REFORMED QUAD-II POWER AMPLIFIERS
Last update, February 2017.
Picture of reformed Quad-II, 2006, KT90 output tubes

Those with keen eyes will realize these amps are not quite like original Quad II amps.
Why are blue and red LEDs glowing?
The output tubes are KT90. The GZ32 tube rectifier is missing, but can be plugged as a feeble
attempt to make ppl think its really original, but tube rectifier is not part of the circuit any more.
I rebuilt this pair in 2006, but also rebuilt others in 1998 and 2010 for my customers.
I own another pair of originals I bought cheaply at a garage sale which await being given
"singing lessons" far more drastic than shown in many schematics below.

This page is about the range of possible improvements between very basic and infuriatingly complex.

Large schematic images installed onto the page may appear too small so feel free to
"open image in new window" or otherwise increase the image. Where possible all images
should look well when printed to fill an A4 page.

Without basic electronic knowledge and experience,
I suggest you limit your efforts to basic improvements.
These will include :-
1. Replace all R and C with modern metal film R and MKP coupling caps and new electrolytics.
2. Replace GZ32 with GZ34, which allows the original 16uF+16uF caps to be replaced with
33uF + 33uF.
3. Remove original 180r and 25uF cap between cathode FB winding CT and 0V and replace
with a wire link.
4. Disconnect ends of cathode FB winding from KT66 cathodes, and connect parallel networks
of 390rx5W + 470uFx63V between each winding end and each KT66 cathode.
5. If the original Quad 22 unit is not to be used, Install an RCA input chassis socket near original
Jones plug and connect to EF86 input grid to facilitate connection of modern RCA cabling to an
alternative preamp.
6. If the Quad 22 unit is not to be used, remove original Bulgin mains input socket plug and
install IEC chassis plug and a mains switch to allow independent operation of both left and right
channel power amps.

Make sure all amp metal chassis are connected to Earth from terminal on IEC input.

For those not wanting to change anything inside a Quad-II, they might seek help elsewhere at
the very wonderful site....
http://www.keith-snook.info/quad-stuff.html 
Keith's website includes a section on replacement PT, OPT and Chokes for Quad-II, now made
by Majestic Transformer Co, in UK.
http://www.transformers.uk.com/valve-audio-transformers/
I was not impressed by MT's replacement OPT. IMHO, it is quite impossible to have a satisfactory
OPT for 2 x KT66 which is so small as the original but I give some thoughts at bottom of this page
about addressing the problem of a tiny OPT.

Before I mention more details about mods for Quad-II, I should include some history......

There was a 'Quad One' amp which preceded Quad-II amps which suited the years following
WW2 when there was only single channel sound. When stereo recordings became possible the
Quad-II mono amps were built to be only usable with the Quad-22 "control unit", aka preamp.
We must respect Peter Walker who worked long and hard to give excellent hi-fi sound to a
world which was otherwise dull, and full of audio products with little real merit.

IMHO, Peter Walker deserved most fame for his production of Quad ESL electrostatic speakers.
These are still very much loved my many audiophiles. Their production and quality control
involved heroic efforts on the production line.
The Quad tube power amps and preamps were very nice kit for the era of the 1950s where
speakers were mainly 16r0 with high sensitivity of over 93 dB/W/M. Few people ever needed
more than 20W of power, and most listeners used less than 0.3W of average power for pleasant
listening to most music.

Peter Walkers dream for ESL57 was realized, but they had much lower sensitivity than dynamic
"dome+cone" speakers needing only 1 x SE 6V6 for 4W, or a PP pair for a luxurious 10W.
But Quad-II amps were quite suitable considering the sensitive hearing of UK people and their
small lounge rooms and humble desires, and their desire to not upset the neighbours with loud
live broadcasts of London Symphony Orchestra. Most of the other ppl preferred inane 1950's
syrupy pop music.

Between 1950 and 1960, arguments raged about which amp was best, the Quad, Leak,
Williamson, Radford, or some USA brands. For many in Australia the cost of such exotic imported
hi-fi gear was extremely high so a small number of hi-fi enthusiasts built their own amps after reading
magazine articles published in magazines such as Wireless World which later became Electronics
World, or Radio, TV and Hobbies which later became Electronics Australia. I have repaired or re-built
some of these amateur efforts which were so often riddled with terrible mistakes, very unsafe
construction, very poor component choice, all indicating that unskilled gung-ho fools with no money
had been at work.

Quad-II and ESL57.
The original Quad-II could make 20W which were needed to get the best sound from Quad's ESL57
which had rather low sensitivity of around 83dB/W/M. The ESL57 were produced after Quad-II amps
and were limited to levels which may not meet modern expectations which includes much higher bass
levels. ESL57 was a "full range" giving barely adequate bass. ESL impedance varies hugely from 33r0
at 80Hz to 1r8 at 18kHz. In the main AF power band 80Hz and 700Hz, average Z is above 20r0, and Z
does not fall to 8r0 until about 5kHz.
Graph 1A. ESL57 Z and input power.

Graph 1 shows ESL57 impedance vs frequency and is fairly accurate.
But I also show the power developed in ESL57 when a constant 5Vrms is applied in order to get
a flat F response. With a constant 5Vrms applied at all F, very little power is developed below 200Hz.

Many dipole ESL panels give limited bass below 100Hz and many will use a dynamic sub-woofer to
augment the panel performance. The few who take the trouble to use a good low bass speaker will
have an active crossover between preamp and power amp to reduce bass signals in Quad-II amps so
both amps and ESL panels just need to cope for all above say 200Hz, so that headroom for all above
200Hz in amps is at least doubled. But this is extremely complex and difficult in an original Quad-II
plus Quad22 amp system.

The power in ESL57 has plateau around 1kHz at 2.4W. Between 3kHz and 20kHz, ESL57 needs
much more Po and to maintain a constant level test signal, 13.9W is needed at 18kHz, with 5Vrms to 1r8.
But Peter Walker knew that above 5kHz, the level of audio energy declines so it didn't matter if the load
between 5 kHz and 20kHz dipped to a low 1r8.
Because Quad-II has output resistance of about 1r0, the damping factor at 16kHz is is only 1.8, and the
Vo at speaker terminals sags about 4dB. It would seem that Peter Walker was aware of this and if you
use ESL57 with some other amp ( solid state ) with Rout typically < 0r1, then ESL57 sound a little hard
edged because the upper treble response has been boosted by +4dB. This was what people told me.
I have never owned ESL57. The problem is solved by using a 1r0 resistance between low Rout amp
and ESL57.

The drop of Z to 1r8 means the ESL57 HF panel becomes far less sensitive above 5kHz.
The main general problem with many ESL is their lack of sensitivity at HF because their load character
is effectively capacitance with decreasing reactance at HF plus some added series and parallel R.

The simplest equivalent dummy load network to ESL57 is ( 1r5 in series with 2uF ) parallel to 15r0.
The simple dummy load isn't quite right because the Z at LF is a maximum 33r at 40Hz.
A pure 2uF load for many amps can cause stability and F response peaking at 20kHz. 2uF has Xc = 4r0
at 20kHz, and just why the ESL57 gets down to 1r8 at between 16-18kHz is a mystery, considering that
ESL57 has some input resistance in its large and heavy step up transformer. So we may assume there is
capacitance in the step up transformer winding as well as in the treble panel.

But 80% of the energy in music is below 1kHz, and with very little above 5kHz, so the ESL speaker Z can
be allowed to be low even when fed by an amp which is designed for maximum class AB Po with 18r0 with
OPT set for 16r0. Allowing a speaker Z to become 1r8 makes no sense, but Z at 10kHz is 4r0, and does
not fall to 1r8 until F > 16kHz where there is extremely little energy in most music.
In 1950s the hi-fi audio bandwidth was considered = 30Hz to 15kHz. Very few audio sources such as FM
radio or LPs or early tape recorders had any content above 10kHz.

My graph below shows 3W is available from amp with 1r8, and that 3W is enough to give adequate
levels of 16kHz. Levels above 3W at 16kHz will produce high THD. But all THD for any F above 11kHz
cannot be heard because lowest H product is 22kHz. The real worry are IMD products produced by
presence of lower F signals amplitude modulating the higher F which people can hear all too well if the
bass signals are clipping. Such IMD makes the sound Grubby & Muddy compared to being in the
audience where London Symphony Orchestra is playing something by Mozart.

Quad-II has 2 linking patterns for its OPT to suit either 16r0 or 9r0 speakers.
For ESL57, the OPT is usually strapped for 16r0. For pure class A, the load must be 32r to get max
Class A Po = 18W, but average load between 50Hz and 300Hz for most music energy = 22r, so you
get 11W of initial class A with class AB up to 22W, and most music is produced by class A power.
But above 300Hz the average load is 10r0 which gives 5W in class A 14W max in AB. So when the
volume is cranked up there is considerable class AB action where IMD and PSU noise becomes higher
than anyone would want. IMHO, it would be better to use the 9r0 OPT strapping with ESL57 or any
other dynamic speaker ever used. Original Quad-II do not like 4r0 speakers, and although there is a
possible OPT connection for 4r0, it was never promoted by Quad and is not shown as a possible
load selection in Quad schematic.

Quad-II amps can be made to produce much better class AB operation and less IMD when using
modern speakers.

But despite all the limitations, when used without clipping and with civilised levels in most lounge
rooms, there are people who say Quad-II plus Quad ESL57 provide unpassable sound quality.
Peter Walker went on to make ESL63 and solid state amps but the initial ESL57 pleased thousands
who paid so much for them. The customer who had me reform his Quad-II amps certainly liked ESL57.
In addition the single pair of ESL57 for these reformed Quad-II amps above, he later had me provide
a stereo power amp with 4 x KT90 per channel to power 3 stacked ESL. There wasn't much wrong
with the sound.

The more anyone thinks about ESL57 and tube amps, the more they should realize ESL57 are easy
speakers to drive, and that a humble tube amp can power them very well, ( providing teenagers are
not allowed near the volume or tone controls ).

While on the subject of Quad ESL there was the later model ESL63 and others ESL989 etc.
Here is the impedance curve for ESL63, after carefully copying and enlarging data from UK by
Hi-Fi News & Record Review, November 1991, and the "recommended retail price was
UK 2,072 Pounds".

Graph 1B. ESL63.

This shows ESL63 to be just as easy to power and much easier at HF than ESL57. But ppl say
Quad-II amps don't have enough power for ESL63 maybe because they are less sensitive, but I
think maybe amp with 2 x KT120 or a quad of EL34 should be fine.
---------------------------------------------------------------------------------------------------------
Quad-II struggle with modern dynamic speakers with low sensitivity and low Z for all frequencies
even when OPT is strapped for 8r0, the lowest setting that is promoted in Quad-II manual.
But there is a usable connection for 4r0 speakers with internal speaker wire taken to point Q on
OPT. This output point on OPT gives 16W of mainly class A Po for 8r0.

Many old amps from 1950s including Quad-II can have troubles with overheating KT66. There is
no active protection except a mains fuse which only blows after a KT66 has become a short circuit.

When you hear about an old Quad-II amp fusing its OPTs or PT, it is usually because old men
hate paying for a set of new tubes. Old ESL develop faults of arcing in panels from failing panel
membranes which inevitably do fail after 50years or less. KT66 data say that KT66 life is 8,000
hours based on being turned on continuously, and with idle heat = 25W. One year = 8,760 hours.
I've known a number of ppl who use their tube amps for 4 hours a day, and they got about 5 years
from their tubes which included 1,825 turn on/off heat cycles. I found all output tubes, KT66, EL34,
6CA7 KT88, 6550 all gave about the same lifetime.

I became irritated by old fellows trying to tell me their KT66 were OK when they were not.
They thought I was trying to overcharge during the service work, just like all the car mechanics,
plumbers, lawyers, doctors. But 4 to 6 years is a good run with many large octal power tubes when
they are idled with Pda close to Pda rating.
I have seen fellows get 40 years from a pair of EL34 - but only because the idle Pda was 17W,
not 25W.
I have also seen KT90 EH in my 8585 amp run for 6 years, and the owner used the amp every day.
When serviced after 6 years, the 8 x KT90 tested like new, with no signs of wear. Idle power was 16W.
Sound was fabulous, and THD and other measurements were excellent. But you should know 4 or 5
years can go past all too quickly. Owners quickly forget to service anything, so their negligence invites
problems.

ESL57 panels did not last more than about 50 years unless you were very lucky. But this was far longer
than many dynamic speakers which developed cone surround breakdown, voice coil jamming, enclosure
degrading etc.
Speakers are perishable items like everything else in this world, including ourselves.
ESL panels have thin stretched polyester membranes which are charged up to thousands of volts dc,
and have applied Vac up to thousands of volts. These panels are fragile, prone to effects of house
dust and prone to arcing after membrane tension relaxes after 50 years. The repair bill always upsets
old men. The treble panels usually to arc first at some low threshold of applied Vac. This may not be
noticed until the amp overheats because of the intermittent short circuit caused during arcing.

Quad-II or any other 20W amp may not drive ESL57 as loudly as expected today. Between 1955 and
2015, average bass content in music has increased. Therefore a sub-woofer makes sense for between
20Hz to 150Hz and powered by its own amp ( usually can be generic 50W solid state amp ).
The difficult part about a sub-woofer is the connection of crossover filter before the power amps to keep
F below 150Hz out of Quad-II amps and ESL and only have F below applied to sub-woofer amp.
Much luck is needed to get the low bass to sound right.
The ESL57 have good performance above 150Hz, and with all below 150Hz kept out of amp and ESL57,
they can both produce higher undistorted levels.
I myself never bothered with a separate sub-woofer for myself. I always preferred to make 3 way floor
standing speakers with bass units dedicated to 20Hz to 200Hz and very well integrated with midrange
and treble.

