High performance power supplies.
A long time ago (around 1996 I think), in a galaxy far, far away I was asked to write a series of short articles about power supplies in an underground audio magazine called Common Ground. There has been some discussion lately regarding power supplies on the AoS forum so I thought it might be of interest to re-publish the articles, as the information therein is still relevant today. I did not want to infringe on or sidetrack Jez’s recent mythbusters thread so I have started a new thread.
If anyone wants to expand the articles with additional information and experiences please feel free to do so.
POWER SUPPLIES
A PURIST’S VIEWPOINT
BY PAUL HYNES ©
The first of these articles is intended to cover the application of high performance power supplies for use with high fidelity amplification. Audio power supplies have been a speciality of mine ever since I first used a regulated power supply designed by Michael Sultzer (Audio Amateur 2/1980). Michael’s discussion on the effects of regulated power supplies on sound quality was sensible and his audible findings correlated well with measured performance. He proved that the power supply impedance should be as low as possible and the bandwidth as wide as possible to minimise any power supply inter-modulation distortion.
There are various methods of achieving good power supply regulation. You can use a passive supply that is grossly over-specified or you can use a reasonably sized power supply with electronic regulation. The passive supply method can prove quite expensive if large low impedance transformers, diodes and capacitors are chosen. The target is to provide a vast energy source relative to the needs of the circuit to be powered. The electronic regulator simulates this vast energy source by attempting to correct any fluctuations in the power supply caused by varying load currents. If your power supply is inadequate, load current changes will cause voltage fluctuations on the supply output that can infect the signal via the amplification stages, which have less than ideal power supply noise rejection capability. Valve circuitry is particularly prone to this problem as typical circuits have very little supply rejection. The resultant signal breakthrough is amplified along with the required signal, playing havoc with the low level information, dynamic range, tempo and sound stage stability. Once this pollution is mixed in with the signal you are stuck with it. Therefore, to be able to track and correct the power supply load current changes, the regulator transient response and settling time should be sensibly faster than the signal processing circuitry.
The ideal regulator would present zero impedance to the load at all frequencies. This implies that the regulator error amplifier should have a transient response with infinite slew rate and zero settling time. At present this is impossible, however it is possible to improve on the usual regulators used in audio applications by a large margin.
Let us examine the regulator parameters desirable for use in audio applications. In essence, a voltage regulator is a DC coupled AC amplifier with its input connected to a fixed reference. A feedback loop senses the regulator output and compares it to the reference. If the load current changes, there will be a corresponding voltage change at the regulator output due to finite regulator output impedance. The feedback loop will attempt to correct this change via the AC amplifier. The time taken for this correction, is governed by the AC amplifier’s transient response and settling time.
Further considerations are the effects of RFI and digital clocking breakthrough on the power supplies. In these instances we are talking multi-megahertz frequencies often with fast waveform rise-times. At these frequencies and speeds, normal regulators are no longer working and can in fact make matters worse as they attempt to correct such errors. They can overshoot their settling target and wobble about at different frequencies vainly trying to catch up with themselves. With this in mind, it should be apparent that we require a regulator that is considerably faster when handling audio signals than first thoughts would suggest. You cannot expect a regulator with a transient response and settling time of microseconds to track a signal waveform with a rise-time of say 100 nanoseconds. However if a regulator can track these waveforms, dealing with audio waveforms should present no problems.
Regulator performance check list for audio use :
1. Speed - Damn fast
2. Settling time - Damn fast
3. Output impedance - very low ( 10 milliohms or better )
4. Bandwidth - As wide as possible to cope with RFI etc
5. Noise - Quiet enough not to interfere with low level signals
There are further considerations regarding power supply distribution, grounding and the fact that we are dealing with a stereo system that is attempting to re-create the original sound field that was present at the recording venue, (studio multi-track recordings aside). Disregarding the controversial arguments about microphone placement, we are essentially attempting to capture the information that our ears would hear at a live performance. Since there are two ears to deal with, the phase and amplitude of information arriving at each ear gives the clues to the three-dimensional sound field that the person is in. This means that it should be possible to give reasonable reproduction of this performance with two channels of information, providing we can keep the phase distortion, amplitude distortion and interference from the outside world to a minimum.
