|
|
| Elliott Sound Products | Bench Power Supplies |
Main Index
Articles Index
A bench supply is one of the most useful pieces of test gear you will ever own. Building one intended for testing preamps and other low voltage, low current equipment is one thing, but making one that's suitable for testing power amps is another matter altogether. In reality, it's so difficult to get right that the likes of the late Bob Pease recommended to his fellow engineers and others that they don't even try. His advice was to buy one from a reputable supplier, and not put yourself through the grief of spending many hours building one, only for it to blow up the many expensive parts used to build it [ 1 ].
In many ways, it's hard to disagree, and doubly so if you want to get voltages of more than 20V at a couple of amps. These days, the problem is doubled, because to be truly useful, the supply needs to be dual tracking, with both positive and negative supplies, with an output voltage that can be varied from zero to perhaps 25V or so. It ideally needs to be capable of at least 3A output, and with current limiting so you don't kill the supply the first time the output leads are shorted together (and that will happen!).
In essence, there's actually not that much difference between a power supply and a power amplifier, except that a power amp has to source and sink current, while a power supply only has to source current to the load. However, where a power amp will be subjected to fairly high dissipation every so often, a power supply has to be capable of providing perhaps 3-5A output into a short circuit, and not fail. This is a great deal harder than it seems.
Consider a supply that can provide 40V at 5A, but is set for an output voltage of perhaps 1-2V and a current of 5A. The internal voltage will be around 50V, so there's nearly 50V across the regulator transistors, 5A of current, resulting in a dissipation of 250W. This might continue for hours at a time or only a few minutes, but that doesn't mean that you only have to allow for a few minutes, because one day you will need 1-2V at 5A for an hour or more.
No-one ever knows exactly what they'll do with a decent power supply until they have one, and it will end up being used to power amplifiers during testing, charging batteries, measuring very low resistances, or any number of other possibilities. I know this because that's what I do with mine (which I built many, many years ago, but it only provides ±25V at up to 2.5A). I've lost count of the number of times the thermal overload circuit disconnected my load, even with a fan for forced air cooling.
It's commonly accepted that bench supplies should be regulated, and herein lies the problem. Regulation adds complexity and can create stability issues that vary from merely vexing to intractable. No-one wants a power supply that oscillates, nor does anyone want a power supply that kills the device being tested (or charged, measured, etc.). In reality, regulation (or at least 'perfect' regulation) isn't essential. Most power amplifiers don't use regulated supplies, and nor do many other high-current loads. You need to be able to adjust the voltage, and it should be reasonably stable, but ensuring that the output voltage only changes by a few millivolts under load is not needed for most applications. It might make you feel better if the supply has perfect regulation, but your circuits mostly won't care.
Current limiting is another matter though. Ideally, when first powered, your latest project needs to be protected in case there's a fault. Like voltage regulation, the current limiting function needs to be adjustable, but it's rarely necessary for it to require extremely accurate current regulation. If we accept that very accurate voltage or current regulation is not essential, that simplifies the design and makes it a great deal easier to build and get working with the minimum of fuss.
Few people want to mess around for ages trying to perfect a regulator that wants to oscillate, and this will be the case if 'perfection' is the goal. If that's what you really do need, then I must agree with Bob Pease absolutely - buy a commercial supply from a reputable manufacturer. However, you'll likely be up for some serious money if you need dual tracking, high voltage (over 30V) and high current (5A or more).
A generally useful supply will have dual outputs, variable from 0 to 25V or so, with adjustable current limiting. Ideally, it will let you use the two outputs in series, allowing a single supply variable from 0 to 50V. 5A output is useful, but not essential. If you use it for testing DIY audio equipment (preamps, active crossovers, power amplifiers, etc.), then you can verify that the DUT (device under test) functions as expected, has no shorts or other major faults, after which it can be confidently connected to the intended power supply. It's uncommon for any competent design to fail with its 'real' power supply if it's been tested at a lower voltage, using a supply with current limiting that protects against damage if there is a problem.
An expansion of the 'basic' power supply is something called an SMU (source-measure unit). These are usually high accuracy, microprocessor controlled supplies, and are able to source and sink current of either polarity. Most supplies only source current to the load, but an SMU can also be used as an 'active load', typically for power supplies or other equipment being tested. These are also known as '4-quadrant' power supplies, meaning they are designed to source or sink current of either polarity. Fortunately, this is not a requirement for basic testing, and is mentioned only in the interests of completeness. I do not propose to cover these supplies in this article.
Please note that this is not a construction article. Although it does show schematics, these are primarily for demonstration purposes, and there is no guarantee that they will function properly as shown. While they have been simulated, this only indicates that the underlying principles are sound, but it does not mean that the circuit will perform as expected in 'real life'. While the circuits described do look as though they will function well, this has not been verified by building and testing them!
It's not an accident that there aren't that many DIY projects for bench power supplies. Most people come to the realisation fairly quickly that it's a very expensive exercise, and that getting a fully working, reliable supply that does exactly what you need is not a trivial undertaking. The circuits shown here are for inspiration, and are provided mainly to give you an idea of the complexities involved - even for apparently simple circuits.
The primary function of any bench power supply is voltage regulation, but current regulation is also very useful. Both are described below.
The first regulated power supplies used valves (vacuum tubes), with a gas discharge regulator as the reference voltage. Predictably, they weren't very good because of the limited available gain available. A few basic examples are shown below, with the opamp version being a fairly good analogue to the modern 3-terminal regulator ICs. These all suffer from a problem that makes them (generally) unsuitable for a bench supply - they can't get down to zero volts output.
When testing something that's just been built, it's important to be able to start with a very low (preferably zero) voltage, and monitor the current as the voltage is increased. If you see the current climbing rapidly with a supply voltage of only a volt or so, you know there's a problem. Including current limiting (covered a little later) means that fault current can be kept to a value where it's unlikely to cause damage.
Figure 1.1 - Basic Voltage Regulator Topologies
The series pass device is V1/ Q1, and the controlling element is V2, Q2 or U1 (valve, transistor and opamp respectively). The voltage reference for the valve circuit is a gas discharge tube, and these typically had a voltage of around 90 volts (depending on the device, voltages from 70V to 150V were available [ 5 ]). The transistor circuit uses a zener diode, and the opamp circuit is shown with an external reference. Feedback is used in each case, and VR1 lets you set the voltage to the desired value. These are the basic versions of a regulator in each case, and there are many variations in practice.
The feedback is arranged so that if the output voltage falls (due to a load being connected for example), the controlling device ensures that the series pass element can pass the extra current needed to supply the load at the desired voltage. The ability of any of the circuits to maintain the desired voltage is called the 'regulation', expressed in percent. For example, if the voltage falls by 1% when the load is connected, that forms the specification for the regulator. Higher gain in the control and series pass devices means better regulation.
There's an extra transistor and resistor in the opamp version. 'Rs' is a current sense resistor, and Q2 is the current regulator transistor. If the current is such that the voltage across Rs is greater than 0.6V, Q2 turns on and 'steals' the base current from Q1 (provided via R1). This is the most basic form of current regulation, and it works surprisingly well in practice. If Rs is 1Ω, the output current is limited to 650mA if the output is shorted (or if the load tries to draw more than 600mA). While basic, this arrangement has been used in countless discrete regulator designs over the years.
