Email : email@example.com
Web: Graham's Web Site
Graham would be interested to see photographs and hear from everyone who builds this amplifier. The e-mail address above is available for direct contact and any related enquiries. Graham now has his own web site dedicated to the GEM amplifier, URL below:
Graham Maynard's GEM Amplifier Web Site
Original GEM class-A//AB circuit dated 8th July 2005.
This latest text update - 6th August 2006.
The GEM is an audio power amplifier embodying simultaneously active Class A and Class AB output stages for 100+ Watts into a 4 ohm loudspeaker. The front end pcb will drive up to 100+W variants, or the 200+ Watts
version into a 4 ohm loudspeaker load.
The circuit for this design developed out of some 35 years of
on/off investigations into the John Linsley-Hood, MIEE, 1969
class-A amplifier. It has been named in remembrance of my late
father, Gordon Ernest Maynard, who supported my interest in
audio/electronics/radio/etc. from a young age. It is not claimed
to be the best amplifier for anyone to use, for indeed there are
so many system design ideals and requirements that not one of
them can be expected to suit everyone, no matter how well any one
might measure and perform in relation to original requirements.
As always there is more than one way of studying any problem and
achieving a given end result.
The original and 'simple' 1969 JLH class-A amplifier design
provides excellent first cycle accuracy through mid and high
frequencies, thus its delivery is both neutral and clean. Being
class-A there are no output stage conduction crossovers, and, if
properly constructed there is no need for the stabilization
components or the series output choke that so often introduce NFB
control delay to a real world amplifier's output terminals when
dynamic loudspeakers are being driven. The JLH not only amplifies
percussion transients and spoken sibilants cleanly, it is also
silent behind voices and notes, such that an artificial
brightness or smear does not affect the reproduction of detail,
and thus its output is instantly recognisable as being correct.
So many 'distortion' analysts study an amplifier's forward
linearity characteristics under steady sinewave drive with a
passive resistor load whilst ignoring the complex circuit
activity that arises when dynamic loudspeakers are driven by
dynamic and asymmetric music signal waveforms. The JLH amplitude
'distorts' more than most amplifiers under forward analysis, and
yet it sounds much better because of the way its circuit damps
without delay or overshoot in the presence of secondary
dynamically generated loudspeaker system back-EMFs which attempt
to reverse drive the output terminal.
For more in-depth information and construction details about JLH class-A amplifiers, see Geoff Moss's excellent Website:-
The two most significant reasons for a JLH class-A amplifier presenting a sonically neutral output relate to it having an open loop bandwidth adequate for audio requirements *before* NFB is applied, then to it possessing a natural closed loop stability without need for additional dominant pole filtering which might then infringe upon those open loop capabilities; hence the closed NFB loop's ability to maintain phase linear control of output terminal potential in the presence of loudspeaker generated back-EMF up to the very highest of audio frequencies.
So often it is a need to add stabilisation components to other amplifier designs where NFB is used to reduce amplitude
distortion, that leads to these very same components introducing a NFB delay which then colours dynamic loudspeaker reproduction in a manner steady sine measurements simply cannot ever reveal. This colouration arises when output stage driven amplifier-loudspeaker system current flow becomes back-EMF modified by the dynamically induced milli-second to milli-second variation of reactive loudspeaker system elements, as their impedance and phase angle change due to the momentary sequential elemental delays related to storage and release of audio waveform energies. If loudspeaker current flow becomes leading with respect to audio amplifier input waveform voltage, and the necessary correcting NFB loop response is fractionally delayed by a bandwidth limiting dominant pole filter or internal series output choke, then the amplifier's output current correction becomes lagging at higher frequencies and the output terminal cannot quickly enough be prevented from developing a fractional error potential. The amplifier does quickly 'catch up', but in the meantime a tiny additional loudspeaker system dependent 'dominant pole and/or choke related' interaction error has already been generated at the amplifier-loudspeaker interface, and no amount of NFB can completely erase this because it was NFB control delay with respect to priorly energised loudspeaker back-EMF that caused it in the first place!
So why then does not everyone use good JLH, Nelson Pass or other class-A amplifiers ?
(1) They run constantly hot when compared to other amplifier types.
(2) Pure biased class-A designs lack dynamic powering capabilities.
(3) Some provide marginal low frequency phase response or damping.