One very important issue here is to keep a solid ground reference for each amplifier stage. Any impedance in the ground reference will create voltage fluctuations, which are dependent on the currents flowing through the ground reference. These voltage fluctuations are damaging to the desired signals, however they can be minimised by using a reference star.
Signals are amplified with respect to the reference star. Only the signal currents that are essential to the waveform transfer pass through the signal common wire.
The regulator references are connected to the reference star and provide stable supply rails for the amplifier stages. If separate transformer windings and regulators are used for each stage of amplification then transformer / diode / reservoir capacitor charging currents do not appear in the reference star (this is not true if one common power supply feeds all the regulators as the power supply return currents would have to pass through the signal earth wires). Thus the reference star remains clean to the information transfer.
For example :-
image002.jpg
This is ideal from the point of view of one stage of amplification. Separate transformer windings are essential to this method of power supply distribution. Poor quality regulators will give poor results. The closer the regulators approach the ideal power source the less detrimental the effect they will have on the desired signal.
You may cascade as many of the above stereo stages as required in your sound system without causing power supply inter-modulation problems. It is not necessary to ground the signal source for correct operation as the optimum ground point from a convenience point of view is the line level pre-amp star. Each stage references its signal to the following stage whilst still maintaining signal integrity and no power supply interaction through the signal ground.
Combine this technique with fast, phase coherent signal amplification and you will have one hell of an amplifier, capable of resolving a stable three-dimensional image of the original performance with excellent musical tempo.
Happy listening, more next issue.
Paul Hynes.
POWER SUPPLY REGULATION
THE SECOND ARTICLE ON POWER SUPPLIES
BY PAUL HYNES
This article covers the subject of regulators in more depth than discussed in power supplies issue 1 and in particular the problems encountered utilizing standard three terminal regulators.
Most audio equipment manufacturers use industry standard regulator devices in their products because they are readily available, cheap and generally easy to apply. Some benefits are offered by these products, notably, reduced power supply ripple breakthrough from the rectifier / capacitor power sources. This allows a much smaller capacitor to be used, which in turn reduces component costs considerably, more than offsetting the cost of the regulator itself. Multiple regulator systems can be applied more easily and cheaply, and once again these regulators can be significantly cheaper than a high quality decoupling capacitor of sensible size. As you can see the main benefit of using these devices is essentially one of cost reduction. Whilst this is a laudable aim, most enthusiasts will generally prefer to look for performance improvement before cost considerations (within reason, acknowledging the fact that few individuals on this planet have unlimited budgets).
So let us look at regulator performance with a typical industry standard, the 317 / 337 type adjustable regulator available from various manufacturers. The performance varies from manufacturer to manufacturer, but not dramatically so, and the following figures are fairly typical of the specification sheets on offer.
We will start with the typical power supply rejection ratio (PSRR) of these regulator devices with respect to frequency. At 100Hz the PSRR is 60-70dB ( 60dB represents a ripple reduction of about 1000 times ) which is quite reasonable, particularly if the circuits to be powered also have a good supply rejection. At 1KHz the PSRR begins to reduce. This is due to the regulator device’s internal frequency compensation causing reduced gain at higher frequencies, leaving less loop gain available for error correction. At 10 KHz, the 317 regulator manages 50dB of PSRR (316 times ripple rejection).
At 100KHz both of the devices only achieve approximately 20dB of PSRR (10 times ripple rejection) and at 1MHz only 10dB of PSRR (3.16 times ripple rejection). We are able to deduce from this information that at low audio frequencies both devices offer reasonable PSRR, but this situation deteriorates rapidly above 1KHz, becoming relatively ineffective at frequencies above 100 KHz where radio frequency interference and digital clocking power surges may have to be dealt with. The output impedance is a useful guide to comparisons of performance, as it shows the regulator’s ability to control the load with respect to frequency. The 317/337 graphs show a respectable 10 milliohms from DC to 1KHz. At around 1KHz the frequency compensation capacitor comes into operation, reducing the loop gain and negative feedback with respect to frequency, to achieve regulator stability. It’s at this point that things start to deteriorate. The output impedance is a function of available negative feedback and as this feedback reduces with rising frequency, the output impedance rises accordingly. At about 1MHz the regulator runs out of gain and is no longer functional.
