Small Power Supplies

Across the Web, there are countless designs for low current (typically 1A or less) power supplies for preamps, small PIC based projects, ADCs, DACs and almost any other project you can think of. Many are very basic, using nothing more elaborate than a resistor and zener diode for regulation, while others are very elaborate indeed.

For most beginners and many experienced people alike, it becomes very hard. One has to decide where extreme precision is needed, how much noise can be tolerated and just how complex the supply needs to be for the application. Some assume that a 'super regulator' of some kind must be better than a readily available IC solution, whether or not it will make an audible difference is neither checked nor tested.

It must be understood that a regulator (in almost any form other than a zener diode) is an amplifier. Admittedly the amplifier is 'unipolar', in that it is designed for one polarity, and can only source current to the load. Very few regulators can sink current from the load, but shunt regulators are an exception!

Since amplifiers can oscillate, it follows that regulators (being amplifiers) can also oscillate. As the bandwidth of a regulator is increased to make it faster, it will suffer from the same problems as any other wide bandwidth amplifier, including the likelihood of oscillation if bypassing isn't applied properly.

There is also an endless fascination by some to build the smallest and cheapest power supply possible. Many circuits can be found that don't even use a transformer, and while some have acceptable or adequate warnings about safety, others don't. Indeed, there is one published design that breaks the wiring code of every country on earth, has no warnings, and is a death trap (this one has its own section in this article - see Cheap Death).

If you are not experienced with mains wiring, do not attempt the following circuits. In some countries it may be unlawful to work on mains powered equipment unless you are qualified to do so. Be aware that if someone is killed or injured as a result of faulty work you may have done, you may be held legally responsible, so make sure you understand the following ...

For anyone who is unfamiliar with the terms 'creepage' and 'clearance' as applied to electrical equipment, they may be defined as follows ...

The distances are measured between high and low voltage circuitry, and between high voltage conductors where the voltage may track or arc between conductors without adequate separation. System specifications such as IEC60950-1 and IEC61010-1 dictate the required creepage and clearance spacing for a given system. IEC60950-1 regulates the requirements for Telecom Equipment, and IEC61010-1 regulates the requirements for Industrial and Test Equipment. In the US and Canada, UL/ CSA standards apply respectively. In many cases, power supply (especially SMPS) makers will cut slots into the PCB to increase the creepage distance. Different applications have differing requirements, but if you allow 8mm (a little under 0.32") that will cover most cases. 5mm (0.2") should be considered the absolute minimum. This is the distance between the pins and PCB pads of most optoisolators for example.

All countries have electrical wiring codes and standards, but compliance may be voluntary, implied or (in a few countries) mandatory (at least for some products). In any case, if a product is found to be dangerous, there will usually be a recall, which may be mandatory if the safety breach is found to be a built-in 'feature' of the product. It is the responsibility of anyone who builds mains powered equipment to ensure that it meets the requirements set in the country where it's built or sold. The authorities worldwide take electrical safety seriously, and woe betide anyone who falls foul of the standards by killing or injuring someone.

Note: IEC60950-1 and EN60950-1 will be withdrawn in June, 2019 (since amended to December 2020), and transferred to IEC62368-1. IEC62368-1 is the standard for safety of electrical and electronic equipment within the field of audio, video, information and communication technology, business and office machines. The Australian/ NZ version will be AS/NZS62368-1 and UL62368-1 in the US.

We'll start with the ideal regulator and work back from there. The ideal regulator has perfect regulation, so the voltage does not change regardless of load. It is also infinitely fast, so infinitely sudden load changes (over an infinite range of current) have no effect. Noise is non-existent (which also means zero ripple), the output is not affected by any variation of input voltage provided it's above the output voltage, and the voltage remains stable over the entire temperature range ... from -50°C to 150°C would be sufficient.

Needless to say, the ideal regulator does not exist. All regulator circuits have limitations, and it is the job of the designer to determine which limitations will have the greatest impact on the device being powered, and work to minimise those at the expense of other parameters. For example, a simple discrete based preamp will have relatively poor power supply rejection, so noise is a potentially major problem. Since the current won't vary much in use (for this hypothetical design), extreme speed is not needed. This hypothetical supply needs to be reasonably stable and have very low output noise - high speed and extremely good regulation are not necessary.

Another supply might be needed for a medical application where the voltage is critical and the load varies in fast steps (a high speed analogue circuit followed by an ADC, and with digital logic control perhaps). Noise doesn't need to be especially low, since the ADC chip has its own voltage reference which includes good filtering. This supply needs to be very fast to keep up with the changing load current, and requires accurate voltage. It will also need to be inherently safe, because it's for a medical instrument. As such, it will have to be fully certified in the countries where it's used.

The above are but two (extreme) examples of possible supply requirements, but there are as many different requirements as there are circuits. In some cases, it is not possible to suggest a supply unless you know exactly what will be powered from it. In others, almost anything will work just fine. Since The Audio Pages are mostly about audio, I shall concentrate on supplies that are applicable to audio projects, however the same basic principles apply for all power supplies, large and small.

Since most hi-fi products are powered from the mains, we need to galvanically isolate the output of the supply from the mains voltage. This is a vital safety requirement, and cannot - ever - be ignored, regardless of output voltage or power requirements. Galvanic isolation simply means that there is no metallic electrical connection between the mains and the powered device. A transformer satisfies this requirement, but is not the only solution. One could also use a lamp and a stack of photo-voltaic cells ('solar' cells), but this is extremely inefficient. Because most of the alternatives are inefficient or just plain silly (such as the example above), transformer based supplies represent well over 99.99% of all isolation methods. Switchmode supplies also use a transformer, so are included in the above.

Transformers only work with AC, so the output voltage must be rectified and filtered to obtain DC. This is shown in Figure 1 - the transformer, rectifier and filter are shown on the left. For simplicity, mainly single supply circuits will be examined in this article - dual supplies essentially duplicate the filtering and regulation with the opposite polarity. The filter is the first stage of the process of noise removal, and deserves some attention.

C1 (the filter capacitor) needs to be chosen to maintain the DC (with superimposed AC as shown in Figure 2) above the minimum input voltage for the regulator. If the voltage falls below this minimum because of excess ripple, low mains input voltage or higher current, noise will appear on the output - even if the regulator circuit is ideal. No conventional regulator can function when the input voltage is equal to or less than the expected output. It can be done with some switching regulators, but that is outside the scope of this article.

In the above schematic, there is about 380mV RMS (1.24V peak-peak) ripple at the regulator's input, but only 4.5mV RMS (14.2mV p-p) at the output. This is a reduction of 38dB - not wonderful, but not bad for such a simple circuit. Load current is 142mA. With the addition of 1 extra resistor and capacitor to create a filter going to the base of Q1, ripple can be reduced to almost nothing. If you wish to experiment, replace R1 with 2 x 560 Ohm resistors in series, and connect the junction between the two to ground via a 100µF capacitor. This will reduce ripple to less than 300µV - 62dB reduction. Alternatively, one might imagine that just adding another large cap at the output would be just as good or perhaps even better. Not so, because of the low output impedance. Adding a 1,000µF cap across the load reduces the output ripple to 3.8mV - not much of a reduction. While simple, this regulator will actually cost more to build and use more PCB real estate than a typical 3-terminal IC regulator. The IC will also outperform it in all significant respects.