Quad-II have been said to offer fine class A performance. The amount of class pure A1 Po depends on
the idle Pda of the tube and the tube having a high anode load.
The highest Ea ( Vdc between anode and cathode ) I have measured with GZ32 = +320Vdc, and Iadc
= 68mAdc, for each KT66. this is based on having Ig2 = 4mAdc, and Ek = +26Vdc, and average 10Vdc
across primary windings, with B+ at OPT CT = +356Vdc.

I recently tested a sample of Quad-II to give results shown in Graph 2 below........

Pda at idle = 320V x 68mAdc = 21.76W, so for 2 x KT66, total Pda = 43.5W, and max class A Po efficiency
= 40%, so expect 17.5W class A from anodes. If the OPT winding losses are 10%, expect 15.6W at OPT
secondary.

Many people found GZ32 was fragile, prone to early failure, especially where a KT66 had bias failure.
The GZ34 is a far better rectifier tube and raises B+ to about +370Vdc, and Ek = 28Vdc, so Ikdc
= 77mAdc for each KT66, Ia = 73mAdc, and Ig2 about 5mAdc. Ea = +331Vdc, idle Pda = 24.2W, and
anode class A = 19.4W, with output class A max = 17W at OPT Sec.

Tannoy made the most impressive dynamic speakers in UK. Who can forger the Dual Concentric 15"
drivers used in well made heavy 180Litre boxes? I spent a night with a fellow who had a pair with
170L DIY ported reflex boxes using 50mm office partition material which had its hollow cores filled with
sand. The fellow had 8W SET amps with 300B made by Allessa Vaic in late 1990s. Tannoy speakers
don't need a sub-woofer, and have efficiency of 95dB/W/M and IMHO, a pair of Quad-II amps can
provide flawless music quality - anything by Mozart or Bob Marley sounds just fine.

Graph 2. Original Quad-II Po vs RL

Graph 2 shows the maximum power levels where THD < 1.5% and available with the two
advertised methods of OPT strapping for either 8r0 or 16r0, and the never ever advertised
strapping for 4r0.

These power levels were measured using two KT66 with idle conditions :- B+ at OPT CT
= +361Vdc. Rectifier was GZ32, and 100Hz ripple at CT = 18Vrms, with PSU caps
= 16uF + screen choke + 16uF. 100Hz ripple at KT66 screens > 50mV.

Ek = 28Vdc, with separate 390r + 470uF from each cathode to ends of CFB winding U-V-W
with V taken to 0V Idc in 390r = 69mAdc, with Ia 64mA and Ig2 = 5mA approx.
The Ea between anode and cathode = +318Vdc, and Eg2 = +313Vdc.

The measurements for Po were slightly vague. Maximum Vo for class A was where idle Ikdc
increased +5% due to charge build up in coupling caps or grid current. The peaks of output
waves had negligible levels of 100Hz hum until the amps worked heavily into class AB where
Ikdc could increase by as much as 50%, and much 100Hz ripple is seen at clipping peaks on
CR0, so Po is taken where the original sine wave is present, but with 1/2 the ripple seen.
Nobody will understand that unless they see it on a CRO.

But there is 18Vrms of 100Hz ripple at OPT CT for class A operation. In class AB the Idc at
CT may increase from 140mAdc to 210mAdc and ripple becomes 27Vac, and in class AB there
is no common mode rejection of the hum at CT so much 100Hz hum is in series with the anode
Va and is applied to each 1/2 primary and so it appears at OPT Sec. The high 100Hz ripple
does cause some IMD.

With separate R&C cathode biasing with 470r+470uF to each KT66 cathode, and with no other
mods, I found Quad-II becomes unstable at LF when no load is connected or when the load is
say twice the nominal strapping value.

Peter Walker must have known this, and it may have been the reason why he used a common
Rk 180r plus 40uF bypass for both cathodes of KT66. This is stable at LF. But Andy Grove
designed Quad-II-Forty with 390r + 200uF to each cathode, and that was stable at LF, despite
having an almost identical schematic to Quad-II. But then the Quad-II-40 has a slightly better
OPT made in China somewhere.
The reasons for the LF oscillation is combined LF phase shift with low value C between OPT CT
and 0V, low amount of OPT Lp at low levels, the C+R coupling from EF86 anode to KT66 grids,
PFB in the phase inversion, and the high open loop gain with high RL values or no load at all,
plus the high amount of GNFB. Possibly, the value of Ck for each Rk is critical, but usually if
you raise Ck to say 2,200uF or reduce it to 47uF, it just changes the frequency of the LF
oscillation.

I cured the LF oscillations with open loop gain shelving network between EF86 anode consisting
of 0.1uF as original plus parallel network of 0.022uF and 2M7 to output tube grid and 680k
biasing Rg. This stopped all LF instability.

Never ever assume that any single modification you perform on any amp you ever work on will
always have a totally positive outcome. All amplifiers are far more complex than the average
DIYer likes to ever be forced to realize !

After measurements were taken, I concluded the KT66 have Rd diode line R value between 100r
and 220r.
The old paper copy data sheets from 1950 have vague Ra curves, and it is difficult to determine
the diode Rd for Ea below 100V.

KT66 in Quad-II have idle Pda = 318Vdc x 64mAdc = 20.5W, each tube. Class A Po from each
KT66 = 9.4W. The class A RLa load for each KT66 = 4.6k. Thus the two tubes make 8.8W, and
RLa-a must be 9,200r for maximum class A for all three 3 strapping patterns, when the right Sec
load is used.

The calculated load for tubes includes the winding resistance of OPT which in effect is like series
resistance with perfect OPT with zero resistance, but with whatever load ratio is has.

Original Quad-II OPT has primary wire resistance for each 1/2 primary = 173r and 125r and including
the 16r6 for each 1/2 of CFB winding. Average Rw for each 1/2 primary = 150r, so the total RwP
= 300r while working in class A.
The winding resistance of secondary when strapped is :-
16r0 = 1.71r, 8r0 = 0.95r, 4r0 = 0.64r.
( There is a table below which deals with winding losses with much more detail. )
In class A and when strapped for 16r0, the RwP + RwS measured at Pri = RwP + ZR x ( RwS )
= 300r + ( 243 x 0.64r ) = 455r.
If the anode class A RLa-a = 9,200r, then OPT load = 9,200r - 455r = 8,745r, and Sec load
must be 36r.

The OPT TR = 3,180t : 204t = 15.588 : 1.0 for ZR = 242.99 : 1.0 giving
Nominal Load Ratio 3,887r : 16.0r.

Total class A winding loss % = 100% x RwP+S / [ ( OPT ZR x Sec RL ) + RwP+S ]
This case, Sec load 16r0, class A loss% = 100% x 455r / ( 3,888r + 455r ) = 10.47%.
This increases to 16.3% for the class AB operation so average total RwP+S at full AB Po for
16r0 is about 15%.

Winding loss% is the percentage of power lost in winding resistance from the total power produced
by tubes.
If the secondary load was reduced to 8r0 for 16r0 strapping, during initial class A Po the loss = 19%.
The class AB with 8r0 will increase to 25%. These high losses are entirely due to the tiny size of
the Quad-II OPT.

Whenever anyone measures an original Quad-II amp, they expect to measure more Po because
they may have measured 30W from other amps using 2 x KT66 or EL34 in RCA, Leak, Dynaco,
Mullard, etc.

For Quad to have ever made a bit more Po, they would have had to have used a bigger OPT with
thicker wire for less losses and used higher Ea, but that would have been a struggle in 1950
because Britain was still much disabled by WW2. Peter Walker made something usable for BBC
studios and for the upper middle class and above.
If the OPT had been bigger, the chassis would be bigger, so costs would have been higher, and
most people would not have heard any difference. 

Fig 1. The original Quad II monobloc amplifier schematic :-
Part numbers are the same as the original schematic.

I scanned a good 50 year old paper copy of the schematic and here is the re-drawn version
I hope everyone finds easier to read. Originally, only two OPT strapping patterns for 16r0 and
8r0 load matches were mentioned on the schematic. I have added the strapping pattern for 4r0;
just move the wire from Vo from T and reconnect it at Q. Graph 2 above shows the difference in
power levels when using the 4r0 load match. But the power has less THD and a better damping
factor, and where 20W is enough Po, there is no point arguing for anything any better from Quad-II
if your speakers are nominally 6r0 or lower.

For Quad-II, only speakers over 8r0 should be used with the 8r0 strapping, and only speakers over
above 16r should be used with 16r0 strapping. ESL57 has average Z of 22r for the most energetic
AF between 80Hz and 300Hz. I would never use the 16r0 strapping for anything and I recommend
all 16r0 speakers be connected to OPT strapped for "8r0".

Where the load = strapping pattern value, the amp makes its maximum class AB Po of about 20W
but better quality sound is had where speaker values are twice the strapping pattern load because
there is about 17W available of pure class A.

For maximum clean sound, 8r0 speakers work best when connected to 4r0 strapping, and 16r0
speakers to 8r0 strapping. 32r0 speakers will be required for high class A at 16r0 strapping.
Nobody makes 32r0 speakers.

For all modern dynamic speakers between 4r0 and 8r0, I suggest use only the 4r0 strapping with
KT66.

When using output from Point Q, the existing links between T, S, R, and Q may all be ignored
because the two windings between T and Q are not connected in circuit if speaker connects to Q.
Between Q to P there are 102t using 4 x 51t windings in series/parallel.
Between T to R there is 1 x 51t, and between S to Q there is 1 x 51t winding.
When strapped for 8r0, all Sec copper is used with equal current density.
With 16r0 strapping all copper is used, current density is unequal.
With 4r0 strapping, two x 51t windings, ie, 1/3 of Sec copper is not used, so winding losses are
highest at the 4r0 strapping.

I think Peter Walker made a mistake by not having one more turret terminal on the OPT to allow
all the secondary turns to be used for a 4r0 load matching, and thus be able to to keep the winding
losses as low as possible. See my table below which indicates OPT losses for original
Quad-II OPTs.

Despite the high winding losses with the "illegal" 4r0 link of speaker to point Q, the power available
is not much lower than for other load matching. Graph 2 shows the measured power and it speaks
for itself.

Go forth ye hi-fi listeners, connect ye 4 ohms speakers to point Q, thy music will delight thee.

Further down this page I have details about how to alter the secondary winding connections inside
the potted OPT. This great amount of hard but skilled work will give you a few more W of maximum
output power, and you can judge if its worth it.

The Quad-II overload behaviour is fairly benign. Most PP output stages with cathode biasing will
suffer increasing Ek when driven hard in class AB. Quad did not intend that anyone would want
to play recordings by Heavy Metal to deafening levels. Real music has an average Vrms level much
below the maximum, and most ppl in 1955 listened to music using average of less 0.5W in each
speaker so all listening was powered by class A. Short lived peaks in music might increase to 25W,
but not for long enough to cause Ek change or B+ change.

During gross overload with a continuous sine wave at say 1kHz, with say 4r0 load connected
with OPT strapped for 8r0, and at clipping, the Ek will rise from +26Vdc to about +40Vdc, the
Idc to KT66 increases from 144mA to 220mAdc, and B+ at each anode reduces from 330Vdc
to 300Vdc, leaving Ea = 260Vdc, with tubes working in class C with grid current and high
crossover distortion due to increase in bias Eg1. Tubes will still will not overheat unless the load
is a short circuit, and then the tubes will definitely overheat and fail as I have seen happen with
use of faulty ESL57 with arcing panels. Nobody in a right state of mind will overdrive Quad amps
or any other amps.

I conclude Quad-II has fairly good inherent ability to withstand BRIEFLY excessive signal levels.
If you use a pink noise source with bandwidth limited from 20Hz to 20kHz, and you crank up
levels while watching the CRO, and you get the peaks of noise to just begin to clip, you will find
the average Vrms measured by a meter is about 1/3 of the maximum where clipping is produced.
Suppose you have Quad-II strapped for 8r0, and have 8r0 speaker load.  Suppose you measure
maximum peak Vo = 20Vpk. Clipping is at 14Vrms for an 8r0. But the Vac meter will read the signal
as 4.4Vrms, ( -10dB ) and the Po = 2.42W, which is 1/10 of the clipping Po. The vast majority of
Po is processed in class A with only peaks making the amp move into class AB operation.

This was why Leak 30 amps had individual R+C biasing networks for all their UL hi-fi amps with
KT66 and EL34.

Most output tubes withstand brief excessive saturation where instantaneous Pda or Pdg2 exceed
ratings temporarily. But if Vdc and Idc conditions change for long enough, output tubes will
overheat badly and a tube or two is doomed unless the amp is turned off. Used sensibly, and
with active protection, tube amps last quite well.

To get lower winding losses requires different OPTs. I've looked everywhere and it seems only
Majestic in UK make replacement OPTs which are supposed to offer better better performance.
They fit inside the original Quad sheet metal pot. But I doubt anything anyone could make which
would fit inside the existing pot could give better results, because the pot size is so difficult to fill
efficiently to get enough wire and iron inside to give low winding losses.

The original Quad-II OPT core has non standard E&I dimensions, ie is is not wasteless pattern.
To get significantly lower winding losses and lower Fsat for Quad-II, very drastic changes to the
whole amp is needed where the screen supply choke and GZ32 are removed and the area on
chassis used for the two EF86.
The chassis plan area for OPT becomes much increased which can be occupied by an OPT
using standard wasteless EI lams with 32 tongue which has plan area = 96mm x 80mm.
The core stack can be 75mm. There is no need for potting, and a bell end cover over windings
may be made to match the top of power transformer pot. With all painted same Quad grey,
it will look fine, and the OPT will be able to handle TWICE the Po produced by Quad-II.
Details of a decent OPT for Quad-II are at bottom of this page. Don't even think of going there
if you do not have good hands-on experience, time, knowledge, patience and money. 
I have a pair of Quads which will be altered to this recipe, but will use 2 x KT88 plus 2 x 6CG7
or similar.
The original Quad-II OPT will be used in another amp I have which has 3 x 6CM5 seen at
2323-triode-integrated-6cm5.html

However, if the original Quad-II OPT is retained, better technical performance and sound is
possible if the original circuit is upgraded with modern R&C parts unavailable in 1955 and
including high value electrolytic capacitors. There is no need for GZ32 or GZ34 which are
best replaced with silicon diodes. The use of a couple of LEDS and a couple of small bjts can
now be used for protection circuits and bias balance indication to make sure an owner knows
how his amp is going, and if there is a faulty output tube. This is prudent in an age where modern
people are just not used to the unexpected and perhaps smoky failure of the primitive amplifiers
of the 1950s, and there is now no Quad Company Support where you can buy replacement PT,
OPT or choke or anything else in tubed Quad preamps, power amps, AM/FM tuners, etc.

