneath
the 3 active binding posts, and thus Rout = 0.4 ohms + 0.2 ohms = 0.6 ohms
which gives a damping factor of 10 with a 6 ohm speaker.

With the volume control turned to the middle 12 o'clock position, approximately
4.9Vrms is needed for clipping power of 80 watts into 5 ohms which will produce
an ear deafening level of approximately 104dB SPL using 3 pairs of stacked Quad
ESL57 based on being able to obtain 86dB using 1 Watt at one meter with
one ESL speaker from one channel.
At 80 watts the amp makes 20Vrms into 5 ohms and it does not exceed the
voltage rating for the speakers. The level with 1Vrms of input which can be
expected from a CD player will reduce the output voltage to 4Vrms and a power
output of 3.2 watts per channel which would give an SPL of 94dB with the 6
stacked Quad ESL. The gain of the internal preamp has been reduced to only
2.6 times or 9dB ( see below ) between its input and output so that excessive
gain will not be a problem yet there will always be a high enough power ceiling,
regardless of the speaker variety used.

The gain of the preamp was somewhat carefully chosen because there is no
ability to delete the preamp from the signal path.

In-built line level Preamp
The 12AU7 integrated preamp ahead of each power amp uses both halves
of the twin triode paralleled, and there is an active constant current source for
the dc carrying load component to reduce the THD to tiny amounts regardless
of the input levels which can be up to 20Vrms before the preamp clips.
Even when the output level of the preamp is at 10Vrms with its input at 3.8vrms,
THD < 0.3%, and at normal levels where a CD player produces 1Vrms on very
high level signals, the THD < 0.06%, and nearly all second harmonic and
thus does not destroy the musical fidelity.

See the graph of THD for 8585 input preamp which shows the 2H, 3H, 4H,
5H and 6H as they rise above the noise floor between 0.4Vrms output and
10Vrms output.

The 12AU7 preamp has a mild 12dB amount of shunt NFB between its anode
output and grid input to ensure channel gain remains constant and well balanced
and to reduce THD and noise and Rout.

The input selector switch is a 2 pole x 6 position silver plated wafer rotary switch
supplied by RS components. The balance control pot is a cermet type supplied
by Farnell, and the gain control pot is a dual 50k stereo Alps "Black" carbon track
pot which has been used in numerous quality amps for the last 30 years at least
and which is available at RS and Farnell Components.

Tube Layout
The four front tubes are 12AU7.
The next row of four tubes from the front are EL84 or 6BQ5, which are exactly
the same type of tube, but with different commonly used type numbers.
The rear eight tubes are the eight matched octal output tubes.

Fuse Replacement and
Home Service
The only fuse which should be replaced by the owner is the 3 amp slow blow
mains fuse near the IEC input mains chassis plug at the rear of the amp.
The amp must have the mains cable removed from the amp before fuse

There are other fuses within the amp and placed close to the appropriate
circuit point, and none should be replaced by the non technically trained
owner without a technician examining why the the fuse blew.
See the list of fuses below the power supply schematic, Fig3.

Fig 3 Power supply for 8585.

Fuse locations and sizes and mains voltages.

Mains fuse = 3A slow blow, type 3AG, for Mains = 240Vrms, 50Hz or 60Hz.
Mains fuse = 6A slow blow, type 3AG, for Mains = 120Vrms, 50Hz or 60Hz.
Changes to Mains voltage settings are done by removing the central transformer
Mains fuses are externally accessible and mounted in rear panel.
Anode dc supply fuses, two 1A slow blow, one 0.25A slow blow; access by
removing bottom cover.
Bias supply fuse, one 0.7A or 1A slow blow; access by removing middle
transformer cover on top of the amp.
Heater filament dc supply, two 3A slow blow; access by removing middle
transformer cover on top of the amp.

Turn on delay
There is a short turn on delay of a few seconds, only to prevent excessive
input current at the moment of switch on, and a relay click should be heard
4 seconds after turn on.

The amp has active protection to prevent one or more of the output tubes
from conducting more than 3 times the 33 mA of idle current for longer
than 4 seconds.
If this ever occurs, a second relay in the power supply will turn off the main
anode dc supply to the output tubes, leaving the amp turned on with heaters
glowing, but unable to make any sound, but harmless in this condition.
Without any anode current the amp cannot overheat any part of itself
During this "fault" condition, the two red LEDs at the front of the amp will glow
to indicate the fault condition.
Re-setting the amp after tripping the protection during an accident such as
turning up the volume with shorted speaker leads is achieved by waiting
20 seconds after turning off the amp and then turning it back on.

