Return to Home Page - www.GMweb2.net
The GEM: A Class-A//AB Amplifier.
Graham Maynard graham.maynard4@virgin.net
- INTRODUCTION -
The GEM is an audio power amplifier embodying simultaneously active Class A
and Class AB output stages. To aid construction, a universal 'star
connection' pcb capable of accommodating either plain or low ESR radial or
axial electrolytic capacitors has been developed, and it may be used to
construct 25W - 200W conservatively rated loudspeaker driving amplifiers
according to individual choice.
The circuit developed out of some 35 years of on/off investigations into the John Linsley-Hood, MIEE, 1969 class-A amplifier, and thus I suggest printing off the following circuit diagrams to assist reading.
(1a) The original JLH class-A circuit diagram.
(1b) The first GEM
(2) Final design 100W (choke version)
(3) Final design 100W (CCS version)
(4) Final design 200W (CCS version)
- ACKNOWLEDGEMENTS -
This amplifier 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 also a partly 'fortuitous' outcome resulting from endless
thinking time afforded me as a result of some serious head injuries which
limited my physical capabilities, and still debilitate if I attempt normal
everyday activities. (See;-
http://gmweb1.net/) I also owe thanks to my Internet friends Carlos
Mergulhao in Brazil, and Daniel Bosch in South Africa (who designed/maintains
this Website) for their honest hands-on testing and reporting, because there
have been problems for those who did not use the specified low path
capacitance 'star' pcb construction layout essential with this design!
The GEM is
not claimed to be the best amplifier for anyone to use, for indeed there
already are so many system design ideals and requirements that not one of them
can alone be expected to suit everyone, no matter how well any might measure
and perform in relation to original requirements. As ever, there is always
more than one way of studying a 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 when properly constructed there is no need for additional
stabilisation components or the series output choke which so often introduce
NFB related control delays to a real world amplifier's output terminals when
dynamic loudspeakers (instead of a test resistor) are being driven. The JLH
not only amplifies percussion transients and spoken sibilants cleanly, it also
is non-reactively silent behind voices and notes, such that an artificial
brightness or smear does not affect the reproduction of detail or stereophonic
imagery. JLH class-A amplifier output is instantly recognisable as being
correct, and yet it can still sound lacking when compared to the typically
amplified sounds we have become accustomed to hearing via every-day class-AB
circuitry.
So many
'linearity' analysts study an amplifier's forward distortion characteristics
under steady sinewave drive with a passive resistor load, ie. in time
dissociated isolation, and thus they ignore (or maybe cannot imagine) the
complex internal circuit re-activity arising when voice coil energised
loudspeakers are dynamically driven by asymmetric music signal waveforms.
Actually, the JLH class-A amplifier 'amplitude distorts' more than most modern
‘hi-fi’ amplifiers when observed under forward sinusoidal analysis, and yet it
sounds so much better due to the way in which its circuit naturally damps
without delay or overshoot in the presence of secondary input at the NFB
(output) terminal. That secondary input is the dynamically generated
loudspeaker system back-EMFs (leading currents) which, when the musically
driven voltage input changes in time wrt output, can momentarily attempt to
reverse voltage drive the output terminal at some signal input frequencies,
whilst generally loading it at others. Thus leading/lagging current flow
through reactive components wrt voltage (including the amplifier itself with
its own closed NFB loop *plus* any internal feedforward or local sub-loop
induced reactivities) induces music-amplifier-loudspeaker dependent frequency
selective shifts of output stage control group delay in time wrt source input
voltage, and where a loadspeaker can render NFB loop control variable in time
then the amplifier cannot maintain continuing output terminal linearity
either.
Though much better than the JLH, the GEM's steady sine determined 'amplitude
linearity' remains inferior to many supposedly 'ultra linear' or 'blameless'
class-AB designs, yet its coherence, or 'waveform amplification linearity in
time', is similar to that of the JLH. The GEM's 'JLH-like' class-A output
device prevents leading loudspeaker current flow from inducing unavoidable
fractional bias voltage crossover spikes before (as happens with all plain
class-AB designs) the amplifier's natural *open loop* capabilities can
internally control the class-AB output devices in real time. Amplitude
distortion due to frequency selective loudspeaker reactance induced group
delay shifts of output terminal control can far exceed sine plus resistor load
measured 'amplitude lnearity' errors, and this is what led to the development
of this more powerful 'JLH' like amplifier !!!