The quiescent bias anode currents of the output tubes change as tubes age. In original amps
the Idc in each KT66 may easily become very different because there is only one shared
"cathode bias" network R12 180r and C5 25uF.
As tubes age the tube grids begin emitting electrons thus conducting small but unwanted grid
currents even at idle so the Eg1 may rise to a positive Vdc above the bias supply point at top
of R10, 100r. The Vdc measured across R10 should be about +0.23Vdc. The chosen value
of grid bias R7, R9, of 680k is much too high.
After 50 years, typical R values go to 750k. The high values were chosen to allow the weak
EF86 to operate without reducing the voltage gain by having RLa too low. But the +Vdc which
appears across the 680k in an ageing KT66 causes the idle current to go higher which raises
tube temperature which causes even more +Vdc across 680k and and more heating. This is an
unwanted thermal positive feedback mechanism.

The Vdc across the 680k measured normally should be < +0.5Vdc. Where Eg1 > +1Vdc above
the earthy ends of 680k, the tubes are nearing the end of their reliability. I have seen KT66 at
near the end of their life with +9V at the grid at idle and with 90mA of anode current with slightly
red hot anodes. This is disastrous for the music, and such a tube continues to overheat
insidiously before finally melting down internally, and perhaps terminally damaging a power
and / or an output transformer. If ONE KT66 begins to conduct too much Ia, the Ek rises with
high current in R12 180r. This rise in Ek tends to turn off the Iadc in other KT66.

In Quad-II input stage with 2 x EF86, the Ikdc of both tubes flows through common Rk 680r and
through NFB network resistor of 100r. This means there is about +0.24Vdc at top of 100r, and
both 680k grid have 0.246V at one end, so you will never measure Eg1 to 0V = less than +0.24Vdc.

The output tubes rarely ever age at the same speed. So while 90mAdc may be flowing in one
tube, there may be 40mAdc in the other and there is a 50mAdc difference in the two KT66.
This Idc can magnetize the core to a high Bdc level and cause bad distortion. The OPT core
has no air gap and was designed for well balanced and equal Iadc = 70mA in each output tube.
The Idc imbalance causes high THD tubes and very limited bass response because Bac max is
much reduced. Everything is worse if the old original Hunts 0.1uF coupling caps from EF86
anodes have become leaky, further increasing the positive grid voltage. I recently found two
such 0.1uF caps had become 400k resistors. The KT66 had high +Vdc at each grid, saturating
both KT66 to have about 500mA flow in R12 180r, so it fused open fairly soon.
EF86 input tubes are set up in the original amps in what is called a "floating paraphase phase
inverter". It means a fraction of the output from V1 anode is applied to V2 grid to achieve two
equal amplitude drive signals to the KT66. The R4, 7, 8, 9 all "float" on top of the global NFB
network. The feed from V1 anode to V2 grid is 6dB of positive FB.
You may think the distortion is increased 6dB as a result of the PFB. But the feed allows V1 and
V2 to be a balanced amp and most even numbered H are reduced. Odd H may increase because
of the PFB. In practice it is not a serious fault, because output tube THD will always be much
higher than drive amp THD even with the PFB. In many Quad-II I have seen, they have not been
serviced anyone qualified and output tubes have unbalanced anode currents and resistance
values have changed and signals to each output tube grids are badly unbalanced. THD can
measure up to10 times more than it should at all levels.
Graph 3. Original Quad-II THD vs Po.

In Graph 3, Curve A is THD for original Quad-II without any Global Negative Feedback.
To measure this,
the R10 100r is shunted to 0V, therefore not allowing the fed back voltage from Vo to appear at
V2 grid via R8 and at under R4 680r.
The amp has 16r0 load connected to OPT with 16r0 strapping, and max Po is class AB1.
There is 9W of initial pure class A.
One can say that because the EF86 driver amp has low THD compared to output stage, then
measured THD without GNFB is mainly due to KT66.
THD rises to to about 2% at 9W, then rises to 5% at 22W and clipping.
It would be worse if there was no CFB winding and KT66 were working in pure beam tetrode without
CFB. The CFB makes KT66 act similarly to being triode connected without CFB, while being able
to give higher Po. Class AB triode without any NFB at all would not perform much better than what
we see in curve A.

Curve B is for the same amp but with normal GNFB, ie, with no shunt across R10. The average
gain reduction factor = 1/7, about 0.143, = -23dB. The THD reduction at 9W is from 1.35% to
0.07% = 0.0518, about -25dB. At 18W the THD reduction = 1/27 = -28dB. The GNFB does a
lot to reduce the gain.

The output stage gain does not vary much with load change between say 8r0 or 16r0 because
of the local CFB which converts the output stage to have very similar overall gain change between
9 with no load at all, and about 3 when RL = 4r0. and about 4 with 8r0 load. So when no load is
connected, the overall open loop gain increases by 2, or +6dB so the total amount of GNFB
increases to 29dB, quite high. The gain increase within the output stage between 8r0 and no load
is about 10, or 20dB, so in effect the total amount of GNFB and CFB without any load is huge, over
45dB. So no wonder LF oscillation occurs with no load connected. I found pure C loads at output
cause oscillations, but never when there is an R load present of at least twice the strapping
value, ie, about the value for pure class A Po.
The gain reduction with GNFB should be equal to THD reduction. But the figures I measured
don't confirm the text books, and maybe you get more or less THD reduction than the gain
reduction with GNFB.

If you connect a CRO to either output grid to view the wave driving output tubes, you will see that
its THD has become about 5% by the time clipping begins, so in effect, the GNFB fed back to
input is amplified by input EF86 by about 39 times so that the drive signal of output tube grids
contains more THD than measured without GNFB at clipping. This sort of thing occurs in all amps
with NFB. 

There are other explanations of just how NFB works elsewhere on this website.

The measurements in Graph 3 are quite good for any tube amp.
THD spectra is mainly 3H at all levels but many higher H are present. When viewing the THD
at low levels on oscilloscope, the THD envelope is much modulated by rectifier noise.
The presence of saw-tooth shaped 100Hz hum at OPT CT makes small changes to gm of output
tubes, which affects their gain enough to modulate the small signal and distortion levels.
Most of the hum Vac at CT is excluded from OPT sec by common mode rejection, but I recall
concluding that the level of noise caused IMD at low level was equal to normal THD production
by tubes, as well as the small amount of HD caused by iron in the OPT.

Peter Walker thought his Quad-II were good enough with a 16uF anchoring the OPT CT.
The quality of electrolytic C was not too good in 1955, and remarkably, I have found the 2 x 16uF
used in Quad-II were reliable, with many lasting until 2016. The old C were well sealed inside metal
cases. By about 1960, better electrolytic C appeared and began to get smaller and more reliable
and a standard mod for Quad-II was to replace the 2 x 16uF with 2 x individual cylindrical 33uF.
But that only lowered hum at CT by -6dB, and at screen Eg2 by -12dB. I found diode switching
noise is present in old Quads.

The undulating shape of both curve A and curve B indicate some THD cancelling is probably
going on, but there is no need for me to bore everyone with further uncertain explanations.
The THD we measure is what we see and that is the certain reality.

I measured the THD with 32r and with OPT set for 16r0. This gave about 17W of nearly all pure
class A Po and with GNFB the THD was below 0.1%. This was what everyone really expected
in 1955. ( But it was not long before many people "upgraded" to Quad 303 and then to 405 which
gave 100W per channel from Squalid Stait, but which were better able to power ESL63. )

Curve C is for a MkIV Dynaco Monobloc I recently re-engineered with 2 x KT88.
See Dynaco-mkIV-reformed.html
Including Curve C is a very cheeky sneaky move of mine. Curve C is like a mongrel dog running
loose at a dog show for poodles.
Well, the mongrel Dynaco certainly has been nicely groomed, and isn't raising a leg on all the
sheilas present, but shows how well it can yodel compared to the productions of the stuffy grey
poodles bred up by the Poms in their little Glorious Island.
IMHO, whenever we consider audio gear, we should always be prepared to compare one amp
or speaker with another, lest we loose all understanding of "better" or "worse", both by comparing
the sound and the measurements. Notice the Dynaco THD curve is almost a straight line, somewhat
typical where the input and driver tubes are all low µ triode such as the 6CG7 I have used in the
MkIV re-build. The 6CG7 is what I think is the King of little triodes. He stole the crown from old
King 6SN7. But other fine princes are 6DJ8, 12AT7 and 12AU7.

The reformed Dynaco MkIV was tested with the same RLa-a as for Quad-II, about 4k3, with the
original Dynaco OPT. The output load is 4r0, less than the minimum 6r0 I recommend. The MkIV
has 33% UL taps with KT88, and 6CG7 input & drivers. Si diode rectifier and B+ ripple noise at
OPT CT is -60dB. The MkIV has 14dB with 4r0 load, some 10dB less than Quad-II, yet the Dynaco
has THD at -6dB lower up to 2.2W which covers listening levels for most people.

The Dynaco gives 38W class AB1 max with 4r0, but only 5W initial pure class A. So while the Quad
has less than 0.1% THD at 9W, the Dynaco has 0.2% because it has moved into class AB1
working. But the THD spectra in re-enginered Dynaco THD is mainly 3H + slight 2H but has much
less other rubbish than in Quad-II spectra.

The majority of listening is done with amp power less than 2W. The Dynaco's KT88 have Pda+Pg2
= 20W, so they run cooler than the Quad KT66. If the KT88 were idled with Pda = 30W, the first
9W would have less THD than Quad's. The class A Po is determined by RLa-a and the idle Iadc.

The talk of Dynaco is not quite right because its circuit has been changed to mine, and the
reformed amp is not a Dynaco any more than it would be a Potato, or Tomato. Is it a Dynaturn? 
Turnerdyn? Turnaco? But the Dynaco has major benefits of large electrolytic C and triode
input / driver. It has original Dynaco UL OPTs, not potted, heavier than Quad-II, but re-strapped
for better load matches. PTs have been replaced for Australian mains at 245Vac. The original
Dynaco mk IV wasn't any better than Quad-II.  

KT88 can have the same 70mAdc as KT66, but have higher Ea, and therefore will make more class
AB Po total. I doubt that "extra heat is worth the ears". Some customers will say they like their music
well cooked, but I would add that needs a good chef who would never over cook anything, and who
lets his sense of taste and good nutrition be the final guide. 
The Amp Master should understand all the numbers extremely well, and never guess a single action
he takes, but when he is finished the music must sound well to a room full of people.

I can conclude that if one is to re-furbish such grand but limited old amps, it is always possible to
improve the circuit behavior to get a better measuring and better sounding amp.

Fig 2. Basic Reformed Quad-II schematic....

The component numbers used here don't relate to any component numbers in the original
schematic except by coincidence. This schematic is a re-drawn version I did in 2007, and includes
slight changes. More improvements could be done but the above has what I consider to be the
minimum, including........

1. Remove old 2 pin mains Bulgin chassis socket to rubbish bin. Install IEC 3 pin mains chassis
plug with 2AG mains fuse included for standard IEC mains cable. This may infuriate those wanting to
keep the Quad-22 preamp arrangement intact. Safety comes first, and if you ignore this step 1, then
don't blame me if you electrocute yourself, or family member.
Install mains on-off DPST rocker switch. Bypass switches with 10nF 2kV ceramics, ( not shown ).

2. Install RCA input socket just near existing Jones socket for Quad-22 preamp. If you doubt you'd
ever spoil your listening with an original Quad-22 control unit, then remove the Jones socket.

3. Note C1A and R1A to deal with Vdc swings in external preamps.

4. Connect R1B between 0V at Jones plug and the chassis. This interrupts mains earth loop currents
which can cause hum with other audio gear connected, preamps, CD players etc.

5. Remove grey box with 16+16uF caps inside and put in rubbish bin. Disconnect HT windings and
5Vac heater windings from GZ32 valve socket.

6. Decide if you really must have a dead tube in your amp to make it look right. Did the Kremlin
function better with Stalin's preserved body left hanging around to be observed by his millions of
dopey followers? At least the Russians have their love of music, and life, and most know when to
leave the worst of the past behind, while hanging on the best, and watching carefully for those
addicted to the past.

7. Install well thought out terminal strips to allow the B+ rectifier circuitry to be built. Do not try to
use GZ32 as a slow turn on series diode for B+ rail. I tried, and it didn't work out because the
KT66 heat up more slowly than the GZ34, and the B+still soars +440Vdc for a few seconds
before the KT88 anode current pulls it low to about +370Vdc with wanted 145mAdc loading by
KT66 etc. This indicates the B+ rail supply output resistance is initially about 480r. The drop of
B+ during class AB Po can be -35Vdc or about 10% when Iadc increases from 145mA to about
200mAdc max, so Rout = 350r.
This is much reduced with minimum added series resistance after the HT winding which means
using Si diodes and CRC with low R and large C values in CRC B+ filter.
The idea of the tube diode B+ delay would require a delayed turn on for 5Vac heater. That is too
complex and also useless, and it is better let B+ rise to +440Vdc within 5 seconds after limiting
inrush current at turn on as shown in other schematics here, and the power tubes will sort them
selves out gracefully while heating up.
Having fixed bias is probably best for output tubes during turn on to avoid high initial peak cathode
currents because the -Vdc for bias is established within 2 seconds after turn on, and well before
KT66 cathodes begin emission which begins after 12 seconds. Turning a tube amp off then on
again after 4 seconds while tubes remain can cause excessive Iadc during the turn on.
Slowing that first 4 seconds is wise to limit the rate of initial current increase. 