image004.jpg
The internal frequency compensation capacitor used in these devices has another, more important, effect on their behaviour. This capacitor has to be charged and discharged by the internal circuitry before the feedback loop can apply error correction. This may sound familiar to those of you who peruse the audio related electronics and HIFI press, as various articles have been published about Transient Inter-modulation Distortion (TID) and Slew Induced Distortion (SID). The same thing is happening here, and due to finite currents available to charge and discharge the compensation capacitor, we are left with line and load transient settling times of up to 5 microseconds with these devices.
One final point is the noise specification, which is typically 0.003% of the regulator output voltage. For a 15 volt regulator this equates to 450 microvolts of noise on the output. This is too high for use in low-level signal preamplifiers without resorting to additional noise decoupling.
More soon.
Paul Hynes
OPTIMISATION OF POWER SUPPLY REGULATION
A PURIST’S VIEWPOINT
The third article on power supplies
BY PAUL HYNES (C)
Having ascertained that power supply impedance should be as low as possible, over as wide a bandwidth as possible, its time to look at the various techniques used to achieve this.
Battery supplies
The simplest power supply is a battery. Lead acid cells can offer very low impedance over a reasonable bandwidth. They are also reasonably quiet and make good power supplies for audio equipment. However for valve HT use they have to be stacked in series to achieve the required voltage levels and this increases the cost considerably. The batteries have to be recharged regularly and recharging multiple battery stacks is complex and costly. Typical lead acid battery life expectancy is 2 years so the cost of battery replacement has to be considered. It’s rather like running a car. You have to expect maintenance bills from time to time.
Basic mains supplies
Mains powered supplies start with a transformer diodes and reservoir capacitor. Usual commercial implementations are far from adequate and suffer from serious power supply inter-modulation and noise problems. This is particularly apparent with simple valve circuits, as they have virtually no power supply rejection capability. There are quite a few valve amplifiers that suffer bad hum and noise problems, together with poorly controlled bass and image instability.
Choke mains supplies
Adding a choke to the supply can offer a useful improvement in noise filtering and also regulation because it adds some energy storage in the choke. However, it falls short of the ideal power supply because the power supply output impedance is too high to prevent power supply inter-modulation problems from occurring, necessitating some form of voltage regulation after the choke filter for good performance. Given that post filter regulation is required and that large power supply chokes are expensive and careful design is required to achieve good performance, this can be an expensive route.
Valve electronically regulated supplies
Electronic regulation was first used with valves, and, as there are reasonably fast devices available, this was quite successful with respect to transient response and bandwidth. However the output impedance is still too high for low power supply inter-modulation.
Solid-state electronically regulated supplies
The older solid-state regulators offer much lower impedance but this is usually over a seriously restricted bandwidth. Transient response is poor due to the way they were designed. These older solid-state regulators are noticeably worse than valve regulators, and indeed transformer, diode, capacitor and choke power supplies.
New techniques in high-speed regulation
The availability of new, very high-speed circuit topologies has allowed the design of regulators that out-perform all the other methods outlined above by large margins in all the sonically important parameters.
Fast error amplifiers with slew rates exceeding 2000 volts per microsecond and with a settling time of typically 25 nanoseconds or less are now available. They can be quiet enough to use with valve moving coil head amplifiers and sturdy enough for heavy-duty applications. Output impedance can typically be less than 5 milliohms from 0 Hz to 100 KHz.
The improvements offered by this new generation of regulator topologies when applied to all circuit systems are not subtle. They include improved dynamic range, rock solid image stability under large dynamic swings, large stable soundstage, quiet operation allowing a wealth of low level information to be perceived, accurate tempo due to lack of delayed power supply reactions and well controlled bass due to very low power supply inter-modulation.
Reply With Quote
All good stuff!
Originally Posted by Mr Kipling
