The Load-Invariant Power Amplifier PCB.

Constructional Notes & Extra Info.

This page last updated: 25 July 98

to The Institute
20 NOV 95                                         

   The Load-Invariant power amplifier is believed to be the first
of its kind, in that it is specifically designed to minimise the
distortion produced when driving loads below 8 Ohms.
   The design offers not only extremely low distortion into 8 Ohms,
but also a performance which is only marginally worse into 4 and 3
Ohms. While the amplifier is not totally Load-Invariant, as THD
into the lower loads is not identical to that into 8 Ohms, it is
probably the closest approach to it made so far. Just how close
this approach is depends on which option is selected for the output
   The theoretical basis of the design is explained in detail in
the Electronics World article "A Load-Invariant Power Amplifier"
published in Jan 1997.
   As in a standard Blameless Class-B power amplifier, the basic
philosophy is the linearisation of each stage with a relatively
high degree of local negative feedback before closing the global
feedback loop; this permits the global feedback factor to be kept
relatively low to maximise HF stability, while still giving
exceptionally good linearity. The main innovative features .

   The PCB follows exactly the design published in Electronics
World for Jan 1997, including the SOAR (safe operating area)
protection against output short-circuits, and the inclusion of rail
fuses. The SOAR protection locus allows output power to be
   Please note that it is not in general possible for us to give
advice on modifications to the circuitry beyond those detailed
here. An apparently simple modification may have more implications
than are immediately obvious, and may require extensive testing to
verify that there are no unpleasant side-effects.

   The board is of high-quality roller-tinned fibreglass, in
double-sided plated-through-hole (PTH) format, which allows the
layout to be optimised to minimise topological distortions, such as
induction of supply-rail currents into signal paths. There is a
full silk-screen component ident and a solder-mask on both sides to
minimise the possibility of solder shorts. Improvements over the
previous PCBs have been made as follows:

  The form-factor of the PCB has been altered so that two will fit
   side by side in a 19" rack chassis.
  All transistor positions have emitter, base and collector marked
   on the top-print to aid fault-finding. TO3 devices are also
   identified on the copper side.
  Alternate positions are provided for the input transistors
   (Q1,Q2 or Q101,Q102) so that devices with differing pin-outs may
   be used without having to cross their legs.
  The PCB is double-sided PTH, so wire links have been eliminated.

   The space available for articles in EW is finite, so extra
information for which no room could be found is given here about
the circuit and its operation. Much greater detail will be found in
my book "Audio power amplifier Design Handbook" (Newnes; ISBN 0-

   The input stage follows my design methodology in running at a
high tail current to maximise transconductance, and then
linearising it by adding input degeneration resistors R5,7  that
reduce the final transconductance to a suitable level. A current-
mirror Q3,4 forces the collector currents of the two devices to be
equal, thus balancing the input pair and ensuring that it does not
generate second-harmonic distortion; even a small Ic imbalance
generates second-harmonic at a much higher level than the
unavoidable third harmonic. The mirror is itself degenerated by
R6,8 to minimise the effects of Vbe mismatches in Q3,4; the value
of R6,8 is not critical and any value in the range 10 - 68 Ohms
should be sufficient.
   As a result of the insights gained while studying the slew-rate
characteristics of this configuration, I have increased the input-
stage tail current from 4 to 6 mA, and increased the VAS current
from 6 to 10 mA over the original Class-B circuit. This
significantly increases the maximum slew rate, though the full
theoretical speed increase is virtually impossible to obtain. One
reason is feedthrough in the VAS current source; in the original
circuit an unexpected slew-rate limit was set by fast edges
coupling through the current source c-b capacitance to reduce the
bias voltage at the worst possible time.
   This effect is minimised in this design by using the negative-
feedback type of current source bias generator, with VAS collector
current as the controlled parameter. Q6 senses the voltage across
R18, and if it attempts to exceed the Vbe, it turns on further to
pull up the bases of Q5 and Q7. C5 filters the rail ripple from the
supply to this circuit and prevents ripple injection from the V+

   The input pair positions Q1,2 on the PCB are laid out for base-
in-the-middle TO92 transistor packages; ie C-B-E. If the high-beta
2SA970 or similar devices are used, they will be found to have the
pinout B-C-E, with base and collector swopped; positions Q101,102
are provided to cater for these.