Predictably, the opamp version will have far better regulation than the other two, because it has extremely high gain. Most modern 3-terminal regulator ICs use a similar (but optimised) topology, and the reference voltage is generally a 'band-gap' arrangement with very high stability. Two values are provided for regulation - 'line' and 'load'. Line regulation is a measure of how much the output changes as the input voltage is varied, and load regulation is a measure of the change of output voltage as the load current is changed. If you look at the data sheet for any 3-terminal regulator, this info is provided, but not always as a percentage - sometimes it's shown as ΔV (change of voltage), usually in millivolts. Most are better than 1% (line and load).
There are many factors that need to be considered in any voltage regulator circuit. One of the hardest to get right is stability, to ensure that the circuit has a fast reaction time, but without oscillation. Using an opamp driving a current amplifier (typically an emitter follower) will usually be stable, but if any additional gain circuits are used within the feedback loop, it will almost certainly oscillate. This means additional components have to be added (usually low-value capacitors), and their optimum location isn't usually immediately apparent. Examples can be seen in Figure 6.1 (single supply, opamp with emitter follower output) and Figure 7.1 (dual supply), where the opamp is followed by a gain stage. Given that most 'ordinary' opamps are limited to a supply voltage of less than 36V, this limits the available output voltage when a gain stage is not included.
In some respects, a power supply is not unlike an audio power amplifier. The only real difference is that amplifiers can source and sink (absorb) current, whereas a power supply only has to source current to the load. Indeed, a perfectly capable regulator circuit can be built using the common power amplifier building blocks. However, power amplifiers aren't expected to drive capacitive loads, where voltage regulators must be capable of driving any load, whether capacitive, resistive or inductive. Of course, a power supply also needs to protect itself from damage (shorted outputs or very low impedance loads), and it must be able to deliver its rated current into any load at any voltage. Series pass transistor dissipation can be extreme, but the supply must carry on regardless. Compared to power supplies, power amps are simple!
A supply with current regulation used to be fairly uncommon, but it's very useful for a number of reasons. If the maximum current can be set for just above the expected current drain of your circuit being tested, if there's anything wrong a current regulator will limit the maximum current (as its name suggests), and hopefully prevent damage to the DUT. There are only a few techniques that are commonly used for current regulation, almost always using a sense resistor. The voltage across the resistor is monitored, and if exceeds a preset value, the voltage is reduced to maintain the preset current.
One of the most common tricks is to use the base-emitter voltage of an ordinary BJT (bipolar junction transistor) as the 'reference', being roughly 0.65V. If the voltage across the sense resistor increases beyond that, the transistor turns on, and (by one means or another) reduces the voltage. Any attempt to draw more than the preset current results in the voltage falling further. This isn't a precision approach, but it's usually 'good enough'. The voltage can be amplified by an opamp (either IC or discrete) if necessary.
Figure 1.2 - Basic Current Regulator
Q3 and Q4 perform current limiting. The voltage across RS (the current sense resistor) causes Q3 to conduct, thus turning on Q2. This reduces the voltage by pulling down the reference voltage, reducing the output voltage, and therefore the current. A number of different sensing methods are shown further below, but this is one of the simplest. Because there are two transistors involved, the gain is quite high, so the current limit is 'brick wall'.
One way to make a very robust power supply is to use a high-power transformer based supply, and control the voltage using a Variac (see Figure 4.1). This is unregulated, but it's the simplest way to create a high power supply that can be used with almost any amplifier (or other projects, including power supplies). There's no over-current protection (other than fuses), but I have a couple of supplies that use this exact configuration. When I need lots of voltage and current, these supplies are invaluable. However, one needs to be certain that the unit under test has no inherent fault(s) first. This ideally requires current limiting. While 'safety' resistors can be used in series with the positive and negative supplies for initial tests, this is a nuisance.
Most (nearly all in fact) of my initial tests are done using a zero to ±25V, 2A dual tracking supply that I designed and built about 35 years ago (at the time of writing, and it's still working). It has current limiting down to about 100mA, and has a fan for the heatsink, along with an over-temperature shut-down. These are needed because it does get used for 'strange' applications, and yes, the output(s) have been shorted many times - usually by accident, but sometimes because there's a fault in the item being tested. Something as simple as a small solder bridge can spell doom for a power supply that can't protect itself.
The dissipation problem was discussed briefly above, and this is the Achilles heel (as it were) of all high current linear supplies. The answer (of course) is to use a switchmode design, but that is so far outside the scope of normal DIY that it doesn't warrant consideration. Every issue faced by a linear regulator is raised to the 'nth' power for a switchmode supply. Those you can buy have undergone considerable development, and use specialised parts that are not suited to a DIY approach. Unless you are capable of designing and building switchmode transformers, then it's out of the question altogether.
If you have a linear supply that can provide up to (say) 50V at 5A, the best case dissipation at full current with a shorted (or low voltage) output is 250W, but in reality it may be a great deal more. If you think that's fairly easy (there are transistors rated for 250W dissipation after all), think again. The SOA (safe operating area) and thermal limits come into play very quickly, and a transistor with (for example) 56V across it may only be capable of 3A or so, based on a case temperature of 25°C. Ultimately, you will need to provide enough transistors to be capable of handling at least twice the power dissipated, and preferably more. My suggestion would be to use a minimum of 5 × 125W transistors, and while that sounds like overkill, in most cases it will suffice - there's some reserve, but not very much! A lower voltage reduces stresses, and I know from many years of experience that ±25V is usually sufficient for most tests.
At higher voltages, if you used 5 × TIP35C (NPN, 125W at 25°C), they can each pass 1A with 50V across the transistor (50W), but only at 25°C. At elevated temperatures, that is reduced, falling by 2W/ °C above 25°. At a case temperature of 75°C, total dissipation is limited to only 25W for each transistor. That rules them out of contention with a simple scheme, because the dissipation will exceed the maximum allowable as the heatsink becomes hotter. Of course, you can use far more robust transistors, but they will be commensurately more expensive. The TIP35C (125W) is around AU$3.00, vs. over AU$5.00 for the MJL3281 (200W) and more than AU$6.00 for the MJL21194 (200W).
All of the available devices have the same limitations - SOA and temperature always mean that you can get far less power from any transistor than you expect. Forced air cooling is mandatory unless you have access to an infinite heatsink, which in my experience are hard to come by. Even using insulating washers may become impractical, because the additional thermal resistance means that the transistors have to be de-rated even further. In turn, that means a 'live' heatsink, sitting at the full supply voltage. Should it come into contact with an earthed chassis, the result will be a very loud Bang ! As you should now be aware, there are so many things that can go wrong that the advice to buy a commercial supply starts to look very sensible indeed.
Then (of course) there's the transformer. After that there's the high current bridge rectifier, followed by filter capacitors. All of these need to be very substantial, with a 500VA transformer, 35A bridge, and at least 10,000µF of capacitance. Just the hardware (transformer, bridge rectifiers, filter caps, heatsinks and power transistors) will probably cost at least AU$200 - or more. You still don't have a chassis/ case, pots, knobs and ancillary parts, including mains and DC connectors, meters, etc. Remember that for a dual supply (the only kind that's really useful), everything is doubled. You'll be up for at least AU$400 just for the basics, and closer to AU$600 by the time everything is included. If this hasn't convinced you that a commercial supply is worthwhile, then nothing will.