The regulator in Figure 1 is very basic - it has been simplified to such an extent that it is easy to understand, but it cannot work very well. This is not to say that it's useless - far from it. It must be remembered that the simple regulator will cost more than a 7815 3-terminal regulator IC though. Prior to the introduction of low-cost IC regulators, the figure 1 circuit used to be quite common, and a very similar circuit was common using valves (vacuum tubes). Early voltage references were usually neon tubes, designed for a stable voltage. These will not be covered in this article.

While a simple regulator may well be all that's needed for many applications, especially for circuits that use opamps, the regulator itself is generally not particularly critical. This is because most opamps have a very good power supply rejection ratio (PSRR) - the TL072 has a PSRR of 100dB (typical). This means that any low frequency signal on the supply (or supplies) is attenuated by 100dB before finding its way to the opamp's output pin. This varies with frequency!

Please note that the above does not apply if there is a connection from either supply to an opamp's input pin. If this is the case, extensive filtering may be needed to remove supply noise. If any supply noise is presented to an opamp input, it will be amplified along with the signal.

Referring to Figure 2, it should be obvious that the filter capacitor C1 removes much of the AC component of the rectified DC, so it must have a small impedance at 100Hz (or 120Hz). If the impedance is small at 100Hz, then it is a great deal smaller at 1kHz, and smaller still at 10kHz (and so on). Ultimately, the impedance is limited by the ESR (equivalent series resistance) of the filter cap, which might be around 0.1 Ohms at 20°C.

It is important that capacitive reactance is not confused with ESR. A 1,000µF 16V capacitor has a reactance of 1.59 Ohms at 100Hz, or 15.9 Ohms at 10Hz. This is the normal impedance introduced by a capacitor in any circuit, and has nothing to do with the ESR. At 100kHz, the same cap has a reactance of only 1.59 nano-Ohms, but ESR (and ESL - equivalent series inductance) will never allow this to be measured. The ESR will typically be less than 0.1 ohm, and is generally measured at 100kHz. Indeed, at very high frequencies, the ESL becomes dominant, but this does not mean that the capacitor is incapable of acting as a filter. It's effectiveness is reduced, but it still functions just fine. Some people like to add 100nF caps in parallel with electros, but at anything below medium frequency RF (less than 1MHz), such a small value of capacitance will have little or no effect. While this is easily measured in a working circuit, few people have bothered and the myth continues that electrolytic caps can't work well at high frequencies.

Contrary to popular belief in some quarters, electrolytic capacitors do not generally have a high ESL. Axial caps are the worst simply because the leads are further apart. ESL for a typical radial lead electro with 12mm lead spacing might be expected to be around 6nH. A short length of track can make this a great deal worse - this is not a fault with the capacitor, but with the PCB designer.

The regulator itself has a number of primary functions. The first (surprisingly) is not regulation as such, but reduction of the power supply filter noise - mainly ripple. Including a reasonably stable voltage as part of the process is not difficult with ICs, so this is included as a matter of course. The regulated voltage is not especially accurate, but this is rarely an issue.

The output impedance should be low, because this allows the voltage to remain constant as the load current changes. For example, if the output impedance were 1 Ohm, then a 1A current change would cause the output voltage to change by 1V. This is clearly unacceptable, and one might expect the output impedance to be less than 0.1 Ohm - however, this is frequency dependent and may include some interesting phenomena with some regulators (LDO - low drop-out regulators can be especially troublesome). For more details of the issues you may face with these types, see Low Dropout Regulators which has information you need to know before using them.

In order to maintain low impedance at very high frequencies, an output capacitor is commonly used. This is in addition to any RF bypass capacitors that may be required to prevent oscillation.

It must also be remembered that in any real circuit, there will be PCB traces that introduce inductance. Capacitors and their leads also have inductance, and it is theoretically possible to create a circuit that may act as an RF oscillator if your component selection is too far off the mark (or your PCB power traces are excessively long).

Bypassing is especially important where a circuit draws short-term impulse currents. This current waveform is common in mixed signal applications (analogue and digital), and the impulse current noise can cause havoc with circuitry - an improperly designed supply path can cause supply glitches that cause false logic states to be generated. Even the ground plane may be affected, and great care is needed in the layout and selection of bypass caps to ensure that the circuit will perform properly and not have excessive digital noise.

In general, linear opamp circuits will not cause impulse currents, because the audio signal is relatively slow. In many cases, the power supply current will not be modulated at all, because the opamp's output current remains substantially within its linear (Class-A) region. Even where the supply current is modulated, it will a relatively slow modulation, and track inductance is generally insignificant within the audio range.

The essential sections of almost all regulators are shown above (in highly simplified form). The voltage reference is most commonly a band-gap reference, because these are very stable, easy to implement during IC fabrication, and have excellent performance. The nominal reference voltage is 1.25V, and this is easily amplified to achieve the required voltage. Alternatively, the band gap reference can be used to control a current source that supplies a 6.2V zener diode. This voltage is chosen because the positive and negative voltage coefficients of the zener cancel, providing a very stable reference voltage over a wide temperature range.

The error amplifier simply compares the output voltage with the reference. If they are the same (the output voltage may be scaled using a resistive divider as shown), then all is well. If the output voltage is low, the error amplifier makes the appropriate correction, and passes this to the series pass device (most commonly a BJT (bipolar junction transistor), and this process continues (extremely quickly) until the output voltage is restored. Should the output rise (reduced load), the opposite occurs. In many circuits, the input voltage and/or output current is constantly changing, so the error amplifier is always working.

The regulator circuit uses feedback to maintain a low output impedance and to maximise noise rejection. Because all feedback circuits have stability criteria that must be met to prevent oscillation, there will always be a frequency above which the regulator cannot function well. A suitably sized output capacitor is used to maintain the low impedance up to the highest frequency of interest.

Because of the amount of feedback used, most regulators have a very low output impedance. As a result, adding a very large output capacitance does not necessarily reduce the noise as much as one might expect - or even at all. Where extremely low noise is essential, a simple resistor/capacitor filter can be added, but at the expense of load regulation.

There are a number of terms that are used to describe the performance of any regulator. These are listed below, along with brief explanations.

There are obviously many more, such as power dissipation, maximum current, current limiting characteristics, etc. These are dependent on the type of regulator, and the specifications and terminology can vary widely. Many of the parameters are far too complex to provide a simple 'figure of merit', and graphs are shown to indicate the transient performance (load and line) and other information as may be required to select the right part for a given task.

One special family of regulators are called LDO (low drop-out) regulators. Where a common regulator IC might need 2 to 5V input/output differential, an LDO type will generally function down to as little as perhaps 0.6V between the input and output. These are commonly used in battery operated equipment to maximise battery life. Some of these devices are also very low power, so there is a minimum of power wasted in the regulator itself.

Few (if any) regulator ICs presently available have poor performance. While there may be 'better' types one can use, this does not mean that a better (more expensive) regulator will cause a system to sound any different.

Very few audio applications really need anything more than the traditional fixed voltage regulators, such as the 7815 (positive) and 7915 (negative). Yes, they are somewhat noisy, but the noise is generally (but not always) immaterial when the circuit is opamp based. See below for the reason.

A 7815 (or 7915) has a typical output range of from 14.4V to 15.6V, so expecting the voltage to be exact is unrealistic. The load regulation (i.e. the change in output when the load current is changed) is anything from 12mV to 150mV when the load current is changed from 5mA to 1.5A. For this test, the input voltage is maintained constant.

