High Power audio amplifier module

Project presentation

Abstract

Audio amplifiers come in a huge variety - from small to big, cheap to expensive, weak to powerful and many other properties. So basically everybody looking for an amplifier can find one that best suits personal preference on the market. Why did I develop my own one instead? The answer is simple: For the fun of learning something from it. I haven't studied electrical engineering or any other science, but learned everything I know on the job. Amplifiers are challenging in many ways and therefore offer great learning opportunity. Plus, they have practical use, namely in HiFi or PA systems and so I can use what I have built for listening to music. I have built another amplifier before using a schematic designed by somebody else and back then, around year 2000, this was challenging. This time, roughly 10 years later, I wanted to develop the schematic partially myself and build a monster amplifier, which is challenging by nature.

Design inspiration

Most amplifiers around follow widely used schemes and mine is no exception here. The starting point was a design by Anthony Holton. Many amplifiers were built using the PCBs he sold back then and worked very well. Also, the front-end is capable of driving a big output stage. Actually it is mostly the output stage that determines the possible output power of an amplifier.

A great resource for inspiration and help is the DIY Audio Forum. I was pretty clueless when I started this project and some guys there got me up to speed quickly and supported my project. Thanks again. Amplifiers based on this kind of push-pull VAS seem very popular on the forum and many improvements are discussed and tested by enthusiasts sharing their ideas and knowledge.

The schematic was developed using LT Spice, now owned by Analog Devices. This seems to be a very common way to develop audio amplifiers. The biggest weakness of simulation is model accuracy. So a critical view on the results is required together with a healthy safety margin in case reality deviates too much from simulation.

Detailed design

Schematic

Assembly

Design considerations

The schematic still has a lot of similarity with the original, just like with any other amplifier of this kind. Major changes: I used different transistors everywhere except the OPS due to the original parts have become unavailable on the market. In most places, transistors can be substituted by similar ones. I used a different type of input LTP tail current source. The two transistor CCS performs better, but is more prone to instability and therefore requires attention in order to keep it stable. C13 and C14 aim to do that. There are countless other ways to do this as well. I added a current source for the input LTP cascode reference so that the design is less dependent on supply voltage. I changed the cascode reference point from ground to the LTP emitters. The idea behind is that the cascode reference becomes floating and thus VCE of Q1 and Q2 becomes constant. I added a cascode (Q7) to one half of the VAS long tailed pair in order to better balance the voltage swing across both VAS LTP transistors. This is still not in perfect balance, but better than before. I changed the VAS current mirror for no reason but others have reported that this arrangement performs better. I changed the loop compensation because back then I didn't understand how it works and I added alternative compensation options to maybe play with and some small details here and there. One thing I'm not sure about is whether the differential input is really useful. Achievable CMRR is somewhat limited and thwarted in case of any imbalance of the source feeding the amplifier. But it is a nice thing to have for just little added parts cost and complexity. And in case two modules need to be bridged, the differential input makes inverting one channel easy.

The design uses SMT components for resistors and some other components in order to compress the physical layout as much as possible. While this is good from a performance point of view, SMT components are not very DIY friendly. I made good experience soldering with two irons equipped with screwdriver tips simultaneously.

The setup of the power supplies enables multiple supplies and separate ground connections. This allows to run the front end at a higher voltage than the output stage, using the power supply most efficiently and therefore, maximizes potential output power. However, I came to the conclusion that running from a single supply has the advantage that the output stage cannot be driven into saturation. Recovery from clipping in case the output devices drain potential hits the rails is likely not very graceful and best avoided, also for safety reasons. According to my circuit simulation, recovery from clipping is difficult to handle gracefully.

Power supplies for this amplifier are +/- 90V. Biased to a healthy 100mA per OPS device, a single channel of the amplifier dissipates 180W idle. Apart from getting stoned to death for wasting energy by climate hysterics nowadays, it is really difficult to get rid of all the energy. The heat sinks I used have a thermal resistance of slightly below 0.3K/W and run at 50°C with ambient temperature being normal. Although the result of the thermal calculation was pretty clear, I was more optimistic in terms of heat management. Clear case of wishful thinking. From a thermal point of view, I would either recommend a bigger heat sink or lower voltage supplies like 80V, which not only helps to manage heat dissipation, but also is much safer considering the possible SOA of the output stage.

Build considerations

OPS MOSFET matching

MOSFET matching statistic
All MOSFETs of the output stage need to be matched in terms of VGS within 10mV at bias current, which is around 100mA in this case. All N-channel MOSFETs need to be matched to each other and all P-channel MOSFETs to each other. There is no need to match the N-channel to the P-channel MOSFETs. Matching is annoying, time-consuming manual work. There are two options: Either be very quick and read the instruments value ASAP prior to the junction heating up or wait an eternity for settling of the value once thermal equilibrium is reached. I used the first approach. In order to have sufficiently large groups of matched transistors in the end, excessive amounts need to be procured in the first place. I bought 100 samples each.

LTP transistor matching

It is a good idea to match the transistors of the long tailed pairs as well. In this case, I matched the transistors of the input LTP. The VAS LTP is dis-balanced by design and therefore, matching is futile. This is a lot of manual work that is only feasible for a DIY project. For the next design, I will consider a matched transistor pair at this place.

