A 120 watt Solid State Amp

Introduction
Again an amplifier at a time when nice class-D amps (Hypex) are available? After the absolute success
of the SSA35 (it measures and sounds so good) I present its big brother. So I repeat:
Against all HiFi-religions I offer a class B amplifier (Douglas Self, others call it class AB) with an op
amp and overall feed back which performs excellent (< 0.01 % distortion) into a 4 Ω resistive load.
Already in a description of the SSA35 in 'Another 35 watt Solid State Amplifier' on this site, some hints
have been dropped about how to build an amp for more than 35 watt, but there are different solutions.
Sometimes more power (it is to say: a higher voltage source for reactive loads) is needed, eg. with
ESL's....

What I learned from others..

Recently I used a 120 watt clone of a Threshold amplifier to drive my ESL's (+ MFB). Together with
Pieter Meijer, I came to the conclusion that the SSA35 sounded better. The question is: why?
I decided to do an investigation.
Threshold is proud of their feed back: on the front of their amplifiers is stated that the power stages
have been left apart from the feed back loop! Stability is not ensured with an overall feed back within
this amplifier concept indeed. The three stage output stage is the problem. Cross over distortion is only
70 to 80 dB down if the amp is loaded with 4 Ω in spite of five power transistors parallel in CFP (com-
plementary feedback pair).
To avoid the problems of the input stage and the VAS, I do use an op amp. Look at my SSA35 for
explanations of this choice. Douglas Self tells in his Audio Power Amplifier Design Handbook that the
output stage should be controlled by a voltage source, so my idea to replace the input stage and the
VAS with an op amp is not
that bad.
For high power there is at
least one problem: the op
amp should operate at vol-
tages larger than +/-18 volt.
To make 120 watt into 4 Ω,
a power supply of more than
+/- 35 volt is needed. To
stay away from hum and
distortion at low frequencies
with high power, I choose for
+/- 45 volt.
Bear in mind that a power
stage is 'a unity-gain output
stage. Here that stage is a
CFP (Complimentary
Feedback Pair) in stead of
an EF (Emitter-Follower) for
better termal stability and
less distortion because the
Vbe of the output devices is
inside the local NFB-loop.
Moreover it cannot reverse
bias the big output devices! Self discourages strongly to build an output stage with gain: “… so
making things worse by seeking voltage gain is not the way forward.... ….my advice would be that you
probably do not want to go this way.” (See page 151 of Self’s book: AUDIO POWER AMPLIFIER
DESIGN HANDBOOK, Fifth Edition).

Bootstrapping?
With the SSA35 (on this website) I suggest to bootstrap the op amp to allow a higher power supply
voltage. Here I would go a different way….
Accoutrement: More simple
Looking after op amps with high power supply voltages I stumbled across the LTC6090 which can
handle 10 mA with a supply voltage of ±70 V, and an OPA445 that can handle >15 mA with a supply
voltage of ±45 V. Is this a candidate for the pre amplifier so that no bootstrap is needed? It’s open loop
output impedance is 220 Ω, so….. Using an unknown op amp, one is handed down to the idiosyncras-
ies of it. On the other hand, I do have excellent results with power stages with 2SA1943, 2SC5200,
BD230 and BD231 which are not too complicated so that I will give it a try. Replacing the OPA134 with
an OPA445 is very attractive indeed, coming to about the same diagram as that of the SSA35 (see
above). Its open loop gain is 100 dB at 20 Hz and 40 dB at 20 kHz.
DC stability
Douglas Self predicts more DC stability with 0.1 Ω resistors in series with ‘the big transistors’ (R14/R15
below). Without emitter resistors with the BD’s we come to the diagram:

D1 and D4 together with the 1000 µF elco’s avoid the power line of the op amp and the quiescent
current from swinging during heavy load of the power stage. 1000 µF charged with 45 volt in a low
power environment however is inadmissible because of damage of small traces on the PCB in case of
a short circuit during experiments. It could be wise to add a small resistor (47 Ω) between the 1000 µF
elco and the rest of the circuit.

Load of op amp
The hfe of the BD230/231 is at least 25 and those of the 2SC5200/2SA1943: 35 so that the peak base
current of the BD’s will not exceed 10,000/25x35 = 11.5 mA at a maximum output current of 10 A.
The OPA445 can only supply 10 mAAC so R5 and R6 should be as high as possible. For biasing the
BD’s the 1N4148’s (D2 & D3) need some 5 mA. If a smaller diode is chosen, this current will be smaller
and consequently R5 and R6 will become larger (I hope…. ).
There is another reason to look for a different diode: the diode best should be fixed on top of the BD’s
for temperature stability (D. Self). With a glass diode this is less simple. The 1N4148’s are replaced by
transistors, an NPN (BC549C) and a PNP (BC559C). Both are in TO-92 version, so that they simply
could be glued on the BD’s with the flat side down.

