This is only a preview of the September 2025 issue of Practical Electronics. You can view 0 of the 80 pages in the full issue. Articles in this series:
Items relevant to "Compact Hi-Fi Headphone Amplifier, part one":
Articles in this series:
Articles in this series:
Articles in this series:
Articles in this series:
Articles in this series:
|
AUDIO
OUT
AUDIO OUT
L
R
By Jake Rothman
Discrete discoveries and a cascode amplifier PCB design
Baxandall balanced input
Since my cascode amplifier circuit
can be configured as inverting, it seemed
logical to try adding an input balancing
transformer using the Baxandall
configuration described last month. This
led to some interesting effects.
It has to be admitted this is a risky
approach, since we are expecting the
limited negative feedback to linearise
three things at once: the output
stage, output capacitor and input
transformer.
In analog design, it’s desirable to
linearise one thing at a time with separate
loops, hence the profusion of op amps
used in modern designs. Trying to make
negative feedback (NFB) do too much
can lead to complex interactions and
odd frequency effects.
In the days of valves and expensive
transistors, though, this approach was
normal, with NFB being taken around
multiple capacitors and transformers.
To get this to work without the circuit
bursting into low-frequency oscillation
(‘motorboating’) involves a lot of
component value tweaking.
This is the sort of design challenge
I enjoy, because you get more out of
less. It’s more creative than stringing
together standard circuits and is a
privilege only afforded to the home
constructor or sole trader today.
Bumps in the frequency domain
As expected, LF oscillation occurred
at around 8Hz when adding the modem
transformer to the input (as shown in
Fig.4(a) in the August 2025 issue). This
was soon tamed by changing the lowfrequency (LF) step network resistor,
70
R3, to 11kΩ. All that remained was a
tolerable peak of 1.5dB at 13Hz.
This sort of tweaking is easy to do
with a signal generator and oscilloscope;
‘posh’ analysers such as the Audio
Precision don’t go down low enough.
Distressing distortion?
Of course, the AP is very useful
for checking the distortion, which
is where some intractable problems
were revealed. The Baxandall balanced
input depends on a low-impedance
virtual ground on the inverting input
to work.
Unfortunately, the input DC blocking
capacitor (C1) and the LF step network
raise the impedance at low frequencies,
which causes the distortion to increase
rapidly below 200Hz, as shown in Fig.1.
It was still much better than connecting
the transformer in the normal way,
though.
Subjectively, the effect was enhancing
rather than destructive, especially on
electro jazz pop, such as Level 42 (“you
old codger”, I bet you’re thinking!).
Switch off or carry on?
The utility of a fully floating (Earthisolated) transformer input cannot be
understated. I found the switch-mode
power supply buzz and monitor noise
from my desktop computer to be
completely eliminated when being used
as a sound source.
With a stereo amp using a single power
supply, the left and right amplifier signal
Earths have to be joined together at the
input jack. This causes an Earth loop,
which invites hum. Putting ground lift
resistors in can fix this, but distortion
then increases. Using transformers
breaks the loop.
The amp also excelled when used as
mid-range and tweeter amps in an active
speaker system with a common power
supply for all three amplifiers. Floating
differential inputs greatly simplify Earth
loop problems. The LF distortion was a
long way from the required bandwidth,
so it had no effect.
Building bridges
Until the boards arrive, I can’t test the
transformer bridge-mode version of the
amp, so it’s time for another diversion
and back to Blumlein. Normally, when
making a bridging driver, a circuit that
gives two anti-phase outputs is used,
such as two op amps, wired as an
inverter and buffer. Very boring; I’ve
already deployed 100 NE5532s in the
last year.
0.5
Total Harmonic Distortion (%)
I
'm waiting for a delivery from
JLCPCB near Hong Kong. When the
PCBs (designed by Mike Grindle)
arrive, we’ll have an efficient way
of building the cascode-input power
amplifier and its unusual power
supply, described in the July issue.
Meanwhile, I can report results of
some further investigations into the
design.
0.2
0.1
0.05
0.02
0.01
.005
.002
.001
.0005
.0002
.0001
20
50
100
200
500
1k
Frequency (Hz)
2k
5k
10k
20k
Fig.1: the distortion curve with a balancing transformer added to the cascode
power amp. Note the rise in LF distortion. The output is 15V peak-to-peak into 8Ω.