Fig 11 further down this page has the protection schematic and notes
about its exact operation.

The LEDs at the front also indicate the bias condition of the amp.
When the bias of each output tube is correctly adjusted, the two LEDs will
remain extinguished. A small variation of bias balance between the two
halves of the PP circuit of each channel will cause the LEDs to light up.

During normal operation, the red LEDs should remain extinguished, but during
extraordinarily loud music, the leds may flash at times due to some temporary
bias current imbalance. Should noticeable distortion be heard, and should the
LEDs flash at an unusually low gain setting, something is wrong with the
speakers or leads, ie, they may be shorted together, or there may be a fault
within the amplifier.

A schematic of the protection circuitry is further down the page in Fig 11.

The amp uses fixed bias. This is misleading, since if it was fixed, how come
it has to be adjusted?
Well, once adjusted to the correct level, it remains fixed at the adjusted level
for at least 3 months, and sometimes for many years without further adjustment
being required.
And there are 8 output tubes which EACH require separate adjustment.
All the 12AU7 and EL84 are automatically biased and need only be checked
during a yearly routine check up.

Output Tube Bias should be checked every 3 months by a
prudent owner!!!

However, owners find that bias seldom needs adjustment unless red LEDs
light up to indicate some bias "drift has occured. 

You will need a simple voltmeter and flat bladed screw driver with a thin
shaft at least 150mm long.
Bias adjustment can be done without moving the amp from the equipment
stand, and speakers and preamp maybe left connected.
Volume control should be turned down to to zero volume level.
There are 4 test points on each side of the front panel, with each test point
corresponding to an adjust screw on the top of the chassis nearby and
corresponding to each output tube, when viewed left to right.

Each test point is a recessed brass plated philips head screw.

It is safe for an unskilled person to perform the bias adjustment.

The positive red lead probe of the volt meter is held against the recessed
philips screw head and the negative black probe is plugged into the hole for
the black lead probe, or simply held against the chassis.

The long flat bladed screw driver is used to reach down through the mesh
tube cover to turn the shafts of the 8 bias adjustments.

Do NOT attempt to turn the recessed philips head screws;
these are not adjustment screws.

The voltmeter is set to a low range of direct voltage, say the 2Vdc range,
and is used to measure the first test point voltage and the adjacent screw
on top of the chassis adjusted in either direction so the test voltage measures
0.7V dc.
Although slightly awkward, an untrained person may hold the test leads
of the voltmeter with red lead to the test point, and the other hand can
adjust the appropriate screw.
Make sure that the red lead probe is not shorting the recessed contact
screw to the chassis.

When replacing a tube or all tubes, always check the bias quickly of all
tubes after turn on and turn the bias voltage measure to lower than 0.7Vdc
if any read higher.

Allow the amp to warm its tubes for 5 minutes and then set the bias of
all 8 test points slowly and accurately to 0.7V dc from points 1 to 8.
After 10minutes, repeat the bias setting from 1 to 8 since the adjustments
of one or more tubes will affect the setting of the remainder. After 20
minutes, repeat the bias adjustment again.
Both the red LEDs in the front panel should both remain extinguished, since
the balance in both channels will be correct because there is equal current
in each tube.
If one or both the LEDs remains alight after the bias adjustment, you have
done it wrong, and turned the wrong screw whilst measuring the wrong
test point, or you have a problem in the amplifier.

When correct, *all* test points should each measure 0.7Vdc between
test point and chassis. I have found the bias voltage may be 0.56Vdc
which means Ia for each tube is about only 28mA and yet the sound
remains excellent.

The relative rotation position of each adjust screw will be slightly different
with new tubes, and as the tubes age, the rotation position will vary increasingly.
If it is impossible to obtain enough adjustment screw rotation to get a 0.7Vdc
reading for any output tube, and Vdc is higher than 0.7Vdc, or below about
0.25Vdc, it is possible that this tube has a fault, and it may need replacement.

The output tubes are those most likely to ever cause problems if they overheat.
The output tubes run at about 150C temperature at the top of the glass, and
each output tube has its filament power liberated as heat, about 10 watts, and
its anode power input of about 16Watts, plus screen grid input power of 1Watt
making a total of 27Watts.
An orange glow should be seen at the centre of the small innermost cathode
electrodes within the larger dark grey colored outer metal box anode electrode,
seen easily just inside the glass envelope.
This larger electrode within each tube should never appear to glow red hot itself.
If it does glow, there is a fault in the tube operation, and a hand held above
the tube will feel that it is running hotter than the tubes around it.
Such overheating should be reported to the amp maker, but the protection
circuit has been designed to turn off the amp before the tube anode glows
red hot due to too much current flow.