   There are positions on the PCB for the input bootstrap
components, (C3,R9) but these may simply be omitted if the facility
is not required; for example, a competent preamp with a buffered
output should be able to drive the 2K2 input impedance without
problems. If desired R4,(input bias resistor) and R15,R16,C8
(feedback network) may be returned to the original values used in
the Class-B amplifier, which gives a high input impedance without
bootstrapping; this will of course degrade the DC offset and noise
performance back to their original (but still very respectable)

   The VAS Q11 is linearised by beta-enhancing stage Q10, which
increases the amount of local NFB through Miller dominant-pole
capacitor C9. (often referred to as Cdom in these writings) Note
that R25 has been increased to 2K2 to minimise power dissipation,
as there is no significant effect on linearity or slewing. Do not,
however, attempt to omit it altogether, or linearity will be
affected and slewing much compromised. I have tried replacing it
with a constant-current source, but this seems to give no benefits
in linearity.
   As described in [3], the simplest way to prevent ripple from
entering the VAS via the V- rail is old-fashioned RC decoupling,
with a small R and a big C. We have some 200 mV in hand in the
negative direction, compared with the positive, and expending this
on voltage-drop through the RC decoupling will give symmetrical
clipping. R27 and C10 perform this function. 470uF is in general
adequate, but the slight increase in output ripple is measurable.
(if not audible)

   The output stage is of the Complementary Feedback Pair (CFP)
type, which gives better linearity and much better quiescent
stability than the more common EF alternative, as a result of the
local negative feedback loops around driver and output device.
Where there is negative feedback there is always the possibility of
oscillation, so this configuration is slightly less straightforward
to apply, which may account for its lack of popularity.
   When oscillation in a CFP stage does occur, it will probably be
at a much higher frequency than global-feedback Nyquist oscillation
(say 2 MHz rather than 400 KHz, but these are VERY rough figures)
and is often very sharply confined to one half-cycle only, due to
the parameter differences that exist even between nominally
complementary devices. I have always found that the 100-Ohm base-
stopper resistors (R28,R29 in the circuit) are effective against
this sort of misbehaviour, but it is possible that with some
combinations of devices this value may need adjustment.
   Several rude things have been said about my output network (ie
the Zobel network C15,R38, and the output inductor L1 with its
damping resistor R39) but I have left it unaltered as while it may
be sub-optimal in some respects, IT WORKS, conferring what appears
to be complete immunity to instability with capacitative loading.
   I suspect L1 could be reduced in value and size, without any
ill-effects, but testing this against all possible real-life loads
is very lengthy, so there is a powerful incentive to stick with the
output network that worked for your Grandfather...

   SOAR (Safe Operating area) protection is given by the networks
around Q12,13. This is a conventional single-slope SOAR system,
which effectively draws a straight line protection locus across the
Vce-Ic graph. This is much more effective than simple current-
limiting. The extra complexity of dual-slope SOAR limiting was
considered unnecessary.
   In the positive half, Q12 monitors the current through Q16,18 
and also the voltage across it. D6,7 prevent spurious conduction of
Q12,13 when they are reverse biased at the collector-emitter by
large output voltage excursions.
   In the negative half of the output stage, Q13 conducts when the
protection locus is exceeded. Q9 is a current limiter for the VAS
transistor Q11; when protection transistor Q13 conducts large
currents can potentially be drawn through Q11. Q13 monitors the VAS
current through R26, and turns on to shunt base drive away from the
VAS when required. This protection is not required for positive
SOAR limiting as Q7 is a current-source with inherently controlled
output current.
   There is actually a subtle trap waiting for those who apply SOAR
limiting incautiously; Q12,13 can conduct just a trifle in normal
operation, and this is usually not very symmetrical so it looks
like second-harmonic distortion. You can drive yourself mad looking
for the source of this distortion in the VAS or the output stage,
and so I strongly recommend that D6,7 are omitted until the circuit
linearity has been checked.