Practical Electronics | September | 2025
R9
3.9kΩ
+40V
+10V
+10V
+30V
+30V
C2
47pF
C3
4.7µF
35V
Output 1
inverting
Output 2
non-inverting
C4
4.7µF
35V
TR1/2
BC549C
+18.5V
+18.5V
0.7V
+
R7
120kΩ
+
C1
330nF
Input
R3
10kΩ
+50V
2.5mA
+
R1
120kΩ
R5
10kΩ
0.7V
R2
100kΩ
R4
1.5kΩ
+1.5V
+1.5V
R6
1.5kΩ
C6
100µF
50V
1mA
+16.5V
1mA
R8
100kΩ
C5
100nF
2mA
2mA
CRD
0V
Fig.2: the LTP phase-splitter seems to have been
forgotten. It can have single-ended or balanced inputs.
It’s an ideal application for Blumlein’s
major legacy, the long-tailed pair.
A circuit that gives two anti-phase
outputs from a single input used to be
commonly known as a phase splitter.
These were an important stage in valve
amplifiers, where they drove the two
output valves in push-pull. The classic
circuit still used today is the Mullard
510, which used a long-tailed pair
built around a double-triode ECC83
valve.
I thought it would be useful to design
a transistor version to add to my discrete
circuit armoury. Older amplifiers used a
single valve or transistor to accomplish
phase splitting, called a concertina phase
splitter.
This had the disadvantage of the load
on one output affecting the other output.
The LTP phase splitter doesn’t suffer
from this defect.
LTP phase splitter
The circuit is shown in Fig.2. The
first version I built set the tail current
Photo 1: the outputs of a phase splitter are the same amplitude.
with an 8.2kΩ resistor, textbook style,
which was a bad idea; not all the
signal was transferred to the second
transistor, some was lost in the resistor.
This resulted in the second output
being 20% lower than the first. The
whole point of a phase-splitter is equal
balance.
The horrid sound and inefficiency
of one amplifier in a bridge clipping,
while the other amplifier is clean,
defeats the whole point of the system.
If the tail resistor is replaced by a 2mA
current regulating diode (CRD), as shown
in Photo 1, the imbalance is fixed.
Alternatively, a constant current sink
built around a JFET or bipolar transistor
could be used.
The op amp phase-splitter has the
advantage of being able to drive low
impedances, down to 1kΩ or so, with
insignificant distortion. This phasesplitter has a high output impedance,
defined by the collector resistors R3
and R5. Replacing these resistors with
a current mirror would double the gain
Total Harmonic Distortion (%)
0.5
0.2
0.1
0.05
0.02
0.01
.005
.002
.001
.0005
.0002
.0001
20
50
100
200
500
1k
Frequency (Hz)
2k
5k
10k
20k
Fig.3: the LTP phase splitter distortion with a 20V peak-to-peak output (unloaded).
It’s quite high, but is essentially innocuous third harmonic at a consistent level.
Practical Electronics | September | 2025
and output current capability for the
same supply current.
I measured the unloaded distortion
with an 8V peak-to-peak signal at
0.03%, as shown in Fig.3. It was
mainly third-harmonic, arising from
the start of the symmetrical soft clip
curvature.
Such distortion is very innocuous
and similar to that produced by
speaker excursion limiting and tape
saturation.
Interestingly, the distortion did not
increase with 600Ω loading, although
the output voltage dropped.
One thing I’ve noticed with
differential operation is that the
distortion of tantalum coupling
capacitors cancels out because it is
equivalent to operating two capacitors
wired back-to-back. The distortion
is further minimised in single-rail
designs such as this, because the
capacitors are polarised. So there is
no characteristic bass tip-up in the
distortion curve.
The gain of the circuit for each
output is 3.5 times, defined by the
emitter resistors along with the collector
loading. With differential outputs, there
is a further gain doubling because one
output goes up as the other goes down.
The frequency response measures -1dB
at 20Hz and 20kHz.
The output rolls off around 300kHz,
with output 2 dropping first because
of the extra losses passing through
an extra transistor. The capacitance
of the constant current sink also has
an effect.
Maybe the long-forgotten Baxandall
bootstrapped complementary constant
current circuit would work well
here, but that’s another investigation.
Alternatively, I could design a cascode
phase-splitter that needs no current
sink at all. A practical, breadboarded
71
version of the phase-splitter from Fig.2
is shown in Photo 2.