Power Consumption
Highest power consumption occurs when the tubes used are 6550 or KT88
which have 1.8A rated filaments, and when the amp is working heavily into
But the power consumption will vary very little during ordinary use with
average power below 5 watts.

Power consumption from the mains is as follows, with KT88/6550.
Output filaments, 8 x 6.3Vac x 1.8A = 91W.
Input 12AU7 filaments, 4 x 12.6Vdc x 0.15A = 8W.
Driver EL84/6BQ5 filaments, 4 x 6.3Vac x 0.8A = 20W.
B+ supply to anode supply, 510Vdc x 0.4A = 204W.
Bias voltage supply, -132Vdc x 0.020A = 3W.
Sub Total power = 326W. Allow winding losses of 6% plus
extra input power for AB operation with music up to occasional clipping =
additional 74 watts.

Total maximum mains input power = 400Watts,
and idle power = approx 350Watts.

Amplifer is Heavy !
The amplifier need a sturdy support bench and should be handled and moved
only when turned off and with great care.
The 8585 is 42Kg, and above the maximum weight allowed by National
standards for unaided lifting in a workplace. But many heavier tube amps
are available from ARC and VAC, and I like to think mine give better
music and less smoke per Kg than my competition :-). 
Never drop the amp to a hard surface because the weight of the transformers
may deform the steel sheet metal cases or aluminium/brass chassis.
Due to the weight of tube amps, lifting them around can produce a back
ache if not done correctly.

Always call a friend to help you move the 8585, I am
sure he'll enjoy
a listen afterwards.

Future 8585 will never be built, but 85 Watt monoblocs will be.
To give minimum weight, ease of handling, and very best low
the two monoblocs have cabling to a third and separate power
supply chassis, so that each chassis is less than 15Kg. 


General notes about 8585 output power,
THD measurements, and active protection.

Also some basics about NFB, and PP amp operation.

The available undistorted power power varies with load. For the 8585
the recommended loading for the frequencies between 100Hz and 800Hz
should average 3 ohms or more.

Fig 4.

The Fig 4 graph curve shows the power output at less than 1% THD for loads
between 0 ohms and 33 ohms, using 4 x KT90 per channel.

Harmonic Distortion products in the 8585.

There are 3 following graphs with comments regarding harmonic distortion

Both channels were tested, the results shown are for the right channel only
since the curves for each channel were remarkably similar.

Fig 5.

Fig 5 shows the distortion components of 2H to 7H where relevant, ie, as they
increase with output voltage levels.
There is considerable 2H produced in comparison to the 3H at below 4Watts
which would cover the average listening levels of 99% of the population.
The amount is all less than 0.03%, and lower than an SE amp, and utterly negligible.
Notice that the dreaded higher number 4H, 5H, 6H, and 7H harmonics
do not appear significantly until the amp is being worked above about 37 watts,
or 13.5Vrms of output, where they try to rise above the 0.005% level.

Fig 6.
Fig 6 shows the difference in harmonic products produced when the input
preamp has its output minimized by turning up the volume control to maximum.
This gave less 3H and more 2H especially at high levels, and Fig 5 & 6 show the
effects of 2H cancellations when the preamp is used to produce more voltage
before the volume control attenuation.
I first thought perhaps the preamp 2H would add to the power amp 2H
produced in the V2 12AU7 but this appears not to be the case, so the
2H is otherwise being produced in the power amp to what i expected.
The point is that the 2H is a small amount of 2H for any tube amp.
In PP amps the 3H is usually the most dominant but in this design of PP
amp the 3H is very low compared to many other designs using such low
Iadc bias levels.
Being able to obtain less than 0.014% of 3H at 4Watts is a good result for
any tube power amp, and regardless of the volume control setting.
In both Fig 5 & 6 the higher order harmonics are negligible at ordinary
very loud levels.

Fig 7.

Fig 7
has 3 curves drawn above for THD for the 8585 with KT90 output tubes.
Tubes are biased lightly with Ia at only 33mA per tube.
Curve A & B were plotted in 2006 after changing from the 2004 schematic for the
input preamp.
Curve C is derived from the 2004 measurements for 4,6,8 ohms in Fig 10 below,
and for a 5 ohm load and with volume control at maximum to ensure the input
preamp creates the least THD possible.