8. The HT is rectified with 1N5408 silicon diodes through current limiting R21, R22, 47r / 5W into
C11 47uF. This gives B+ = +405Vdc approx when mains are 245Vac and mains taps are soldered
for maximum possible.
I removed the the mains adjustment switch on original amps because only the highest Vac input
should ever be used. Having 245V applied to say 220Vac will make B+ after Si diodes reach +445V
and the KT66 Iadc be too high and idle Pda will be exceeded. So with KT66, the B+ should not be
increased to more than about +30V above what it is with GZ34, so added series resistances I show
are quite forgivable. With C11 47uF, 100Hz ripple voltage = 6.5Vac, and after R20 135r and at C10
470uF the 100Hz = 0.17Vac, and 1/100 of the level in original Quads. This is a vast improvement
on original amps where Vr at CT = 18Vac. If the amps are to always be used for low volume below
10W the high B+ output resistance does not matter at all. All Vdc rails will not move much during
pure class A operation with music.

9. Install the 470uF laying on side under chassis and alongside 47uF. The 47uF is minimum C value,
and you could have any value above up to 470uF, because the 47r helps prevent excessive peak
currents charging the caps. The 47r are not essential but add to the effect of winding resistances
of PT to lessen peak charge currents. The GZ32 or better GZ34 with C higher than 33uF makes
peak charge currents in diodes rise close to quite low limits for tube diodes. Use of 47uF will cause
GZ32 to arc internally, and GZ34 barely cope. 100uF will destroy both easily. But a pair of IN5480
will last maybe 500 years, even if there is no series 47r and C11 = 470uF.
 
The 470uF Xc = 11.2r at 30Hz. 16uF Xc = 331r, so 470F is a better to anchor the OPT CT to 0V to
minimize rectifier noise getting into signal path. Although B+ soars to +440V at turn, it settles back
to about +390V at the OPT CT after KT66 draw current within 20 seconds. Modern 450V rated caps
will cope OK with the temporary rise of B+ to +500Vdc. 

10. For best noise free operation of tubes in tetrode mode with fixed Eg2 screen supply, this B+ rail
be very well filtered and stabilized. While operation is all class A the high noise at OPT CT did not
matter much when it was so difficult and expensive in 1955 to make rail hum lower than we find it.
Because Ra is high, there is common mode noise rejection by the balanced primary winding with CT;
the hum at CT is applied equally to each KT66 anode, so little hum current flows across the OPT
winding. Original Quad-II choke is 20H x 600r with 16uF, and the attenuation factor of 100Hz = 0.008,
so the 18Vac 100Hz at screens in original Quad was reduced to 0.14Vrms at least. But it is far from
perfect where I would prefer less than a few millivolts. Using 47uF instead of 16uF means 0.047Vac,
better, with some way to go. With the 47uF + 135r + 470uF, 100Hz at 470u = 0.17Vrms, and 20H
choke + 47u reduces 100Hz at screens to < 0.5mVrms.

11. The two KT66 cathodes are individually biased with R12+C6 and R13+C7 networks of 470r / 5W
and 470uF respectively. These networks are connected between pin 8 on KT66 sockets and the
wires leading from ends of the cathode feedback winding on OPT. Make sure the POSITIVE end of
cap connects to pin 8 for cathode on KT66 socket. Under dynamic music conditions the very slow
time constant of the cathode bias networks prevent much movement of the cathode Ek even when
music peaks occasionally reach up to clipping levels in class AB1. The balancing of the KT66
cathode currents is automatic with the two bias networks, and as the tubes age they keep their Ek
constant and Ikdc imbalance is negligible compared to the original biasing with a common 180r+25uF.

12. The original 180r+25uF bypass cap are removed, and a wire link soldered across the two turrets
for these 2 parts. This connects the cathode feedback winding CT to 0V. The theoretical Rk for each
tube should be 360r, twice 180r, but 470r gives just slightly less Ikdc, and good control of Ek. The later
Quad-II-Forty has individual cathode biasing with an almost identical schematic to Quad-II, but with 6SH7,
KT88, and KT88, and each Rk = 390r, and working B+ at about 450Vdc. The KT88 run too hot, and so
did 390r, and I have had to repair / re-engineer a couple pairs of Forties - which barely make more than
33W.

The original common original Quad 180r 180r was wire wound, and rated for 3W working which having
Ek = +27Vdc for 150mAdc which I have often measured with Ia+Ig2 = 75mAdc per tube. 180r Pd = 4.0W.
If Idc were to to rise to a total of say 200mA, the 180r Pd = 7.2W and it gets real hot, and I have seen
them go open because owners did not replace old KT66 beginning to conduct excessive Idc after their
10 years of constant use. When 180r goes open, Vdc across 25uF goes high, so it rapid fails to become
a short circuit, so Ikdc then increases hugely, with both tubes each conducting 300mAdc+, and maybe
the mains fuse blows, but not if only one tube has Idc sitting at 120mA and other at 50mAdc, and giving
a red-hot anode and BAD music in one channel which isn't noticed for months.

In theory, each Rk for original Quad should be 2 x 180r = 360r, thus allowing nearest standard value
= 390r, and definitely 5W rated. But IMHO, Ia + Ig2 is too high in original Quads and may be reduced
slightly, hence my choice of 470r to reduce Ikdc to about 70mAdc.

The Iadc +Ig2 dc is largely determined by Eg2. And Eg2 is close to B+, so that if B+ is raised +30Vdc
the Idc from B+ rail supply may increase +20mA, making KT66 Pda rise a little too high. Therefore
always use individual Rk, and NEVER less than 470r.

13. The stability of Quad-II with its high total amount of NFB is not unconditional, and pure
capacitance loads of 0.22uF with links set for 8r0 will cause HF oscillation. To ensure unconditional
HF stability, the network of R7 & C4 is connected between V3 and V4 grids. This reduces open loop
gain and phase shift of V1 and V2 above 20kHz.

14. I found the original amps oscillate at LF if separate R&C cathode biasing is installed without
adding the mods for larger C values in PSU. The original common Rk of 180r may have been
thought to be necessary, even when Peter Walker may have liked to use individual cathode biasing.
The common Rk gives less LF phase shift. The cathode input resistance of both KT66 is about 125r
and is in parallel with 180r for total R = 74r, and with 25uF the pole is at 86Hz, which does not matter
in class A. But in class AB it matters a lot. For each tube, cathode Rkin = 250r, and with 470r total Rk
= 164r, and with 470uf the pole is at 2Hz. It may well be better to use 47uF to get pole at 20Hz, which
may reduce LF gain below 20Hz where amp wants to oscillate at LF.
I found use of gain shelving R+C network of 2M7 parallel with 0.022uF between 0.1uF and output
grids stopped all LF oscillations with individual R+C cathode networks where the rest of the mods
were not attempted.  But Fig shows what I used which gave unconditional stability at LF and HF all
without gain shelving networks

15. The Fig 2 basic schematic for V1 and V2 unchanged from original. But R8+R10 680k may be
reduced to 470k without much lessening of EF86 gain, and to give a lower biasing Rg for KT66.
In many old Quads, the 680k have risen to 800k, and all other old carbon composition resistors
have also increased in value, and unequal values so hence all original Quad R are to be considered
toxic to enjoyment of Mozart or his Jamaican great great great grandson Bob Marley. With Rg each
470k, R9 2k7 must be changed to 1k8 to retain the theoretically correct feed to V2 grid. With an
increase to B+ of +30Vdc, B+ applied to EF86 stage is also increased, so Eg2 is higher so Iadc
and Ig2dc both increase which increases flows, so gm of EF86 increases so open loop gain
increases and effective amount of GNFB increases. Thus the margin of stability is reduced, so
the use of 470k metal film R for biasing V3+4 grids is not just wise, its necessary.

16. The bottom part of Fig 2 schematic shows the active protection provided by the group of solid
state parts which have ZERO effect on the sound. The circuit is powered by a small 5VA 240V : 12Vac
transformer which creates a +16Vdc rail to power SCR, green and red LEDS and relay coil.
The circuit has it own mains Active and Neutral lines after the on-off switch and mains fuse, but its
Neutral line does not include the relay contacts in series with Neutral line to large T2 power trans.

The circuit has no loading effect on tube operation and only reacts to excessive Ek at either output
tube cathode. Signal Vac at each KT66 cathode is filtered away by RC networks below R23 & R25.
If cathode Vdc, Ek, at one or other or both KT66 cathodes rises from approximately +33Vdc to 52Vdc,
the gate voltage at SCR rises to 0.65Vdc. This gate threshold voltage turns on the SCR which stays
turned on until the amp is "reset ", ie, turned off, then back on again after 2 seconds at mains switch.
The SCR switches the relay coil on which opens the contacts in series with Neutral mains line to the
large T2 mains transformer thus turning off T2 and the amp.

While the amp operates normally, the green led shows protection circuit and amp is turned on and
OK. If the SCR is tripped, it turns off the green LED and turns on the red LED which tells an owner
something is wrong. Such circuits work far better and more reliably than a fuse. Fuses are often
replaced by owners who choose the wrong type of fuse and the wrong current value. Many of my
past customers came to enjoy how my protection circuits work because they saved huge repair
expense on replacement OPT, PTs, and tubes etc. In the more elaborate modification to Quad-II
amps bias balance indication could be added, but its slightly more complex as shown shown in the
next schematic....

Fig 3. Reformed Quad-II for 2005........

Fig 3 SHEET A component numbers do NOT relate to any numbers in original Quad-II or any
other schematic except by coincidence.

For this design, I intended to retain the original working conditions for KT66, but I offered my
customer KT88, 6550, KT90. He chose KT90. I found the KT90EH made in Russia to be very
reliable providing idle conditions were much lower than the 50W Pda rating.

For this modification, I decided to use EF80 or 6BX6 to replace original EF86. There are a lot
of NOS 6BX6 around and they tend to have lasted OK without getting gassy. 6BX6 is a general
purpose pentode with twice the gm of EF86, and used in countless TV sets.

The operating conditions for 6BX6 have slightly higher Ia but lower RLa with 120k // 220k C
coupled grid bias R for output tubes. The use of original Quad-II 680k to bias KT90, KT88, 6550
is NOT wise because any idle grid current will generate too much positive grid Vdc when the tubes
have slightly aged.

The 6BX6 are set up as a true differential balanced amp, aka long tail pair, LTP. I did not want to
use the original Quad-II paraphase splitting arrangement which is positive feedback.
To make the LTP work best, the common cathode resistance should be a CCS. To get the
2 x 6BX6 to work in pure pentode mode, the 2 screens should be bypassed to common cathodes
and each with separate series supply R8+9, 270k. But the common cathodes are loaded by these
R because screens are bypassed by C2+C3 0.1uF, so a CCS is not really possible. Screens fed
Idc via R10 56k and C4+C5 bootstrap the R8+9 270k to NFB Vac at top of R2 100r.

V1 6BX6 g1 is for signal input, V2 6BX6 g1 is for GNFB input from the GNFB R divider R20 470r
and R2 100r.
The 100r is a low enough R to happily bootstrap the cathode R tail of R7 66k and screen feed
R8+R9. The bootstrapping with C4+C5 from R2 100r reduces the Iac in the cathode and screen
R so that the effective cathode R tail value is about 10 times the 44k static value of
66k // 270k // 270k, ie, 496k. If there is 36.00Vrms at V2 anode, there will be 36.3Vrms at V1 anode,
and balance is within 1%, and a lot better than in countless amps I have worked on.

The forward signal to V1 grid creates balanced drive to both output tubes, and the NFB applied
to V2 grid creates balanced application of correction signal. Despite the circuit using many more
parts than in original Quad-II, the operation of LTP is PURE and BETTER. An interesting tube
which may be used is 6EJ7, with lower anode RLa and which allows higher Iadc = 3mA to get
gm higher and get gain of 200. Radford used 6EJ7 and afaik, there were many ppl who thought
Radford tube amps to be better than from Leak or Quad or many USA and European
manufacturers.

There are quite a lot of 6EJ7 around. It is a quite gutsy frame grid pentode. Another good old
pentode is 6SH7, but is octal, and is in Quad-II-Forty. Don't use 6SJ7 which is like older 6J7 and
EF86 with low gm. 6CA7 is also another high gm octal pentode. Sadly, it is difficult to find finding
NOS 6SH7 or 6CA7 in good working condition and which are not microphonic or noisy because
of gas in the tube; both are metal envelope tubes developed in WW2 to prevent tubes in radios
being shattered - as did happen when a grenade was thrown into a radio operator's room.
The radio could be repaired, but the operator was usually blown to bits.

The KT90 give output resistance that is much lower than KT66, slightly lower THD and better
maximum current for class AB operation with RL down to 1/2 the value used for nominal strapping
value. If Quad-II is strapped for 8r0, an 8r0 load gives max AB Po of 23W, but class A Po is limited
to 7W. But where the load is 4r0, with strapping for 8r0, then the Po is poor quality because THD
is high and damping factor low. However, KT90, KT88, 6550 will cope better than KT66, 6L6GC,
EL34.

Biasing KT90.
The two KT90 cathodes are individually biased with R18+C11 and R19+C12, 270r+1kuF, and
with Ikdc of 62mAdc, expect Ek to be +18Vdc. The anode Vdc-0V is +375Vdc at idle, so idle Ea
= +357Vdc.
This is about +37Vdc more than in original Quad-II so the Va peak swing can be much higher.
The high capacitance values of PSU and bypassing means the rail Vdc do not change much and
non sustained power can be 30W.

The +18V cathode bias is far too low to be useful for biasing on its own because all large octal
tubes require bias Eg1-Vk to be about -36V. To achieve this, I have a fixed bias supply of -18Vdc
to each V3+4 grid via the Rg 220k. The pair of 270r used for R+C biasing is sufficient to give
fair regulation of idle Iadc. If Iadc increases with class AB then the rise of Ek will be much less
than use of Rk = 470r or 560r.