For more in-depth information and construction details about JLH class-A
amplifiers, see Geoff Moss's excellent Website:-
http://www.tcaas.btinternet.co.uk/index.htm
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 the additional dominant pole
filtering which then so often infringes upon and degrades those full bandwidth
open loop capabilities in other ‘low amplitude distortion’ designs. Thus the
JLH has an inherent closed NFB loop ability to maintain phase linear damping
control of output terminal potential in the presence of loudspeaker generated
back-EMF throughout the audio frequency range.
So often it is this 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 and plus in an unavoidable
colouration of dynamic loudspeaker reproduction in a manner that resistor
loaded steady sine measurements cannot ever reveal. This colouration arises
when on-going output stage driven amplifier-loudspeaker system current flow
becomes back-EMF modified by the dynamically induced milli-second to milli-second
variation of reactions within composite loudspeaker system elements, as their
impedance and phase angle change due to audio waveform energy transfers
through the momentary sequential electrically lagging-leading storage-release
within crossover components and electrodynamic driver-airspring-cabinet-reflection
energy transductions.
Never forget that the current flowing through a cable/loudspeaker system for a freshly starting sound waveform always lags the amplifier's initial voltage output, before becoming modified by the natural cable/crossover/driver/cabinet/room defined reactivities that lead to generation of the complex electrical loading responses we so often see illustrated as a loudspeaker's lagging/leading phase angle versus frequency characteristic.
Where a loudspeaker characteristic is shown as being leading within a stated frequency range, it does not become so until *after* one, two or three half cycles of ongoing energisation.
An initially leading amplifier output voltage wrt current can thus become loudspeaker system modified within the first and second cycles of transduced reproduction, such that the ensuing complex combination of on-going amplification with lagging or leading resultant loudspeaker current flow can lead to distortion at audio frequencies due to NFB control having to alternate between attempting to drive and attempting to damp loudspeaker current. Positive and negative 'servo' ripples arise on a sub-cycle basis due to fractional closed NFB loop stabilisation component (C.dom) rendering the amplifier’s output stage inductive, or to a series output choke delay wrt to the reactive loudspeaker system, and this actually causes the zero current drive/damp voltage shifts (crossover distortion) we hear from class-AB amplifier designs. The outcome is a fractional group delay disturbance arising within the audible passband such that the loudspeaker terminal takes on a system defined error superimposition with respect to what should be an amplified version of the original audio waveform, and this occurs on a per channel basis, such that image position within the stereo field becomes de-focussed by loudspeaker reactivities in response to the music itself. That same NFB activity delay induced error and the resultant frequency dependent disturbance arising in real time, does not arise when a passive test resistor load is used on the test bench because the amplifier is not having to alternate between driving and damping; ie. no back-EMF current ever leaves the resistor in advance of on-going audio voltage output !
When loudspeaker current flow becomes leading with respect to amplified audio input waveform voltage, and the necessary correcting NFB loop response is fractionally delayed by the bandwidth limiting dominant pole filter rendering the amplifier output terminal fractionally inductive, or there is an internal series output choke, or there is amplifier to loudspeaker cable inductance, all of these inductances being series additive, then the amplifier's output current correction becomes fractionally delayed with respect to *loudspeaker* back EMF at higher frequencies, and the loudspeaker terminals cannot quickly enough be prevented from developing a fractional error potential wrt linearly amplified input. This is like having a damped but oscillating servo error superimposing a fast +/- error waveform at the amplifier's output terminal with respect to what should be the ideal closed loop controlled audio output. The amplifier does quickly 'catch up', but in the meantime a tiny additional loudspeaker system dependent 'dominant pole and/or impedance delay' related interaction (interface) error has already been generated at the loudspeaker terminals, and no amount of NFB can completely erase this because it was NFB control delay (phase shifted damping) with respect to priorly energised loudspeaker developed back-EMF that allowed the error to arise in the first place!