Protection of C2
To keep the output offset voltage low, C2 has been planned: a 1000 µF / 6.3V capacitor so that the
gain is 1 at subsonic frequencies. During power on the voltage across C2 could rise to an unknown
value. Because the elco is of a low voltage type (small, cheap) it should be protected towards voltage
overload with two diodes anti-parallel. (C2 and the diodes will be shortcircuited later.)

Oscillations
MicroSim does not ‘detect’ oscillations but as a precaution the RC-combination across the output
terminals has been chosen to 4 Ω and 330 nF.
Biasing

Without R9, R10 Douglas Self states on page 183 of his book: Audio Power Amplifiers Design
Handbook, fifth edition, that the voltage across the two bases of the BD’s should be 1297 mV with R14
= R15 = 0.1 Ω. The Iq in the power transistors should be 15.3 mA.
To get an idea of the behaviour of TO-92 transistors (BC549 & BC559) in our biasing circuit:
For temperature stability R7 & R8 should be replaced by a parallel circuit of a resistor and an NTC-
resistor (thermistor) as in the diagram above inherited from the SSA35. After careful investigations with
MicroSim8, the addition of R9 & R10 (1 Ω) and R24 & R25 (15 Ω) are proven to be necessary for a
large R5 and R6, to keep the load of the op amp small!

Supply line voltage of the OPA445

The OPA445 has been designed for +/- 45 volt supply. In the last decades the mains voltage has been
risen from 220 to 230 volt AC in Europe and is planned to rise to 240 volt. To avoid a too high voltage
on the op amp a resistor between D1 and D4 and the corresponding 1000 µF elco’s could be of help.

Non-linear junction capacitance distortion

If a ladder attenuator (also on this website) is used at the inlet, the input circuit R1, R2 and C1 should
be changed to 10 Ω, 100 kΩ and 560 pF respectively, to match to the attenuator and to avoid this ‘nljc-
distor-tion’.
Current limiter
If shortcut protection should be served by a transistor, somewhere in the circuit a voltage of > 0.7 volt

should arise when the output current becomes too large. We make the statement: the highest output
current will be 7 ARMS or 10 Apeak for ≥120 watt into 4 Ω.
With the 0.1 Ω resistor in series with the collec-
tors of the power transistors (R14 & R15) and the
1 Ω resistors at the emitters of the BD’s (R9 &
R10) a simple current limiter circuit will satisfy.
The rates of R27/R30 and R28/R29 define the
current limit. For long-term shortcuts, the 6 A
fuses should serve.
R31 & D9 (BAT83) as R32 & D10 (BAT83) isolate
the bases of the BD’s from the non-linear para-
sitic Ccollector of the BC639 and BC640 respec-
tively. Mind that the Vce and the Vcb of these
transistors will meet nearly 75 volt (see to the
left) so that BC639/640 is a must! Erroneously
the types in the picture at the left are called
BC546 and BC556…... Mind that the numbering
of some components has been changed meanwhile…..

With R14, R15, R9 and R10 some feed back will be added which enhances the temperature stability
and enlarges the input impedance. MicroSim tells us that R5 and R6 should be enlarged to 15 kΩ (with
R24 = R25 = 15 Ω) which will relieve the op amp! We’ll see….

Number of power transistors

The β of the power transistors 2SA1943 and 2SC5200 remain constant until a collector current of at
least 3 A. For 120 watt into a 4 Ω load an output current of 5.5 A is required which is 7.8 Apeak so that
three transistors would satisfy. However with some more transistors in parallel the cross over distortion
decreases, which effect is not yet clarified. I take four transistors in each leg. Too many power transis-
tors will undoubtedly limit the bandwidth of the power amp so that much feedback could become
difficult.

The Realization
The amplifier will be built into the cabinet of my Threshold clone which has two nice 45 volt power
supplies with big elco’s.
All power transistors are assembled on aluminum angel sections which in turn are fixed on heat sinks.
Both channels have their own PCB with the power transistors positioned at one side.
The main PCB’s became 239 x 89 mm.
The main printed circuit board
Below the top view is shown. The most components on top are in white, the copper traces at the
bottom are dark green (the board is transparent here), the four copper traces at the top are light green
and the solder tags are gold-coloured.
Mind that initially two BC640’s and BC639’s had been planned for a current source to the bias diodes
to replace R5 and R6 enlarging the impedance at the output of the op amp. The diodes however have
been replaced by T11, T12, R24 and R25 so that these current mirrors are not needed any more.