I have VR1 arranged as an adjustable 10k trim pot accessible with thumb nail or kitchen knife at
side panel of amp. It is located close to 2 x red LED near the output tubes. This pot is used to
adjust the balance if of Ikdc for each tube within 5% accuracy very easily, and visually, without
needing a volt meter.

The network around VR1 includes 2 x 10V zener diodes, which may be 5W. After turn on, and
with VR1 set to its centre position, -18Vdc to both KT90 grids is immediately established before
tubes warm up. Because most pairs of output tubes will have different warm up times, the Idc in
each is unequal and one LED will appear on with other off.
But once the tubes have warmed up after say 10 minutes, there may be 10mAdc difference, with
Ikdc = 75mA in one KT90 and 62mA in the other, and despite the presence of separate R+C
cathode biasing. To get balance, VR1 is turned in either direction slowly until both red LED
appear equally bright.

VR1 adjusts the fixed bias Eg1 applied to either output tube until equal Iadc is attained when the
two LED are equally bright. This balance adjustment is different to all others I have used where
one the turn of the pot makes Eg1 applied to one grid rise, and Eg1 to other fall. to tube one.
Where the Iadc at idle is low, the Pda will be low, so it does not matter if Iadc in one tube is slightly
increased with Eg1 less positive, and Iadc in other tube slightly reduced with Eg1 made more
negative.

But where we want a lot of class A, then we may find we do NOT want to make a tube which has
ideal Pda get any hotter while adjusting the other tube to get down to being equal the first.
Then you have two tubes both slightly too hot.
So, the best simple solution is to prevent the Eg1 to ever rising positively above the nominal fixed
bias value, while allowing Eg1 of either tube to be reduced. This may result with two tubes slightly
cooler than intended, but that is far better than having two tubes slightly too hot.
The arrangement shown has 10V zener diodes to limit Eg1 rising above -18Vdc, on one tube while
allowing the other tube to have Eg1 reduce to -24Vdc max,, and the possible 6Vdc swing is usually
enough to cure whatever Iadc imbalance may exist. Once VR1 is set for balance of Iadc, it may be
left alone for months. The LED may flash or change brightness in high volume class AB Po,
especially when teenagers are allowed to "see how fah-king loud dad's old hi-fi amp can go."
Usually, heavy sustained clipping with a low speaker load will cause amp under attack from
teenagers to behave like teenagers, just become sullenly silent, with a red "fault" turning on, telling
the invaders to get lost. Active protection has saved many of my amps from teenagers, speakers
that are kaput, shorted speaker cables, or randomly failing tubes which suddenly decide to conduct
way too much Idc. ( Some owners buying expensive NOS tubes have been surprised by a tube
quitting a month after buying it. ). But 40 years on a shelf in storage does not do a tube any favours.

In the previous 2014 edition of this page, I faithfully showed the schematic of 2005 which had -5Vdc
applied to output grids and which allowed each Eg1 to be adjusted +5V or -3V, which worked OK
when I tested the amp, but this page for 2016 has the better circuit arrangements shown.

The schematic for PSU and balance monitoring is below, and has YELLOW LED nominated form
balance, and this avoids confusion where someone gets frightened by 2 red LED turning on.

Fig 4. Reformed Quad-II 2005, PSU and balance + protection.

The balance pot in SHEET A is mounted on an internal bracket so only the short 6.3mm shaft
appears at hole in chassis panel, allowing easy turning by thumbnail or a screw driver. I make no
apology of having 4 LED on this amp; all are there to allow a civilized relationship between owner
and amp. Any resemblance to Christmas tree lights is entirely unintended.

HT is rectified with silicon diodes. There is really is no need for additional series R between HT
ends and C8 100uF because peak currents are limited by the output winding resistance of about
90r, and 1N5408 have 3A continuous rating. I have 4 x 1N508 in 4 pairs of two in series to give
higher peak reverse voltage rating. You might use 1N4007, but I most certainly will not; I want the
diodes to last, and even when a tube or cap becomes a short circuit which would cause the mains
fuse to blow.

The original Quad-II had 16uF after tube diodes so there was 18Vrms of 100Hx ripple at CT.
With my arrangement, there is 2.9Vrms at C8 100uF. The R3 100r plus C4 470uF have 100Hz
attenuation factor = 0.034, so Vripple at top C4 = 0.1Vrms, which is 1/180 times less than in
original amps. There is no further betterment of sound to be had if Vripple is reduced by having
a choke instead of R3 100r.

The output dc resistance of the Vdc output from C3 470uF is lower than found with use of GZ32.
Heavy class AB operation is fine, with maximum B+ reduction with say 4r0 at 8r0 strapping giving
rail sag of 10%. The 470uF has Xc = 11r2 at 30Hz, the old 16uF was 331r.
After removal of 2x16uF boxed electrolytic caps from 1955, there is plenty of room for 100uF+470uF
rated for 450Vdc.

The GZ32 was disconnected, but left in this pair of amps, having no useful purpose except to fool
dumb onlookers.
The amp design called for -28Vdc rail so I used the 5Vac GZ32 winding in series with one end
of 6.3Vac heater winding to get 8.2Vac. This is applied to a half wave voltage tripler rectifier to
get -30Vdc, then reduced to -28V after RC filter. The size of 470uF caps for low Vdc is tiny; and
there is plenty of chassis space.

The 8.2Vac is also rectified to make +10Vdc which is filtered down to +6.3Vdc at low ripple for
6BX6 heaters, each needing 0.3Adc.

There is a -390Vdc rail for the low 4.3mAdc supply to cathode current of V1+V2 6BX6.
In SHEET A the Rk between V1+V2 cathodes and -390Vdc have 33k + 33k + 19k in series,
which limits the amount of Vdc across metal film resistors rated for 0.75W. Never tempt fate by
having more than 200Vdc continuously across any resistor, I have replaced so many which just
went open for non apparent reason.

The amp protection and bias balance circuit requires a small 5VA 240V : 12V transformer, T3,
mounted somewhere conveniently under the chassis. This provides power for the small circuit
board used for the
solid state devices. There are two red LED on the chassis top beside each KT90 in the picture
at the top of this page. They look a little dull, but have same brightness. If one goes out, the other
goes brighter, indicating the nearby KT90 has more Ikdc than the other. Yellow LED also look OK.

With KT90 I found you get 25W AB into 8r0 class AB with slightly more into 4r0, and Rout = 0.78r.
KT66 could never achieve this. But the KT90 has much higher current ability for the adverse
loading condition.
 
Once the bias balance LEDs have been adjusted for equal brightness, any change in output
tube Iadc will always make one LED glow more brightly than the other, which may tell an owner
to re-balance the bias, so he may just poke a finger to make LED equal brightness. 
When the tubes continue aging and becoming more unmatched, the amount of turn to get
equally bright LED may not be available, and this tells an owner it is time for a new tube or two.
 
If the owner ignores the LEDs, then Iadc in one output tube may rise into the dangerous region
where a tube is too hot. Many owners ignore such things and do not hear the immediate
degradation of music. When a tube has gone past poor Idc balance, and has Idc more than twice
the idle Idc for longer than 4 seconds, the Ek will rise enough to send a signal to the SCR which
turns off the amp automatically to save a huge expense on PT or OPT etc. The owner just
cannot ignore this.

There is no danger to OPT with use of larger tubes than KT66. The protection circuit will
prevent any damage. The idle Idc for KT90, KT88, 6550 etc is less than for KT66.

The 2A mains fuse in original Quad-II only ever protected against short circuits in PT.
The fuse would not blow even with one KT66 becoming totally red hot and melting down from
bias failure. Much damage could and did occur to many Quad-II amps before the damn fuse blew.

I found 0.5A slow fuse worked OK. It may fail after a year from repeated heating cycles at turn
on but 0.5A means the mains input power must be about 140W before it blows. A 2A fuse blows
when Pin = 480W.

But KT90 may conduct 500mAdc. If there is a continual 500mAdc in 1/2 the primary of Quad-II
OPT which has Rw = 167r, then 41W of heat is generated in the 1/2 primary winding, so it will
fuse open !
Unless the amp is turned off well before Iadc reaches 500mA, the OPT winding will fuse before
you can feel the OPT getting HOT.
Even KT66 with max 300mA Idc will produce 15W of heat in 1/2 the primary which will damage
or fuse the winding. 

It is no good to use a fuse between cathode to CFB winding because it will have to be 200mA
rated to avoid nuisance blowing. If additional fuses are to be used, place one 0.5A slow blows
between each end of HT winding and subsequent R and diodes.

It is ALWAYS better to have an active protection circuit.

The Quad-II set up for Fig 3 above schematic draws 88W from mains, so with 245V mains the
input current
average is 0.36A. This is less than in the original Quad-II which has the GZ32 heaters consuming
about 12W more. KT90 are perfectly interchangeable with KT66 and draw the same idle current.
But KT90 can produce a outright maximum of 30W instead of 22W with KT66. KT90 heater
current is 1.6A instead of KT66 at 1.3A but it is OK because these amps were designed to have
add on tubed preamps and tuners which will never be used with this pair of amps.
The musical performance includes tighter bass and more controlled and detailed treble, so I have
to say KT90 in Quad-II sounds better. KT88 or 6550 may also be used.
Stop being nervous. The bigger tubes have a very good sonic flavour.

Two amps on bench in 2006....

These are two Quad-II amps on bench with KT90 output tubes.
The reformed integrated preamp is right side of amps. In the shelf below there is a manufactured
black PSU for preamp with same box shape and size as preamp.

This picture shows the whole system being trialled for a couple of days before sending it off to my
customer. The GZ32 are not plugged in to keep it looking original, but may be if wanted.
Each amp has its own red rocker type mains switch and blue "on" LED and a mains fuse that is
accessible without mucking about behind the bench. The aluminium panel for mains switch covers
holes for original mains voltage settings. I have a fixed highest mains Vac setting because Australian
mains is usually over 240Vrms unless high load power pulls it lower on hot days or freezing nights.

 
This shows the rear end of amps with new IEC chassis plug to replace original Bulgin.
The original Quad-II recessed plastic 4 mm banana sockets had become cracked and were
replaced with something from some retired HP test gear.

The new signal input socket is Canare 75 ohm RCA which replaces the original 6 way "Jones"
socket for use with Quad-22 preamp which had umbilical cables to each power amp. The terminals
are mounted on a white fibreglass panel. The appearance was not important. If I had done a better
job it would have included much more metal work and repainting etc, and all of that would make
no difference to the sound, and would have doubled the low price I charged for this work.

Shielded interconnect cables from a preamp should only be used since the speaker output
cables are close to the amp input. I did try using unshielded dual foil cables which were close
to the speaker cables. No HF oscillations occurred, probably because the live input cable is
tied to the low impedance of the cathode follower in the preamp. But I do not like unshielded
twisted pair or flat foil interconnects because they often pick up switching noise, mobile phone
noise and other noise from nearby mains cabling. Cables using twin 4mm wide x 0.1mm thick
copper foils and inside a polythene hose are fragile, and always break sooner or later.

I quite like RG58 coax cable for interconnects. A very much modified Quad 22 control unit is
described in my page on Quad22mods.
For each power amp I used a blue "on" LED and red mains on switch

Blue LED are too bright with say 4mAdc, and they run tolerably bright with about 0.25mAdc.
Owners prefer blue, but I prefer green for "on" using a plain diffused 5mm dia type which runs
on 4mAdc.

The Po from the pair of KT90 with other mods is...
Graph 4. Po versus RL for 3 strapping patterns of OPT Sec....

This shows slight increase in Po for KT90, and was prepared using continuous 1kHz sine
waves up to clipping. At lower RL, the B+ sagged 10% and fixed bias of -5Vdc was used with
470r for each Rk for KT90 cathodes. 41V Zener diodes limited Ek rise during testing.
My conclusion is that with schematics above, instant power of up to 30W+ is easily possible.

For those who worry about THD, here is....
Graph 5. THD for 2005 Quad-II with KT90.

The above graph 5 is drawn on logarithmic axis for both THD and output Power. The test is for the
Fig 3 REFORMED schematic of 2005 reformed Quad-II, and shows results for KT66 and KT90 with
same loading in the same circuit with same GNFB which is much less than in original Quad-II amps.
(( However, in 2014 I tested an original Quad-II amp with exact original schematic and in fair
condition, and the THD is plotted in Graph 3 above on this page. ))
In Graph 5 you can see that at onset of visible waveform flats on CRO, ie, amp clipping, both KT66
and KT90 produce about 1% THD at 21W and 24W into 8r0 respectively. KT90 have an average
of 1/2 the THD produced by KT66, but make 1/3 of THD at 3W which covers most listening levels.
KT90 make 0.03% at 3W, 0.1% at 14W. KT66 make 0.1% at 3W, 0.2% at 14W.

The curves with all their kinks are typical class AB tube amp measurements when driven with
pentodes which produce more THD than triodes. But at clipping the THD and IMD does not matter
as much as at normal levels < 3W. This is where our focus on the soloist is intense, but really,
anything below 0.05% is not too bad, considering tube amp artefacts are less objectionable than
solid state's.

The trend for the KT90 to have half the THD of the KT66 continues below 1/4W. In 2006, it was
difficult for me to measure THD accurately because at 1/4W, Vo = 1.41Vrms, and if THD = 0.01%,
then the THD = 0.141mV, and noise can be easily 0.5mV which obscures observation on CRO of
the THD. The noise becomes the dominant artefact at low volume. If noise = 0.5mV, and does not
increase at say 14W where Vo = 10.6Vrms, then SNR is said to be -86.5 dB which is quite
acceptable.

If you hold an ear close to a bass or midrange speaker cone, and with input shorted to 0V, and you
barely hear anything, then you have what 0.5mV noise sounds like in average speakers. The noise
is of no concern. But I have had to fix amps which you heard humming from across the room.