If you were to look at a major supplier (such as RS Components, Element14, etc.) you'll find dual supplies that can do 0 to ±30V at 5A, or 0-60V if the two outputs are wired in series. These may not be in the same league as Tektronix, Keysight or other 'laboratory' equipment makers, but the cost is less than for the major parts alone if you were to try to build your own. While the maximum voltage is less than ideal, I know from years of experience that up to ±30V is quite sufficient for basic testing, and all power amps shown in the projects section were tested with my ±25V supply before being connected to my 'monster' Variac controlled supply (which can deliver up to ±70V at around 10A or more).
Many of the true lab supplies use digital (via a keypad) entry for the essential parameters. For general use, this is an absolute pain in the bum! Using ordinary knobs and pots is a far better option most of the time, because the effect is immediate. It's common for lab supplies to use a rotary encoder to control the current or voltage, but you have to select the function first, and it may take several full turns to cover the full range.
If something starts to get hot in your test circuit, the last thing you need is to have to press a multitude of buttons or turn a knob ten times to reduce the voltage. With a standard pot, one twist anti-clockwise and the voltage is back to zero. You will never know just how frustrating keypad entry really is until you need to change something quickly. Ideally, there would be a 'ZERO' button to turn off the output, but I've not seen a digital supply that has one. Reading rapidly changing current on a digital display is simply not possible unless it features an averaging function (which will be buried three levels down in a menu - somewhere).
Having used bench supplies all of my working life, I can say with certainty that 'ordinary' pots are more than satisfactory for normal test purposes. Extreme accuracy is rarely essential for most testing, and if by some chance you do need a very accurate voltage or current, it's easy enough to build a separate regulator. Mostly, you won't need it, and if the supply is accurate to within a volt or so, that's almost always good enough. Obviously you need to be careful if you need 3.3V or 5V for logic circuits, but they will often have their own regulator, and will work with 7-12V quite happily.
Digital displays and controls can also give a false sense of security, because we tend to believe the meters because they display voltage and current down to a couple of decimal places. However, unless they are properly calibrated (with a known and calibrated accurate meter), then they could easily tell you that the voltage is 5V when it's really 5.5V or 4.5V. Because all digital systems ultimately rely on DACs and ADCs (digital to analogue and analogue to digital converters), they require an accurate reference voltage. If that goes awry for some reason, then all readings are meaningless.
For this reason, I do not cover digital control systems here. Control of voltage and current remain in the analogue domain - they are analogue functions, and adding another complication is not necessary. Fairly obviously, at least some of the ideas shown can be adapted to digital control, but I don't show any examples.
The simple arrangement shown in Figure 1.2 belies the likely difficulty of implementing a good current limiter, and this is where things can get difficult. There are two choices - 'high-side' and 'low-side' sensing. 'High-side' means monitoring the current in the positive and negative outputs, and is complicated by the fact that this voltage is not only variable, but also at a voltage that's usually incompatible with opamps. You can't expect an opamp to have its inputs at perhaps 30V or more, since that's generally the maximum operating voltage. This isn't a trivial issue to get around, and it's generally better to monitor the current before the series pass transistor(s) so the voltage doesn't vary so much. However, this makes the voltage problem worse, because the unregulated supply will typically be around 35V or more - well over the range for any low cost opamp.
A simple 'high-side' current limiter is shown in Figure 1.1 ('Opamp' version), but it's not as simple as it looks. It's difficult to make it variable without using an unrealistically large sensing resistor, and accepting that you will lose significant output voltage across the resistor, which will also get hot. A switched scheme is shown in Figure 7.1, and while this certainly works, it's not particularly accurate and nor is it the most practical.
'Low-side' sensing gets around that problem, but it can only be used for a single supply. Sharing a low-side sensing circuit between the positive and negative supplies won't work, because most of the supply current flows between the +ve and -ve outputs, often with little flow in the common connection. It can be done, but it's far from ideal, especially if a single pot is to be used for setting the voltage (a dual tracking power supply). The Figure 6.1 circuit uses low side sensing, and it will still work on both polarities of a dual supply because the outputs have their common point after all regulation.
There are specialised ICs available to get around the high-side current sensing problem. Three 'demonstration' high-side current sensing circuits are shown below. However, these are all shown with a positive supply only. The first two can be used in the negative supply (assuming a complementary design such as Figure 7.1), but the IC version cannot. There doesn't appear to be a solution for that particular problem.
Figure 3.1 - High-Side Current Sensing Circuit
A current mirror (Q1 and Q2) is used to sense the current across the sense resistor (R1, 100mΩ), and the output is level-shifted by the resistor network. The output is monitored by opamp U1, which is set up as a differential amplifier. VR1 is included so that the zero point can be set (i.e. zero output voltage with zero current through R1). The opamp is deliberately set up with a bit more gain than it needs, and the output is scaled with VR2. As shown, the circuit will provide an output of 1V/A, so at 2A current, the output is 2V. The arrangement shown is fine for up to 5A, and for higher currents, the value of R2 and R3 need to be increased.
While this circuit is capable of high accuracy, it's also very susceptible to temperature variations between Q1 and Q2. Ideally, these would be a 'super-matched pair' in a single package, but these can be difficult to find and while inexpensive, most are now available only in an SMD package. Naturally enough, a similar arrangement can be used without the current mirror, but sensitivity is reduced and the maximum allowable voltage is also lower. The current mirror can handle an input voltage of 50V easily, but the simple differential opamp circuit is limited to about 40V. Higher voltage is possible by increasing the value of R2 and R3, but that reduces the sensitivity even more.
If you were to use the Differential Amplifier circuit, the output voltage varies between zero and 250mV for a current between zero and 2.5A. Sensing current below 100mA (10mV output) is difficult. Of course, you can increase the value of the sense resistor, but at the expense of power dissipation. At 2.5A, a 100mΩ resistor dissipates 625mW, but to get the same sensitivity from the differential amplifier you'd need to use a 1Ω resistor, which will drop 2.5V and dissipate 6.25W. This is clearly a fairly serious compromise. There's also the ever-present issue of opamp DC offset, which may also need to be addressed if you need to regulate to low current (anything below about 100mA is a challenge).
In case you are curious as to the use of a -1.2V supply for the opamps, this ensures they can get to zero volts at the output. The LM358 can (allegedly) get its output to almost zero, but in reality it doesn't quite make it. The small negative voltage allows it to get to zero easily. Most other opamps will not allow such a small negative supply, and will require around -5V to work properly. This will take many above their recommended operating voltage if a 30V supply is used as shown.
In all cases, it's imperative that the input voltage remains within the specified range for any opamp used in this role. With a 30V supply, the inputs should always be at least 4V above the minimum supply voltage, and 4V below the maximum. Whenever possible, the input voltage should be close to 15V (assuming a 30V supply).
A simple solution that can be applied to the simple (one opamp) high-side sensor is to use switched resistors instead of a single fixed value. For example, 100mΩ is fine for higher currents, and you can switch to a 1Ω resistor to allow accurate setting for lower currents (less than 1A for example). This adds another switch, but it also simplifies the design, and opamp DC offset is much less of a problem when you need a low current limit.