The magnitude of this loudspeaker driven error at an amplifier's output terminal is easily imagined, for it is proportional to the sine of the NFB loop controlled damping angle at any frequency, divided by the damping factor. Unfortunately however, whilst high NFB amplifiers provide impressive damping factors these also tend to be the ones having the poorest damping angle due to the fitting of the circuit stabilising components essential for unconditional stability when the high circuit gain is used to degeneratively ‘improve’ output linearity. Thus, at the typical 1kHz specification frequency, an amplifier retaining a damping factor of as little as 20 at 2 degrees should be as capable of output terminal damping loudspeaker system generated back EMF at 1kHz with respect to audio input, as can a high NFB one offering an impressive damping factor of say 550 though at a not uncommon angle of 90 degrees. Phase linear amplifier damping is much more essential than for simply attaining an impressing headline figure, for firming up a bass response is not the only requirement. Phase linear damping is essential for controlling mid/bass *loudspeaker* section generated back EMF and thus minimising the drive/damp control error which can independently interfere with treble section drive via the amplifier's output terminal in a manner the amplifier circuit itself cannot prevent when its NFB loop control is running out of phase (lagging). This is why the tweeter output of high damping amplifiers can be heard sounding much more harsh (or indeed be blown) when compared with say a JLH or an old tube amplifier. The amplifier ends up dynamically generating and imprinting a recognisable system dependent sonic signature as it attempts to control loudspeaker generated back-EMF.
Some famous
designs, as by JLH, Nelson Pass, Sugden and a few others, retain low angle
damping to 10kHz, and notably their designs tend to be class-A; so why then
do we not all use this type of amplifier ?
(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.
- STEPPING STONES -
During the early 1970s I constructed a 2ft(60cms) long JLH class-A monoblock.
It had a genuine 100W measured sine output and sounded very clean, and whilst
at maximum input it could generate some surprisingly noisy short circuit
sparks which raised thoughts about output stage survival. (It never blew, and
still runs, though became a
test bed for many variations !) However when it was compared with the
physically smaller and cooler running 2x KT88 Ultralinear Leak TL50+ classic
monoblock chassis 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 high quality 100W class-AB solid state
designs.
The reason for this 'weakness' relates yet again to loudspeaker current flow,
whereby dynamically induced momentary requirements can far exceed the peak
sinusoidal output current capability of a pure class-A biased output stage.
Where you see a bass resonance impedance peak on a loudspeaker characteristic,
this is due to a synchronously timed electromechanical storage and release of
energy that often takes one or more half cycles to fully develop *after* an
initially increased drive plus energy storage current loading at that resonant
frequency - a drive that started out as a lagging current via the voice coil
resistance. Now when a bass crossover/driver section has already been priorly
energised and the amplifier attempts to drive it against resonant
crossover/driver motion the output current necessary to maintain linearity can
momentarily peak at a much higher level than would be the initial peak
sinusoidal current at that frequency.
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,
because amplifier output becomes prematurely current clipped at all
frequencies as it fails to the output NFB loop sensed demand necessary for
maintaining bass linearity. 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 that same positive going peak output current
limitation will still set in at frequencies where the dynamic loudspeaker
current loading becomes significantly leading. Phase shift induced current
demand can also increase when crossover network slopes are 12dB/oct or more,
so it little surprise that compact mini-monitors cannot normally be
satisfactorily driven by small amplifiers.
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 genuine class-A output for 100W of quiescent heat.
Generally though the resulting amplifier was not temperature stable through
different audio duty cycles; or the modified biasing arrangements had an
audible impact upon transients; or the circuitry became unacceptably complex.
More recently I tried numerous arrangements whereby individual output devices
were replaced by identical composite sub-circuits running in never-off class-A
at low level, but conducting as if class-AB during periods of increased output
demand. This type of arrangement simulated well. Test amplifiers also worked
and were less complex than circuits having separate bias control arrangements,
but they sounded 'punchy' as if the amplifier could not keep itself from over
reacting to loudspeaker phase shifted current demands; as if the phase
splitting JLH driver could not maintain linear control when the current gains
of the individual but dynamically self adjusting output device characteristics
became externally altered on a per-half basis due to varying loudspeaker
system demand.