Ripple rejection is quoted as a minimum of 54dB to a typical value of 74dB. These figures can be bettered by using the LM317/337 variable regulators. They have lower noise and better ripple rejection than the much older fixed regulators, but in most circuits it makes no difference whatsoever. Claims that there is some 'quality' of DC that is somehow (magically?) audible are usually nonsense. The use of super regulators is usually unjustified for any opamp circuit, and has marginal justification at best even with very basic discrete designs. For lowest possible noise, a cap is required from the adjustment pin to earth (ground), and this should have a discharge diode fitted between the adjust and output pins (both oriented appropriately for polarity of course).

There are quite a few other regulator types on the market, but the National Semiconductor types seem to have the lion's share of the market as far as normal retail outlets are concerned. Not that there is anything wrong with them - they perform well at a reasonable price, and have a very good track record for reliability. While one can obtain more esoteric devices (with some searching), many of the traditional manufacturers are concentrating on switching regulators, and don't seem to be very interested in developing new analogue designs.

While there are many discrete or semi-discrete regulators to be found in various books, websites (including this site) and elsewhere, they are usually only ever used because no readily available IC version exists. An example is the ESP P96 phantom power regulator - this design is optimised for low noise and the relatively high voltage needed by the 48V phantom system. Regulation is secondary, since the phantom power voltage specification is quite broad. It is still quite credible in this respect, but it has fairly poor transient response, which is not an issue for the application.

LDO (low drop-out) regulators are becoming much more popular, because people like to be able to have regulated supplies from a battery supply. Users would also like to be able to use batteries down to the last drop (as it were). The low dropout regulator achieves this by using a PNP (or P-Channel MOSFET) series pass transistor (for a positive regulator), and the voltage differential between input and output can be less than 0.6V, compared to a couple of volts or more for a traditional regulator. There are some caveats when using LDO regulators though, because they are far less stable than their conventional counterparts.

The series pass transistor operates with gain because it's not an emitter/ source follower. This introduces additional output impedance, so the external load has more influence than with a conventional regulator. Capacitance, ESR (equivalent series resistance) and inductance at the output pin have to be within specified limits to prevent oscillation, so there is some loss of flexibility. A normal 78xx regulator can usually have anything from 100nF to 10,000µF across the output and it will work perfectly happily regardless, but no such liberties can be taken with the LDO version.

In many cases, just substituting the output cap with a another having a lower ESR can convert a stable and happy regulator into an RF oscillator. It is essential to get the data sheet for any LDO regulator and make sure that you follow all recommendations to the letter. Instability often results if the output cap isn't large enough or has an ESR that is too high or too low. LDO regulators are not inherently stable, and manufacturer data sheets must be used to determine the stability criteria.

Virtually all LDO regulators rely on the ESR (and perhaps ESL - equivalent series inductance) of the output capacitor to correct the phase response of the internal circuitry to ensure stability. This is a complex area and will not be covered in any detail here. Also, be careful with selection. Many LDOs are designed for low input voltages, and are generally used for providing low voltage (1.2V - 3.3V) supplies for microprocessors and the like. For the most part, they are not suitable for use providing typical opamp voltages (±15V for example). Negative versions are available, but making a selection of either positive or negative parts is difficult because there are so many different types.

Because noise is not just 100/120Hz supply ripple, we also need to look at regulator (wide band) noise. Common 78xx/79xx regulators have pretty good ripple rejection, but are usually quite noisy. The noise is predominantly high frequency, and is at frequencies where opamp PSRR is nowhere near as good as it is for low frequencies. As a result, some opamp circuits may produce audible noise that comes directly from the power supplies. In general, this is a non-issue and will not cause any problems at all, but for those occasions where noise is audible, the fixes are quite simple.

One solution is to use adjustable regulators such as the LM317/337. These are much quieter than the 78/79 series ICs, and the difference may be audible, especially in high gain circuits. As an example, the original version of the ESP P37 discrete preamp has a PSRR of around 31dB for wide band noise. 10mV of supply noise will result in 297µV of output noise. This may be audible under quiet listening conditions, although few (if any) regulators will be that noisy. 10mV was a convenient reference level - the data sheet for an LM7815 says maximum noise level is 90µV. In reality, most off the shelf regulators will be fairly similar.

If the noise floor is audible, then two possible causes need to be addressed. If it's caused by the opamp itself, then replacement with a different (low noise) type is the only solution. If the source is power supply noise, the easiest way to get rid of the vast majority of this noise is simply to use a simple RC (resistance, capacitance) filter at the output of the regulators.

Using 10 ohm series resistors from the supply with 1,000µF caps to ground for each polarity, noise is almost completely eliminated. The supply voltage is reduced by only 100mV for each 10mA of current drawn which will not affect any audio circuit. This is a far cheaper option than using a relatively expensive discrete power supply that requires exotic opamps and costly 'audio grade' capacitors and other components. The noise can be expected to be reduced by at least 60dB with this simple filter. High frequency noise (the most intrusive, and least affected by the opamp's PSRR) is affected the most by the filter. Note that it is pointless adding a large cap without a series resistor - the output impedance of most regulators is so low that it will have almost no effect.

High frequency noise from the regulators can be reduced by adding a capacitor from the ADJ terminal to earth/ common. It is then essential to add a diode from ADJ to the output to discharge the cap should the output be shorted. There are very few opamp circuits that will genuinely benefit from the extra filtering though.

The easiest way to make a super regulator is to use two regulators in series, with the first one at a higher voltage than required at the output. For example, a 15V output might have an input to the second regulator of perhaps 22V, and additional filtering (as shown below) may be added as well. While ripple will be reduced to virtually nothing at all, will doing any of this improve the sound? Almost certainly, the answer is "No". While many have claimed superior performance (with the usual superlatives and a complete lack of any objective evidence), it is unlikely that anything changed. Note that only the positive side is shown in Figure 4. Refer to the article for complete details.

One popular version is the Jung 'Super Regulator' (a modified version is shown below). While I have no doubt whatsoever that its performance is exemplary, the level of performance achieved is simply not necessary in most audio circuits. The general arrangement is a pre-regulator (an LM317), followed by an opamp based error amplifier, precision reference diode and series pass transistor. In other words, two cascaded regulators. Although it also allows remote voltage sensing in some versions, this is of little use when the power supply and the audio boards are only 100mm or so from each other. The use of a fast opamp and optimised circuitry will certainly give excellent transient response, but no normal audio signal has a high enough frequency to make transient response an issue.

Superlatives abound on many sites describing the circuit. Some people have noted that it may be prone to oscillation (so has to be made slower) in some configurations, and I have received emails from people complaining that this has happened (and no, I don't know why people would complain to me about someone else's circuit). Meanwhile, no-one seems to have noticed that the vast majority of opamps being powered don't actually care one way or another if the DC has 1 or 100µV of supply noise.

Naturally, since the Jung version is popular, others have jumped on the bandwagon. As a result there are several versions of alternative super regulators, many of which will be prone to oscillation, and will almost certainly not provide any measurable improvement in audio performance ... unless they do oscillate of course. Predictably, regulator oscillation can never provide a positive outcome in any audio circuit.