Curve A is for the 2006 above amp schematic and with the volume control set at
the middle position. This means that any incoming signal is amplified 2.6 times,
then reduced in level by a factor of 0.156 times by the volume potentiometer.
The curve A was plotted by varying the input signal at 1 kHz until the amplifier
clipped at just over 20Vrms into 5 ohms.

Curve B was plotted with the same increasing input signal but with the volume
control set to the maximum level so to reduce the amount of amplification by
the V1 12AU7 preamp and negate its probable effect on the thd measurements.

Notice that there is a difference between thd levels  of curve A and B and it is
due largely to cancellation effects of the second harmonic distortions produced
in the preamp and power amp.

Curve C is taken from measurements made on the older schematic presented in 2004.
There is not a huge difference between any of the curves, but  there is a consistent
reduction of THD of at least 6dB between the 2004 and 2006 versions of the amp,
ie, between the curve C and A respectively.
This is mainly due to a reduction of 2H in the V1 12AU7 preamp by using a CCS
load for the anodes and due to 12dB of shunt NFB to reduce the 12AU7 gain.
The older 2004 preamp relied only upon about 4dB of current NFB from the
unbypassed 1.5k cathode R of the V1 12AU7.
At low levels the reduction in distortion due to a large increase in NFB does not
result in an exactly proportionate THD reduction when the preamp is tested with
the power amp.

Fig 8.

Fig 8 shows the THD result for the V1 12AU7 triode line level preamp built
into the 8585.
I measured the harmonics up to 10Vrms output only because when testing
high output resistance circuits above 600 ohms my test gear has to use a
high input impedance low distortion buffer between the device under test
and the analyzer whose maximum input voltage is limited to 10Vrms.
There is no point in measuring  output voltages above 10Vrms because it is
extremely unlikely the input preamp would ever have to produce more than
this voltage level. It is of course capable of about 60Vrms of output.
As can be seen the THD at 2.6 Vrms of anode output which would be a
high level from a CD player input signal, THD = 0.04%, and mainly
2H with ALL other harmonics below the noise floor.

There are those who despise and discourage the use of NFB to reduce noise,
output resistance and distortion on the grounds that a small amount of it
such as I have used around this preamp will significantly raise the higher
order harmonic products above the 2H to become a serious sonic pest.
There is NO evidence that they are correct. I am convinced that 3H or
other harmonics all at below the 0.001%  level do not have any effect
on the sound quality at all


The older 2004 schematic for one channel of the 8585
is included here as Fig 9 for reference.
Part numbers used in Fig 9 have no similarity to numbers used in Fig 1
and Fig 2 above.
Fig 9.
Fig 9 above is the 2004 schematic of the 8585 for the record.

Fig 10.
Fig 10 is the THD measured in 2004 with the 2004 schematic of Fig 9.
Notice that even without the improvements to the input preamp section of
the amp the THD is below 0.1% for any load above 4 ohms and any power
below 30 watts.
The further analysis of the THD spectra in 2006 in Figs 6 & 7 shows that
there are very small and insignificant amounts of high numbered harmonic
content above 3H.

The THD measurements were taken with Ia = 40mA at idle per each output tube.
Should the bias Iadc be increased to 60mA, the THD into 4 ohms which is the
worst would be reduced from 0.04% at 4 watts to 0.03%.
The best fidelity is gained by using loads above 4 ohms.
It is thought that this reduction of THD by biasing more heavily into class A
would not be audible, and not worth the cost of wearing out the tubes sooner,
and dissipating an extra 76 watts of heat within the amplifier.
The 2006 version of the amp measures better than the 2004 curves indicate
yet the idle bias current levels are slightly lower than the 2004 levels.

Protection circuit.

Fig 11

Fig 11 shows the schematic on a circuit board under the chassis about
150mm x 100mm with the above circuit to be able to make this amp well behaved.

Filtered Vdc signals from each output tube cathode of L and R channels are fed to
a pair of differential amps.
The K1, K2, and  K3, K4 signals are from of each side of the PP circuit.
Considering K1, K2, Left channel, the two cathode signals are applied
through R1, D1, and  R2, D2 to then be applied through 4k7 R3 and to
the base of Q1.
A 100uF cap filters out nearly all the ac signal from the cathodes
and only the dc signal is applied to the Q1/Q2 differential amplifiers.
The highest dc signal from K1/K2 is what determines the dcV at the
Q1/Q2 bases.
When all bias currents are correctly adjusted in each output tube, the dc
voltages at Q1/Q2 bases will be very close to equal but should one tube
vary its current by more than about +/- 10mA, there will be a dc change
to Q1/Q2 inputs, and this will be amplified to produce a difference
in collector voltages, and thus turn on the red LED through D4 to D7 diodes.
This will tell an owner immediately if there is a bias problem.