Bridging amplifier impedances
Photo 2: a
breadboard version
of the LTP phase-splitter
circuit shown in Fig.2.
c be
R3
C1
0V
C2
Input
R1
C7
R8
C6
Speaker +
Speaker Gnd
R12
R13
R11
C9
C3
R17
C11
+
C10
TR4
R14
LED2 ZD1
+ +
TR5
TR7
b
c
e
TR6
R19
C13
c be
R15
R25
Dirty Gnd
IC1
R10
LED1
+
VR2
R21
R22
Clean Gnd
R9
R6
C8
L1
C2
+
C4
R23
R24
Speaker –
C5
TR2
VR1
TR3
R20
R7
c be
b c e
R2
R5
TR1
R4
R16
R18
+
TR9
TR8
Metal tab
e c b
e c b
54V Unreg 50V Reg
From transformer 36V AC
AC
CLTP - POWER SUPPLY
+50V Reg1
TR2
+
C6
e
c
+
R
6
Metal tab
e
b
c
TR1
C5
+54V Unreg1
AC
b
C4
+50V Reg
Mains E
D4
C7
R
2
+
C2
F1
R1
D1 D2
R R
5 4
R
3
D5
+54V Unreg
D3
D9
D8
+
C3
72
Speaker Gnd1
Speaker Gnd2
Dirty Gnd1
Dirty Gnd2
Clean Gnd1
Clean Gnd2
Fig.5: the power supply board
overlay. Note that sufficient pins are
provided for stereo use.
+
C1
D7
D6
COMP-LT PAIR POWER AMP
Fig.4: the
cascode
amplifier
PCB
overlay.
The voltage loss on the discrete
LTP phase splitter outputs could be
reduced by making the power amplifiers
non-inverting, giving a higher input
impedance, thereby reducing the
loading. This would be quite tricky with
the PCB given here.
One of the great advantages of bridging
is that the output voltage is double that
of a single amplifier. For a given speaker
impedance, say 8Ω, the output power
is theoretically quadrupled. This is
because each amplifier has to deliver
twice the current as well as there being
twice as many outputs.
The problem is that the speaker
impedance ‘seen’ by the system is
halved, ie, our 8Ω speaker looks like a
4Ω one (that’s why the current doubles).
This doubling of current can stress the
amplifiers, increasing the distortion and
dissipation. One way around this is to
reduce the transformer voltage a bit and
double the VAS stage (TR3) current.
In my case, this is great because I
have loads of 15-0-15V 60VA toroidal
transformers in stock. These should give
around 50W RMS into 8Ω bridged for
short periods with no stress. Another
single amplifier could also be run for a
tweeter amplifier, forming the basis of an
excellent active speaker system.
When using two amplifiers in bridge
mode, it is important to use a single
joint power supply; strange distortioninducing Earth loops can occur if
separate power supplies are used with
a common Earth.
An interesting effect of using ClassAB amplifiers in bridge mode is that
the current pulses drawn from the
supply are full-wave rectified versions
of the signal rather than half-wave. In
theory, this means that the smoothing
capacitors have an easier time because
there is an effective doubling of the
frequency of the signal-derived ripple
on the supply rail.
There is an advantage in dual-rail
bridge systems in that no speaker
current goes into the Earth; it’s entirely
confined to the power rails. In this case,
where the amplifiers are single-rail, the
‘negative’ rail and speaker Earth are one
and the same. I suspect there is little
advantage in practice, because supply
wiring impedances are going to be more
significant.
Ding dong; the boards have arrived,
so I can get the soldering iron out at
last. Still, the LTP phase-splitter is a
useful circuit block. Did you know
it can form the basis of an all-pass
filter or phase-shifter? Now there’s an
interesting idea.
Practical Electronics | September | 2025
Photo 4: the power
supply board (C6 has been
increased to 1000μF). The odd shape
may provide space for a transformer.
Building the cascode power amp
The PCB overlays for the amplifier and
power supply are shown in Figs.4 & 5.
When initiating component insertion,
remember that putting electrolytic
capacitors in the wrong way is more
catastrophic in single-rail circuitry
because of the full polarisation voltage.
Reversal of the output electrolytic, C10,
will cause a big explosion and take your
loudspeaker and output transistors
with it! Positive is denoted by the
square pad on the PCB. The same goes
for the LEDs.