For anyone who wants to make a super regulated system, a far cheaper option would be to use a pair of cascaded LM317s (for example, a pair of P05 boards). At an output current of 150mA, the first regulator reduces the input ripple from 680mV peak-peak (206mV RMS) to less than 470µV P-P (143µV RMS), a reduction of 63dB. The following filter (R3, C3) reduces this to 123µV P-P (42µV RMS), another 11dB. The second regulator reduces this to 116nV P-P (42nV RMS), 60dB - at least according to the simulation. The total is almost 134dB ripple rejection, but a single misplaced track or wire could easily degrade that badly.

Remember that this is the voltage on the power supply, and the PSRR of any opamp circuit hasn't been considered yet. Discrete circuitry, and especially low feedback designs, are less tolerant of supply ripple, so some circuits of this type may benefit from the additional ripple filtering offered by a cascaded regulator circuit. However, unless you are amplifying exceeding low level signals, it is unlikely that any of the above will be necessary. Add the 70dB PSRR of any reasonable opamp, and the expected output noise is so far below the noise floor of any system that no further improvement will yield any audible difference.

It is also worth remembering that even straight wires have resistance and inductance, so even if transient response and regulation were perfect at the power supply, 100mm of wire will instantly introduce losses. Remote sensing can be used to counteract this, but for an audio circuit ... complete overkill to achieve no useful purpose.

Figure 4A shows my simplified version of a Jung (et al) 'super' regulator. There are many variations on the basic theme, but many are similar to the original. One notable common part is the D44H11 series pass transistor. This is described as a fast switch, and has a rated f_T of 30-50MHz (speed depends on the manufacturer). The opamp (AD825) is also common in many alternative versions, as it is also very fast and can provide more output current than many other opamps. It only appears to be available in a SMD package, and is not a cheap part. Other suitable devices include the AD797, which has lower noise but is considerably more expensive. The LM317 is set up so that its output voltage is about 2.6V higher than the final regulated output. I have eliminated the E96 resistor values that are often specified (499 ohms for example) because they are simply not necessary in this application. 1% or 2% metal film resistors are expected regardless, not for accuracy but low noise.

The output voltage is set by the voltage divider using R6 and R7, and all significant voltages are shown on the circuit. R6 is bypassed by C4 so the AC gain of the circuit is unity, ensuring minimum noise. I haven't built one of these, but a simulation shows that it has extremely low output impedance, but like most regulators is it still unipolar. It cannot sink current from the load, but this is rarely a requirement for any internal power supply. All 100µF capacitors should be low ESR types. The opamp gets its DC from the regulated output. Note that it is quite possible that the circuit shown may oscillate, depending on the devices used, PCB layout, etc. Fast opamps can oscillate easily, and may only need a few millimetres of (unbypassed) PCB track in a supply line to introduce enough stray inductance to cause problems.

As noted earlier, there is no hard evidence to show that the use of this (or any other) 'super' regulator will affect the output from any opamp based circuit. Claims include 'better bass' and/ or 'improved soundstage', but any opamp can supply an output signal to DC, and the output is largely independent of the power supply. There is no reason to expect that having 'perfect' DC will make any audible difference ... provided of course that any comparative test is double blind. Sighted tests are fatally flawed, and while measurements may well show that the DC from a 'super' regulator has lower noise or better regulation than a simple LM317/337 regulator, that does not automatically translate to improved sound quality.

Shunt regulators have some advantages over traditional series regulators, despite their low efficiency and comparatively high power dissipation. The advantages of shunt regulators are as follows ...

The simplest shunt regulator consists of nothing more than a resistor and a zener. If designed properly, this is a very simple power supply arrangement, and offers acceptable performance for many applications. For example, the P27B guitar amplifier preamp has a pair of zener shunt regulators on the board, and these give hum free performance despite the very high gain of the preamplifier.

There are very few shunt regulators used in modern equipment. This is not necessarily a good thing, since almost no-one designs in an over-voltage crowbar circuit, so failure of a series regulator is often accompanied by wholesale destruction of the circuitry that uses the regulated supply. This is especially so with logic circuitry ... 5V logic circuits will typically suffer irreparable damage with a supply voltage above 7V.

In the two circuits shown above, it is quite obvious that the high performance circuit will outperform the simple zener. As a quick test (which is by no means conclusive, but gives a good indication), the circuits were simulated. The DC input was deliberately 'polluted' with a 2V peak (1.414V RMS) 100Hz sinewave to measure the ripple rejection of each version. The zener alone was able to reduce the ripple to 11mV RMS, a reduction of just over 42dB.

If R1 and R2 are replaced with a single 100 ohm resistor (omitting C2), ripple rejection falls to 25dB (82mV RMS ripple). This technique for ripple reduction used to be very common when people built discrete regulated power supplies. The two resistors and the 47µF capacitor form a low pass filter, with a -3dB frequency of 14.4Hz. Note that a split resistor is essential - if the 470µF cap were simply in parallel with the zener, there is very little improvement - the RMS ripple voltage is only halved to 40mV, rather than reduced to the 11mV measured using the split resistor method.

Why? Because the zener has a low impedance, and this acts in parallel with the cap's impedance. By splitting the resistance, the capacitor works with the effective impedance of the two resistors in parallel - this is much greater than the impedance of the zener, so the cap has more effect. Needless to say, a larger capacitance gives better ripple performance - doubling the capacitance halves the ripple voltage for example.

The opamp based version achieved 2.3µV RMS - over 116dB rejection. This figure must be taken with a (large) grain of salt of course - simulators and real life don't often coincide. In reality, I'd expect about 80-90dB reduction for a 'real' circuit. Please be aware that the opamp based regulator circuit is shown as an example - it is not a working circuit, and would almost certainly oscillate if constructed as shown.

For the simple zener version with full load, the zener dissipation is 440mW. This rises to almost 1.7W with no load. If a 1W zener were used, it would fail if the circuit were operated with no load for more than a few seconds. Resistor dissipation remains the same whether the circuit is loaded or not, but it increases if the output is shorted to ground. The two resistors need to be at least 1W, since each dissipates about 500mW.

The high performance version needs a 5W resistor for R3. Transistor Q2 has maximum dissipation with no load, and this will be around 3.5 Watts. Dissipation is around 2.3W with the rated load of 75mA. While the shunt current can be reduced from the 250mA used in Figure 5, performance will suffer if it falls below about 150mA. This can be reduced by using the same split resistor scheme used for the simple zener regulator, and this will improve ripple rejection performance further as well.

It's worth noting that most shunt regulator designs (whether opamp or discrete based) regulate their own supply voltage. This gives an inherent advantage, in that the supply to the circuitry is stable, thus ensuring that the overall performance is optimised without any requirement for pre-regulation.

Finally, a version that has been used by many constructors is shown in Project 37. This is a simple shunt regulator, but the zener power is boosted by adding a transistor as shown above. Note that the resistor is split, with a cap between the two. As noted in the article, noise is extremely low - 100/120Hz hum can be expected to be less than 20µV or so. I found that it was almost impossible to measure hum in the prototype, since normal circuit and test equipment noise was predominant. Although the latest PCB for P37 now uses ±15V, the regulator is still useful for those who wish to experiment.

If you need a negative version, simply reverse everything and use a PNP transistor (BD140 for example). For different voltages, you change the zener, but remember that the output voltage will be between 700mV and 1V higher than the zener voltage because of the transistor's base-emitter junction. The actual voltage depends on the current. For more information on the use of zener diodes in general, see AN008 - How to Use Zener Diodes on the ESP website.