Details of the PCB assembly

There are some places with unusual arrangements. Here are a few illustrations plus mounting instructions:

Custom made inductor Mounting of floating heat sinks to transistors
Output stage MOSFET mounting Bias V<sub>GS</sub> multimplier mounting on OPS transistor
Transistor mounting hardware Heat sink hovering above lower assembly

Connection scheme

Instead of merging grounds on the amplifier PCB, I added a connector for each in order to segregate currents. Main reason for doing so was uncertainty how to ground the amplifier correctly. Below illustration shows how I (should) have wired the modules. Fuses are not shown. The point of the illustration is to show separation of the two grounds from the rectifiers all the way until they join at the output of the capacitor banks. The output of the capacitor banks is also the star point where all supply and ground wires of the module and the speaker connect to. I wired it differently but believe that the scheme shown is the best way to do it. On the mains side of the transformer, wiring mostly depends on legal regulations.

Performance

It is difficult to assess performance of the amplifier and compare performance to the original design. Both is done based on a lot of assumptions and simulated using LTSpice. Supply voltage is 90V for both and the same models used mostly. While the numbers obtained by simulation are most likely wrong, it helps to get an idea how performance might be in an ideal case.

THD comparison

THD comparison 1kHz
Power THD % my design THD % original design
1W 0.0005 0.004
10W 0.002 0.011
100W 0.005 0.036
200W 0.008 0.056
400W 0.011 0.347
THD comparison 20kHz
Power THD % my design THD % original design
1W 0.004 0.054
10W 0.013 0.167
100W 0.031 0.549
200W 0.043 0.858
400W 0.142 2.224

Spectrum

1kHz

Plots of the spectrum at 1kHz and 200W output power. Plotted in blue color is my amplifiers spectrum. The original spectrum of the Holton amplifier is plotted in red color (hover over image to view).

20kHz

Spectrum at 20kHz and 200W output power.

While it is obvious that my design has lower THD, the original amplifiers harmonics level to the noise floor beyond 1MHz, but mine do not. I haven't found the root cause yet. Maybe it is a simulation artifact, maybe some minor instability somewhere.

Loop gain

Back when I adapted the design, I didn't understand how the compensation works and therefore, came up with a pretty simple one. The plot shows that the amplifier is likely stable. Loop gain drops to unity at roughly 300kHz with -88° phase shift and 24dB gain margin. The plot shows that the amplifier is likely overcompensated and the crossover frequency could be pushed higher maybe, but I don't trust my design and simulation that much and find stability is paramount. It seems that the original design has a crossover frequency at only 80kHz with -96° phase shift and 28dB gain margin. Capacitor C8 is for phase lead compensation and "gains" some phase margin at higher frequency. This can be seen in the plot around 1MHz, where the phase shift suddenly recovers. Without C8 and with C5 and C6 reduced to 22pF each, unity gain frequency would be at 600kHz with -106° phase shift. Installing C8 again results in -85° phase shift in this configuration. As Bob Cordell pointed out (book page 379), phase lead capacitors across the feedback voltage divider resistor may open the door to HF ingress from the speaker terminal, so it is a trade-off. Well, back then I found it a good idea but the amp probably would do fine without capacitor C8 as well.

Again, my design is plotted in blue color, the original design is plotted in red color (hover over image to view).

Square wave response

Blue plot shows actual response, green plot shows possible improvement with different compensation discussed above and plotted in red is the original design.

Clipping

Clipping of the positive half-wave is somewhat graceful but the negative half looks bad. Below example waves result from common rails for front-end and OPS.

As mentioned earlier, higher supply rails for the front-end allow higher output voltage swing. However, the OPS saturates and therefore, the waveform shows some overhang from sticking to the rails.

Summary and outlook

As of year 2020, the amplifier still performs well, although I never got a chance to make use of the enormous power it can provide. For me it runs in class A only with incredible headroom available in case I ever need it. Best suited to drive super inefficient speakers, like many HiFi speakers are, or high impedance speakers or earth-shattering sub-woofers. With a low voltage supply, low impedance speakers can be driven as well.

I made up my mind how to possibly improve the circuit, reviewed it again, also gave it to a former colleague for review and came up with quite a long list. I made an assessment of potential benefit versus risk and in the end just changed a few component values. Result of the reviews is that the design currently is pretty good. I've been investigating some ideas of major changes for the next revision, but found most of them risky to implement. Below table contains some ideas for a future revision. Some of the non-risky changes can be implemented in the current hardware. The most obvious way to further improve the scheme is detailed in Bob Cordell's book "Designing Audio Power Amplifiers". This is a fantastic resource. I simulated circuits based on his paper MOSFET Power Amplifier with Error Correction but didn't succeed to get this any stable.

Idea Improvement Risk
Remove R39 Part is not assembled, useless and ill advised in any case None
Remove D9 and D10 Components are useless None
Increase size of C16 and C17 Better power supply decoupling None
Add resistor between Q13 and D3 like tail CCS has it May improve stability of current source None
Add capacitor parallel to D3 Reduce Zener diode noise (only with additional resistor mentioned above) Low
Add series resistors to base of Q3, Q4 and Q7 May improve cascode stability Low
Increase capacitance of C11 Improves LF distortion Low
Degenerate Emitters of Q5 and Q6 More linearity of VAS LTP but lower OLG Medium
Cascode Q5 and Q6 Better balance of VAS LTP High
Add cascode to current mirror Better balance of CM High