I also tried measuring THD with loads between 4r0 and no load at all. With a 16r0 load connected
to the the amp set for 8r0, at 1W the THD is 6 dB less than with 8r0 and with 4r0 it is about twice
what the 8r0 produces. At low levels below 2W which covers much listening for many people, any
load down to 4r0 is OK, and KT90 give around 1/2 the THD of KT66 for all loads.

The measured output impedance with KT66 in my circuit is 1r2, and with KT90 it is 0r9 approx.
I also tried Russian 6550EH which gave similar THD to the Russian KT90EH, and Rout = 1.0 ohm.
The other thing to bear in mind is that I have 15dB GNFB used in my reformed Quad-II.
In the original Quad-II with paraphase circuit for EF86, there is about 21dB NFB, or 6dB more.
Now from graph 3 above, and with KT66 in original amps, I got 0. 07% THD at 9W.
In Graph 3 with KT90, and at 9W, I get 0.06%, and Rout is lower than original.

So, I get better THD results with 1/2 the applied GNFB.

Fig 5. Quad-II with KT88 triodes and triode input / drivers.

I reformed my first Quad-II in 1998-1999.
An earlier version of this page of 2014 included the hand drawn schematic from 1999 which
I now do not need to show because it was messy, difficult to read, and likely to lead someone
astray. In 2014 I re-drew the amp reformation in Sheet 1 and for PSU in Sheet 2 which also
has Ia balance indicator, active protection and delayed turn on. The two sheets give all the
information needed to make a superlative triode amp with KT88, KT90, or 6550. In this case,
I have the general ideas applied in old Quad-II using original OPT and PT. Part numbers in
Fig 4 have no intentional resemblance to any other schematics on this page or website.

The V1a+b input triodes in a 12AT7 are paralleled for gain of about 30 for SET input stage.
It acts as a differential amp with grid input from signal source and cathode input from GNFB
from OPT sec. The anode output needed for V2 = 6.3Vrms- The Vk GNFB = 1.0Vrms+, so
Va-k for 12AT7 = 7.3Vrms and with gain of 30, Vg-k = 0.2433Vrms+. The gain between input
and output at OPT = 12.5V / 0.2433V = 51.38. The fraction of output fed back, ß,
= 1.0V / 12.5V = 0.08.
Gain with GNFB = 12.5V / 1.2433V = 10.05, so gain and THD reduction factor = 10.05 / 51.38
= 0.195, ie about -14.5dB.

V2a+b triodes within a 6CG7 form an LTP with drive to only one grid at V2a, and with V2b
grid taken to 0V. V2a+b generate equal amplitude Vo with opposite phase and with low
THD < 1% at 50Vrms. The CCS with MJE340 is used between commoned cathodes and a
negative rail of -17Vdc to ensure that each Va has equal amplitude where anode loads of
R15+R18 and R16+R21 for each triode are equal, easy with modern resistors. The 2 triodes
do not need to be exactly matched. The finite resistance of the MJE340 collector exceeds
1M0, and balance of Vac drive to output stage is within 1%.

The LTP drives the KT88 grids which each have fixed bias of about -47Vdc. This -Vdc is
balanced by the R network associated with VR1 10k0 wire wound pot. The idle bias is actually
fixed, and VR1 is adjusted so that idle Idc in each KT88 become equal. While both KT88 are
healthy, idle Idc in each will both be between about 55mA to 60mAdc. The anode and screen
are connected together, and the Iadc is the sum of Ia and Ig2.
The VRI can change Eg1 by +6Vdc and -4Vdc max. it is unlikely that 2 samples of KT88 would
require Eg1 difference of 10Vdc to achieve the same Iadc. Highest Iadc expected = 60mAdc,
so Pda at idle = Ea x Ia = 390Vdc x 0.06mAdc = 23.4W, and this is well below Pda rating of
42W, so KT88 will last a very long time.

A pair of bjts in a simple LTP operate two matching LED ( easy to find when you buy say 20
in a bagful. ) I found I could use any pair I picked from 20. It is easy to see that both LED are
equally bright, and after several tests I found the difference between Iadc of each KT88 was
less than 3mAdc after setting VRI. Operation of bjts is explained below Fig 5 Sheet 2 below.

The Quad-II OPT may be a toy but it does manage to work OK with KT88 in triode. There is
very little difference Idc and Vac between KT66 in tetrode and KT88 in triode for 20W to 8r0.
The KT88 cause no additional stress to OPT.
KT88 triode µ = 6.7, and Ra = at idle = 1k2 approx, so with 10% CFB, ß = 0.1.
Ra effectively with CFB = Ra / ( 1 + [ µ x ß ] ) = 1,200r / ( 1 + [ 6.7 x 0.1 ] ) = 718r.
The reduction of Ra is not much with triodes, but welcome. The amount of FB applied by
CFB is only about 3dB. The Ra at sec = Ra effective / ZR = 718 / 431 = 1.66r, and if we
add total Rw of 1r6 we get sec Rout =  3.26r, which is 1/3 of the Rout with KT66 without
GNFB. The 14.5dB GNFB reduces the 3.26r down to 0.64r, so DF = 12.5, and quite good,
even with much less GNFB than in original Quad-II.
 
I first tried  EL34 in triode which made about 13W max and with only 6dB GNFB but the
owner said music had less vitality and dynamics when compared to his other 10W amp with
tiny triode-tetrode 6GW8 output tubes in ultralinear with about 16dB of global NFB.
That was a previous job of mine to rebuild a very badly made kit, but with my schematic. 
KT88 in triode and 12dB GNFB delivered the kind of vibrant and accurate dynamics he
was looking for, plus the higher and comfortable power ceiling.
Quad-II amps need global NFB.

This modification sounds very well even with 4r0 speakers, and with OPT strapped for 8r0
setting. The use of the 4r0 strapping will give the best sound for both 4r0 for 20W class AB
and for 8r0 where Po will be about 15W, but mostly class A.

A pair of amps with this mod was used every day from 1999 to 2012 in a house at Cooma,
NSW where summers are not cool. KT88 in Quad II amps do not cause PT overheating
unless the full range of added AM and FM tuners and Quad-22 control units are used.
The PT of each Quad-II is rated to provide the extra current for the tubes in the attached
gear for 1960. The KT88 do not cause the PT to overheat. In these two amps I drilled lots
of holes in the bottom covers and attached 16mm feet at each corner of chassis bottom
plate. I reversed the position of socket plates below surface of chassis to allow more air flow
and this helped to keep the amp cooler. Without the GZ32, and because KT88 draw slightly
less idle current than KT66 in original amps, there is less heat generated in the power
transformer and overall temperature is always cooler than originals. I have never had to
repair anything in these amps.

This schematic requires that the OPT anode and cathode connections
are reversed to get correct phase for GNFB.

Fig 6. 1999 amp PSU, balance indicator, protection. turn on delay.

Fig 6 has the PSU for triode operation to get the highest possible Ea from original PT
with 310V-0-310V HT windings. With CLC input using 220u + 3H + 220u, and Si diodes,
I was easily able to get B+ = +400V and with very low 100Hz ripple < 6mV OPT CT.
The original potted screen supply choke contains 20H x 600r. The choke was removed
and replaced by a choke 3H x 40r for the anode supply. The resonant F between 3H and
200u = 6.2Hz, and well below bottom of 20Hz AF band.
The space for removed GZ34 was filled with the electrolytics. An aluminium box about
85mm high was made to fit between PT and OPT to enclose the choke and caps, with
holes for ventilation. 2 x 220uF x 450V rated caps are far smaller than the original boxed
2 x 16uF caps designed in about 1952. Plenty of room became available under the
chassis.

Sadly, I have no picture of the amp, reformed some 2 years before I went online and
everyone began demanding pictures and videos. 

With mains at 250Vac I found I had B+ = + 410Vdc after Si diodes into 100uF C18 with
KT88 triodes as I show.

With better electrolytic C quality since 1952, R+C filtering of B+ rails for input tubes is
quite excellent and better than in original Quad-II.

The box I made to cover the choke + electrolytics was left natural with fine sanded finish.
But it is now possible to buy paint which will match the exact colour used by Quad. A tiny
sample of the colour can be scanned by a gadget at a paint store and the paint colour will
match. Many samples of this amp can be found with a good paint finish, and Quad's stocks
of grey paint have been all used up. These were bought by Quad from the British Navy after
WW2, who had planned for WW2 to not be sorted until 1955.

Outside the UK, there are no jail sentences given to miscreants who persist with painting
Quad-II amps any other colour except boring pharking grey. I have been told bright pink
with orange decorations are now popular in the USA where ppl want their amps to resemble
the The Donald, accidentally elected in 2016.

Fig 6 is the PSU for Fig 5 amp and contains more parts than I used in 1999, such as the
delayed turn on relay which prevents high inrush currents.
The use of CLC for B+ for triodes or the CFB set up is nice, but not necessary, and having
CRC with 100uF 100r, and 470uF will work well enough for anyone. The B+ with CRC
will be only 10Vdc lower.

For those who are frightened by the possibility of fusing secondary windings of the old
PT with high current from charging shorted capacitors, I suggest use of 0.5A slow fuses
to heater rectifier for input triodes.

The 8.2Vzd + 10Vzd are 5W types and used to stabilize the Vdc to CCS for driver triodes.

The 47nF C15, C16 shunt the ends of HT winding to 0V and suppress switching noise
from diodes.
Going beserko with extraordinary mods has now been identified as the disease called
'Moderitis-Sindrome'

Fig 7. Sheet A, for 2010 Quad-II amps reformed in 2010.....

Fig 7 schematic is similar to Fig 3 above, but has two input 6BX6 set up with a real
CCS for commoned cathodes with Q1 MJE340 taken to a -17Vdc rail. After much testing
and measuring of THD, I found there was no real need to have each screen bypassed to
the commoned cathodes, and I could use Quad's original method for the screens where
each has its own screen feed resistance but both are commoned with C4 0.27uF.
This was the generic way screens were dealt with in countless pentode differential amps
in 1950s. Thus there is no need to bootstrap the screen feed resistors to get the common
cathode resistance as high as possible for accurate natural balancing.

The measurements for 2010 show the differential input stage gain = 220 with the total
RLa = 73k. The Ra of 6BX6 is unknown, but probably > 500k, and if it was 500k, then
RLa // Ra = 63.7k, and gm = A / ( RLa // Ra ) = 220 / 63.7 = 3.45mA/V, and 2.4 times the
gm of EF86 in Quad-II.
Thus the lower RLa can be used, and thus be able to reduce the value of the grid
resistance to bias the output KT88 to 270k.

Use of 6EJ7 with higher gm could give similar gain with slightly lower R feed to anodes,
68k and biasing Rg of 180k.

The original Quad had Rg 680k for KT66, and this led to the inevitable idle grid current
making grids positive which made more Iadc flow and made them idle hotter, which led to
a more positive grid in what is a positive FB effect due to inevitable tube ageing where
gas in the tube is not all absorbed by the gettering.
It only requires 6 micro amps, 0.006mAdc to cause Vdc across 680k to increase +4.1V,
and I've seen that in many old amps with nearly dead tubes. 

This ageing effect is best countered by use of individual R&C biasing of output tubes
or some other method which allows balance to be maintained, AND with low value biasing
resistances. 50k is recommended for 6550 with fixed bias. AFAIK, KT88 and 6550 made
in Russia have identical internal construction and I found no strict need for 50k, and found
120k was OK even with fixed bias where there is no self regulation of Iadc with an R&C
cathode network.

Dynamic Bias Stabilization is used in this 2010 Quad mod, as explained at
300w-5-bias-stabilizer.html
That page deals with having 12 x 6550 in one monobloc, and I wanted the benefits of
cathode biasing and added benefit of stable Ek during high Po with class AB. I did not
want to sell someone a stereo amp system with 24 bias adjustments to worry about - for
2 channels.

Dynamic Bias Stabilization is the stabilization of Ek for amp ac working, and Q2+Q3 are
TIP31C which worked well to bypass Iac exceeding 2 x idle current during class AB rather
than allow this current to charge up the cathode bypass caps C13+C14 1,000uF.

I invented this method of keeping Ek constant for class AB amps with cathode biasing.
The regulation of dc operation at idle is maintained for a wide enough range of Ia values
up to twice normal idle Iadc.

The use of the DBS scheme looks terrible to anyone not able read a schematic properly,
ie, understand immediately how it works. People see a bjt and scream "HOW DARE YOU !
- in a TUBE amp, you rotten sod".
Water off the duck's back afaiac. I can assure everyone the DBS circuit is a another use
of a bjt to be a slave to the tubes to allow them to better handle your precious music
without so much THD, IMD etc.

The BJTs still allow auto biasing to occur which save having any bias adjustment to
confuse and worry any owner. With Eg2 equal to Ea = +350Vdc, Eg1 grid bias needed
was -37Vdc. With Ia+Ig2 = 61mAdc I would have needed Rk = 590r with pure cathode
biasing with Eg1 = 0V. I figured 470r was OK for Ikdc regulation and Ek could be +30Vdc
if I had a fixed bias applied of -7Vdc.
Fig 3 above shows Rk 270r with Ek at +18V and fixed bias = -18V, with a pot to balance
accurately. Fig 4 above shows pure fixed bias with balance pot and Iadc balance
monitoring. Which is best?
Probably the best is the fixed bias option with balance adjustment. The KT88 or KT90
or 6550 do very well in the initial class A operation and then can also make quite a
high amount of class AB despite the limits of the OPT wire resistance.

Fig 8. PSU and Protection for 2010 amp mods.

In the 2010 amps I modded for a customer, and in the previous edition of this page,
I showed HT windings driving 1N5408 diodes through 47r to charge 470uF for the
B+ applied to OPT CT. This gave 100Hz ripple of 0.7Vrms, which I found quite OK and
to not contribute to noise at high levels of class AB.
However, my inner perfectionist of 2016 demands I leave out the 47r because the
existing HT winding Rw is high to limit charge currents of GZ32. It is better to place 50r
+ 220uF ahead of the 470uF and thus have Vr at OPT CT = 0.1Vrms, to further reduce
the Vr. If there is a short circuit in a diode or 220uF then the peak Iac in a 1/2 winding
> 2Apk, and at mains input it is > 2.6A, so the mains fuse will blow.

