General design aspects for audio amplifier power supplies

Abstract

In this article, I collect some ideas and resources for audio amplifier power supplies in general. Most of this is also applied to the power supplies I designed for my modular audio amplifier.

Example Schematic

This is the schematic of my high capacitance medium size power supply with filter. This is a good example to discuss each feature in detail.

Interface

I don't like screw terminals because I've observed that torque applied to the screw damaged the solder joints quite often. Also, a screw may become loose or even lost and rioting inside the enclosure. My preference is cage clamp terminals. They are easy to operate, provide good contact, accept any kind of wire end, are DIY friendly and reliable as long as the spring does not fatigue or break. The cage clamp terminals I prefer are from Wago 236 series and come in 5 mm pitch. They can be swapped for some other kind of connectors instead, like 6.3x0.8mm Faston connectors (although I don't like them).

adding some more terminals adds versatility to both the input and the output. The transformer terminals accept a lot of different configurations, providing clean point to point wiring, avoiding any messy loose wires around. Transformer secondary windings can be connected in parallel or serial. The output offers many terminals in order to connect either multiple modules or to accept segregated ground and supply connections from the amplifier modules.

Rectifier diodes

The rectifier arrangement is a bit unusual as one of the bridge rectifiers appears to be superfluous. However, for highly powerful applications, this arrangement has slight advantages, discussed controversially. Since both Nelson Pass and Bob Cordell agree that the dual bridge rectifier arrangement has advantages over the single bridge, this is what I prefer. The downside of this arrangement is higher component count and power loss.

Basically any diode should work well as a simple rectifier. My preference for the rectifiers are soft recovery diodes. Foremost important is forward voltage drop across the diodes. The FFPF30UP20STU diodes have exceptionally low forward voltage drop even at high currents and this helps to lower power loss. Diode turn off causes substantial noise emission. Different diodes have different turn-off characteristic, but the resonance between diode capacitance and transformer secondary winding is in inherent problem and the best way to deal with it is to add snubbers. This is explained later on.

For low to medium power amplifiers I would use diodes in TO-220 chassis and for higher power TO-247 chassis seems more adequate. Suitable components start at 1€ each up to 2€ each. Diodes can be either mounted to a small heat sink or to the metal chassis.

Another option would be integrated full bridge rectifiers, but I find those parts a bit inconvenient to design in. Some people claim that discrete diodes are superior over integrated bridge rectifiers, but I haven't investigated this yet.

Snubbers

I added plenty of R-C snubbers to my designs in order to mitigate resonance between diode capacitance and the transformers secondary windings during diode turn-off. This resonance causes a lot of ringing of the secondary voltage, which may cause RF emissions. The components required to build snubber networks are low cost, but help to avoid unnecessary noise emission and therefore allow to dramatically improve the behaviour of the power supply. The snubbers can be tuned for best performance and the values in my schematic are rather brute force over-damping. I don't see any downside over-damping the network an this is a good starting point.

Optimization of snubber networks is very complicated in theory. Mark Johnson has presented a simple test jig that allows to optimize damping of the resonance. This is the best way to achieve critical damping. The solution proposed by Mark uses less parts than my brute force approach and this may have a slight cost advantage.

Some power supplies may have small capacitors across the diodes, which does not completely solve the problem, but shifts it to a lower frequency range instead - at least a partial improvement. Many commercial designs I have seen so far, don't feature any snubbers at the rectifiers. Obviously it is possible to pass EMI compliance without snubbers. For commercial products built in huge quantities, shaving off a few cents is pretty common. However, I strive for best performance with high end audio performance in mind and don't want any unnecessary radiation inside my amplifier that can be tamed with part worth just a few cent.

Resonance

Capacitor

Snubber

Capacitors

The capacitors are the single most expensive parts of the whole assembly in any case. I prefer long life high quality capacitors. The capacitors in all of my power supply modules have 40mm diameter and four terminals with a pitch circle of 22.5mm. For terminals ensure that the pins can support the huge mass of the capacitors. I just like to built things durable. The form factor of the capacitors restricts the variety pf available components a bit, but I find that the components offered fit any of my needs.

Filters

My very first amplifier, which I built when I was roughly 15 years old, featured chokes in between the PSU capacitors in order to form π filters. In my second amplifier I omitted the chokes, but used stainless steel sheets to connect the capacitors together and those have both high resistance and also inductance. Later research and experience made me become more cautious regarding inductors in the power supply, which I had to overcome by more research. I still plan to write an article about power distribution networks since this is a very important topic. Most engineers I got to know are totally incompetent in this field and mindlessly throw random capacitance into the schematic creating power supply problems that are super easy to avoid.