There are several special purpose ICs available for high-side current sensing, with one shown in Figure 3.1. These include the LT6100, INA282 and several others, but they are only available in SMD packages, making them rather unfriendly for DIY applications where a PCB is not available. These are very accurate, and allow the voltage of the current monitored supply line to be much higher than the IC's supply voltage. In common with most SMD ICs, they are often only available in packs of five or more, and they aren't exactly inexpensive. If you wanted a dual supply (±25V for example), there is no negative version of these current shunt amplifiers, and this creates additional complexity. The INA282 can (apparently) sense a negative voltage, but it can't exceed -14V. The gain is 50V/V, so a much smaller shunt resistor can be used (0.02Ω shown). That means the output changes by 1V/A, so for 2.5A output, the output voltage will be 2.5V. Because it's an active circuit, it will introduce phase shift, which might make the current regulator unstable. This has not been tested.
The current sense IC datasheets also contain useful information about the proper connection to a current sense resistor. You must ensure that there is effectively zero PCB, Veroboard or hard wiring included in the sensing circuit. The sensing leads must come directly from the current shunt, avoiding any other wiring. This is known as a 'Kelvin' connection, which ensures that track or wiring resistance is not included in series with the current sense resistor.
Figure 3.2 - Low-Side Current Sensing Circuit
Low-side sensing is a far simpler option, but there are circumstances where it can't be used. For example, you can't use low-side sensing in the Figure 7.1 circuit, because the common is literally common to both the positive and negative supply. In a balanced circuit or if you only draw current from between the two outputs, nothing will register regardless of the current drawn. This method is used in the Figure 6.1 circuit, and there it's not a problem because each supply is a separate entity until the two are connected by the series/ parallel switching.
I haven't shown any of the options that can be used. For example, if you use a very low value sensing resistor, the small voltage across it can be amplified with an opamp to get more voltage. 100mV/ A as shown is fine for loads up to around 5A or so, but with more current the losses become too high. For example, even at 5A, a 0.1Ω resistor will dissipate 2.5W and you lose 0.5V across the resistor. With higher currents this quickly gets out of hand. At 7A, the resistor dissipates almost 5W, and it will get extremely hot. These caveats also apply to high-side sensing of course, as the physics are identical.
The current sense resistor (whether high or low side) must be inside the voltage regulator's feedback loop, or it can't compensate for the voltage drop across the sense resistor. In reality, it usually doesn't matter, because very few circuits that you will test will care if the voltage 'sags' a little under load. For an amplifier that uses a conventional power supply (unregulated), the actual voltage will change far more than it will with a bench supply, even if the current sense resistor is outside the feedback loop.
If you have the bits and pieces needed to build a robust power amplifier supply, then with the addition of a Variac (see Transformers - The Variac if you don't know what that is) you can build a 'monster' supply that will suit high power testing with almost any load. You don't get regulation, nor is there any current limiting (not even short circuit protection), but with the right parts it's a formidable piece of test gear.
I have a couple, one of which really does qualify as a monster. The circuit is shown below, and it's literally what I use for high power tests. Any piece of equipment that's connected to it has already been verified to be functional, and that's essential because it can destroy almost anything given the opportunity. It's an extremely useful piece of kit, and all project amplifiers published on the ESP site have had their final test with this very supply.
Figure 4.1 - Variac Based Power Supply
The supply is just a 1kVA transformer, two bridge rectifiers (35A each), and a bank of capacitors salvaged from a very ancient hard disk drive many years ago (the drives that were as big as a washing machine!) It's set to the desired voltage with the Variac that I have on my workbench as a matter of course. The supply isn't regulated, but can supply enough current for any amplifier that I have ever tested with it. Long ago, a Variac was a very expensive piece of kit, but Chinese variable auto-transformers are now surprisingly affordable.
This also means that the applied DC is very similar to that normally provided by a linear supply, but with better regulation due to the oversized transformer and filter capacitors. This is obviously not a cheap option, but it cost me almost nothing because I had everything I needed in my 'junk box'. The 10,000µF caps shown should be considered a minimum - mine uses around 20,000µF on each supply. If you have them available or can afford them, use as much capacitance as you can! Note the inclusion of 'bleeder' resistors - without them, the voltage can remain at a dangerous level for many hours. I normally don't use them because the connected amplifier )or other circuitry) discharges the caps, but that's not necessarily true with test equipment.
The continuous output current is around 7A, but with an amplifier load it can handle 25A peaks (and more) with ease. Do you need something similar? Only you can answer that, but it doesn't need to be as big as the one I use. Of course, there's no current limiting, so you need to be sure that the circuit works before using the 'monster' supply! The output fuses protect against shorted outputs, but will not save your project from damage if it's faulty. A supply such as this is applicable for final tests, not for initial testing or fault finding. There is no current limiting, so a fault can cause significant damage (the fuses only protect the supply, not the load!). Shorted outputs are obviously a cause for some concern, so care is required.
One approach that has been used in many supplies is a simple transformer 'tap switching' scheme. If you only need (say) 15V or less, the transformer's output is switched with a relay so the AC output is only 15V AC, rather than the full 30V AC needed to get a clean 30V DC output. If the output is run at a low voltage but high current, the dissipation is reduced because there's less voltage across the regulator. When a voltage of 16V DC or more is selected, the relay switches to the full output (30V AC). This can be extended with more taps of course, but that would require a custom transformer, dramatically increasing the cost.
Tap switching supplies have been around for almost as long as I can remember. The most impressive I've seen used a motorised Variac to maintain the AC input at just enough to prevent any ripple breakthrough on the DC side. These were very large, extremely high current, and would have cost a fortune when they were made (sometime in the mid 1970s). This isn't something I'd suggest anyone try to build, as the cost and difficulty of setting it up would be well beyond the budget of even a well-heeled DIY fanatic.
Simple tap switching supplies use two AC voltages, so for a dual supply you need two tapped windings, plus an auxiliary winding to provide the normal ±12V or so for the control circuits. Finding a suitable transformer will be next to impossible, so you'd need to have a transformer custom made. This isn't a problem for manufacturers because they will build many supplies and the cost can be amortised over a complete production run. Hobbyists don't have that luxury.
The use of tap switching reduces the demands on the series pass transistor(s). For a dual supply, you'd need at least two power transformers (and realistically you'd also need a third transformer to provide the control circuit supply voltages). This would increase the already significant cost of building a dual power supply. There's also additional components needed to sense the output voltage, and switch from the low to high voltage tap automatically (and vice versa) using relays. While building any power supply is a challenge, adding tap switching just adds another layer of complexity. I don't propose to go any further with this, as it makes an already complex and difficult job that much harder and more expensive.
There are some savings too of course, particularly in the number of series pass transistors needed and the amount of heatsinking. However, these are not sufficient to offset the cost of the transformers, and the power transistor(s) can still be subjected to short-term conditions that push them outside of their safe operating area. Such excursions may be brief, but a transistor can fail in a millisecond if the SOA is exceeded - especially if its already at an elevated temperature. I recall a friend who built a fairly basic tap-switching power supply from a kit many years ago, and he had nothing but trouble from it. This was a semi-commercial product, complete with case and everything needed to put it together. It failed so many times that he eventually gave up in disgust. No-one wants to go through that!
There's another method that's worth a bit more than a passing mention, even though it does have some serious challenges. Using 'phase cut' circuitry (similar to that used in lamp dimmers), it's possible to vary the input voltage prior to regulation, simply by adopting fairly simple low frequency switching. However, it also imposes far greater than normal stresses on the transformer and the filter cap, but these are not insurmountable problems.