- A NEW CROSSING -
More recently it occurred to me that the JLH current splitting transistor
could be used to conventionally drive a modified JLH class-A output stage,
where the single collector connected device could be replaced by a biased
Darlington connected class-AB output stage; ie. the upper half of that
class-AB output stage could then simultaneously be used as the upper half for
the lower emitter connected class-A output device. In other words, both
class-A plus class-AB output stages in a single circuit, both sections
operating simultaneously and acting upon a common output termination, with the
faster emitter driven class-A output device maintaining transconduction
continuity through slower collector driven low current class-AB crossovers, no
matter at what voltage angle any loudspeaker current might momentarily flow.
This was my first circuit 1b.
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. Some time later I reasoned that further
improvement would be possible through providing a 'stand alone' upper half
collector load for the now separately operating lower class-A output device in
order to fully separate class-A output 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!
Resistor current flow between the positive rail and the class-A collector will
work, but it cannot remain constant through large output voltage amplitude
swings; this means that the A to AB bias balance would be correct at zero
output potential only, and whilst adequate for high quality at low output
levels, bias interaction would increase through loud asymmetrical music
waveforms; also the class-A collector resistor could not be bootstrapped due
to the need for it to have a relatively low value.
Transistor current sinks can introduce their own amplitude/slew induced non-linearities unless they have separate very low impedance and temperature compensated bias reference, for without this there could be a drift in the quiescent bias current null with varying audio output duty cycles.
An output choke is simple and realisable, and although winding heat dissipation could be a problem I still felt that this option might be successfully implemented, as indeed it was for the 100W version. However additional heat dissipation from a choke suitable for the 200W version would require this component to be specially wound, so yet again, although I had a perfectly functional base design, my thinking was still not over.
Eventually
I realised that the VAS bias chain could itself simultaneously set the
reference potential for both a positive rail based PNP constant current source
and the lower active NPN class-A output device, as is shown in the higher
power circuit. Both of the above GEM class A//AB circuits have been fully
tested, thus either a choke or transistor constant current source class-A
output stage option may be chosen; say with a single 2SA1387 type device and
0.47 ohm emitter resistor being used for the 100W version, with the 27 ohm
resistor in parallel with a low voltage 10mF inserted in the VAS bias chain as
per the 200W circuit diagram.
So now, and at very - very - long last, I actually have a trustworthy solid
state 'audio' power amplifier capable of the low level
refinement normally available only via genuine class-A amplification, yet
retaining class-A like 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 afficionados might say it is the silence between the notes that
makes a performance, however when it comes to audio reproduction it is the
lack of cerebral distraction due to silence *behind* the notes that leads to
the revelation of unadulterated original detail, and from this an imagination
that we are hearing as if 'live' at the recording performance itself.
- KEEPING DRY -
I have always thought of amplifiers as having two input terminals, the second
one being the output terminal connected NFB sensing node. I also recognise
music waveforms as being dynamically irregular series of 'splashy' and ever
changing asymmetrical first cycles; not the smoothly liquid streams of
sinusoidal components most theoreticists so often encourage us to dip and
waggle our toes in whilst we are encouraged to fixate upon their academically
correct but time isolating 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 us that we become distracted
from more meaningful fundamental matters.
It is here worth noting;-
(1) that the original JLH class-A has no additional signal or NFB path capacitance capable of delaying transient response capabilities, also,
(2) that
its integral NFB cannot become positive at a high frequency because so few
active devices are actually enclosed by the NFB loop.
Of particular importance is the method of NFB application; ie. 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 boil 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 then develops a need to compromise between stability
and open loop bandwidth control, and this can end up audibly impacting upon
first cycle (transient) response capabilities and loudspeaker damping.
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 ensured.
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 a stable low damping angle and good linearity.
It is a 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 induced asymmetries have passed and the
waveform has become steady; whereas a low first sinewave cycle distortion
figure cannot be achieved without the overall total harmonic distortion figure
already being better at the same frequency.
- AVOIDING SPRAY -
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 this additional device
usage could introduce unavoidable phase changes within the closed NFB loop and
otherwise induce 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 current source !
The separately acting but simultaneously driven and parallel output connected
single ended class-A output stage now 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 driving/damping
crossover error correction 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 200W 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 + 470 ohms for the 100W,
and 100pF + 220 ohms for the 200W circuits. These last values might seem
high, but the 27k resistors increase their turnover frequency.