During the early 1970s I constructed a large JLH class-A monoblock. It had a genuine 100W measured sine output, sounded very clean, and could generate surprisingly noisy short circuit sparks to raise thoughts about output stage survival. (It never blew, and still runs!) However when it was compared with a physically smaller and cooler running 2x KT88 Ultralinear Leak TL50+ this solid state monster lacked dynamic attack. It also sounded like a purely voiced but wimpish choirboy when beside the maturely rocking muscle outputs of other typical 100W class-AB solid state chassis.
The reason for this 'weakness' relates to loudspeaker current
flow, whereby dynamically induced momentary requirements can far
exceed the peak sinusoidal output capability of a pure class-A
biased output stage. This, combined with an inability for the JLH
upper output transistor to conduct as deeply as the lower
transistor, leads to what sounds like a pop-rock music output
linearity weakness developing as soon as half power levels are
reached. With modern H-pak plastic power transistors it is
possible to obtain up to 50W of pure class-A output from a single
pair of JLH connected output devices, but there will still be
that positive going output current limitation when the amplifier
is used to drive dynamic loudspeaker systems.
I returned to this problem many times, and at first attempted to
overcome it by upping the class-A rating whilst implementing
different dynamic biasing arrangements to hold down the quiescent
dissipation. These designs worked, and I achieved 100W of class-A
output for 100W of quiescent heat. Generally though the resulting
amplifier was not temperature stable through different audio duty
cycles; or their biasing arrangements had an audible impact upon
transients; or they were unacceptably complex.
More recently I tried numerous arrangements where individual
output devices were replaced by identical composite sub-circuits
running in class-A at low level, though conducting as if class-AB
during periods of increased output demand. This type of
arrangement simulated well, they also worked and were less
complex than with additional biasing arrangements, but they
sounded 'punchy' as if the amplifier was over reacting to
loudspeaker generated back-EMFs; as if the phase splitting JLH
driver could not maintain balanced drive splitting control when
the individual composite output device dynamic characteristics
became externally altered on a per-half basis by varying
loudspeaker system demand.
A NEW CROSSING.
More recently it occurred to me that the JLH current splitting
transistor *collector* could be used to drive a conventional
class-AB output stage, whilst its *emitter* controlled a lower
JLH class-A output device exactly as before. Also the upper half
of that class-AB output stage could then simultaneously be used
as the upper half for the lower class-A output device; in other
words, both class-A plus class-AB output stages in one circuit,
with a common output termination, each operating simultaneously,
with the class-A connection maintaining transconduction
continuity through low current class-AB crossovers at no matter
what voltage angle any loudspeaker current might momentarily
Now this did work, and well too, but I was still not convinced
that the sound could hold its own against good tube power
amplifiers, so I was still not able to hang up my imagineering
hat. I reasoned that further improvement would be possible
through providing a 'stand alone' upper half class-A collector
load in order to fully separate class-A current flow from the
class-AB biasing arrangement. My options for this were resistors,
a transistor current sink, or an output choke similar to those
that went out of fashion long before transistors were invented!
Now resistor current flow between the positive rail and the
class-A collector could not remain constant through large output
voltage amplitude swings; this means that the A-AB bias balance
would be correct at zero output potential only, and whilst
adequate for high quality at low output levels only, bias
interaction would increase through loud asymmetrical music
waveforms; also the resistors could not be bootstrapped due to
their need for low value. Transistor current sinks can introduce
their own amplitude/slew induced non-linearities plus an
inconstant reference bias variation with temperature, this
resulting in a quiescent A-AB bias null offset variation with
temperature. An output choke is simple and realisable, and
although winding heat dissipation would be a problem I still felt
that this option could be successfully implemented. As indeed it
was for the 100+W version. However additional heat dissipation
from a choke suitable for the 200+W version would require this
component to being specially wound, so although I had a perfectly
functional base design my thinking was still not over. Eventually
I realised that the VAS bias chain could simultaneously set the
reference potential for a positive rail based current source, as
is shown in the higher power circuit. Both of the above circuits
have been fully tested, thus either the choke or transistor
constant current source class-A output stage option may be
So now, and at very - very - long last, I actually have a most
capable solid state 'audio'
amplifier that is capable of the low level refinement normally
available only via genuine class-A amplification, yet with equal
refinement throughout its excellent high power class-AB drive
reserve, plus, and this too is at all levels, a 'blackness'
behind notes and voices that is more often associated with top
flight tube amplifiers only. Some might say it is the silence
between the notes that makes a performance, but when it comes to
audio reproduction it is that lack of cerebral distraction due to
the silence behind the notes, which then allows us the pleasure
of imagining ourselves as being 'live' at the recording
I have always studied music waveforms from a dynamic viewpoint -
as if they are an irregular series of 'splashy' and ever changing
asymmetrical first cycles; not the smoothly liquid streams of
sinusoidal components that theoreticists so often encourage us to
dip and waggle our toes in whilst we are encouraged to follow
their academically correct but time isolated examination
methodologies. Unless our thoughts stay with initially coherent
audio wavefronts and the turbulently reactive myriad of circuit
and interface responses subsequently arising, simplistic
applications of established theory can so blinker that we become
distracted from more meaningful fundamental matters.