Diodes D3, D9, Q14, D20 from each of the base inputs of all 4 bjts Q1 to
Q4 are taken to a common rail which detects the highest Vdc being applied
to any base input. The voltage on this rail is applied to Q7 emitter follower
buffer base through R22.
Should the output of Q7 emitter rise to cause more than 0.7Vdc at the
SCR gate, it will latch on and the protection relay will open to cut the
HT on the power transformer.

Turning the amp off then on after 20 seconds will reset the amp,
but if the problem causing the excess cathode current re-appears, then the
HT will be shut down again. With the shut down, both LEDs are turned on
through D8 and D19.

Q5 and Q6 act in a 4 second delay circuit to shunt R41 on the power supply
schematic which limits the current input to the main B+ caps at turn on.
After 4 seconds the surge in charge currents when the relay closes to bring
the B+ up to its maximum is slightly less than the initial surge.
This allows a more sensitive mains fuse to be used.


Negative Feedback explained, NFB

Negative feedback occurs in many systems employed by men and women
in their daily lives.
There is even NFB used in principle in the toilet. Every time you flush the loo,
the water empties from the small tank ( aka cistern ) on the wall, and water
from the supply plumbing then begins to flow into the tank to refill it.
There is a float in the tank which sends a message via a lever to the valve
controlling the input flow of water to the tank.
When the water level raises the float, the lever sends a message to the valve
to turn off the water, and the level of the water is cut off.
NFB is controlling the water level. Even while the water evaporates and
its level slowly drops, the level will be maintained by the network of
Thermostats for air-conditioning work in the same way; a temperature
sensitive device reads the temperature, and if too cool it turns the heater on,
and too hot it turns on a cooler.
The message from float lever or thermostat is called negative feedback,
and negative feedback is where that fed back message is applied in a way
which opposes the action of the input message.
If the water supply pressure increases, the valve will be forced open only
slightly and the float will rise to apply more closing pressure on the input valve.
So even if input pressure doubles, the water level in the tank may only rise
very slightly.

Fig 12.

Fig 12 shows the whole of one channel of the 8585 in the form of a
Block Diagram Schematic which is much easier to understand than trying
to read off what I have posted above in Fig 1 and Fig 2.

NFB in preamp.
The line level preamp within the
8585 is a single 12AU7 inverting amp which
has a "shunt resistance" NFB network. Such networks have (R1) between input
signal coming from a low resistance source and the tube input grid. (R2) is from
the anode output and the grid. The 8585 block diagram omits the coupling caps
which have such low impedance in the audio band that they don't need to be
displayed; only the essential gain setting elements are shown. In the 8585, R1
is effectively approximately 40k0, when R2, R3, and VR1 are considered
together. (R2) is actually R8 and 270k.

Beta = 40k / ( 270k + 40k ) = 0.129.

Suppose the THD of the preamp with operating voltages as shown is
measured to be 0.03%.
With output voltage at anode = 1.8Vrms, the distortion
voltage = 0.54mVrms, and it isn't much, and let us say its phase is +0.54mV.

This distortion voltage is divided by the (R1) and (R2) resistance divider so that
Beta x VDn appears at the grid, ie, 0.129 x 0.54 mV = +0.07mV.

This is amplified by 14 times by the 12AU7 to make a resulting distortion
voltage of -0.98mV appear at the anode. This seems absurd, because I just
said there is +0.54mV at the anode. But what is actually happening is that
without the NFB (R1)&(R2) network connected, there was already
+1.52mV of Dn at the anode for the same anode output and grid input signals.

What is happening is that the -0.98mV is the Correction Signal which reduces
the distortion without NFB from 1.52mV to 0.54mV, ie, the NFB has reduced
the open loop THD from 0.84% to 0.03%, a very welcome reduction
in THD. And the effective anode output resistance Ra' of the tube is
considerably reduced, and may be calculated,
Ra' with NFB = Ra without NFB / ( 1 + [ x B ] ).
With paralleled 12AU7, Ra' = 6k0 / ( 1 + [ 17 x 0.129 ] ) = 1k3,
and the bandwidth of the 12AU7 stage will be in excess of 300kHz.