The design of shunt regulators in general isn't difficult, but there are quite a few things that need to be calculated. The unregulated input voltage must be higher than the desired output, and this includes any ripple. For example, if the minimum voltage is 16V and the maximum 20V (4V peak-to-peak of ripple) you can't expect to get 15V output because 1V headroom just isn't enough. The minimum voltage should be not less than 25% greater than the desired output. For 15V out, that means no less than 18-19V input. Remember too that the incoming mains will vary and this has to be taken into account as well.

The feed resistance (R1 and R2 in Figure 5A) should pass a minimum of 1.5 times the maximum load current. If your circuit draws 50mA then the resistors need to pass 75mA. The voltage across the feed resistance is the input voltage minus the output voltage. You then need to work out the power dissipation of the resistors, zener and shunt transistor. Some general approaches to determining capacitor values is available in the article Voltage & Current Regulators And How To Use Them. I do not propose to explain the complete design process here - most of it is based on nothing more complex than Ohm's law.

So-called 'transformerless' power supplies can use a resistor or a capacitor to drop the AC mains voltage to something usable by electronics. The resistor approach is not covered here, because it's very rare that it will have low enough power dissipation to be usable in most cases. A capacitor provides a 'lossless' voltage drop, because it's a reactive component. While it has a very low (i.e. 'bad') power factor, these supplies are generally only used for limited output current, and the poor power factor is not an issue.

To some, the idea of making a power supply that does not use a transformer is appealing. Even relatively small transformers are bulky and heavy, and they will always radiate a small amount of magnetic interference. However, supplies that don't include a transformer are not isolated from the mains supply and are inherently extremely ~~dangerous~~ lethal.

There are several safety points that you'll see repeated here. This isn't because I like repeating myself, but to make absolutely sure that potential (sorry

) constructors don't miss them. These supplies are lethal in the wrong hands (inexperienced constructors in particular) and if my repetition only saves one life it's worth it.

These supplies are usable in a limited range of products, and they can have no direct input or output connections. This limits their usefulness somewhat, since most projects require some connection to the outside world. While isolation is possible using opto-couplers, these are often slow and not very linear, so hi-fi applications are ruled out. A remote sensor (for example) can be used, provided that the sensor, lead and connector are all fully insulated, rated for mains voltages, and have no accessible metal parts.

Where such circuits are used, they will be completely enclosed, and may have circuit functions accessed by well insulated push-buttons, infra-red or radio remote controls. Well insulated (plastic shaft) pots can also be used. Typical applications are wide-ranging, and include motor speed controllers, 'high tech' light dimmers, temperature controllers and many others. Audio is not included in any common usage.

While it would be possible to isolate inputs and outputs using transformers, no-one makes 'line level' transformers that are rated to withstand mains voltages. Even if they were available, the cost would be far greater than a small mains tranny and a basic conventional power supply.

Consequently, the applications are strictly limited to areas where the necessary inputs and outputs can be opto-isolated, or where there is no direct connection to the outside world at all. Many PIC based projects are intended for controlling mains appliances, and these can use a transformerless supply without problems. Naturally, external probes or other sensors must also be insulated in their entirety. They must withstand the full mains voltage, safely, and for well beyond the expected life of the apparatus.

Now, looking at the circuit, it is obvious that one side is referenced to the neutral, and neutral is connected to the building's safety earth or to safety earth at the local mains distribution transformer (this varies by country). Therefore, you might think that the circuit should be safe. However, the regulatory bodies in every country insist that the neutral is a 'current carrying conductor', and it is recognised everywhere that the possibility exists for active (aka live or phase) and neutral to be interchanged. This may occur in old buildings (wired before any standard was applied), or could be caused because of an incorrectly wired extension lead. Many countries have non-polarised mains plugs that may be inserted into an outlet either way.

Any one of the above makes the circuit deadly. The output becomes referenced to active, not neutral, so all connected circuitry is at mains potential. For this reason, circuits such as those shown may only be used in such a manner that no part of the power supply or its connected circuitry may be accessible to the end user. This means no connectors for input or output, and all components must be fully insulated to prevent accidental contact.

Now that the necessary disclaimers are completed, we can look at the circuit itself. The fuse (F1) is obviously intended to guard against the risk of fire, by opening if the current exceeds that expected. R1 limits inrush current, which can be very high if power is applied while the AC input is at its maximum value. R1 needs to be at least 1W, and it is intended that its value is considerably less than the capacitive reactance of C1. In some cases, R1 may be a fusible resistor, thereby eliminating the separate fuse. I consider this to be a poor protection mechanism, but it's cheap.

C1 is the actual current limiter. By using a capacitor, there is almost no lost power - capacitors used within their ratings have extremely low losses. R2+R3 is intended to discharge the capacitor when mains is disconnected, and two are used to obtain a satisfactory voltage rating. Without this, C1 can hold a significant change for several days, so anyone touching the pins of the mains plug could receive a very nasty shock. R2+R3 must be rated for the full mains voltage. It may be necessary to use 3 or more resistors in series to ensure they will withstand the applied voltage continuously.

C1 will have almost the full mains voltage across it (230V RMS for the circuit shown), and cannot & must not be a DC rated capacitor. A 400V DC cap will work with 120V mains but this is most unsatisfactory and it will fail eventually. The cap voltage should be a minimum of 275V AC if used with 230V mains. In general, it is unwise to use DC rated capacitors where high AC voltages will be across the cap - the use of AC rated components is highly recommended in all cases. X-Class capacitors are designed to be connected across the mains, and are the only type that should be used.

D1 and D2 form the rectifier. D2 must be installed to prevent C1 from charging to the peak of the mains voltage (340V, via D1). Without D2, the circuit will not work! C2 is the filter cap, and needs only to be rated at slightly above the zener voltage. A 6.3V electrolytic will be quite acceptable. Finally, D3 (a 5.1V zener diode as shown) provides regulation. The DC will have significant ripple - in the circuit shown and at 50Hz input, there will be about 325mV peak-peak of ripple on the supply. This is normally quite acceptable for a PIC circuit, provided it is not expected to perform any accurate analogue to digital or digital to analogue conversions.

Ripple can be reduced by adding a resistor (R3) between C2 and D3, but care is needed to keep the voltage across C2 within ratings. For example, a 33 ohm resistor reduces ripple to about 63mV peak-to-peak and keeps the voltage across C2 just below 6.3 volts. Figures shown are for a 220 ohm load at 5.1V - about 23mA. The available current is reduced if the voltage is increased, so 330 ohm loads are shown in Figure 6B. In general, there should be a second capacitor in parallel with D3 to allow for higher than normal peak current in the powered circuit.

A common variant of the circuit shown above is to use a zener diode in place of D1, and D3 is not needed. This reduces the component count, but the output voltage will be 650mV lower than the zener voltage, and there will be more ripple on the DC supply. A resistor and a second electrolytic cap can be used for better filtering, but the output won't be as well regulated because of the series resistor.

There is no real limit to the voltages available, but the highest voltage normally used will be around 24V or so. If you need multiple voltages, simply add series zeners as shown in Figure 6B. As shown, you have ±5V, but that can just as easily be +5 and +10V, simply by deciding which terminal is 'common'. It is critical that you understand that 'common' is NOT the same as earth/ ground. No part of the circuit can be touched safely, and the supply can only be used for fully enclosed applications with no input or output connectors that can be accessed by the end user.