   It is often stated in hifi magazines that both valve and
semiconductor amplifiers sound better after hours, days, or
possibly even months, of warm-up time. However, regular readers
will have concluded by now, much as I have, that if an unsupported
statement appears repetitively in hifi magazines it is almost
certainly not true, and I fear this is just another example. It
would certainly be possible to come up with a transistor amplifier
with bad crossover distortion and a very poorly-controlled
quiescent current, which might produce audible changes over time,
but the effect would need to be so gross that it would be
impossible to miss in even the most casual measurements, and I have
never seen this reported as an explanation. Such a case would have
to represent plain incompetence in design, for there are a wealth
of electronic techniques for rendering the circuit performance more
or less immune to variations in beta and temperature.
   I think it important to state emphatically that in this design
the operating conditions stabilise in a few seconds, giving the
full intended performance. No "warm-up time" beyond this is

   The published design is not intended as a purely cookbook
project- such are not the domain of Electronics World- and
therefore omits the important ancillary of relay muting and DC
offset protection under fault conditions. It is VERY strongly
recommended that a DC protection system is added; eg a ready-
designed version such as the Maplin version, which is known as
Velleman Kit K4700 (Order Code VE24B)

   If you intend to design your own relay protection system- and
this is fairly straightforward- then here are some points to
ponder. Having paid for the relay, for DC protection, it seems
sensible to use it for system muting as well, to prevent thuds and
bangs from the upstream parts of the audio system from reaching the
speakers at power-up and power down. The amplifier itself, being
dual-rail (ie DC-coupled) does not generate large thumps itself,
but it cannot be guaranteed to be completely silent.
   Your relay-control system should:

1 Leave the relay de-energised when muted. At power-up, there
should be a delay of at least 1 second before the relay closes.
This can be increased if required.

2 Drop out the relay as fast as is possible at power-down, to stop
the dying moans of the preamp etc from reaching the outside world.
My own preference is to do this by sensing the incoming AC; when
this disappears, the relay is dropped out within 10 msec, which
should be long before the various reservoir capacitors in the
system can begin to discharge. However, if the mains switch
contacts are generating RF that is in turn reproduced as a click by
the preamp, then this system may not be fast enough to mute it.

3 Drop out the relay as fast as is possible when a DC offset of
more than 1 - 2 V, in either direction, is detected at the output
of either power amp channel; the exact threshold is not critical.
This is normally done by low-pass filtering the output (47K and 47
uF works OK) and applying it to some sort of absolute-value circuit
to detect offsets in either direction. The resulting signal is then
OR-ed in some way with the muting signal mentioned above.

4 Don't forget that the contacts of a relay have a much lower
current rating for breaking DC rather than AC. This is an issue
that doesn't seem to have attracted the attention it deserves.

   The relay control system in the Precision Preamp 96 (EW July/Aug
1996) meets all these requirements, but obviously much larger and
more capable relays will be required. These will need more
operating current and this must be considered (uprate Q206) if you
adapt this circuit.

   Please note that the HT rail fuses are intended only to minimise
amplifier damage in the event of output device failure. They must
not be relied upon for speaker protection against DC offset faults.

   Fuses in series with the output line are sometimes recommended
for DC offset protection. It appears to be true that they have a
better chance of saving expensive loudspeakers than the HT fuses,
but there are at least three snags:

1 Selection of the correct fuse size is not at all easy. If the
fuse rating is small and fast enough to provide some real
protection, then it is likely to be liable to nuisance blowing on
large bass transients. A good visual warning is given by behaviour
of the fuse wire; if this can be seen sagging on transients, then
it is going to fail sooner rather than later. At least one writer
on Class-A amplifiers gave up on the problem, and coolly left the
near-impossible task of fuse selection to the constructor!

2 It has been widely reported that fuses running within sight of
their rated current generate distortion at LF due to changes in
resistance caused by RMS heating; this is presumably third

3 Fuses have significant resistance (otherwise they wouldn't blow)
so putting one in series with the output will degrade the
theoretical damping factor. This must be the most mis-named
parameter in audio, and the real effect on any normal loudspeaker
will be completely negligible.

   There is a great deal more that could be said about amplifier
protection, and I hope to deal with some of the lesser-known issues
in a future article.

   The amplifier design as published presents a set of output stage
options rather than a fixed design, and may be configured for the
power desired by fitting an appropriate number of output devices
and by appropriate choice of supply rail voltage and current

   This design has excellent supply-rail rejection, and so a simple
unregulated supply is perfectly adequate. The use of regulated
supplies is definitely unnecessary, and I would recommend strongly
against their use. At best, you have doubled the amount of high-
power circuitry to be bought, built, and tested. At worst, you
could have intractable HF stability problems, peculiar slew-
limiting, and some expensive device failures. Just say no!