Current limiting printed circuit board

The ‘protection circuit’ has been fixed on a separate small PCB
which is assembled on top of the main board.
Meanwhile the circuit has been changed: C8 (10 nF) is no longer
needed and replaced by resistors of 100 Ω to get 60 Ω for R15 and
R16. R11 and R12. R31 & R32 became 39 kΩ.
T17 & T18 (ahead T15 & T16) must be BC639 and BC640 because
of the high voltages they will meet during large audio periods.
The trace between the 100 Ω resistors and the connectors c and e
should be erased. The 100 Ω resistors (R18 & R19) are fixed directly
to the emitters of the BD’s.

Implementation hints
To the left a detail of
the main board is
shown. Initially two
current mirrors (BC640
and BC639) had been
planned which are re-
placed by R5 and R6
(R5 and R8 on the
board).
The light green traces
are on top. If one could
prefer a one sided
implementation, these
connection could be
made with wires.
The red encircled
connections ‘b’ – ‘f’
correspond with those
of the current limiting
board to mount it on
top. The connections
are made with 0.8 mm
mounting thread. ‘a’
and ‘g’ are connected
with thin cord.
Main board and heatsink

The op amp, the protection board and ‘the changes’ are not applied jet.

Detail of T11 (BC549)

and T12 (BC559) fixed
on top of T1 (BD230)
and T2 (BD231) res-
pectively.
Mind that the collectors
and bases of the BC’s
are soldered to the
bases of the BD’s as
short as possible to
help the temperature
transmission.
The 1Ω emitter resis-
tors (R9 & R10) and
those between T11 and
T12 are not fixed yet!

Implementation
Measurements and tuning
Hitherto all values of the components has been designed by simulation. It became clear that the
biasing of the final stage was much more complicated than with the SSA35!

Oscillations
MicroSim does not always detect oscillations. As soon as the amplifier had been loaded (with 4 Ω), it
oscillates in the peaks of an output voltage of a few milli volts already! 100 nF in series with 10 Ω
across the output connectors helps, but 330 nF with 4 Ω in series has been installed.
Output power
Without current limiter the unloaded output voltage is 80 Vtt. Loaded with 4 Ω, the output voltage is 64
Vtt because of the voltage drop in the power supply. This counts for all frequencies between 1 Hz and
16 kHz (the used LF-generator stops there). This results in >120 watt output. There are no oscillations,
not even when the amp is extremely overdriven.

The tuned diagram

The collector resistors of the BD’s should be 15 Ω indeed. There is no need for an NTC resistor to
keep the quiescent current in the power transistors in the order of 25 mA independent of their
temperature.

R5 and R6: MicroSim predicted 15 kΩ but in practice 8.5 kΩ (10k//56k) should be used. This means
that the op amp is loaded with ~ 4 kΩ which points out to be no problem.
With the diagram above the quiescent current in the BD’s and the power transistors has been
measured: Immediately after power on (‘cold’): Iq of the BD’s is 43 mA and Iq of the 2SC/2SA’s is 35
mA.
After 1 minute: Iq of the BD’s is 42 mA and Iq of the 2SC/2SA’s is 25 mA.
After 25 minutes: Iq of the BD’s is 41 mA and Iq of the 2SC/2SA’s is 24/25 mA.
After 1.5 hours: Iq of the BD’s is 41 mA and Iq of the 2SC/2SA’s is 24/25 mA.
After 5 minutes full power (70˚C): Iq of the BD’s is 34 mA and Iq of the 2SC/2SA’s is ~9 mA.
After 1.5 hours rest with power on: Iq of the BD’s is 41 mA and Iq of the 2SC/2SA’s is 26/28 mA.
After 7.5 hours rest with power on: Iq of the BD’s is 41 mA and Iq of the 2SC/2SA’s is 24/26 mA.

NB.: because of R9, R10, R24 and R25 the currents in the NPN and PNP-branch are equal within 3%.