It is here worth noting (1) the original JLH class-A has no
additional signal or NFB path capacitance capable of delaying
transient response capabilities, also, (2) the integral NFB
cannot become positive at a high frequency because of the
tailored overall gain-bandwith product with so few active devices
being enclosed by the NFB loop.
Of particular importance is that NFB is applied to the emitter of
the first transistor with respect to the input base plus any
audio input carried thereon. This is a classic series connected
voltage feedback arrangement. So, although the JLH class-A has
two distinct 180 degree phase changes along its signal path, both
of these are not then encompassed by the closed NFB loop, and
this is why with sensible construction topology and loading,
these amplifiers cannot splash over into device induced phase
shift instability at higher frequencies. The high frequency
output voltage does not become fully out of phase with potential
at the input transistor *emitter*!
Unfortunately any circuit more complex than the basic bipolar JLH
class-A naturally introduces additional high frequency phase
change, whether this is through Mosfet gate capacitance or the
utilisation of additional bipolar devices. Generally there is
then a need to compromise between stability and open loop
bandwidth control, and this can end up audibly impacting upon
first cycle (transient) response capabilities.
Thus, opting to use a differential input stage in order to
minimise input transconductance distortion and output zero offset
drift; or, mirroring the differential input stage to reduce
power-up thump and maximise open loop gain plus NFB - which
further minimises amplitude non-linearity; or, running an output
stage using Mosfets or separate drivers and output transistors;
can, individually or together, be said to introduce 'audible'
change - if - the dominant pole turnover frequency must
subsequently be pulled down to, or be reduced to a lesser open
loop audio frequency in order that closed loop stability be
Yet I implement all three of these individual circuit
arrangements whilst still retaining excellent first cycle and
signal to (noise + control delay induced error) figures, plus
good stability and low distortion. It is fact that a good total
harmonic distortion specification cannot guarantee a good first
sinewave cycle response because sine measurements are not taken
until after the first cycle has passed and the waveform has
become steady; whereas a low first sinewave cycle distortion
figure cannot be achieved without the overall thd. figure already
being better at the same frequency.
To overcome additional semiconductor device induced phase change
at high frequency I implement a base emitter connected 10nF
capacitor at the differential input pair NFB sensing node, plus a
220nF base-emitter connected capacitor on the NFB leg of the
differential mirror. These values are chosen to have minimal
impact upon the forward audio frequency signal path with regard
to the established tail current plus output stage loading of the
differential pair. At higher frequencies however, where
additional device usage could introduce unavoidable phase changes
within the closed NFB loop and cause closed loop instability,
these capacitors make the differential pair behave like an
original single JLH input transistor with series emitter voltage
feedback, and make the current mirror behave like an inactive
The separate but simultaneously driven and parallel output
connected single ended class-A output stage actively minimises
inherent transconduction variation and switching delays through
the relatively low current class-AB crossovers at moments when a
dynamic loudspeaker load presents the output terminal with
leading load current (momentary reverse current drive).
Additionally, the local, plain 27k resistor derived, A//AB output
stage degeneration sets up a constant minimum of crossover error
damping without reliance upon the global NFB loop! The output
stage is further additionally stabilised in its own right via a
Miller connected 22pF plus 1k series capacitor-resistor pair at
the VAS/splitter transistor, yet again though, using component
values which cannot impact upon the open loop audio bandwidth.
With higher rail voltages and twice the number of output devices
used in the 200+W circuit, these values become 47pF and 470 ohms,
though when equivalent / fake / non-Toshiba / older devices are
used, then the Miller connected values should become 47pF + 1k
for the 100+W, and 100pF + 470 ohms for the 200+W circuits.