For CFB, having low Vr at screens is important, and I retained the Quad choke of 20H,
600r. But there is no need for the choke because the Vr at OPT CT = 0.1Vrms, and with
1k5 to replace L1 and with C5 = 100uF, The Eg2 remains high enough and Vr at output
tube screens < 1mV.

I see no reason for fuses between each end of HT winding and diodes. If you do use
them, I suggest 0.25A slow blow types. But these fuses may not do much because
whatever may cause them to blow would also cause the mains fuse to blow if it has
a low enough value less than in original Quad.

The amp needs to be turned off well before any tube fault may cause a fuse to blow.
After 18 years repairing amps, virtually all causes of mains fuse blowing was due to
serious bias failure, ie, tubes overheating and becoming a short circuit. Once or twice
an electrolytic reservoir C right after the diodes became a short circuit which immediately
increased Idc and mains fuse would blow. Unfortunately, failing tubes can fuse fragile
PT HT windings or OPT primary windings before making a fuse blow so an active
protection circuit is necessary to avoid high repair costs by detecting excessive Idc in
one or both KT66 / KT88 etc, and turning off within 4 seconds.

I have a 0.5A fuse in series with all B+ anode Idc after C8 470uF. This is a precaution
against a sudden shorting of anode circuit to 0V. The OPT has average Rw for each
1/2 anode primary = 144r. If anode goes to 0V, you have 400Vdc from 470uF + 220uF
able to discharge through 144r, with peak Ia = 2.77A, some 40 times more than the idle
 Idc. The time constant for 690uF and 144r = 0.1Sec, and this may be long enough to
increase wire temperature and fuse the winding wire. The heating in 144r for 0.05Secs
= 324W! Use the fuse and don't risk wrecking an OPT.

I used 6.3Vac for ALL tube heating in 2010 but in this 2016 schematic I show the
5Vac heater winding for unused GZ32 connected to one end of 6.3Vac heater winding
to get 8.3Vac to make Vdc rails of approximately +10V and -20Vdc.
The 10Vdc is RC filtered down to 6.3Vdc at 0.6Adc for 6BX6 heaters in parallel.
There is a 6.2V x 5W zener diode to clamp Vdc so that if one 6BX6 filament went open,
or was removed from socket, the Vdc applied to other 6BX6 does not increase.

The -18Vdc is RC filtered for the CCS for 6BX6 common cathodes and fixed bias for
output tubes.

The protection circuit has been simplified to what I found works well. The Vdc signal
for excess Idc is taken from each end of CFB winding U-V-W, which has Rw = 33.2r,
and with V = CT. Each 1/2 winding Rw =  16.6r and with Ikdc = 68mAdc, the Vdc = +1.13Vdc.
If this rises to +1.8Vdc, it is because Idc = 108mAdc and without any output Po being
produced the Pda+g2 = 38W, and tube would be hot, with Pda near its 42W limit. This is
enough to indicate there must be a fault and the amp must be shut down.
With +1.8Vdc at U or W, Idc flows through R1 or R2 2k2. Vac is filtered away by C1 or
C2 470uF and then Idc flows in diode to R3 2k7 and R4 10k . The input gate current
current for tripping = 0.03mAdc at gate-cathode = 0.7Vpk, so SCR trips when 0.7Vdc
appears at top R4 10k. The current in R3 = 0.1mA total.

Usually one output tube endures bias failure before the other, so we may design for where
one tube conducts excess Iadc. If both conduct excess Idc simultaneously, then the SCR
will trip at a slightly lower Ikdc limit.

I found that when the amp has the same sec RL load as the strapping value, say 8r0 load
for 8r0 strapping, and for class AB, the sine wave clipping level is tolerated without tripping
the SCR. But where load is reduced to 4r0, then the amp will be turned off before clipping.
With music Vac, and use of 4r0 load, the amp will not turn off when clipping begins on
highest peaks. But with sustained higher levels with a sine wave, the SCR trips before
clipping is reached.

We most certainly need our gadgets to stop working and turn off when output tubes get
sick or a speaker lead is shorted or someone expects the same high sound levels which
only a 100W amp could deliver.

If the cathode bypass caps were to ever fail to become a short circuit, as I have seen often,
then the protection would work more reliably than if  the sample Vdc is taken from cathode
at top of bypass caps. However, I have never ever witnessed electrolytic cap failure unless
the Vdc across it went too high for too long. The protection circuit prevents the cause of
the electrolytic failure. Electrolytic C failure can happen if the temp > 85C and and the
liquid inside boils. I have seen them explode in front of me and have been lucky to have
been wearing glasses to avoid bits of metal wrecking my eyes.

Fig 9. Reformed Quad-II amp for 2014 with fixed bias.

Fig 9 has no cathode bias R+C for KT88 and uses fixed -Eg1 bias = -47Vdc. The Ikdc
of each KT88 is equalized by adjusting VR1 10k which moves the -Eg1 applied to each
in opposite directions of a volt or two. No two output tubes require exactly the same Eg1
to give identical Ikdc, even when they are a matched pair. Older tubes vary more, and
when the Ikdc is exactly equal you may find Eg1 are -48Vdc and -46Vdc. The the Ikdc are
thus balanced, and the accuracy of balance can be seen with two LED which will appear
equally bright when balance is within 3% which is good enough.
The use of fixed Eg1 bias means that Ea is maximum for the KT88 without the inclusion
of cathode bias Vdc. Thus Va swing is increased by 20Vrms more than for cathode bias.
There is no Ek variation due to charge up of Ck with increasing Ikdc flow with class AB.

Instead 2 pentodes for input & driver stage, I have a similar arrangement to the Fig 5
above for fixed bias with triode connected KT88.
V1 = SET 12AT7, and V2a+b = LTP 6CG7 with MJE340 CCS. The triode input and
driver stages give less THD than EF86 and probably less than EF80 / 6BX6.  I have not
built the Fig 8 amp, but it should work better than the Fig 4 amp, which I did build in 1999. 
The total amount of CFB + GNFB exceeds 20dB so there will be a tendency for HF
oscillations, and there are 3 amp stages so LF oscillations are also likely. To avoid all
oscillations, and for unconditional stability and wide bandwidth with a resistance load,
I show all required R&C values for "critical damping", ie, open loop gain and phase shift
reduction below 20Hz and above 20kHz by parts R10+C5, R9+C4, R28+C13, and C11
270pF. The amp should not oscillate at any F with any pure L or pure C load.
The amp must be tested with a 5kHz square wave at low level, say 2Vrms output and with
pure C loads between 0.05uF and 2uF. Square waves should not have more than 6dB of
overshoot, and not more than say 4 ringing waves declining to a flat line within 100uS,
or a 1/2 wave time for 5kHz. If tested with sine waves with a C load, the response
should not have peaks in the response exceeding +6dB above the 1kHz levels.
Everyone building any tube amp MUST overcome the tendency of all amps to oscillate
when ANY NFB is applied. The amp is an active bandpass filter device with NFB
applied and there are limits to how much NFB can be applied and how much bandwidth
is possible when NFB is applied. Everyone needs to understand phase shift basics.

Fig 10. 2014 PSU with balance indicator, protection, B+ delay.

Fig 10 is the PSU schematic for Fig 8 amp.
In this one, I have HT winding charging 220uF via 1N5408 without current limiting
resistors because the HT winding Rw is already high to suit GZ32. The CLC filter with
220uF, 3H, 220uF which is adequate to get ripple low and keep size of parts no larger
than needed. The Quad-II 20H choke is not needed and room is needed for the 3H
choke.

This PSU may be used for KT88 in triode if the screens are connected to anode.
It is also possible to feed each KT88 screen through 2k2 and bypass each to the cathode
of the opposite tube with 100uF which then has each KT88 effectively working in 20% UL
mode but with the existing 10% CFB. This slightly reduces KT88 gain and Ra and THD.
Whether this is worth the extra pair of R+C may be argued, and I have never tried it.
In other amps I found class A1 operation of pentodes or tetrodes to sound best and measure
best when both adequate CFB is used in conjunction with some screen signal derived from
UL tap in anode winding. In this case there are no available UL taps on Quad OPT unless
one alters the OPT potted in tar, a dreadful effort. Bypassing each screen to g2 to cathode
of opposite tube provides adequate UL signal which is correctly phased. The only worry
may be HF stability.

Both KT88 screens could be commoned at +300Vdc via 4k7 from +394Vdc, and bypassed
with 100uF and shunt regulated with 4 x 75V x 5W zener diodes. This will slightly reduce
the THD for listening levels. Eg1 -Vdc for bias will have to be reduced to get the Ikdc to
about 60mAdc.  

At turn on, there is a 4second delay before R17 270r is shunted by Relay 2, and the delay
is long enough for B+ to rise to near +350Vdc and peak input Iac is kept low. When R17
is shunted by Relay 2, there is a second Iac peak inrush to bring B+ up to about +445Vdc
where it waits about 12 seconds before the B+ is pulled down by KT88 idle current to
about +400Vdc.
The mains peak inrush Iac are roughly equal and 1/2 what Iac input would be without the
delay. This allows a lower current rating for F1 mains fuse so that the fuse is more likely to
blow when a genuine fault occurs.

The circuit board needed for small parts for delay, protection, and balance indication will
need to be kept as small as possible and installed with 2 screw fixing off spacers so that
access to parts is by removing 2 screws, and folding out the board on flexible cabling.
----------------------------------------------------------------------------------------------------------------------
Output Transformer issues for Quad-II amps.
The picture shows OPT removed from its can.
Core is a 25mm stack of E&I
with C&T pattern. The core lams are mounted vertically.
Fig 11. Picture of Quad-II OPT out of its pot.

This picture from Keith Snook's website.

Fig 12. Quad-II original OPT connections and details.

Fig 12 shows the original Quad-II OPT details as best I can without pulling one to pieces.
The winding losses and winding resistances agree with countless measurements I have made.

For use with modern speakers with low average impedance and low sensitivity, I suggest that
for the best 15W you may ever have, strap the OPT for 4r0 for 6r0-10r0 speakers, and strap
OPT for 8r0 for12r0 - 20r0.
The connection for 16r0 is almost useless.

With speaker Z = twice the strapping value, the Po is mainly class A and winding losses are
below 9.4% for 8r0 or 16r0 strapping. They are 11.7% for 4r0 strapping.

The winding losses for class A operation are minimum for class A operation, but for class
AB Po exceeding the initial class A, the losses increase for where each output tube takes
turns to drive each 1/2 of OPT primary in class B with the other tube cut off. For where the
load = strapping value, the maximum Po is due to mainly class B and Rw loss% increases
by factor of about 1.4, where RwP and RwS are equal % of the P and S loads respectively,
while in class A.

If we consider the OPT strapped for 4r0 and driving a 2r0 sec load, then nearly all Po is class
B and if the tubes produce say 12.5W for 2r0 as my graph above shows, then that is 51.5%
of the power generated by anodes, because losses are 48.5%. The power at anodes would
be 12.5W x ( 100% / 51.5% ) 24.27W, so using a speaker load of 1/2 the strapping value of
4r0 will give only about 1/2 the Po which tubes are capable of producing.

The THD, IMD and damping factor are both quite horrid at high levels for secondary RL less
than the strapping value. However, as long as average levels do not exceed 1W used by
most people on most evenings, the sound is listenable even with 4r0 speakers used on OPT
strapped for 8r0. This is the forgiving feature of the initial class A Po. Better sound will be
had if the OPT is strapped for 4r0, and better still if speakers are 8r0, and the 4r0 strapping
is used, and where high levels and winding losses are of no concern.

The use of KT88, KT90, 6550 instead of KT66, and with higher Ea will generate higher Po
at the tubes and we need not be so worried with winding losses, especially if the amp has
had at least the basic reforms of PSU. I found KT88 in triode in Quad-II and as in Fig 4 above
sounded very well compared to the original amps.
 
Quad's OPT losses are about 3 times what I would ever wind myself. The lower Rw means
wire size must be much increased which means the OPT should weigh at least twice that for
original Quad-II. RDH4 gives recommendations for weight / Watt of power, and Quad failed
to meet the good book's ideas about OPT. The cost of GOSS transformer iron and copper
winding wire was extremely high in 1953, when the UK was impoverished by the efforts of
WW2. It was a miracle that Quad amps were ever made. UK bravery did not stop after WW2.

The winding losses for the 4r0 strapping can be reduced if L2 and L5 can be connected in
series and then paralleled to windings between P and Q. This would make 4r0 winding RwS
= 0.39r instead of the original 0.62r, and Rws at P = 379r, and total RwP+S = 713r, and thus
total losses would be nearly the same as for 16r0 strapping.

But to be able to arrange sec windings to allow better 4r0 strapping the OPT must be very
carefully removed from its pot and an additional turret connection point installed and one
internal sec wire connection altered. This kind of work is normally done by Mr Zealous
Perfectionist, and he's a difficult fellow to deal with.

Fig 13. Quad-II OPT core dimensions.


Fig 14. Quad-II OPT with improved OPT winding strapping.

Figures 11+12+13 show what is inside the original Quad-II OPT pot.

Fig 14 shows a small alteration to OPT secondary winding connections sealed inside pot
containing the OPT to allow all sec windings to be better used for use with 4r0 speakers.

There must be one added turret terminal plus a change to a wire connection after removing
the OPT from its pot which means gently heating the potted OPT upside down and suspended
in an oven on wires. 100C would probably be plenty to melt the tar based potting mix which
may be poured out into a can for re-use later.