The switching element can be a MOSFET, IGBT (insulated gate bipolar transistor) or an SCR (silicon controlled rectifier), with the switching synchronised to the mains with a simple zero-crossing detector. The idea is to impose a delay, starting from the zero crossing (time zero). It's usually easier (and adds fewer additional challenges) to wait until the input voltage has fallen to the desired voltage, so a 'leading edge' configuration is used. When the input voltage has fallen to just below the threshold voltage, the switch is turned on, charging the main filter capacitor. A simplified block diagram is shown below.
Figure 5.1 - Phase-Cut Pre-Regulator Block Diagram
The challenges mentioned earlier include extremely high peak currents, especially with a low output voltage at a high current. These can be mitigated by adding an inductor and flyback diode (shown as 'Optional'), with the greatest issue being that the inductor has to carry a large DC component without saturation. This means a low-permeability core has to be used, so more turns are necessary for a given inductance. This adds resistance and increases losses (meaning more heat is generated). However, including the inductor will give better results than you'll get otherwise, and it reduces the high current stresses otherwise imposed on the transformer, bridge rectifier and filter capacitor. The diode (D1) must be a high-speed type, rated for the maximum output current.
This technique has been used in several commercial products, and while it does do exactly what's intended, it makes poor use of the transformer's VA rating if the inductor and diode aren't used. Without these, you can expect the transformer's output current to be up to four times the DC current. That means that for 3A DC output (and using a 25V transformer), the transformer needs to be 300VA, where normally a 150VA transformer would be sufficient. To make matters worse, the inductor has to be fairly large - around 10mH is needed, a large and expensive component.
The circuit works by comparing the input control voltage to the ramp, created by the ramp generator and synchronised to the mains frequency with a zero-crossing detector. When the AC voltage reaches the required amplitude, the switch turns off, preventing the capacitor from charging any further. The 'idealised' waveform is shown (assuming no inductor or storage/ filter capacitor), and it's apparent that the voltage and current supplied to the output is reduced depending on the phase angle. This process and waveforms can be seen in more detail in the Project 157 - 3-Wire Trailing-Edge Dimmer project article. It's a different application, but the process itself is pretty much identical.
I actually have a power supply that uses this arrangement, but its 120V AC input makes it pretty much useless unless I power it from a Variac. At no load, the voltage jumps up then slowly falls until it's below the threshold, when it jumps up again and the process repeats (in a somewhat random pattern). Under load it's not too bad, but this is not a technique I'd recommend. Apart from the fact that the one I have is rated for 150V at 5A, it also weighs in at around 40kg, and has one very large main transformer, a smaller auxiliary transformer to power the electronics, and a large filter choke (inductor). It is very 'old school' in terms of layout and construction, and it never gets any use. I don't even recall how I came to own it! If I need that sort of voltage and current, I use my Variac controlled 'monster' supply.
Yet another approach is to use a switchmode step-down (buck) converter as a tracking pre-regulator. You can think of this as a 'high tech' version of the phase-cut pre-regulator described above, which provides the advantages, but fewer disadvantages (in terms of transformer utilisation at least). Some fairly high powered modules are available surprisingly cheaply, and the idea is to ensure that the voltage provided to the series-pass transistors is only a couple of volts greater than the output voltage. This can improve efficiency so you can get away with much smaller heatsinks, and thermal management isn't such a challenge. A suitable feedback mechanism has to be provided that controls the output of the switchmode converter, such that it is always just great enough to ensure regulation.
The pre-regulator reduces the series-pass dissipation to only a few watts, even at full current. It should go without saying that this approach requires some serious development, and while it's probably the best all-round solution, it's far harder to get right than any of the other options examined so far. This is the electronic equivalent of using a motorised Variac (as mentioned above), but is cheaper to make and easier to control. The design challenges can be extreme if you try to build your own, and keeping switching noise out of the final output can also be difficult. If you need very low noise (for performing noise or distortion measurements for example), the switching noise will almost always intrude on the measurements. This is an option that won't be covered further here.
A single supply might be attractive for some people, and it's certainly simpler than a dual tracking version. Of course, if you only have one polarity that limits your options as to what you can test, but they are commonly available from any number of suppliers. The circuit shown below is adapted from one that's shown on a number of different websites [2, 3, 4]. As such, it's difficult to know which one was 'first', and there have been many improvements (or at least changes, which aren't always the same thing!) made to it over the years. The basics haven't changed much, and the one shown below dispenses with one voltage regulator in favour of a simple diode regulated negative supply. Because I used LM358 opamps, the negative supply only needs to be around -1.2V at fairly low current.
When the supply is in current limit mode, the LED will come on, indicating 'constant current' operation. It's normally off, so you can tell at a glance if the load is drawing the preset current with a reduced output voltage. Constant current operation is particularly useful for testing high power LEDs or LED arrays, as that's the way they are meant to be driven. You also need an 'on/ off' switch, which reduces the output voltage to zero when in the 'off' position. This is an essential feature (IMO) as it lets you make changes without having to disconnect the supply. The best arrangement is to provide the switching at the output of the supply, as that lets you set the voltage while the DC is turned off. Consider using a relay (or two) for the switching, otherwise you need a heavy duty switch. Wile the voltage can be reduced to (near) zero by pulling the non-inverting input of U1B to ground, there may be 'disturbances' when AC power is first applied. This is avoided by switching the output.
The supply shown below is fairly basic, and you'd need to add meters for voltage and current, and thermal management (a fan and over-temperature cutoff) at the very least. There are countless improvements that can be made, but they would make the circuit more complex, more expensive, and provide more 'exciting' ways to make a seemingly minor error and cause the supply to blow up the first time it's switched on.
Figure 6.1 - Single Supply Schematic
U1 is a 7815 regulator, but with a 15V zener from the 'ground' pin to raise the voltage to 30V. D10 ensures that there can never be more than 15V input/ output differential across the regulator. Additional reference zener current is provided by R3 to ensure a stable output. U2A is the current regulator. When the voltage at the inverting input (U2A, Pin 2) is greater than that on the non-inverting input (Pin 3), the output goes low, pulling down the reference voltage provided to U2B (the voltage regulator). The voltage is reduced by just the amount required to ensure that the preset current is provided to the load.
The current limit is variable from (theoretically) zero to 2.5A. VR4 allows adjustment to ensure the reference voltage for U2A (TP2) is as close as possible to 825mV (825mV across R18 (0.33Ω) is 2.5A output current). It may be possible to increase the output current to 3A (990mV reference voltage), but you would need to add another series pass transistor to keep the transistors within their SOA at minimum voltage and maximum current. Some ripple breakthrough at maximum output (voltage and current) is likely unless you add more capacitance (C1).
When in voltage mode, U2B compares the reference voltage from VR2 with the voltage at the output, reduced by R16, R11 and VR3 (voltage preset). If the output falls due to loading, U2B increases the drive to the output series-pass combination (Q3, Q4 and Q5) to maintain the desired voltage. The upper output voltage limit is imposed by the opamp (U2), which can't force its output to much above 25V with the typical output current of around 2mA (this depends on the gain of the output section, Q3, Q4 & Q5). Note that the reference voltage is itself referred to the negative output terminal - this ensures that the regulator will correct for any voltage drop across R18. If it were otherwise, regulation would be badly affected, especially at maximum current.