Overall then, if this circuit is physically constructed using the
recommended star ground, star rail and star output nodes which
prevent current peak induced hf voltage drops along the connector
to one sub-circuit interconnect from co-coupling into another,
the resulting GEM amplifier will present a low and flat phased
output impedance which renders it highly impervious to composite
dynamic loudspeaker system impedance variation and the back-EMF
induced interface errors that can so often arise with global NFB
amplifier designs due to their dominant pole filters delaying
output current generated control of output terminal voltage.
In spite of what some designers claim, there is no way of
completely eradicating or displacing crossover distortion arising
when a lone class-AB amplifier dynamically drives a reactive
loudspeaker. NFB might reduce the resultant load induced
distortion, but cannot fully eradicate it because loudspeaker
back-EMF generated currents can reverse drive the class-AB output
stage through a fraction of its crossover bias potential before
dominant pole delayed NFB control can attempt correction. When
the dominant pole is at an open loop audio frequency, the effect
upon the closed loop response becomes audible because the control
of loudspeaker generated back-EMF becomes phase shifted, and thus
fractionally delayed, with a new additional momentarily
uncontrolled output terminal voltage error arising that has
nothing to do with the original input signal waveform. Thus
output terminal voltage error generated due to say a bass or mid
driver section back-EMF can become directly coupled into mid and
It is loudspeaker system current flow causing the development of
these output terminal error voltages which the amplifier cannot
quickly enough prevent, that then becomes recognisable as a
typical 'solid state sound'. This often manifests as a falsely
bright, occasionally a more desirable 'live' response, a
perception like 'glassy' or 'ice cold', a jittery high treble, or
occasionally as an uncomfortable upper mid-range peak. It can
'inexplicably' arise after a set of loudspeakers has been
changed, yet actually be due to flawed amplifier design! Thus
there is considerable difference between designing an amplifier
capable of amplifying audio frequencies at low distortion when
resistively loaded, and designing an 'audio'
amplifier capable of competently driving real-world loudspeaker
systems! If you doubt this statement, then please compare in a
directly switched A-B fashion, this class-A//AB amplifier with
other conventional class-AB types.
During the 1970s the Quad Hi-Fi company patented their '405' Current Dumping design. It was powerful, good sounding, compact and reliable. I used one myself and marvelled at their theoretical ingenuity, though for me the 2x KT88 Leak TL50+ still provided better reproduction. Since then that Current Dumping circuit has been repeatedly refined for professional or home use, and yet whilst updated models remain available today, so does a more recent Quad 2x KT88 II-40 monoblock chassis! The Current Dumping circuit cleverly combines both class-B and class-A output stages through a reactive output bridge which allows for the slower class-B switching. With my own circuit however, I have combined class-AB and class-A outputs in real time, this to provide powerful audio reserves with a class-A like cleanliness, whilst retaining the kind of transparency at higher output powers more often associated with expensive push-pull triodes or those ultralinear kinkless tetrode (KT) designs. (Prior to this my favourite power amplifier had been my own hybrid 100W 4x KT88 UL PP-AB1). I also believe the performance of the GEM is future proof and will not be superceded by digital amplifier designs, because dynamically energised loudspeaker system back-EMFs cannot fail to interact with the integral filters necessary to prevent their switching output stages from becoming RF noise transmitters.
This circuit was intended for use with modern 2SC5200-2SA1943
power transistors, yet it has been successfully constructed using
other device types, including a single Sanken 2SA1216+2SC2922
pair in place of the paralleled 100+W AB outputs. Feel free to
use whatever transistors are to hand or in your 'salvage' box
because the parallel class-A plus class-AB operation will make
better use of obsolete / scrap devices than can conventional
amplifier circuits, including the old and almost indestructible
industrial 2N3055 based series.