   When selecting the value of the mains fuse in the transformer
primary circuit, remember that toroidal transformers take a large
current surge at switch-on. The fuse will definitely need to be of
the slow-blow type.

   As before, we have attempted to configure the PCB to use easily
obtainable components, in particular the following, which are all
available from Maplin Electronics. Maplin order codes are given for

1) Driver heatsinks. The PCB has mounting holes suitable for
   heatsink Type-SW38-2                       (Order Code JW28F)
2) Fuseholder clips. 20mm Fuse Clip Type 1,   (Order Code WH49D)
3) Quiescent-adjust preset. Cermet preset 1K, (Order Code WR40T)
4) Wirewound resistors. 3W "WW Min"           (Order Code W+value)
5) Non-electrolytic capacitors; Polyester.    (eg Order Code WW41U)
6) Output inductor; 18 swg enamelled copper wire.(Order code BL25C)

   The mechanical design adopted depends very much on personal
taste and resources, but I will offer a few points that need to be
taken into account:

1 An amplifier giving significant power into low impedances
requires effective heatsinking with a free convective air flow, and
this points toward putting the sinks on the side of the amplifier;
the front will carry at least the mains switch and power indicator
light, while the back carries the in/out and mains connectors, so
only the sides are completely free.
   The internal space in the enclosure will require some
ventilation to prevent heat buildup; slots or small holes are
desirable to keep foreign bodies out. Avoid openings on the top
surface as these will allow the entry of spilled liquids, and
increase dust entry. BS415 is a good starting point for this sort
of safety consideration, and this specifies that slots should be
not more than 3mm wide.

2 A toroidal transformer is strongly recommended because of its
low external field. It must be mounted so that it can be rotated to
minimise the effect of what stray fields it does emit. Most
suitable toroids have single-strand secondary lead-outs, which are
too stiff to allow rotation; these can be cut short and connected
to suitably-large flexible wire such as 32/02, with carefully
sleeved and insulated joints. One of our prototypes had an ILP
toroid mounted immediately adjacent to the TO3 end of the amplifier
PCB; however complete cancellation of magnetic hum (output level
below -90 dBu) was possible on rotation of the transformer.
   A more difficult problem is magnetic radiation caused by the
reservoir charging pulses (as opposed to the ordinary magnetisation
of the core, which would be essentially the same if the load
current was sinusoidal) which can be picked up by either the output
connections or cabling to the power transistors if these are
mounted off-board. For this reason the transformer should be kept
physically as far away as possible from even the high-current
section of the amplifier PCB.
   As usual with these transformers, make certain that the bolt
through the middle cannot form a shorted turn by contacting the
chassis in two places.

   The distortion performance of any Class-B power amplifier
depends almost as much on the topology and layout of the power and
ground wiring as on the subtleties of the circuit design. This has
been taken into account in the PCB layout, but the external wiring
has to be the responsibility of the constructor. We therefore give
a recommended wiring scheme that has been approved by the designer.
(The assumption is made that a simple unregulated supply is used;
as noted above, a regulated supply is quite unnecessary and may
cause unforeseen complications)

1 There are several important points about the wiring for any
power amplifier; see the attached wiring diagram:
 a: Keep the + and - supply wires to the amplifiers close together.
This minimises the generation of distorted magnetic fields which
may otherwise couple into the signal wiring and degrade linearity.
Sometimes it seems more effective to include the 0V line in this
cable run; if so it should be tightly braided to keep the wires in
close proximity.
   For the same reason, if the power transistors are mounted off
the PCB, the cabling to each device should be configured to
minimise loop formation. 

 b: The rectifier connections should go direct to the reservoir
capacitor terminals, and then away again to the amplifiers. Common
impedance in these connections superimposes charging pulses on the
rail ripple waveform, which may degrade amplifier PSRR.
 c: Do not use the connection between the two reservoir capacitors
as any form of star point. It carries heavy capacitor-charging
pulses that generate a significant voltage drop even if thick wire
is used. As the drawing shows, the "star-point" is tee-ed off from
this connection. This is a star-point only insofar as the amplifier
ground connections split off from here, so do not connect the input
grounds to it, as distortion performance will suffer.