The components are spaced out on
the boards for educational use and
experimentation, as shown in Photos
3 & 4.
It is a good idea to space some resistors
a few millimetres above the board
Photo 3: the completed
cascode circuit board
with small heatsinks.
Practical Electronics | September | 2025
Photo 5: it’s a good idea to space the
resistors a few millimetres above the
board, even if they normally run cool.
(Photo 5) in case they burn up during
fault conditions, causing damage such
as holes in the PCB and possible fires.
The Zobel resistor (R25), constantcurrent source resistor (R14), VAS
emitter resistor (R24), driver transistor
resistors (R16 & R17), output transistor
resistors (R18 & R19) should be raised in
all power amps for this reason.
It’s quite reliable to do this on boards
with plated-through holes (eg, doublesided PCBs). However on single-sided
PCBs, it can result in broken pads and
tracks if the components get pushed
down, unless the leads are supported
with special crimps or ceramic beads.
The sealed Bourns-pattern TO5 trimmer
resistors have very small holes; you can’t
ram normal skeleton presets in. These
tend to crack up and oxidise, anyway.
A transistor ‘sandwich’
This is not lunch but a thermal
construction where the driver transistors
are bolted either side of the thermal sense
(VBE) transistor, TR5, as shown in Photo
6. Instead of butter, thermal paste is used
to ensure good coupling. They are held
together using a 12mm-long M3 machine
screw with a lock nut. Note that the metal
tabs all face the same way; left, towards
the nut/output capacitor.
No insulating washers are needed
because the inside of the device holes
are insulated.
I had changed the original Zetex
transistors used for the more common
BD139/40 devices for the drivers, TR6 &
TR7. TR5 can be any NPN TO-126 device.
I used a BD135 because it was at hand.
I thought the original copper foil
Photo 6: the ‘transistor sandwich’. I
thought it was the best thing in thermal
compensation, but there were side
effects.
73
amp are made, it’s essential to check for
oscillations with a scope on the output.
Because there is less high-frequency
open-loop gain due to the increased
compensation, the HF distortion is
worsened slightly from 0.02% to 0.03%,
as shown in Fig.6. This could be fixed by
using second-order compensation, but
that’s for another investigation. I didn't
notice any subjective differences.
I’ve always agreed with Baxandall
and Peter Walker (of Quad) that all
competent amplifiers sound very similar
up to overload.
Simplifications
Photo 7: soldering the original ZTX
transistors to the PCB. This isn’t ideal,
so I am designing a different board.
mounting arrangement was too tricky. If
you want to use the original transistors,
they can be fitted as shown in Photo
7, but because they are centre-base
rather than centre-collector, a bit of ‘leg
crossing’ is called for.
Component substitutions
Transistor substitutions can be tricky
with power amplifiers, and this proved
to be no exception. Two component
values had to be changed. Firstly, C8,
the compensation capacitor, had to be
increased to 56pF because the BD140
(TR7) is slower than the ZTX751, causing
high-frequency oscillation.
Also, the BD135 has a slightly lower
(0.69V) switch-on voltage than the
ZTX300 (0.78V) used for TR5. This
meant that R12 had to be changed to
3.9kΩ. Interestingly, the ‘transistor
sandwich’ actually over-compensates,
the thermal coupling being too good.
This could be beneficial if monolithic
Darlington transistors were used for the
output stage.
Whenever substitutions on any power
Output inductor L1 is only needed
to prevent oscillation with capacitance
from long twin-flex speaker leads. If
the speaker leads are less than, say,
250mm (10 inches), such as in active
speaker cabinets, it can be omitted and
its position linked out on the PCB, along
with resistor R21.
Biasing
If the board is to be used with a
regulated driver supply, the bias chain
can be replaced by resistors and regulator
IC1 omitted. In this case, the zener diode
supply still has a bit of hum. Remember
that the bias input is a signal pin as well,
so any noise on here is amplified.
Heatsinking
The output transistors are designed to
be mounted onto a heatsink or a thick
aluminium chassis. Insulating washers
and bushes must be used since the metal
tabs are at +23V. A thermal resistance no
higher than about 3.4°C/W is needed.
Power supply
The main reservoir capacitor C1 should
be 63V, not 50V as originally specified.