One problem that's often faced is the low current available. Yes, the capacitance for C1 can be increased, but the cap will be physically large, and the cost may be prohibitive. Even the 1µF cap shown will be rather bulky, and will most likely be a pair of 470nF X-Class caps in parallel (close enough to 1µF). The inrush current also has to be considered. Because a discharged capacitor acts rather like a short circuit when voltage is first applied, the 100 ohm inrush limiter shown is the only thing that limits the current. Worst case is when the mains is switched at the peak of a half-cycle, which will cause the peak current to be 2.3A with 230V mains (an instantaneous dissipation of 530W!), or 1.2A with 120V. If the value of R1 is increased, inrush current is reduced but continuous dissipation is increased. This dissipation is real power that you pay for and is wasted as heat.

The standard circuit is full wave as far as the mains is concerned, but rectification is only half wave. The negative half cycle of the mains current is not used, so output current is limited. The maximum current you can expect depends on the voltage, but as shown above will be around 25mA for each microfarad used as C1 (at 230V input). For example, with 1µF you'll get about 25mA, or 10mA for 470nF. The available current is roughly half with 120V mains. We can do better ...

A major improvement is to use a bridge rectifier as shown above. The diode bridge means the available output current is increased, and smoothing is easier because the ripple frequency is double the mains frequency. The disadvantage is that the output common isn't referred to the neutral which is a limitation if the circuit is intended to drive a TRIAC for mains switching (for example). In most cases this is the version that should be used, and it can supply enough current to operate a relay. The circuit shown in Figure 6C can supply up to 55mA, compared to ~28mA for the supply shown in Figure 6A.

By using a bridge rectifier, you make full use of the capacitor current. The current you can get depends on the capacitance and supply voltage & frequency. It can be calculated easily enough by using the formula for capacitive reactance ...

If you need 70mA from 230V mains, XC is 3.2k, and the capacitance is 1µF. If you don't need that much current, 470nF is a standard value, and will provide 34mA. Always aim for a bit more current than you need because the mains voltage will vary. The only capacitor recommended is an X-Class type, as these are rated for mains voltage. Over time, expect to replace the cap, as its value will decrease depending on the 'hostility' of the mains in your area. Every time there's a voltage 'spike' (from other equipment, lightning, etc.) the cap will lose a tiny bit of its metallisation, gradually reducing the value (this is designed into X-Class capacitors to ensure protection against fire).

Transformerless power supplies such as the one shown are fairly efficient. The mains current draw is predominantly capacitive, so while you may draw (say) 5.5VA from the mains, the power is less than 600mW (assuming an output of 24V at 25mA). The zener diode is very important. Without it, if the load is less than the calculated 25mA, the voltage will rise. With no load it can (theoretically) rise to 325V DC, but the filter cap will fail well before that.

In general, if you need to drive a relay (for example) use a 24V coil, as that requires less current than lower voltages. That means your supply voltage will be 24V, and it doesn't matter if that falls when the relay has activated, as it won't drop out until the voltage falls to around 2-5V. Your circuitry has to be able to tolerate the voltage change of course, but that's usually not hard to achieve.

As should be quite apparent, this type of power supply is completely useless for general purpose work and cannot be used for audio because of the serious electrical safety issues. There are only a few applications for circuits such as this, and these are generally control systems and the like. Remember that all external connections, probes, etc, must be isolated to the standards required for the full mains voltage.

Instead of a capacitor to limit the current, you can use a resistor instead. However, this technique results in a high dissipation in the resistor - almost 4.6W for a 20mA output from 230V or 2.4W at 120V. The heat has to be removed somehow, and that's difficult when the power supply and all other circuitry must be fully enclosed for electrical safety. There are some applications where a resistive circuit is the only one that will work properly, but in general it's not viable.

It has been suggested several times that the 120V mains (as used in the US and Canada) could be rectified and used directly as an amplifier power supply. This will give an effective supply voltage of about ±85V with the use of a suitable splitter circuit. While the idea seems plausible ...

The problem is that the entire amplifier is at mains potential, and all inputs and outputs have to be isolated using transformers. This scheme used to be quite common for radio receivers and TV sets - they were commonly referred to as 'hot chassis' sets. Because it's relatively easy to isolate the antenna connection with a high voltage capacitor, these sets were popular because they were cheaper to make. None that I know of ever had auxiliary audio (or video) inputs or outputs, as these would have to be transformer isolated.

Figure 6D - Improved Transformerless Power Supply + DC-DC Converter

If an isolated DC-DC converter is added, you can then refer the output to earth/ ground. There are many different SIP (single inline pin) converters available, and they can be obtained now reasonably cheaply from major suppliers. The isolation working voltage must be a minimum of 1,000V (1kV) - note this is the working voltage, not the isolation test voltage, which will be 2kV or more! Great care is needed to ensure that there is adequate creepage and clearance between the live (mains) side and the low voltage output.

These little converters are fairly efficient, and can supply 50mA or more (depending on the output voltage). Most are rated for 1 or 2 watts at most, but you'll find that unless you use one with an input voltage of 24V, you probably will be unable to get the rated current output. Have a look at the website of your preferred supplier. Input and output voltage can be selected to suit your needs. Remember to verify that the working voltage differential is at least equal to your mains voltage.

This arrangement is only suitable where (very) low current is required, otherwise, there are AC-DC modules available from many suppliers that only require an AC input (usually 80-277V, 50/ 60Hz), and have either single or dual outputs at the desired voltage(s). While they are larger than the DC-DC converters, they are actually not much bigger than a 1µF X-Class capacitor by itself, and provide the smallest available power supply you can get. However, they all require EMI (electromagnetic interference) filters at the mains input, or unacceptable interference may be experienced. Some complete 2-3W AC-DC converters are less than AU$25 each, but others may be much more expensive. You can also get small (Chinese made) AC-DC switchmode supplies from various on-line sellers, and they are usually very cheap. Whether they are any good or not is another matter. Some I've used are actually very good, others not. (See section 8 below for another solution).

With all these nasty limitations, a chap called Stan D'Souza at Microchip Technology decided that there had to be a way to make the circuit 'safe'. Thus, in 2000, a technical bulletin (TB008) was issued that claimed to overcome the inherent safety issues of the traditional transformerless power supply. According to Stan, his circuit could be used just like any normal transformer based supply, but without the expense of a transformer. To state that it wasn't thought through properly is a gross understatement!

What he (and various others at Microchip) completely failed to recognise is that the circuit described violates the wiring rules of every developed country on the planet! No-one else before or since has ever suggested such a dangerous circuit. The circuit is shown below - this is not the exact same circuit as described in TB008, but is based (for clarity) on that described in Figure 6A. The overall concept is identical.

At first glance, it seems to be alright. Look closely! It uses the earth pin of a 3-pin power connector as the return path for the circuit - this is not allowed in any country that I know of. The earth pin is for safety earth, and is intended to carry fault current away from the appliance to prevent electrocution. The earth (ground) pin must not be used as a current-carrying conductor - ever! All current-carrying conductors must be insulated from earth and/or chassis with wiring suitable for the mains voltage used. No country's wiring rules will consider the neutral conductor to be 'safe', because there will always be situations where active and neutral are swapped over - perhaps because of very old wiring, inexperienced persons failing to appreciate the difference, incorrectly wired extension leads, etc.