2 TO3P power transistor installation. Two or four devices may be
fitted, depending if the doubling strategy is used. The large flat
plastic devices are intended to be mounted on a metal surface to
which the PCB is bolted through FIX3,4,5. The devices fit into the
two cutouts on the long side of the PCB, with the leads soldered to
the large rectangular pads on the top (component ident) side of the
PCB. Spring mounting clips to hold down the TO3P packages are
recommended rather than bolts as these give better thermal contact
and put less stress on the package. Note that insulating thermal
washers are essential.

3 TO3 power transistor installation.
The PCB layout assumes that TO3 devices will be mounted off-PCB
with wires taken from the rectangular TO3P pads on the PCB to the
remote devices. These wires should be fastened together (two
bunches of three is fine) to prevent loop formation; see above. I
cannot give a maximum safe length for such cabling; certainly 8
inches causes no stability problems. The emitter and collector
wires should be substantial, such as 32/02, but the base
connections can be as thin as 7/02 without problems.
   Any number of TO3 devices may be connected in parallel, though
the EW article only deals with device doubling. Tripling is
expected to give an even closer approach to complete Load-

   For TO3 power transistor mounting it is recommended that the
heatsink is drilled with suitable holes to allow bolts to pass
through the TO3 fixing holes, through the heatsink, and then be
secured by nuts and crinkly washers which will ensure good contact
with the PCB mounting pads. Insulating sleeves are essential around
these bolts where they pass through the flange; nylon is a good
material for these as it has a good high-temperature capability.
Depending on the size of the holes drilled for the two TO3 package
pins, (and this should be as small as practicable to maximise the
area for heat transfer) these are also likely to require
insulation; silicone rubber sleeving carefully cut to length is
very suitable.
   An insulating thermal washer must be used between TO3 and
flange; these tend to be delicate and the bolts must not be over-
tightened. If you have a torque-wrench, then 10 Newton/metre is an
appropriate upper limit for M3.5 fixing bolts. Do not solder the
two transistor pins to the PCB until the TO3 is firmly and
correctly mounted, fully bolted down, and checked for electrical
isolation from the heatsink. Soldering these pins and then
tightening the fixing bolts is likely to force the pads from the
PCB. If this should happen then it is quite in order to repair the
relevant track or pad with a small length of stranded wire to the
pin; 7/02 size is suitable for a very short run.

4 Driver transistor installation. These should be mounted onto
their separate heatsinks with silicone thermal washers, to ensure
good thermal contact. Use the spring clips intended to hold the
package firmly against the sink. (See also Section 4 below)
Electrical isolation between device and heatsink is not essential,
as the PCB makes no connection to the heatsink fixing pads, but you
will get it anyway unless the washer is damaged.

5 Bias-generator transistor Q8 mounting. A previous design (The
Class-B amplifier; PCB-001) used a double emitter-follower or EF
output stage; for this the optimal place to mount the sensor for
effective thermal compensation was the top of the TO3 cans, to get
as close as possible to the output device junction temperature. 
   The Load-Invariant design uses a CFP output stage rather than
EF, to increase both output efficiency and linearity. The output
device junction temperature is now almost irrelevant, being servo-
ed out by the local CFP feedback loop, and in Class-B mode the
temperature-sensor Q8 must now aspire to reach the temperature of
the drivers instead, which is mechanically much simpler. A position
for mounting Q8 on the same side of the heatsink (HS1) as driver
Q14 is therefore provided. A second thermal washer must be inserted
between Q8 and Q14 for good heat transfer. The same spring clip
(CLIP1) is used to secure both devices.
   This method of mounting The standard spring clip has enough give
(just) to allow the extra transistor and extra thermal washer to be
slipped between it and the driver package. The implications of this
improvement will hopefully be further explored in a future article.

   NOTE: Make sure Q8 is properly in contact with Q14. Without
thermal compensation the quiescent stability in Class-B will be
degraded, though almost certainly not to the point where thermal
runaway is possible.

   The amplifier design presented here is inherently safe in that
all the DC voltages are too low to present any kind of electric-
shock hazard. However, there are a few points I think the
constructor should consider.