Overall then, if this circuit is physically constructed using the recommended
PCB with star ground, star rail and star output nodes that prevent current
peak induced hf voltage drops along the copper conductors from co-coupling
between one sub-circuit and another, the resulting GEM amplifier will present
a low and flat phased output impedance highly impervious to the 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 crossover displacing the loudspeaker system induced crossover
distortion that arises when a lone high NFB 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 still non-linearly reverse drive the class-AB output
stage through a fraction of its fixed crossover bias potential *before*
dominant pole delayed NFB control can attempt correction. When the dominant
pole turns over 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, with a new additional momentarily uncontrolled
output terminal voltage error arising that has nothing to do with the original
input signal waveform. This is how the output terminal voltage error
generated due to say a bass or mid driver section back-EMF can become directly
coupled into mid and tweeter drivers, or, at some frequency where there is
additional crossover section phase shift, there can be a reduction of higher
audio frequency damping control with this leading to the generation of an
uncomfortable sounding reproduction peak !
It is loudspeaker system current flow causing the development of these
loudspeaker 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, or a perception like 'glassy' or 'ice cold', a jittery high treble,
or sometimes 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 and not to the loudspeakers themselves!
Thus there
is considerable difference between designing a high specification amplifier
capable of low distortion audio frequency amplification when resistively
loaded; and designing an 'audio' amplifier capable of
competently driving real-world loudspeaker systems. That is why I do not list
this amplifier's THD specifications, for once lower than say 0.1% this level
can easily be exceeded by the uncontrolled fraction of loudspeaker back-EMF
level an amplifier fails to control anyway. And if you doubt my statements,
then do please compare the final class-A//AB designs with any other
conventional class-AB types of your choice in a directly switched A-B fashion,
to hear first hand why it really is a 'GEM'. Actually it is often possible to
load an existing class-AB amplifier with a resistor plus potential driver to
feed the GEM, whereupon the GEM will illustrate improved class-AB amplifier
performance when it is *not* loaded by a loudspeaker!!!
- THAT'S IT !!! -
During the 1970s the Quad Hi-Fi company patented their '405' Current Dumping
design. It remains as powerful, good sounding, compact and reliable as it was
then, and so remains a decent second-hand buy. I used one myself and
marvelled at their theoretical ingenuity, though for me the 2x KT88 Leak TL50+
still provided better dynamic reproduction. Since then the Current Dumping
circuit has been repeatedly refined for professional or home use, and yet
whilst updated models remain available today, so too does a more recently
released Quad 2x KT88 II-40 monoblock chassis, and for 2007 - a 4x KT88
Quad II-Eighty
!!!!! That Current Dumping circuit cleverly combines both class-B and class-A
output stages via a reactive output bridge which allows for the slower class-B
switching. With my own circuit however, I have combined both 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 the
ultralinear kinkless tetrode (KT) designs. (Prior to this my
favourite/reference power amplifier had been my own 1975 homemade hybrid
100Wmax 4x KT88 UL push-pull AB1, with SS differential input/phase splitter
feeding a pair of ECC82s, each one running as a gain stage plus cathode
follower per push-pull half.).
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 at higher audio frequencies with the integral filters necessary to prevent digitally switching output stages from becoming RF noise transmitters. Class-A//D is a possibility, but that would seem to be a contradiction in terms of both complexity and modal efficiencies.
- CONSTRUCTION NOTES -
Do not re-draw the circuit diagrams, nor attempt to view them through
*conventional* eyes.
Get your head around the layout as it is shown, for this is the exact way in which it needs to be wired in real life.
If you build the GEM differently, with say a ground plane double sided pcb or with an all-in-one PCB including output device connection, it might still work, but it might also oscillate, and then when stabilised it might not perform as well as originally intended because the C.dom must be increased to a point where NFB loop controlled damping turns over within the audio passband! So often it turns out to be the constructor's choice of layout and fabrication which lead to performance and instability problems, for cutting corners or attempting to make the finished amplifier look 'neat' or 'impressive' can be as deleterious as fitting the wrong parts.