Note that the heavy tracks are critical, and any significant resistance in these sections will upset the current sensing. Also, be aware that the points indicated with a 'ground' symbol are marked 'Com' (Common). They are not connected to chassis or any other ground. The 'Com' designation means only that all points so marked are joined together. Also note the diodes with an asterisk (*), which must be 1N5404 (3A continuous) or better. All other diodes are 1N4004 or equivalent (other than the 25A bridge rectifier of course). Bench power supplies often get connected to 'hostile' loads, and the high current diodes (D8 and D9) are to protect the supply.
The supply uses 'low side' current sensing, so it needs some tricks to use it as a dual tracking supply with both positive and negative outputs. The current sense resistor (R18) is a compromise between voltage drop and dissipation. At maximum current (2.5A), R18 will dissipate a little over 2 watts, which is easily manageable using a 5W wirewound resistor. Both voltage and current regulation are very good (at least according to the simulator), and there's no sign of instability. In theory (always a wonderful thing), the current can be regulated down to a couple of milliamps, but in reality it will not get that low. Expect around 50mA or so, but it might be a bit lower than that (depending on the opamp's own DC offset). Another trimpot can be added to correct for opamp DC offset, but it shouldn't be necessary (and adds something else that needs adjustment).
All of the alternative versions specify a single 2N3055 for the output, but with a shorted output and maximum current, the dissipation will be about 80W, and maintaining the series pass transistor(s) at 25°C will be impossible. The TIP35 devices have a higher power rating (125W) and a good SOA (safe operating area), but there is still a case to be made for using three, rather than the two shown. The BD139 also needs a heatsink, but a simple 'flag' type will normally suffice. In common with any transistor that dissipates significant power, excellent thermal bonding to the heatsink is essential, and you will need to use a fan. This can be thermostatically controlled, and can use PWM (pulse width modulation) for speed control, or it can just turn on and off. Figure 8.1 shows a suitable circuit for both operating the fan and shutting down the supply if it gets too hot (which in this context is no more than 50°C heatsink temperature).
If you did want to use the Figure 6.1 circuit for a dual supply, the transformer needs two separate windings. The second supply (#2) is identical to that shown above, and the positive output is connected to the GND (or to be more accurate, 'Common') connection of supply #1. Most of the time, power supplies are used with the outputs floating, with no connection to the mains protective earth. This lets you use the supply as a normal positive and negative supply, or the outputs can be used in series, which will give an output of 50V at up to 2.5A. This way, you can ground any terminal you wish to get the supply configuration you need.
To build it as a dual supply, the 'Voltage Set' and 'Current Set' pots will be dual-gang linear pots, with one section of each for the separate supplies. Tracking will not be perfect, but dual-gang linear pots are usually fairly good in this respect. Using two supplies also lets you connect them in series or parallel. The latter is handy if you have a single supply load that draws more current than one supply can provide. Many commercial dual supplies use this scheme, and it can be very useful. While 'proper' dual tracking would only use a single gang pot with electronic coupling to ensure the voltages are identical, this makes the circuit more complex.
Figure 6.2 - 'Dual Single' Supply Connections
When the switch or relay (double-pole, double-throw or DPDT) is in the series position, the negative of the upper supply is connected to the positive of the lower supply, and both connect to the common terminal. You can have 0 to 50V output, and the common is the centre tap for ±25V. In the parallel configuration, the two positives are joined, along with the two negatives (the common terminal is disconnected). This allows for 0-25V at up to 5A output. Note that the negative terminal is the negative output of the lower regulator. Because the outputs are floating, either the positive or negative terminal can become the system earth/ ground if this is required.
One advantage of using 'dual single' supplies is that they can be used independently (with different voltage and current limit settings), connected in series (usually with tracking) or in parallel for more output current. Unfortunately, if you wanted to use the two supplies independently, you can't use dual-gang pots, and each supply must be set individually. This is a serious nuisance, and fortunately it's not a common requirement.
The arrangement shown let you connect the supplies in series (0 to ±25V or 50V at 2.5A) or in parallel (0 to 25V at 5A). The 'common' terminal should normally not be earthed, so the supplies are floating. This lets you operate the supply without creating ground loops. When in parallel, one supply will usually be at a slightly different voltage from the other, but the current limiter ensures that the current from each supply can't be above the limit (2.5A). There may be a small change in voltage as the current is varied, but this shouldn't create any problems in normal use.
This design means that there is no common circuitry - both regulators are completely independent, and no parts are shared - other than the dual-gang pots used to set voltage and current. This increases the overall cost, but allows greater flexibility. The circuit above doesn't allow for independent supplies, but that is unlikely to be a limitation. A well equipped workshop will have at least two supplies (for example, I also have a separate independent ±12V supply, plus an independent 5V supply). None of these supplies share a common ground - all are fully floating.
The 'on/off' switching is at the final output (just before the output terminals). This lets you set the voltage with no output (meters will be connected before the output switch). A relay (or a pair of relays) lets you use a mini-toggle switch rather than a heavy-duty toggle switch, and is recommended for maximum performance. The relay(s) can be mounted on the front panel, right next to the outputs.
Now we can look at another 'sensible' option. Again, that means an output of around ±25V DC, at a maximum current of no more than 3A or so. Believe it or not, it's still cheaper to buy one! I know that this isn't the 'DIY way', but it's more practical than building it yourself. I've looked at countless different designs over the years, but few are worth the parts it would take to make them. There remain issues with stability (i.e. not oscillating at any output voltage or current, or with 'odd' loads). This might not sound like a problem, but the interactions between voltage and current regulators can make an otherwise well behaved supply suddenly think it's an oscillator. It goes without saying that this is undesirable (to put it mildly).
Project 44 has been around for quite some time (since 2000), and although the maximum output is only ±25V, it a fairly good option for running initial tests. It doesn't have adjustable current limiting, so output current is set by the LM317/ 337 regulators, at around 1.5A. It's usefulness has never diminished since publication, but you must use 'safety' resistors in series with the outputs so that nothing is damaged if there's an error in the wiring of the DUT. The value for any given ESP project is generally specified in the project article or construction notes (available when you buy one or more PCBs).
One of the things that's expected is that a bench supply needs very good regulation. In reality, this isn't actually the case. Power amplifiers usually don't have regulated supplies, and preamps (and similar low current projects) draw a fairly consistent current, so regulation within the allowable range is easy. If a power supply's voltage falls by (say) 0.5V when heavily loaded, it really doesn't matter, because that's a great deal less than it will have to cope with when connected to a 'normal' power supply. The thing that is critical is current limiting, and while this might appear to be simple enough, it's actually difficult to get it to operate reliably. The current limiting circuitry introduces additional gain into the circuit, and maintaining stability can be irksome at best, and next to impossible at worst.
Often, the critical aspect of any current limited supply is at the transition between voltage and current regulation, where the two different forms of regulation interact. At the onset of current limiting, you have the voltage regulator trying to maintain the preset voltage, and at the same time, the current regulator is trying to reduce the voltage to maintain the preset current. For those who really want to build a power supply, John Linsley-Hood presented a design way back in 1975. An updated version is shown below, but modern transistors have been substituted for the originals, and two series pass transistors are included. Adding a third series-pass transistor on each supply makes cooling easier and imposes less stress on the transistors. In the original circuit, the opamps were µA741s, but if you have them to hand the 1458 (essentially a dual 741) is a better choice. You can also use an LM358 in this circuit.