1 The supply rails are low-voltage, but the reservoir capacitors
store a significant amount of energy. If they were to be shorted
out by a metal finger-ring then a nasty burn is likely. Metallic
bodily adornment should be removed before diving into amplifiers.

2 Any amplifier using the mains as a power supply is potentially
lethal. The risks involved in working on a powered-up chassis must
be considered. The metal chassis MUST be securely earthed to
prevent it becoming live if a mains connection falls off, but this
has the snag that if one hand touches live, there is a good chance
that the other is touching chassis ground, so your well-insulated
training shoes will not save you. All mains connections (neutral as
well as live) must be properly insulated so they cannot be
accidentally touched by finger or screwdriver. My own preference is
for double insulation; for example, the mains inlet connector not
has its terminals sleeved, and there is also a plastic boot fitted
over the rear of the connector, secured with a ty-wrap.
   This is a more severe requirement than BS415 which only requires
that mains should be inaccessible until you remove the cover.

5 Readers of hifi magazines are frequently advised to leave
amplifiers permanently powered for optimal performance. Unless your
equipment is afflicted with truly doubtful control over its own
internal workings, this is quite unnecessary. While there should be
no safety hazard in leaving a soundly-constructed power amplifier
powered permanently, I see no point and some risk in leaving
unattended equipment powered.

1 By far the most important step to successful operation is a
careful visual inspection before switch-on. As in all power
amplifier designs, a wrongly-installed component may easily cause
the immediate failure of several others, making fault-finding
difficult, and the whole experience generally less than
satisfactory. It is therefore most advisable to meticulously check:
  That the supply and ground wiring is correct.
  That all transistors are installed in the correct positions.
  That the drivers and output devices are not shorted to their
respective heatsinks through faulty insulating washers.
  That the circuitry around the bias generator Q8 in particular is
correctly built. An error here that leaves Q8 turned off will cause
large currents to flow through the output devices and may damage
them before the rail fuses can act.

2 The second stage is to obtain a good sinewave output with no
load connected. A fault may cause the output to sit hard up against
either rail; this should not in itself cause any damage to
components. Since a power amp consists of one big feedback loop,
localising a problem can be difficult. The best approach is to take
a copy of the circuit diagram and mark on it the DC voltage present
at every major point. It should then be straightforward to find the
place where two voltages fail to agree; eg a transistor installed
backwards usually turns fully on, so the feedback loop will try to
correct the output voltage by removing all drive from the base. The
clash between "full-on" and "no base-drive" signals the error.
   When checking voltages in circuit, bear in mind that C10 is
protected against reverse voltage in both directions by diodes
which will conduct if the amplifier saturates in either direction.
   This DC-based approach can fail if the amplifier is subject to
high-frequency oscillation, as this tends to cause apparently
anomalous DC voltages. In this situation the use of an oscilloscope
is really essential. An expensive oscilloscope is not necessary; a
digital scope is at a disadvantage here, because HF oscillation is
likely to be aliased into nonsense and be hard to interpret.

3 The third step is to obtain a good sinewave into a suitable
high-wattage load resistor. It is possible for faults to become
evident under load that are not shown up in Step 2 above.
   Setting the quiescent current for any Class-B amplifier can only
be done accurately by using a distortion analyser. If you do not
have access to one, the best compromise is to set the quiescent
voltage-drop across both emitter resistors (R36,37) to 10mV when
the amplifier is at working temperature; disconnect the output load
to prevent DC offsets causing misleading current flow. This should
be close to the correct value, and the inherent distortion of this
design is so low that minor deviations are not likely to be very

4 It may simplify faultfinding if D6,78 are not installed until
the basic amplifier is working correctly, as errors in the SOAR
protection cannot then confuse the issue. This demands some care in
testing, as there is no short-circuit protection. 

   Two mounting holes (FIX1,2) are provided on the input edge.
These accept standard plastic pillars. Three further fixing holes
(FIX 3,4,5) are provided for fixing the PCB to the heatsink; if the
power transistors are mounted off-board then the outer two of these
can be used for two more mounting pillars.

   Since the component selection, construction, and usage of this
PCB are entirely outside our control, we can accept no
responsibility for the functioning or performance of amplifiers
constructed with it. We are therefore unable to enter into
correspondence regarding faultfinding, substitute components, etc.
   We can accept no liability for loss, damage or injury incurred
by the construction or operation of this design.

to The Institute