Keep VAS, Zobel and output related wires at least 5cm/2"
away from input devices and wiring. Use a star earth, star power
distribution points from each fused psu rail low ESR 10mF at the
pcb, and a star output node connection. It is also essential to
use separate wires between the negative star node, the PNP output
collectors, and the class-A emitter resistor to prevent class-AB
current peaks from voltage modulating the class-A stage via wire
impedance. See the illustrations for a universal pcb layout
kindly contributed by Daniel Bosch in South Africa. This board
has been designed to use any locally available axial or radial
capacitor types, including larger low ESR variants. Parallel all
of the large electrolytic capacitors with lower value components
to minimise the risk of effects due to an unexpected series
impedance peak. Do not twist any extender e-b-c wires used to
connect out to heatsink mounted output stage devices. Mount the
VAS, the drivers, also the Vbe multiplier on the output heatsink
for automatic temperature compensation. During assembly
check/adjust the pre-sets sliders for 50% setting; note that
these should be 15 or 10 turn components. If long wires cannot be
avoided with your choice of layout, then fit additional 1uF
capacitors between each class-AB output device collector and the
heatsink ground. The series input capacitor comprises two 470uF
components in parallel, though mutually connected plus to minus,
my own being low ESR types.
I recommend, after checking the circuit wiring for correct
assembly, first powering up using two 9V transistor radio
batteries, then with 22 ohm per rail power resistors in place of
the fuses as protection for any error at initial switch-on.
However, do not try to set any bias with either of these testing
options. If everything is okay the amplifier should present an
unbiased open circuit zero output potential within 100mV. It
should also cleanly drive a test loudspeaker for low level signal
testing without any bias being set up, as the two amplifier
output stages will automatically compensate for each other's lack
of bias; you should not hear anything through the loudspeaker
after a possible low level initial power-up charge. I suggest
your testing input could be taken from the headphone output of a
portable CD player or an i-Pod, whereupon the unbiased amplifier
should produce limited audio, even on rails as low as +/-9V!
Click the image for a larger view.
If all is well, power up with a fully fused psu connection. Clip
a pair of test multimeter wires to class-AB output emitter
resistors as indicated on the circuit diagram. Slowly increase
the value of the class-A bias trimpot until the class-AB
'imbalance' reading nulls to zero; then adjust the class-AB
trimpot for an average 40mV of quiescent bias per output pair.
Re-set the bias from zero current after two hours of normal use.
For 70W/4R=2x35W/8R use 30V rails. For 50W/4R=2/25W-8R use 25V
rails with just a single pair of class-AB power transistors.
Daniel has my thanks for checking out every modification update
as it arose, whilst running his 200+W GEMs to drive Apogee
loudspeakers throughout the last 12 months.
Always re-bias from zero current if you alter the rail voltages.
If necessary insert a 0.22 ohm resistor in series with the output
terminal to maintain stability when driving capacitively reactive
loads, or use any series output resistor value up to 2.2 ohm if
you wish to imitate different types of tube power amplifier
output impedance and damping. Also don't forget to bi- or
tri-wire out to composite loudspeaker system sections and drivers
in order to obviate single cable developed dynamic voltage drops
due to varying crossover-loudspeaker system generated back-EMFs,
no matter how expensive your loudspeaker cable might be!
Better still, use these amplifiers as they are intended - as line
driven and loudspeaker sited monoblocks! A simple line driver
circuit appears on another page, this may be used to buffer the
output of a tube DAC or pre-amplifier, or be combined with the
GEM amplifier to raise its stand alone input impedance. Pcb
outlines are also shown.
Finally and very importantly, in relation to the single ended output choke option!
Whilst it is possible to run this amplifier without any choke, say by series connecting four 22 ohm heatsink mounted resistors in its place, for the lower power version I use two 230V 50VA mains transformers primaries connected in series. These have a resistance of approximately 40 ohms each, and do become rather hot, so free air ventilation is essential. My transformers were cut apart, then re-assembled as a thick paper gapped twin 'E' core assembly. See photo.
The links below are all pictures of the actual prototype. Some of these images are quite large and may be resized automatically to fit in your browser. Clicking on the actual image restores full size or images can be saved by using a right mouse click.
Beware of fake Toshiba Transistors. The genuine article is shown below on the left. The center and right most image are fake Toshiba transistors. The fakes have higher junction capacitance and limit the high frequency response having an adverse effect on sound quality.
Do not manually attempt to connect or disconnect the output choke once this amplifier has been powered up. If you are holding the connecting wire and the wire insulation breaks down under back-EMF potential, it will not be the amplifier that is damaged, but *YOU*.
I have enjoyed music listening through several fine amplifiers, but for me this development supersedes all because it imparts notable loudspeaker control without generating the secondary amplifier-loudspeaker interface current flow related error components which so often spoil solid-state amplification.
Thus I wish everyone who constructs this GEM - the very best of listening....
..... Graham Maynard.