Conclusion from my research is that inductors may cause resonance and overshoot of the PSU voltage in some cases. Basically this forms an R-L-C circuit that resonates in case the inductance is too high or the resistance or capacitance is too low. Remedies to get rid of resonance are to increases the inductors series resistance, decrease the inductance or increase the capacitance. Shunt compensation by adding a snubber network is also a good option, but may be impractical in some cases.

In general, it is a very good idea to use multiple large capacitors in the PSU and arrange them as π filters. With the very large capacitance usually present in audio amplifier power supplies, even very small resistance or inductance will result in filters with very low roll-off frequency. Sometimes it is possible to form resistors using the copper of the PCB so this effect even comes free of cost. Also, the wires connecting the PSU with the output stage may contribute to the filter and by choosing the lowest possible square section, this may even safe a few cents and both improve the power distribution network.

Active Filters

When talking about active filters for the power supply, I mean capacitance multipliers. Those circuits really help to obtain a quiet power supply in case the voltage drop can be accepted and the power loss is not too high. There is a limitation of effectiveness for high frequency attenuation and this is why I would always precede an active filter with a passive one.

Voltage Regulators

Voltage regulators are useful for obtaining a precisely regulated supply voltage, but are less useful for obtaining a quiet power supply. PSRR decreases with increasing frequency. So it is a very good idea to precede voltage regulators with passive filters in order to mitigate loss of PSRR.

Active rectification

Simple bridge rectifiers are perfectly fine for class AB amplifiers that usually have around 1A bias current. Even at full load, the power dissipation of the PSU rectifiers is no mayor concern because the RMS power of music is not that high. The situation is different in high power class A amplifiers. One Ampere of current causes roughly 500mW of power loss in each FFPF30UP20STU diode and those diodes have exceptionally low voltage drop. At 4A RMS current, power dissipation would be 2W per diode, which is still fine. Assuming 25K/W heat sinks for each diode, temperature rise would be roughly 50°C. At 10A RMS current, power dissipation is around 4.5W and this is where heat dissipation becomes challenging unless large heat sinks are being used.

During building the first variant of my modular amplifier, I found biasing the output stage into class A a great idea. The transformers are good up to 8A output current and I estimate the output voltage at this load being somewhere between +/-20V to +/-22V. This is good for class A operation with 8Ω speakers. Once I estimated how the PSU performs in class A operation, I concluded that better had made up my mind about rectifier power dissipation in this case earlier.

A technically elegant way to reduce rectifier power loss is active rectification. Analog Devices offer the easy to use LT4320 IC that drives four N-channel MOSFETs as ideal diodes. Given that the on resistance of the MOSFETs is sufficiently low, power loss can be lower than with diodes.

The IC has an internal charge pump without any external capacitance for the charge pump and this limits possible gate drive to the high side MOSFETs. MOSFETs having low on resistance usually require a lot of charge at the gate in order to achieve this low resistance. In case the driver cannot provide enough current drive, the switching speed is reduced, what increases power loss again. This is a difficult trade-off.

I investigated the LT4320 and found that simulation is required to optimize the design. In order to sink 10A RMS from the PSU and not use heat sinks for the MOSFETs, the MOSFETs need to have 5mΩ maximum on resistance in order to have sufficiently low dissipation.

My general opinion regarding the LT4320 is that this device is great, but limited by two factors: Maximum supply voltage and a rather weak built-in charge pump. This restricts the IC to low / medium power applications. I would like to see a high power variant with over 100V supply voltage tolerance and more powerful charge pump. However, the IC seems useful and could perform well in case the design is optimized carefully.

Power Factor Correction

A typical audio amplifier PSU presents a terribly low power factor to the mains voltage net because conduction only happens for a very short period at the peak of the mains voltage sine wave. This is especially disturbing in case of high power amplifiers. Other equipment may even be affected by this to some extend in case the power supply is poorly filtered.

Active power factor correction would be the cure, but is very challenging to implement. I would place the active PFC at the transformer secondary side. There are further benefits beyond a high power factor. Since active PFC is basically a kind of buck-boost regulator, the power supply ripple is greatly reduced and much less capacitance is required. Downside is that the PSU now needs careful filtering to deal with typical switching PSU problems like conducted common mode emissions.

Summary

The seemingly simple power supplies for audio amplifiers has a rich spectrum of problems associated. Some of them can be solved and doing so is neither complicated nor expensive. Some inherent problems require dramatically different approaches that are impractical for the average DIY enthusiast. However, by getting the basics right, there is a lot of room for improvement. In this article, I merely pointed out what needs to be considered and leave solving the problems to the reader.

Links and references

Here are some links for further reading or entertainment:

  • Texas Instruments AN-1849 shows a basic power supply with some very basic, but maybe useful additional control circuitry and a terrible PCB layout example not to be copied in any case.
  • Rectifier snubbers explained in depth
  • DIY Buck-Boost PFC designed without dedicated controller IC. This method is basically the best way to draw a lot of power from the mains. Downside is very high complexity