Distortion varies approximately with output voltage level for small signal triodes.
For Hi-Fi, the maximum voltage from a CD player might be 1.4Vrms, and
average levels = 0.14Vrms, and we might might expect THD < 0.01%,
and well below audibility. I have always found the use of a single triode
preamp like that shown to be entirely wonderful sounding.

NFB in Power Amp.
The power amp within the 8585 has 3 cascaded tube stages and an output
transformer which create a NON inverting voltage amplifier with gain =
40x times without any NFB connection.
There is a "series voltage NFB Network" which is set up with (R1) and (R2)
which in this case are R15, 47 ohms plus R28, 680 ohms between the amp
output speaker terminal and 0V. The junction of the 2 R are taken to the
inverting input terminal of the amp, V2 cathode.

Beta = 47 / ( 680 + 47 ) = 0.0646.

Suppose the THD of the power amp with operating voltages as shown is
measured to be 0.025%. With amp output voltage = 2.0Vrms, the distortion
voltage = 0.50mVrms, and it isn't much, and let us say its phase is +0.50mV.

This distortion voltage is divided by the (R1) and (R2) resistance divider so that
Beta x VDn appears at the V2 cathode, ie, 0.0646 x 0.50mV = +0.0323mV.

This +0.0323mV is amplified by -40 times by the open loop gain factor with
no NFB to make a resulting distortion voltage of -1.29mV appear at the output.
This also seems absurd, because I just said there is +0.5mV at the output.
But what is actually happening is that without the NFB (R1)&(R2) network
connected, there was already 1.29mV + 0.5mV at the output for the same
levels of output and grid input signals, ie, THD without FB = +1.79mV = 0.0895%.

What is happening is that the -1.29mV is the Correction Signal which reduces
the distortion without NFB from +1.79mV to +0.50mV, ie, the NFB has reduced
the open loop THD from 0.895% to 0.025%, a very welcome reduction
in THD.  The THD is reduced by a factor = 0.025 % / 0.0895% = 0.279,
or by -11dB approximately. This reduction in THD is also given as CLG / OLG,
or A' / A.
The output resistance of the amp without NFB is reduced by at least the same
factor or more. I don't need to show how Rout is calculated here but my other
web pages do contain the formulas. With power amps, the NFB extends the
bandwidth without NFB, and calculation of exactly how much is not within the
scope of ths page on 8585 basic issues.

For those still confused about relative phases of signals within amps,
the + or - sign before Vrms indicates the relative phase of the signal voltage,
and there is 180 degrees difference between + and - phases.
The two phases are like two kids on a see saw, and while one ascends,
the other descends, much to their merriment.

The phase inversion or non inversion of the phase of signals from a source
has zero effect on the sound, despite what many people say.

There are a number of different types of feedback used in amplifiers,
both positive and negative, current or voltage, shunt or series. Books have
been written about it, and deeper explanation won't fit into this very brief

Push Pull action and Distortion
Push Pull amps have inherently low distortion even without any use of NFB at all.
If one imagines two men each side of a log trying to cut it with a long bush saw,
their motion is smooth and regular with one man pulling, and the other pushing
the saw, and the individual motional irregularities of one man is cancelled by the
other man's similar irregularities.
One man using the saw will tend to pull the saw with more force than he can push it,
so the cutting power delivered to the log is uneven for each half of the cutting cycle.
This is like the second harmonic distortion in electronic sine waves where the top
half of a sine wave has a different amplitude or height than the bottom half with
respect to the centre point about which the wave moves.
The irregularities of the two men's combined action when cutting tends to give equal
power reductions at the end of the saw's motion, so it is as if
there is a flattening of each +ve and -ve peak of the sine wave, and this is akin to
third harmonic distortion in an electronic sine wave. The overall effect of having a
man on each side of the log gives a much less distorted motion than having both
men working on the one end of the saw from one side of the log. So the overall
effect of PP action with 2 tubes gives much less THD than you get with the same
two tubes in parallel. 

The power in a push-pull circuit is liberated into the load in two ways simultaneously,
with one tube increasing its current into its R load, and the other reducing its current
into its R load. The instantaneous current direction in 1/2 of the PP OPT is oppositely
phased at all times and one tube is said to be pulling while the other is pushing,
although in fact currents do no such thing as we imagine things in our visible world.
Power liberated in a load resistance = load voltage change x load current change,
ie, P = V x I.
Or Po = current change in load squared x load resistance, ie P = I squared x R.
Either way of calculating PO is correct.