Next, there is a fuse joining earth and neutral. Again, this is not permitted under any wiring codes. By joining the earth and neutral, it will instantly trip any electrical safety switch (aka earth leakage circuit breaker, ground fault interrupter, core balance relay, etc., etc). Some countries use what is called the MEN system (main/ multiple earth neutral) albeit by a different name, and a link between the incoming neutral conductor and the earth (safety ground) stake is permitted (or required) at one location per installation. In other countries there will often be no connection between earth and neutral at the premises, as the connection is made at the distribution transformer. While it is possible that the rules elsewhere might allow multiple connections, the connection shown will never be allowed in any appliance. There are very good reasons for this, and the following is only one of many possible scenarios ...

What happens if the active and neutral in a wall outlet are reversed (and the earth is connected)? Firstly, the fuse will blow (violently), and the loud bang and bright flash will give the poor user a terrible fright. This in itself is unlikely to be deadly. It is after the fuse has blown that things become really dangerous, because if plugged into another outlet (provided the earth connection is sound and the safety switch doesn't operate), the circuit will continue to work.

Most householders will be baffled - "Gee, it just blew up, but everything still works!" The next thing will be to try it in the original outlet again ... "Hell, why not, it still works." But remember, this outlet has active and neutral reversed, so it won't work in that outlet - there is no connection to active. The poor user is now flummoxed, so chucks the (whatever it might be) in a corner and forgets about it.

In the US, Canada and some European countries, it is quite common for appliances to have a 2-pin non-polarised mains plug, with no earth pin. If the owner of this particular appliance with its 'Cheap Death' power supply decides to simply replace the 3-pin plug with a 2-pin version, the real fun can start. Whether the fuse is intact or not is more or less immaterial, because only one of the two ways a 2-pin plug can be inserted is 'safe'. Should the wrong choice be made (and the poor user has no knowledge that there is a 'safe' and 'unsafe' way to plug the supply into an outlet), the entire circuit is now at the full mains potential. Anyone touching any part of the circuit - chassis, connectors, etc. - is now connected directly to the mains via the capacitor. The cap is capable of passing more than enough current for a fatal electric shock.

Even assuming that nothing untoward happens (and there is no installed safety switch), if there is a significant load on the branch circuit where this monstrosity is connected, there is a very real possibility that the fuse will blow because of the potential difference between the neutral and earth connections. An electric kettle or a heater can quite easily elevate the neutral lead by a couple of volts with respect to earth, and the fuse will blow. Any pretense of 'protection' afforded by the fuse is now gone.

There are many other 'what if' possibilities that I urge you to explore, any one of which could result in the chassis becoming live, resulting in death or injury. Remember that it doesn't matter how unlikely a given scenario may seem to be, it will happen somewhere, sometime, if there are enough devices using the technique available. A one in a million chance becomes a certainty if there are a million users.

Because of the extreme danger posed by the circuit scheme, I contacted Microchip's technical support group in 2008 with the following information (both messages are verbatim, including errors, grammatical mistakes, spelling, etc.) ...

A few days later, the following was received (with no name or direct contact details). That the response is unsatisfactory is to put it very mildly. This was Microchip's 'proposed resolution' (again, the text is verbatim, but the yellow highlights are mine).

Unbelievable! Remember, this is verbatim, replete with spelling and other errors. Although I immediately posted a response through the tech support contact form, no further reply was ever received - Microchip's tech support people seem to think the issue is 'resolved', simply by stating that they see no issues in the circuit. A far as the person who examined the problem is concerned, it is perfectly alright. What makes this far worse is that it was claimed that "engineers would be viewing it" - well, TB008 can be found all over the Net. Because it was produced by a large and well known company, a great many beginners will assume that it must be safe, and that any criticism is unfounded and has no credibility (at least by comparison). There was even one site that had a link to the article for use by schools!

Interestingly, Microchip has also released AN954, which also describes a transformerless power supply. It has many highlighted warnings throughout the text, and in a bizarre twist of logic the evil TB008 is cited as a reference. The versions described do not attempt to use safety earth as a current carrying conductor, and quite correctly have no connection between earth and neutral. TB008 seems to have been buried in the Microchip website, and I was unable to find the original. An admission of wrongdoing would have been nice, but I suppose that's too much to ask for.

Over nine years have passed (as of this update), and no satisfactory response from Microchip has ever been published. That really isn't good enough.

Where a physically small power supply is required for a project (including audio, but not necessarily for true hi-fi use), one can use the intestines of a miniature 'plug-pack' (aka 'wall-wart') SMPS. Although only small, some of these are capable of considerable power, but installation is not for the faint-hearted. Quite obviously, the circuit board must be extremely well insulated from chassis and protected against accidental contact when the case is open.

The advantage is that the project does not require an external supply. This is often a real pain to implement, because there is always the possibility that the wrong voltage or polarity can be applied if the external supplies are mixed up (which is not at all uncommon). The disadvantage is that the unit now must have a fixed mains lead or an approved mains receptacle so a lead can be plugged in.

The photo in Figure 8 shows a typical 5V 1A plug-pack SMPS board. As removed from the original housing, it has no useful mounting points, so it is necessary to fabricate insulated brackets or a sub-PCB (made to withstand the full mains voltage) to hold the PCB in position. Any brackets or sub-boards must be constructed in such a manner that the PCB cannot become loose inside the chassis, even if screws are loose or missing. Any such board or bracket must also allow sufficient creepage and clearance distances to guarantee that the primary-secondary insulation barrier cannot be breached. I shall leave the details to the builder, since there are too many possible variations to consider here.

This arrangement has some important advantages for many projects. These supplies are relatively inexpensive, and the newer ones satisfy all criteria for minimum energy consumption. Most will operate at less than 0.5W with no load, and they have relatively high efficiency (typically greater than 80% at full load). The output is already regulated, so you save the cost of a transformer, bridge rectifier, filter capacitor and regulator IC.

The SMPS pictured is a 5V 1A (5W) unit, and for most PIC based projects this will provide more than enough current. Consider the safety advantage compared to a transformerless supply - the finished project can have accessible inputs and outputs, and is (at least to the current standards) considered safe in all respects. Personally, I would only consider it to be completely safe if the chassis is earthed. However, it is legally allowed to be sold in Australia, and we have reasonable safety standards for external power supplies. They are 'prescribed items' under the Australian safety standards, meaning that they must be approved before they can be sold.

There is no more effort required to install a supply such as this instead of a transformerless supply, and at least you can work on the secondary side without having to use an isolation transformer. While it is more expensive, how valuable is a life? Far more than any power supply, and that's for certain.

For a variety of reasons, you may find that the regulator you need is only available in the positive version. This may be because it's a high current type, and for reasons best known to the manufacturer it was decided that people don't need a negative version. For example, you may be able to get LT1038 regulators (up to 10A, but now obsolete), but there is no negative equivalent. Much the same applies for other IC manufacturers, and it's now difficult or impossible to find complementary high current regulators.

You may find yourself in the same position if you use off the shelf switchmode supplies. These usually only have a single output, and while dual output types might be found, they will not be as cheap as the single output versions. There are many places where symmetrical (positive and negative) supplies are needed, but your options can be very limited if you need high current.