Instead of a turret, a 15mm M2mm brass bolt with a nut could be used with bolt pointing
out like other existing turrets. It is labelled QA. The wire from L5 to Q is found, and disconnected
from Q and taken to QA. Just which wire you re-locate must be triple checked to make sure
you have got the right wire to move. Then it is tested again for when
strapping is for 4r0 and including QA, and DC resistance measured with 100mAdc applied
through windings and DMM set on Vdc used to measure Rw. 4r0 winding should measure 0.39r.

Once the mod is done the OPT is re-fitted to pot, and warmed up. The potting mix is re-heated
and poured back in around the OPT. If it looks like there will not be enough potting mix to fill the
pot after pouring in 1/2 the potting tar, fine sand may be mixed with tar to increase its volume.

Extreme care must be used for this modification, because inexperienced fools could so easily
ruin a working OPT. There are virtually no replacement OPTs now made for Quad-II which will
fit in the same available space as the original and which offer performance as good in terms of
F response and winding resistance.

Also possible is the creation of OPT UL taps from anode primary turns and from joins between
N1-2 to N2-3, and N6-7 to N7-8.

Notice R?
I figured out it must be 0.36r and it is to make Rw of L1 and L3 in series equal to L4 and L6 in
series so that when paralleled, the current density in the two windings is equal. Peter Walker
had a reason for each and everything to be found or not to be found in Quad-II amps.

L2 + L5 = 0.47r + 0.62r have total Rw = 1.09r. and in theory, we should add 0.16r in series to
make the Rw of L2+L5 = L4+L6 = 1.25r, but I don't think the R difference is high enough to worry
about.

For a better OPT, a much larger core and completely different design would be used for better
overall performance while being easier to wind, with less than half the high resistance which
can fuse open so easily. If anyone can find someone to custom build OPTs,
here are a few suggestions:-

Fig 15. Small size OPT for 8k3 : 4r0, 7r1, 16r0......

This OPT has a core plan size which cannot fit inside the original pot, but this OPT
should fit on chassis to replace the original, but without being in any pot, with bell end
cover on top and painted same Quad grey. The windings at bottom will project beneath
the chassis, so internal wiring will need total revision.

A bell end top cover would look silly, but a sheet steel lid made to appear like the top
of original can could be screw fixed to angles clamping E&I lams. The transformer
laminations and lid can be painted to match the Quad grey. In 2010, the British Navy
finally sold its remaining stock of  Battleship Grey paint made during WW2, which was
expected to go on until about 1985. Peter Walker bought barrels of it, but sadly his
wife got sick of them stacked up in their garage and she sold them off to make way for
a car of her own.

Fig 16. Larger size OPT 6k6 : 4r0, 7r1, 16r0.

Fig 16 is the biggest practical OPT possible for Quad-II. The core plan sizes are
96mm x 80mm. The original amp must be totally re-engineered. The 20H screen
choke and GZ32 are removed, the two EF86 sockets are moved to the available
space, the new OPT can fit across the chassis end where and the to be beside
the two KT66 instead of where the GZ32. The EF86 sockets and circuit board are
removed. A metal cover over replacement OPT should be made to be similar to
the shape of power transformer pot.

For the DRASTIC mods to Quads, with larger replacement OPT, removal of screen
choke and GZ32, and much better parts layout, see further down at bottom of this page.

The benefit gained with better OPT become obvious when the original Quad-II OPT
properties are examined :-
The original Quad-II OPT has well interleaved windings with 7 primary sections and 6
secondary sections. This gives good HF response, and yet the amp can oscillate at HF.
The main cause is the lack of critical damping R+C networks to shelve the HF gain
above 20kHz.
EF86 pentodes have high Ra, and their high RLa of 180k // 680k means a small
amount of shunt C causes enough phase shift to become a problem above 20kHz.
For 4r0 strapping on original amp, 2 of 6 secondary windings are not used. The original
OPT LF performance is not as good as many other OPTs. The original OPT is only 1/2
the weight it should be.

Consider the famous OPT designed by Mr D.T.N Williamson and which has all details
published in RDH4, page 748, half way down. We see it used GOSS E&I laminations with
32mm Tongue x 44mm Stack of "Super Silcor" E&I lams with non wasteless pattern and
window size of 75mm x 25mm. Everyone laughed at Mr Williamson, but Mr Walker
certainly did not, because he would have known the Williamson design was better, but
for commercial reasons Walker used a toy sized OPT full of Williamson ideas. Electronic
commercialism meant prices for amps of any quality did not have toy prices for the toy
parts within. However, the vast majority of music listeners never used more than a watt
from each channel. 

For best bass performance the Fsat of OPT should be below 20Hz for the Va-a for max
Po at clipping at 1kHz. RDH4 has some wise comments which infuriated all accountants
and bosses of any companies making amplifiers in 1950s, so they mainly ignored this
great book. If you want better, go make something yourself, and yes, its easily done!

The maximum primary inductance is not the most important LF parameter for bass F.
For best bass with low THD the LF should extend to below 20Hz at full Vo without core
saturation, ie, Bac < 1Tesla. If the OPT has this property, the primary inductance is high
enough.

If you test a tube amp using a pink noise source test signal with randomly varying F and
constant average amplitude from say 5Hz to 30kHz, you will see there is repeated core
saturation each time some low level LF signals between say 4Hz and 20Hz reach high
enough to saturate the OPT core.

Most music does not have much signal below 30Hz, but we live in a world with deep bass
and "sub-woofer" signals in music, or movie sound tracks so the amps must be "ready for
anything". Core saturation is independent of loading or signal current and is a "voltage
caused" phenomena, with magnetic field about proportional to the applied Vac and F.
 
The older core material had a low Bac max, maybe 1 Tesla for poor E&I lams, but even
with modern GOSS it is not wise to go more than 1.2T when harmonic distortion, mainly 3H
very rapidly exceeds 3%. at some F below 100Hz. So if the core saturates at say
Va-a = 400Vrms at 40Hz, Bac = 1.2T, it means that 20Hz the B = 1.2T at 200Vrms, and
10Hz the Fsat occurs at only 100Vrms, a very low level of Po.

Therefore all tube amps should have C+R input high pass filtering to exclude F below 10Hz,
and have open-loop gain and NFB arranged to give sharp cut off below 5Hz, and have an
OPT with low Fsat < 20Hz at full Po rated Va-a. Nearly every tube amp manufacturer has
tried to avoid the issues of quality determined by best engineering principles.

Fsat can be calculated = 22.6 x Vrms x 10,000 / ( Afe x Np x 1.2Tesla )
where Vrms is across OPT primary, 22.6 and 10,000 are constants, Afe is square mm,
Np is primary turns, and 1.2 is the maximum magnetic field strength, Bac.

For original Quad-II, maximum Va-a is in the class A condition with RLa-a about 9k0, with
Ea about = +350V, and Ia = 70mAdc in each KT66. Va = 317Vpk, 224Vrms, so Va-a max =
448Vrms giving 22W into 9k0. Winding losses reduce this to about 19W at output terminals.

For determining Fsat, winding losses are ignored because the Fsat is considered where
tubes are not loaded with a Sec RL so there's no RLa-a, and only primary inductance.
For class A, Va peak max = 0.9 x Ea approx = 0.9 x 350V = 315Vpk = 630Vpk-pk = 445Vrms.
Original Afe = 25mm x 25mm = 625 sq.mm, and Np = 3,180 turns.
Fsat = 22.6 x 448V x 10,000 / ( 625sq.mm x 3,180t x 1.2Tesla ) = 42Hz, about what is observed
in practice.

For a lower Fsat, Afe must be increased, or turns increased. But the primary turns already
have excessive winding resistance which can lead to a fused OPT. Increasing Afe is the
easiest solution, but that means turn length increases and higher winding resistance.
So we will have to reduce Np, but make Afe 2,240sq.mm instead of 625sq.mm.

Consider Fig 15, small 40W OPT, when used in identical conditions to original with KT66 :-
Va-a = 445Vrms max, Core Afe = 28mm x 80mm, Np = 1,500t, Bac max = 1.6Tesla,
Fsat = 22.6 x 445V x 10,000 / ( 28mm x 80mm x 1,500t x 1.6T ) = 19Hz. The primary wire
copper section area is 1.6 times larger, Np less than 1/2 of original, and RwP is 142r, be
less than 1/2 the original 334r, and winding loss % much less. There is less interleaving in
40W OPT, but with less turns the Lp inductance remains quite high enough because the
Core Afe center leg area is much bigger. 
The leakage inductance is also proportional to Np squared, so with 1/2 the Np of original
OPT the LL reduces to 0.25 x original for the original 7P x 6S interleave pattern. In practice
the 40W OPT has similar low LL to original with interleave pattern = 5P x 4S. The shunt C
is low enough. If you disagree, then feel free to use my pages on OPT design to compare
the figures for yourself at output-trans-PP-calc-3.html
The Secondary has thicker wire than original and with linking pattern all Sec turns are
used for the 4r0, 7r1 and 16r0 load matches and RwS is low so total winding loss % is less
than 5%, under 1/2 that for original.
With Ea +350Vdc, the same 445Vrms is possible but winding loss is lower so RLa-a load
is about 8k5 so Po = 22W, not much more than original.

But to get more Po, fixed bias with Si diodes for B+ can increase Ea = +400Vdc, so max
Va pk = 360Vpk, and Va-a = 509Vrms, and with 8k5, Po = 30W, and initial class A = 19W !
But if the Sec load has a dip from say 7r1 to 4r0 the RLa-a = 4k6, and max Va-a is about
450Vrms, and with KT88, you get 45W and there is still an initial 10W of class A.  

Consider Fig 16, larger 45W OPT :-
With fixed bias, Ea +400V, RLa-a 6k6, max Va-a = 480Vrms max, Po = 34W. Core Afe =
32mm x 75mm, Np = 1,820t, Bac max = 1.6Tesla,
Fsat = 22.6 x 480V x 10,000 / ( 32mm x 75mm x 1,820t x 1.6 ) = 15.5Hz, another better
result than original OPT. But if RLa-a reduces to 3k6, KT88 make Va-a 450Vrms for Po =
50W+

Consider the Williamson, from 1950, in same conditions as Fig 15 :-
Fsat = 22.6 x 480V x 10,000 / ( 32mm x 44mm x 4,400t x 1.2Tesla ) = 13.6Hz. This is better
than nearly everything made after 1970. The Williamson OPT was meant for 10k RLa-a
load, and to give a high amount of class A and 480Vms makes 23W into 10k0.
The Rw loss % will be worse than my designs here when the OPT is used for RLa-a
< 10ka-a. But with Np = 4,400t and Afe = 32mm x 44mm, 1,408sq.mm means Lp max is
very high. But the Lp varies with applied Vac, and Williamson's design ensured Lp = 100H
minimum with Va-a = 6.3Vac at 50Hz, using a spare 6.3Vac heater winding for a test signal.

Many tube amps with GNFB oscillate at LF without a load and the oscillation amplitude
remains low because as the Vac rises the inductance increases and phase shift reduces
which prevents oscillations. In some amps there can be mysterious B+ rail LF signals of
less than 2Hz, and it is usually due to combination of amp gain, low Lp at low Vo, GNFB,
and RCRC filtering of B+ between PSU and V1 input stage.

In all tube amps I have worked on, the LF gain shelving network avoids all such problems
of LF instability. It is entirely vain to expect Lp be such a huge value that it will prevent LF
oscillations, and the C+R couplings between phase inverter and driver and between driver
and output also to have excessively long time constants. Huge LP, and huge coupling C
merely reduce the oscillation F and I have see a number of amps with LF oscillation
below 0.5Hz.
Many DIYers and manufacturers copied Williamson's amp design without using an OPT
equal to Williamson's design so they all had terrible troubles with parasitic LF and HF
oscillations.

Williamson's high Np = 4,400t meant LL would be high. In the W OPT, dual bobbins side
by side are used, each with interleaving pattern = 5P x 4S, and and the two 1/2 primaries
are in series so for the whole Pri the interleave pattern becomes 10P x 8S, and so the W
OPT had good HF properties up to 100kHz. Its shunt C wasn't so low though.

In 2017, GOSS, aka CRGO is much cheaper and has max µ up to 3 times the 1947 max
of 5,000. Plus the max Bac can be 1.6T, higher than 1947. With a larger core Afe than 1950,
Np can be lower for the same outcome, and the larger wire size is easier to wind and has less
Rw. Williamson Afe = 1,408sq.mm, more than twice Afe for Quad-II, but my 45W OPT
has Afe = 2,400sq.mm, some 3.8 x Quad's Afe. RwP for 45W OPT = 145r, and if load is 6k6
then P loss% = 2%, way below Quad-II. The secondary losses = 2.2%.
Total winding losses = 4.2%.

If the Chinese had known how to extend their minds a lot further when they built the
Quad-II-Forty in 1990s, they could have done a far better job. Instead, they allowed insane
mediocrity to prevail.

Fig 17. Original QUAD-II chassis layout.
 
This gives the sizes I found on a Quad-II amp within +/- 0.5mm. The original chassis
was about as small as it ever could be with a choke and GZ32 rectifier. It would have had
to be 45mm longer for a larger OPT and to fit the 2 x EF86.

Fig 18. Re-engineered QUAD-II with new parts layout 2017.

When the choke and GZ32 socket are removed, the holes in chassis must be covered
with a 1mm steel plate with two 19mm holes for new position of the 2 x mini-9 pin sockets
for input and driver tubes. There is no room for a pot for the 45W OPT. if the 40W OPT is
used, used with T28mm, S80mm, L42, H14mm, and plan area 84mm x 70mm. A pot can be
exactly the same as the pot for PT but with 80mm x 100mm area.

T38mm E+I lams cannot be used because they are just too large.
The OPT will need sufficient hole cut to allow windings to project below chassis.
There should be room for smaller auxiliary 7VA trans for protection circuit and relay.
There is room for the audio circuitry around the tubes.  Large 470uF 450V will fit beneath
PT with other C and some larger R for CRC filtering, plus an IEC socket with a built in
fuse for mains entry.

Happy soldering.

To Re-engineered amps

To index page