The great surprise……

The amplifier is unity gain stable! If the feedback-resistor R4 (= 10kΩ) is replaced by 100 Ω, the
amplification is nearly 0 dB without tendency to oscillation, not even during heavy overdriving.
One of the greatest advantages would be the increase of the feedback at 20 kHz. It will become
95dB@20Hz and 35dB@20kHz.
After careful listening, this ‘enhancement’ did not give an audible advantage so it has not been
implemented.
(Forget about the article https://www.temporalcoherence.nl/cms/images/docs/FeedbackHvMnl.pdf in
Dutch. That is absolutely rubbish!!)
Measurements
Maximum output power into 4 Ω: 118 W
Maximum output power into 8 Ω: 80 W
Power bandwidth with source impedance of 10 kΩ: 0 - 50 kHz
Open output offset voltage: < 5 mV
Distortion at 1 kHz 110 watt into 4 Ω: -80 dB = 0.01 % *
Distortion at 10 kHz 60 watt into 4 Ω: -72 dB = 0.025 % *
Double tone test with 18 and 20 kHz 110 watt into 4 Ω: Left: -80 dB, Right: -77 dB
Unloaded gain: 28.5 dB
Loaded gain: 28.5 dB !
Danamic range: 97 dB
Channel separation: 80 dB
Frequency characteristic: -0.5 dB at 20 kHz
Channels equality: within 0.05 dB
Ladder attenuator 3 dB-steps: within 0.05 dB
Unloaded maximum output voltage: 80 Vtt
Loaded maximum output voltage: 61 Vtt
Internal resistance Ri: < 0.01 Ω
nd rd
* only 2 and 3 harmonics. All higher harmonics <0.001 %.

Distortion
The distortion in the left channel is some 10 dB lower than in the right channel, because the BD230 &
BD231 have been paired. The figures in ‘Measurements’ are those of the worst channel.
Obviously the cross over distortion must be very low because the higher harmonics had not been
noticeable on the analyzer, which means < 100 dB (= < 0.001 %).

Observances
With the SSA35 (also on this website) R14 and R15 are not implemented and the BD139 and BD140
(in this case) do have an emitter-resistor of 0.5 Ω. R5 and R6 (there, here R7 and R8) consist of 18
Ω//100 Ω NTC-resistors. In the SSA35 the quiescent currents still enhances a bit with temperature!
Douglas Self promised sufficient DC-feedback from R14 and R15 (0.1 Ω) but at first it took me days to
find the right combination of the ‘diodes’ (T11 & T12), R5/R6 and R7/R8……. The voltage across the
diode replacement BC559C and BC549C is about 605 mV with 1 mA as with the 1N4148. With a
BC546B and BC556B this voltage is about 635/645 mV! That makes a lot! Only with 30 Ω between the
biasing ‘diodes’ (R24 & R25), R5 and R6 could be taken large enough as to the load of the op amp!
MicroSim promised me 15 kΩ but in practice they became 8.5 kΩ (10k//56k)! Only when I did put a 1 Ω
resistor in the emitter duct of the BD’s, every simulation and every practice became handy. Moreover a
simple current limiter came in sight!
The collector resistors of the BD’s (R7 = R8 = 15 Ω) satisfy without any NTC resistor!
nd rd
The distortion consists mainly of 2 harmonics due to the unmatched NPN/PNP transistors. The 3
th
harmonics are always smaller often more than 10 dB. Sometimes a 4 could be noticed on the
software spectrum analyzer TrueRTA. All higher ones were unnoticeable (<100 dB).
The distortion of the unloaded amplifier was < 95 dB at any time at any frequency. I’m even not sure if
this figure is not due to the signal source: the CD-player and/or the disk. Anytime the distortion of the
loaded amp became >10 dB worse, except at 10 kHz in the right channel. Don’t ask me why…….
Mind that the current limiter presented here is not a real shortcut protection indeed! The heat sink is
too small to output continually 120 watt into 4 Ω for longer than about ten minutes. Moreover a fully
driven shorted amp would be overheated.

Ordered from EuroCircuits:

The two boards have been ordered from EuroCircuits in Belgium.
The ‘main board’ is: 238.76 x 88.9 mm and the limiter board became 38.74 x 38.74 mm

Listening tests
The final test is of course a number of listening tests. For these I invited two younger men to help my
old ears! On this website I gathered a number of recordings for this purpose (Muziek- en Geluidsfrag-
menten). Sorry for the listening hints in Dutch, but…. Listen to them. These are all .WAV- files! First
burn them on disk.
What me stroke first is the total absence of any background noise, his or hum, even when listening with
the ear on the speakers. Secondly: the ‘rest’ in the performance and the details in the stereo image.
Third: the basses seemed to be weaker, which is often a good sign.
The used speakers are those described in: ’2 BMRs in een Baffle’, the best speakers I have built up till
1-11-2018.
How it sounds? In one word: awesome. Never mind what music is reproduced. The precision of the
fantom sound stage, the depth of it, the dynamics and vitality, it is all there up to 100 dBSPL!
Used equipment
The source was the forceful updated Philips CD624 with digifi on FPGA as described in ‘Ombouw van
een CD624 met FPGA en PCM1792’ in Dutch.
The speakers are as described in: ‘2 BMRs in een Baffle’’ with the MFB-box.’