This
circuit was designed for use with modern low capacitance 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 100W AB outputs. Feel free to use whatever transistors are to hand
or in your 'salvage' box, including the old and almost indestructible
industrial 2N3055 based series, because the parallel class-A plus class-AB
operation will make better use of obsolete and scrap/recovered devices than
can conventional amplifier circuits. Simply increase the value of the
stabilising C.dom capacitor value if necessary.
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
power rail pcb star node, the PNP output collectors, and the class-A emitter
resistor; this in order to prevent class-AB current peak induced voltage
drops from modulating the class-A stage via interconnecting wire impedance.
See the illustrations for a universal pcb layout kindly contributed by "GEMlover" 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.
All components to the right of the dotted line on the circuit diagram should be hard wired on the heatsink, with just one fine wire linking the pcb input stage to the base of the VAS transistor, whilst all other current carrying wires should be much more substantial. Do not twist any extender e-b-c wires used between heatsink mounted output stage devices. Mount the Vbe multiplier directly on the output heatsink for automatic temperature compensation; it tracks accurately without any need for additional diodes.
During assembly check that the pre-set sliders are adjusted to a 50% resistance; 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 which should be separately connected to the pcb ground star. The series input capacitor comprises two 470uF components in parallel, though mutually connected plus to minus; my own being low ESR types.
A 600 ohm
(studio like) input loading assures accurate transient and high frequency
responses via screened feeder cables, whilst also improving NFB control;
don't forget that NFB acts with respect to the differential input transistor
base, and this is in series with any impedance apparent at the input. (John
Linsley Hood himself recommended low impedance driving for this type of
circuit way back in 1969!) The low value input resistor also reduces input
hum and noise, so when quiescent this amplifier is quieter than any available
audio source. The low impedance input stage with bypassing capacitor between
the NFB sensing base and emitter would be affected by series emitter resistors
at the differential input pair, thus the input pair should be physically
shielded together to prevent the output zero drift that would occur if one
transistor received more radiated heat from the output stage.
I recommend, after checking the circuit wiring for correct assembly, to first
power 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. (This reduces considerably when
the biases are eventually balanced; should be less than 20mV.) The unbiased
amplifier should also cleanly drive a test loudspeaker for low level signal
testing due to the two amplifier output stages automatically compensating for
each other's lack of bias. After a possible initial low level power-up charge
you should not be able to hear anything through the loudspeaker. I suggest
your testing input could be taken directly 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!
If all is well, power up with a fully fused psu connection. Clip test
multimeter wires to a pair of 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. If
necessary, re-set the bias from zero current after two hours of normal use,
and always re-bias from zero current if you alter the rail voltages
For 70W/4R=2x35W/8R use 30V rails. For 50W/4R=2x25W-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 200W
GEMs to drive Apogee loudspeakers throughout the last 12 months.
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 about 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. (!!! As with amplifier performance you cannot measure these dynamically energised voltage drops by measuring between cable ends when driving a loudspeaker with steady sinewaves !!!) Separate runs of quality loudspeaker fig-8 twin to each crossover section or driver are much better than one run of expensive loudspeaker interconnect.
Indeed, any
crossover should actually be at the amplifier output terminals with one twin
then extended out to each driver. Better still, use these amplifiers as they
are intended - as line driven and loudspeaker sited monoblocs. A simple line
driver circuit appears on another page. It has a voltage gain of 4 and 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 for
this 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.
WARNING!
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 the thousands of volts back-EMF potential
... it will not be the low impedance amplifier that is damaged ... but high
impedance
*YOU*.
FAKE TRANSISTORS
Beware of fake Toshiba Transistors. The genuine article is shown
here. An example of fake Toshiba transistors is shown
here. The fakes have higher junction capacitances; this alters closed
loop stability and limits the high NFB frequency response, so they could cause
oscillation and adversely affect reproduction quality. All is not lost
though, because they should still work to the limits of their capabilities,
simply 'up' the value of the VAS connected C.dom until stability is assured.
- MY CONCLUSION -
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 amplified
reproduction.
Thus I wish everyone who constructs this GEM - the very best of listening....
..... Graham Maynard.
PS. I should be interested in seeing photographs and hearing from everyone
who builds this amplifier. The e-mail address above is available for direct
contact and any related enquiries.
Return to Home Page - www.GMweb2.net