Figure 7.1 - Bench Power Supply (After JLH, 1975) [ 6 ]
The above is adapted from the original, which used a single 2N3055 and MJ2955 TO-3 power transistor (one for each rail). Not only were they subject to excessive dissipation in the original (up to 93W at maximum current into a shorted output), but TO-3 devices are rather expensive today. They are also a pain to mount, where flat-pack devices are far simpler in this respect. The TIP35/36 devices specified have a higher power rating (125W vs. 115W each) and a higher collector current, but I've modified the circuit so that it provides a maximum of ±25V and uses a lower voltage transformer. This keeps the series pass transistors to a manageable power level, at no more than 40W each. Feel free to add another series pass transistor for each polarity, reducing the thermal load even further. Q3 (a and b) must have a reasonably good heatsink, as the power dissipation is much higher than it may appear at full output current (and at any output voltage).
The current limit switch is less than ideal, since the switch contacts need to be able to handle the maximum output current (about 2.4A), and it's less convenient than a pot that allows continuously variable current limiting. The 0.27Ω resistors need to be rated for at least 3W, with 1W for the 1.5Ω resistors. The remaining current limiting resistors are 0.5W. While the switch is not as versatile as a pot, the limiting thresholds are designed to protect your circuitry. When first testing, you'd normally use a low current to ensure that nothing is drawing more than it should. The 5mA setting is too low for most circuits, but it can be useful. It can be omitted if you don't think you'll need it.
The output needs either a heavy duty toggle switch or a relay to turn the DC on and off, and this disconnects the supply completely when you don't need any output (such as re-soldering a missed joint etc.). Metering isn't shown - see below for details of adding a voltmeter and optionally an ammeter as well. The two 20k trimpots let you set the maximum voltage (nominally ±25V). They should be roughly centred to obtain the correct voltages. Although not shown on the circuit, you may need to add resistors in series with C4a/b if the supply oscillates when in current limit mode. They weren't included in the original, but the simulated circuit oscillates if they aren't there. A value of around 100 ohms should be sufficient.
The circuit is far from 'perfect' (and nor was the original), but it should work well in practice. The voltage set pots will ideally be a dual-gang pot, so both supplies are varied at the same time. Likewise, the switch (Sw1a/b) will be a 2-pole, 5-position switch. Note that I have not built and tested this circuit, but it has been simulated and it performs as expected. The benefit of a simple arrangement as shown is that it can most likely be built for less than a commercial supply.
The series pass transistors (Q1a/b and Q2a/b) need a very good heatsink, and optimal thermal coupling. If used at low output voltages and high current, you will need a fan to keep the transistors cool enough to ensure they don't fail due to over temperature. The driver transistors (Q3a/b) will also need small heatsinks. The circuit is symmetrical, so while it may appear complex, it's largely repetition. I cannot guarantee that it will be completely stable when in current-limit mode - the simulator tells me it is, but that may just be the simulator itself - reality is often very different from a simulation.
While there is an expectation that a power supply shouldn't ever oscillate, in reality it takes serious engineering to maintain stability along with good transient response. Mostly, a small amount of oscillation usually won't do any harm, and the current limiting is there to ensure that your latest creation doesn't self-destruct if there's a wiring fault. It can also be handy for battery charging (amongst other things), and the limiter's primary purpose is to protect your circuit and the power supply against 'mishaps'. Many supplies will show signs of high frequency instability, rarely when in 'constant voltage' mode, and most often when in constant current mode.
In case you have started thinking that building your own supply doesn't look too daunting, there are some other things needed as well. The transistor temperature is critical, so it's important to include a thermal shutdown mechanism. This can be a simple thermal switch that disconnects the mains if the heatsink gets too hot - simple but not very sophisticated. It's usually better to include an 'over temperature' indicator, and a thermal fan that turns on if the heatsink goes above a predetermined temperature. 'Store bought' supplies may have a variable speed fan, with a final shutdown if the heatsink doesn't cool down. This can happen if there's a sustained high current into a short, a blocked fan filter, or placement on your workbench restricts airflow.
This is a critical part of any power supply. Ideally, if the thermal limit is reached, the supply should turn off, but this is easier with some circuits than it is with others. For example, the Figure 6.1 circuit is easy, as it's simply a matter of pulling the voltage reference to zero (essentially in parallel with the 'on/off' switch). This can be done with a transistor, relay contacts or it can even be made 'proportional' so the maximum output current is reduced as the heatsinks become hot. Thermal limiting is a bit more difficult with the Figure 7.1 circuit, as the 'set voltage' pots are not referenced to ground, but to the output supply rails. Due to the need for complete isolation, a relay is the best choice, and it simply shorts the set voltage pots. You need a double-pole relay because the two pots are separate from each other (electrically).
The next thing is to decide how best to sense the heatsink temperature. The obvious choice is an NTC (negative temperature coefficient) thermistor, and these are readily available in a range of different values (the value is usually specified at 25°C). Unfortunately, thermistors are a nuisance to mount to the heatsink, unless you can get one with an integral mounting assembly. You can make your own, using a miniature bead thermistor and using epoxy to attach it to a wire lug. Naturally, you need to be careful to ensure that there is no electrical connection from the thermistor to its mounting. You can also use diodes or transistors for thermal sensing, but they are less sensitive than thermistors (only -2mV/°C) and more irksome to set up. A transistor can be configured to provide greater sensitivity (because it has gain), and you can get up to -100mV/°C easily. However, the transistor needs a trimpot (preferably as close as possible to minimise noise pickup), and the sensor requires three wires instead of two. They are also fiddly to set properly. A more-or-less typical 10k (at 25°C) NTC thermistor will show a change of roughly -250 ohms/°C.
Because thermistors vary widely in terms of their value change with temperature, it's essential that a method of adjustment is provided. Ideally, you need an accurate thermocouple thermometer to measure the heatsink temperature, as close as possible to one of the series-pass output transistors. You'll need to use thermal 'grease' to get an accurate reading. Typically, the resistance of a thermistor will have fallen to around 30-40% of the 25°C value at 50°C, but this depends on the material used. The datasheet for the thermistor you buy will usually provide the exact details. Make sure that the thermistor(s) are not installed too close to the fan. If they are, the fan will cool the thermistors easily, but may be unable to keep the heatsink to a safe temperature. This can cause failure.
A cheap opamp is the easiest way to get reliable detection of an over-temperature 'event', and several thermistors can be used, with the hottest one triggering the cooling fan(s) or shutting down the supply. You can use a two stage system as shown below, where a mild over-temperature starts the fans, but if the temperature continues to increase then the supply is disconnected from the load altogether. The two trimpots are used to ensure that the initial voltage across each thermistor is around 5.8V at 25°C, which means approximately 65% of the total resistance or VR1 and VR2. Should the voltage across either thermistor fall to about 5.4V, the fan will turn on. The fan turns off again when the voltage has returned to the 5.4V threshold. If the supply cuts out because the temperature keeps rising, the fan will keep running.