The two tubes swap "push" and "pull" roles as the signal voltage swings positively
and negatively each side of zero volts.
This principle has been used since about 1910, and it confuses all non technical
people although two kids on a see-saw are using the principles of push pull action
without knowing what it is, at least until they grow up and study physics.

The complementary contribution to the output power is enabled by the push pull
output transformer designed to take the oppositely phased signals from two tubes,
wheras the SE OPT is designed to accept only one signal from one tube
with only one phase of signal.

The driver stage of the 8585 has very low distortion, and the two input SET
stages also are set up for best linearity, and contribute a negligible amount of even
order THD, ie, 2H, 4H, 8H, thus allowing the low measureable THD seen in
PP amps, without using a high amount of NFB.

Loadline analysis for FOUR x KT90 in PP class AB1 with 12.5% CFB.
Fig 13.

Fig 13 is not a determined attempt to confuse everyone, but you *will need*
to think about what it all means!
I suggest those who don't understand load line analysis should read the basics
in the Radiotron Designer's Handbook, 4th Ed, 1955, or my other pages
on load matching to PP tubes.

Above, the 8585 amp is described and has an OPT with load match ratio,
aka impedance ratio, aka ZR = 4,172 ohms : 5 ohms, to suit 4 x KT90
arranged to power 3 parallel stacked pairs of Quad ESL57.

The load lines for the 8585 load as specified may be plotted as follows :-

1. Draw up the combined Ra curve for TWO KT90 in parallel.
This is shown as a curve from 0.0 thru K, C G and swinging over to the right
below Ia = 1,000mA. This curve is generated by doubling the Ia change
for a given Ea change along the Ra curve for Eg = 0V shown in data sheets
for ONE KT90. The KT90EH I have used is very similar to KT88EH and
6550EH and old original 6550 and KT88.

2. Draw the Pda limit curve for 84Watts, which is twice the maximum
allowed Pda for KT88 and 6550. The 8585 is able to use 6550, KT88
or KT90, so the load lines above are valid for all 3 tube types, as well as
the more recent EH KT120. 

3. Plot Point Q at Ea = +475V and Ia = 60mA. This is the Quiescent idle
point for 2 x KT90 on each side of the PP circuit.

4. Calculate the B RLa which is the Class B load when the operation
has moved above the Class A1 to AB1 threshold.
B RLa = 4,172 / 4 = 1,043 ohms.

5. Plot Point A vertically below Point Q on Ea axis.

6. Calculated Ea / RL B = 475V / 1,043ohms = 455mA.

7. Plot Point B on Ia axis at Ia = 455mA.

8. Draw line from B to A. This is the loadline for 1,043 ohms.

9. Plot Point C where Line B to A intersects the Ra curve for Eg = 0V.

10. Plot Point D on Line B to A where Ia = twice idle current = 120mA.

11. Read off the Ea vertically below Point C = +110V. This is the
point which sets the limit for Ea load swing with this RL, and is E min
for the load voltage swing.

12. Calculate Peak Load voltage swing for RLa-a = 4,172 ohms.
Peak load voltage for RLa-a = 2 x ( Ea - Ea min ) = 2 x ( 475 - 110 ) = 730Vpk.
Calculate Va-a in Vrms = 730 x 0.707 = 516.11Vrms.

13. Calculate maximum output AB1 power = Va-a squared / RLa-a
= 516 x 516 / 4,172 = 63.8 Watts.

14. Read off the Ea vertically below Point D = +350V.
This is the voltage point where the class A1 operation ceases, and class AB1 begins.

Calculate ( Ea at Q ) - ( Ea at Point D ) = 475 - 350 = 125V.

Add this to Ea at Q, 475 + 125 = 600V.

Plot Point E on Ea axis at 600V.

15. Draw straight line from D thru Q and it should proceed thru Point E at 600V.
Line DQE is the load line for the class A portion of power produced by 2 x KT90,
and is equal to a load of 2,086 ohms.

Each pair of KT90 on each side of the PP circuit have the same Class A and AB loading.
The voltage between Point D and E is the peak to peak load voltage for class A for
2 x KT90. 

16. Calculate Class A portion of power for 4 x KT90.

Peak to peak voltage for 2 x KT90 = 2 x 125V = 250V.
Va rms for 2 x KT90 = 250pk-pk / 2.82 = 88.65Vrms.
PO for class A RL a, 2 x K90 = Va squared / RLa = 88.65V x 88.65V / 2,086 ohms
= 3.77 Watts.
Class A portion of PO = 2 x 3.77Watts = 7.54Watts.

Summary so far.