With some mains transformers, the rail
voltage can rise to 55V off load. The
capacitor is unlikely to short out because
Total Harmonic Distortion (%)
0.5
0.2
0.1
0.05
0.02
0.01
.005
.002
.001
.0005
.0002
.0001
20
50
100
200
500
1k
Frequency (Hz)
2k
5k
10k
20k
the electrode foil oxide is formed to the
surge voltage rating (20% higher), but it’s
best to design conservatively.
Wiring
The output transistors can be wired
off-board with quite long leads, up
to 200mm, before instability occurs.
Note there are three Earth wires from
the board. To keep impedances low,
16/0.2mm stranded wire is used for the
high-current connections. The regulated
supply and transistor base connections
can be thinner, such as 7/0.1mm. The
completed amplifier test jig is shown
in Photo 8.
Testing
The first thing to do is put the
trimmers in the correct position. The
bias, VR1, should be set mid-way
and the quiescent current, VR2, fully
anticlockwise to maximum resistance.
No speaker or load is ever connected
with initial power amp testing.
If you have a proper bench power
supply that goes up to 50V (most only
go up to 30V) with current limiting, one
can take a cautious approach to testing
with no blow-ups. If not, one can put
100Ω ¼W resistors in the power rails
for protection. The smoke will soon tell
you if there’s a fault!
Remember that the output transistors
can be un-powered by not connecting
the 54V rail. This will enable low-level
tests to be performed safely.
DC tests
Once the DC conditions are fine,
with the output at somewhere around
half-rail, the bias can be set to 20V with
VR1. This can be checked across C5; the
output emitter resistors (R18 and R19)
are a good test point for this. If you want
to check the VAS current source, there
should be just over 1V across R14.
The rail currents can next be checked
with an ammeter connected in series with
the power supply. The regulated 50V rail
should draw around 7mA. The quiescent
current can be set next, with an ammeter
in series with the +54V rail. This should
be zero until VR2 is advanced. There
will be the usual point where it suddenly
jumps up, so turn it slowly.
Set it to around 100mA, and as the
‘transistor sandwich’ warms up, it will
drop to the required value of 15-30mA.
It will take around 10 minutes to settle
down. This could be sped up by adding
heatsinking, but that’s for the next
board version. If you have a sensitive
voltmeter, the voltage drop across R18
and R19 can be measured; it will only
be a couple of millivolts.
Fig.6: the distortion is slightly higher using BD139/40 driver transistors compared to
the original ZTX751/ZTX651. The output is 12.7V peak-to-peak, 2.5W RMS into 8Ω.
Signal testing
74
Practical Electronics | September | 2025
Now it’s time to get the signal
generator out and give the amp a bit of
a thrash. With no load, check it clips
symmetrically and cleanly on a 1kHz
sinewave. If it’s okay, try it at say 20Hz
and 10kHz. Next, try square waves,
looking for excessive ringing.
Load testing
The next test is to attach an 8Ω 30W
load resistor. The output voltage will drop
a couple of volts, but it should be possible
to obtain over 40V peak-to-peak. VR1 can
finally be trimmed to obtain symmetrical
clipping. Only leave it powered for a
minute or so at a time, or the resistor and
heatsinks will get very hot.
I got a slight increase in power to
28W/42.5V peak-to-peak compared to
the Veroboard version because I used
schottky diodes in the bridge rectifier
rather than bog-standard ones.
Bridge testing
Now that we have two working boards,
it will be interesting to explore bridgemode operation and the distortion effects
of things like wiring layout. Initial tests
were not promising, with distortion at
HF more than doubling, confirming my
experience that bridge-mode amplifiers
generally sound inferior.
It is possible to omit the output
capacitors in a bridge amplifier, but do
this at your peril, since both outputs are
at +23V. A single short to ground and
bang go the output transistors. I’m sure
with further R&D, improvements will be
made. Then we’ll build them into a nice
case, but I’m out of time for now.
PE
Photo 8: the test jig at the prototype PCB stage – less precarious than the old Veroboard version. A pretty enclosure awaits.
“You always say that”, says Kate, as she sweeps piles of mouldy circuits from the dining room table.
1591 ABS flame-retardant enclosures
Learn more:
hammondmfg.com/1591
uksales<at>hammfg.com • 01256 812812
Practical Electronics | September | 2025
75
|