Accommodated in:

This cabinet has been ‘designed’ for a Threshold clone from the 1980s, which sounded worse...

A part of the left and right channel PCB’s. The op amp and the current limiter board over the BD’s are
not implemented jet.
The power supplies in the basement ….

March 8, 2014 updated: 26-11- 2014

,, : 23-10-2018
Conclusions
Thanks to the OPA445, this 120 watt amplifier could have been built with the same principle diagram
as the awesome sounding SSA35. Compared to the SSA35 the number of components ought to be
increased because of the current limiting.

(R2 has been enlarged to 100 kΩ for a better match to the input attenuator.)

November 2018.

During the weeks I listened to the amp I had the idea that the SSA35 sounded different from the
SSA120, so I took it with the other equipment and switched so now and then and indeed.......
Eventually the omition of C2 with the protection diodes gave an enhancement in sound. In my opinion,
the unstable voltage on them induces movements in the sound stage.
Without C2 the off set voltage at the speaker terminals could become too high (> 10 mV) but perhaps
sorting out the OPA445 will give a solution.
Still I, and Gertjan and Pieter, prefer the SSA35 over the SSA120! Could the OPA445 still be the bad
guy?

Replacing the OPA445 with an LTC6090 or OPA134?

At the beginning of this article I found two high voltage op amps: the OPA445 and the LTC6090 from
Linear Technology. This one could handle ±70 volt power supply! The output current however is only
10 mA which is no problem with an 8 Ω application.
If this does not give the desired improvement, the current limiter circuit could be disconnected with the
OPA445 as well as with the LTC6090.
If this does not help, the op amp should be replaced with an OPA134 (with temporarily ±18 volt power
supply) with and without current limiter.

‘Back’ to a bootstrapped op amp

With the SSA35 a bootstrap experiment has been carried out to enlarge the output power of the am-
plifier. A side effect was that the amp sounded a little bit better because the parasitic capacitors of the
input FETs are eliminated. (See also: ‘Condenser microphone with bootstrapped op amp’).

Simply replacing the OPA445 with an OPA314 will destroy it. This possibly could be fixed after lower-
ing the zener voltages to 18 volt. Then the maximum output voltage of the op amp becomes nearly
2.18 = 36 volt. So Veff-max = 0,707 x 36 = 25,5 V. With measurements before we saw that for 118 watt
into 4 Ω Vtt-max = 61 volt or Veff-max = 0,707 x 30,5 = 21,6 volt so it should be possible with R30/R31 = 1
say 500/500.
If R26/R4 = 1 too, the Miller-C’s of the input FETs become invisible (if the amp stays unity gain stable
with the bootstrap!). The total gain of the amp (R4 + R26)/R26 = 2 = 6 dB. It should be ~28.5 dB for the
compensation of the 6 dB input attenuator. So the implemented R4/R26 = 10k/1k will satisfy. BUT: the
bootstrap will overcompensate the ‘Miller-feedback’ with 20 dB! What will be the effect on the sound?
With R26/R4 = 1, a 20 dB preamp should be implemented which could supply 36 Vtt from rail to rail!
Again an OPA134-with-bootstrap should be implemented which has the same problem: 20 dB over-
compensation! Apart from this, the implementation of the separate bootstrap becomes rather compli-
cated! This is not the way out…..

The consecutive tests

1. Disconnect the current limiter circuit.
2. Just replace the OPA445 with an LTC6090.
3. Connect the current limiter again.
4. Replace the LTC6090 with an OPA134 with ±18 volt separate power supply.
5. Disconnect the current limiter.
6. Implement the 18 volt bootstrap.
7. Replace the OPA134 with an OPA445.
Not all steps will be executed. My hope is going on the LTC6090…..

Conclusions

Summary of Features: LTC6090

• Supply Range: ±4.75V to ±70V (140V)

• 0.1Hz to 10Hz Noise: 3.5µVP-P
• Input Bias Current: 50pA Maximum
• Low Offset Voltage: 1.25mV Maximum
• Low Offset Drift: ±5µV/°C Maximum
• CMRR: 130dB Minimum
• Rail-to-Rail Output Stage
• Output Sink & Source: 20mA Minimum
• 12MHz Gain Bandwidth Product
• 21V/µs Slew Rate
• 11nV/√Hz Noise Density
• Thermal Shutdown
• 4mm x 6mm 16-Lead QFN Package