The speaker load to be used with the 8585 in this case has load ohm value
equal to 1/3 of the load value for 1 Quad ESL57.
The ESL57 has Z = 32 ohms at 50Hz, reducing to 8 ohms at 1kHz, then
down to 1.8 ohms at 18kHz, so that the average Z of a single ESL57
for where most power is to be made is between 100Hz and 500Hz and is
about 15 ohms. The ESL57 was designed to be able to be driven adequately
by the Quad-II amp produced before the ESL57 was marketed.
The Quad-II amp produced about 22Watts into 15 ohms.
Therefore an amp capable of 66Watts into 5 ohms should be capable of
adequately powering 3 parallel ESL57. When the ESL are paralleled and stacked
one above the other, their sensitivity increases, and for the same sound level,
less power is needed for the 3 stacked ESL. The 8585 described here has been
used with the stacked ESL57 since 2005, and the KT90
EH test as new, despite
being used for long periods daily.

Now the original Quad-II amps relied on their pure class A ability to provide
fidelity, because their power supplies are rather inadequate. When the Quad-II
OPT is strapped for 15 ohms, the RLa-a is about 4k0, and the extent of
class A power = 9 Watts approx, with an additional 13 Watts in class AB1.
The KT66 struggle to achieve the 22 Watts and because my 8585 has
a much better power supply and input driver stage than Quad-II, and
because stacked pairs of ESL are more sensitive, the amount of pure class A
power does not need to be greater than one Quad-II amp, and for the transients
and crescendos there is 66 watts available from the 4 x KT90. The 8585 could
have higher Ia idle current at say 50mA in each KT90 which would double the
class A PO to 15Watts before the amp moved into class AB, but with the low
bias it uses now it performs flawlessly.

At low bass frequencies, the stacked ESL Z rises to 10 ohms, and there is
no shortage of drive plus the amount of class A power increases. At above 7kHz
there is very little energy in music, and the stacked ESL load reduces to only
0.6 ohms. I have 0.6 ohms in series with each of the 3 stacked ESL
so the minimum RL for the amp = 0.8 ohms.
Because the power is so little where Z is so low, the power remains as class A.

Fig 13 has load lines for RLa-a = 2,086ohms : 2.5 ohms and also
8,344 ohms : 10 ohms. These load lines may be drawn using a similar sequence,
numbers 1 to 16 as for 4,172 ohms : 5 ohms.

Fig 14.

Fig 14 is a photo taken in October 2011 of the 8585 under the chassis,
after a full service following 6 years of trouble free operation, even though
the owner said he uses the amp daily.
All KT90EH tested similarly to brand new tubes, and showed no signs
of wear, ie, gassiness, cloudy gettering, or positive grid current at idle.
All input and driver tubes measured perfectly, and gave no noise.
The servicing was required because four zener diodes I had used
had degraded to having much less than their rated zener voltage of 75Vdc,
so the shunt regulator for the KT90 screen supply produced Eg2 = 200V
instead of +330Vdc, and this turned off Ia in all KT90. I have replaced the
simple 4 x series zener diodes with the BJT based shunt regulator mounted
on the AL angle flange seen at the top of the picture.
Unfortunately, even though zener diodes may be rated for 5Watts, they
can fail if subjected to only 1.5Watts of heat production, and in practice
it is wise to never let their Pd rise above 1/10 of their official rating.
Other zener diodes in the 8585 have been retained since they were
originally installed in 1996, and they show no sign of change.

The point to point circuitry is somewhat messy, but quite reliable, and
it is easy to figure out where parts shown on the schematics are
Amplifiers I have made in recent years are a lot neater under the chassis,
but will never be as neat as a well done printed circuit board.
Unfortunately, the use of printed circuit boards obstructs the flow
of air up and around the parts and up through the holes drilled in
the top plate of the amp and past the glass of the output tubes.
If one looks into the finest samples of circuit production used in
professional test equipment as manufactured by Tektronix, or Lavoie,
you will find 3 dimensional pint to point circuitry, and neatly bundled
wire looms of multicolored wiring, with much thought and design
given to the location and placement of each item to maximize component
density, yet disallow adverse stray signal coupling, and to minimize
operating temperatures of everything. Fans were often used in tubed test
gear to reduce heat, but fan noise is always intolerable for hi-fi amps.
There is no need for a fan to keep things cool in the 8585, because all
transformers and tubes have been designed to run at low temperatures.

Without the heat stress, there is only the music to behold.

For more ideas on load matching and how tubes work
go to my list of educational and DIY pages.

To Index Page