Provided the transformer has separate windings (not a single winding with a centre tap), you can simply build two identical positive regulators and wire the outputs to get positive and negative outputs. This may work out to be a cheaper and batter option than trying to make separate positive and negative regulators. It may be counter-intuitive, but a regulator doesn't have a designated 'common' connection, other than that created when the supply is wired. It makes no difference whatsoever if the regulated (positive) output is deemed to be the output or common. Figure 9 shows how a positive regulator can be deemed to be a negative regulator, simply by moving the 'common' connection.

The above certainly looks 'wrong', but it's not. The regulated output has no sense of 'positive' or 'negative'. As long as all parts operate with their correct polarity, any part of the circuit can be connected to earth/ ground without altering the performance of the regulator in any way. There are several places where it simply wouldn't make any sense, but as shown the negative output voltage is smoothed and regulated in exactly the same way as it would be with a dedicated negative regulator.

The arrangement shown above can be used with any type of regulator. You can wire a pair of 20A switchmode supplies the same way, and they can have the same or different voltages. The only requirement is that the outputs are floating - with no fixed connection from either supply line to earth/ ground. This lets you connect either supply line to the system common (earth), or you can build a 'stacked' power supply, providing (for example) +12V and +24V. Any voltages can be used, but make sure that the output current will always be within each supply's ratings if you build 'interesting' combinations.

Unfortunately (and as you will find if you look), high current linear regulators are now hard to get, and those you find are likely to be insanely expensive. It's now expected that all high current applications will be met by using switchmode regulators, but for many applications the switching noise will be intrusive and they are not considered an option for things like mixers and other audio applications where large numbers of opamps are used. The decision now is whether to go back to using discrete regulators as discussed briefly above, or to use multiple smaller 3-terminal regulators, with each pair of regulators powering a single section of the circuit. Linear regulators can be 'boosted' by using external transistors, but you lose the inbuilt current limiting that's provided in most 3-terminal devices. This is shown below.

It's certainly not as easy as it once was, but there are ways to get around any limitations you may face. It just requires you to have a greater understanding of the principles, and (almost) anything becomes possible.

As noted above, high current linear regulators are no longer readily available. However, it's fairly easy to boost the current from a smaller regulator IC to get as much current as you need. The disadvantage is that the external pass transistor(s) have no current limiting, and thermal shutdown only works on the regulator IC, not the external transistors. While it's certainly possible to add both current limit and thermal shutdown, this adds complexity to a circuit that was once as simple as a single TO-3 regulator IC plus a couple of support components. Very basic current limiting may only involve a couple of diodes and a resistor as shown - D8 and D9 clamp the voltage across R1 to ~1.3V, so Q1 enters constant current mode at about 2.7A and the regulator will then be able to limit the current it supplies. It's not perfect, but it does work. Add another diode in series with D8 & D9 to increase the output current. Three diodes limits the current to about 4.5A.

Figure 11 shows how to add a current booster with rudimentary current limiting. I've only shown the positive regulator, but a negative version is much the same. You only need to change the transistor for an NPN type (and a negative regulator of course). Diodes and capacitors have to be reversed as needed for a negative version. The component values shown are representative only. Only one series pass transistor is shown, but two or more can be used if needed. If transistors are operated in parallel, a resistor must be used in the emitter circuit of each. 0.22 ohms is generally satisfactory.

The circuit works by means of Q1 sensing the voltage across R1, which will typically be around 2.2 ohms and will dissipate less than 1W. At low current there's very little voltage across R1, so Q1 remains off. As the current is increased, the voltage across R1 increases and Q1 will turn on just enough to maintain the preset output voltage. The output voltage is set by the regulator, and the transistor acts only as a current booster. If the regulator is set for 12V output, Q1 will dissipate around 35W at an output current of 5A. The regulator IC will typically pass about 650mA, and will dissipate about 5W. Naturally enough, the dissipation of Q1 and U1 will depend on the regulation of the transformer as well as the load current. The arrangement shown will work with a wide range of different regulator ICs (including fixed voltage types) and external series pass transistors.

Without D8 and D9 there is no form of overload protection, and care must be exercised when this type of supply is used. For example, a short at the output will almost certainly kill the series pass transistor, and a fuse is of no use because the transistor will blow first. In critical applications, additional circuitry will be needed. This may include more complex current limiting, over-temperature sensing and perhaps an over-voltage 'crowbar' to save the load circuits should the regulator or series pass transistor fail. None of these functions are especially difficult to add, but they do increase overall complexity and component count.

This arrangement has been used in many circuits (and products) from all over the world. It appears to be in decline now, because all new high current designs use a switchmode buck converter. They are far more efficient than a linear design, but are only suitable where their noise will not cause problems. We can expect linear regulators to remain the circuit of choice for low noise applications for many years to come.

Decisions, decisions. The main purpose of this article is to provide some general information about small power supplies, regulation, their application and potential dangers. There is no doubt that the traditional transformer based supply is the safest. It is extremely easy to ensure that no live connections are accessible, often needing nothing more than some heatshrink tubing to insulate joined wires. Note that if possible, two layers of heatshrink should be used to provide reinforced insulation over joined wiring.

A transformer has full galvanic isolation and requires little or no EMI filtering, leakage current is extremely low, and a well made transformer based supply is so reliable that it will almost certainly outlive any equipment into which it is installed. While certainly not the cheapest option, a transformer provides a reasonable attenuation of common mode mains noise, and the final supply can be made to be extremely quiet, with virtually no hum or noise whatsoever.

The next best option is a modified plug-pack SMPS or a purpose built chassis mounting SMPS. These are useful where high efficiency is needed, along with very low standby power requirements. They are rather noisy though, and the full range of voltages is not available. There are few (if any) ±15V SMPS available for example, so powering preamps and other low power audio equipment will be easier, quieter and ultimately cheaper with a transformer.

As a last resort, a transformerless supply can be used, but only where the current drain is low (typically less than 25mA or so), and only where there is no possibility of contact with any part of the connected circuit. There is no such thing as a 'safe' transformerless power supply, and they are potentially lethal. There are so many limitations and so few advantages to this approach that IMO it is usually a pointless exercise, unless one has a mains powered appliance that needs a low current supply that can remain completely isolated from contact with the outside world.

The high current option described in Section 10 is the odd man out here - after all, the title of the article is 'Small Power Supplies'. Nevertheless, it's sufficiently useful that it warranted inclusion, especially since it's now quite difficult to get high current IC regulators.

Change Log: Page published and copyright © 12 May 2008./ Updated 18 May 08 - added shunt regulator section and additional reference./ 04 Jun 10 - removed Microchip link to TB008, added LDO info./ Jun 2013 - minor changes./ Sep 2015 - added Figure 6B, Figure 11 and text./ Apr 2017 - Added Fig. 4A & text./ Dec 2017 - added Fig. 6D & text.

	Parameter	Explanation
	Load Regulation	A percentage, being the change of voltage for a given change of output current
	Line Regulation	A percentage, being the change in output voltage for a given change of input voltage
	Dropout Voltage	The minimum voltage differential between input and output before the regulator can no longer maintain acceptable performance
	Maximum Input Voltage	The absolute maximum voltage that may be applied to the regulator's input terminal with respect to ground
	Ripple Rejection	Expressed in dB, the ratio of input ripple (from the unregulated DC supply) to output ripple.
	Noise	Where quoted, the amount of random (thermal) noise present on the regulated output DC voltage
	Transient Response	Usually shown graphically, shows the instantaneous performance with changes in line voltage or load current

Small, Low Current Power Supplies - Part I