Figure 8.1 - Thermal Sensing, Fan and Relay Cutout
U1A is a buffer, included to ensure that the hysteresis resistor on U2B doesn't disturb the first comparator. At low temperatures, the comparator U1B has its output low, and U2A is high, so the fan doesn't run and the relay contacts are closed (provided the DC switch is closed). As the temperature rises, one or both thermistors will drop to a lower resistance. When the thermistor voltage falls to ~5.2V, the fan will start, and if the temperature continues to rise, the supply output relay will be turned off when the thermistor voltage falls further. This arrangement ensures that the temperature should never reach a dangerous level. It will be necessary to adjust the trimpots to preset the initial thermistor voltage to an appropriate level to ensure that the fan comes on when the heatsink temperature reaches about 35°C. The LED is there to let you know why everything has suddenly stopped working (the output transistors are too hot !). The last trimpot (VR3) should be set for a cutout temperature of around 45°C. Both comparators have hysteresis, so the fan won't turn on and off rapidly, and nor will the cutout relay. (Note that U2B is not used.)
Thermistors are not precision devices, so you will need to run your own tests with those you can get. It may be necessary to experiment with resistor values to obtain sensible (and safe) temperature thresholds. You may be wondering why I suggest such a low heatsink temperature (45°C). Bear in mind that the thermal resistance from transistor case to heatsink may be around 0.5°C/W, so if the transistors are operating at 35W, the case temperature will be 17.5°C hotter than the heatsink. That means a case temperature of over 60°C. If your mounting techniques aren't good enough, the difference may be greater, leading to a risk of failure. If you can't place a finger on the transistor and keep it there, then it's probably too hot.
Maintaining a safe operating temperature and shutting down the supply (or disconnecting the load) if the power transistors get too hot is a critical part of any power supply. It's the nature of any variable supply that you never know what you'll eventually use it for when it's first built, and every eventuality needs to be catered for. It's far better for the supply to shut down prematurely than to allow the transistors to get so hot that they fail. Transistors fail short circuit (at least initially), which will put the full unregulated supply voltage across the DUT. The damage that can cause may be catastrophic.
All power supplies need meters. These are normally included for voltage and current, and the most common now is digital. However, 'traditional' analogue moving coil meters are not only cost effective (you can get them surprisingly cheaply), but are also easy to read at a glance. Many digital meters don't provide sensible supply and metering connections (for example, some require a floating supply). This makes the circuitry more complex, and the accuracy that's implied by digital meters is often an illusion. With analogue meters, 'FSD' means full scale deflection.
My preference has always been for analogue meters. If you can get a meter with a dial that's calibrated from 0-30V (for example), one can be used for voltage, and the other for current (0-3.0A). The required shunts and multipliers can be determined easily enough - see the article Meters, Multipliers & Shunts for all the details. It might be possible to use the current-sense resistor as the meter shunt, depending on the sense resistor value and the sensitivity and internal resistance of the meter. In most cases, a 1mA meter movement is a good compromise, and that will let you use the current sense resistor shown in Figure 6.1. Yes, connecting the meter and external resistor will affect the shunt ever so slightly, but the error will be very small (to the point of being infinitesimal).
Figure 9.1 - Current And Voltage Metering
Basic stand-alone metering circuits are shown above. The current meter is a pain, because the polarity has to be reversed depending on whether it's monitoring the positive or negative shunt. It looks convoluted, but it will work exactly as intended if wired as shown. The total meter resistance assumes the use of a 1mA meter movement, calibrated for 30V (voltmeter) or 3A (ammeter), and assuming an internal coil resistance of 200Ω. If the meter used is more sensitive (or its resistance is different), the resistances will need to be calculated. It almost always easier to use trimpots to set the range than fixed resistors, and suitable values are shown. For a voltmeter (calibrated for 30V FSD) ...
Rm = ( V / FSD ) - Rinternal
Rm = ( 30 / 1m ) - 200 = 28.8k
If the shunt resistors for an ammeter are different from the values shown the calibration will be different. The 'total resistance' shown includes the meter's internal resistance (typically around 200Ω for a 1mA movement). Note that if you use a 1mA movement, the shunt resistor will need to be no less than 0.1Ω. A 67mΩ shunt is called for, but this assumes that the meter's resistance is exactly 200 ohms, and there is no provision for adjustment if the reading is off. Whether the same shunt can be used for both current sensing and the ammeter depends on the final topology of the design. It's not always practical, but does reduce voltage losses slightly.
Note that if you use the Figure 6.1 circuit, the two shunts have the same voltage polarity so the reversal shown above isn't necessary. To look at positive or negative output current, the meter is simply switched from one shunt to the other, and the polarity is unchanged. That takes away the crossed wiring shown to the negative shunt in the above drawing.
While a switched ammeter is shown (and that's what my old supply uses), it's better to use a separate ammeter for each output. Provided you have enough front panel space, this removes the tedium of switching the meter, and means that if you forget (and that will happen), you may be monitoring the negative supply, but using the positive supply. Needless to say, that means that you can't see the current and the DUT may be damaged before you realise your mistake. The use of current limiting can mitigate that of course, provided it's set for a non-destructive (low) current when you start testing.
The voltmeter can be switched to measure either positive or negative voltage, or it can simply be wired across the dual supplies (50V for the circuits shown here), and calibrated to show 30V FSD ('Voltage Meter (Alt.)). The implication is that the voltage will be ±25V, or other lower voltage as selected. There may be some small error if the supplies don't track perfectly, but this is usually not a major issue unless you are expecting a precise voltage for some reason. If that's the case, it's better to use an external meter - those on the supply are 'utility' meters - they show the value of voltage and current, but expecting better than around 5% accuracy is unrealistic.
Digital meters are either the best thing since sliced bread, or a blight on the landscape, depending on your viewpoint. Personally, I prefer analogue (mechanical) meters, but they are usually fairly large and unwieldy, taking up more panel space than digital readouts. The greatest benefit of analogue meters is that you can watch the pointer moving, so an increasing (possibly runaway) current is seen quickly, and varying currents can be averaged by eye quite easily. Digital meters are particularly useless if the current varies quickly, because the display just becomes a blur of digits, and you cannot average a digital readout by eye.
However, digital meters are usually cheaper than analogue movements now, and most are fairly accurate. Because they take up less panel space, they are a good option, provided a few simple precautions are taken. In particular, and especially for the current meter, you need to include averaging circuitry that stops the display from showing a bunch of seemingly random digits when the supply current varies rapidly. This can be as simple as a resistor (1k is always a good starting point) and a capacitor to average the reading. With a 1k resistor, a 100µF capacitor means that you have a 1.59Hz low frequency -3dB point, so most rapid variations will be smoothed out so you can read the current. Failure to include this will provide readings that you can't decipher. It's fast enough to ensure that a trend is easily visible.
No details for digital meters are shown here because they depend on the meter itself. Some are auto-ranging, others use switchable ranges, and the simpler ones just give a reading from '000' to '199', with the option to select a decimal point at the desired position (often via a jumper or link on the meter's PCB). For current measurements, it will often be necessary to use an opamp to boost the small voltage across the current shunt. For example, if you have a 0.33Ω shunt, you'll need to amplify or attenuate the voltage across that to suit the range. For 2.5A full scale, that means you only get 825mV with a current of 2.5A, and that needs to be amplified so the meter shows '2.50' (2.5V into the meter). The amount of amplification or attenuation depends on the meter's sensitivity. For example, a 200mV meter will need to have the shunt voltage reduced by a factor of 33 with a voltage divider. It will read 2.5 (25mV) with the decimal point selected by whatever means are provided. Resolution is only 100mV (±2%, ± the meter's final digit 'uncertainty factor', which can be up to two 'counts'). This (IMO) is not good enough.
Ideally, if you decide to use digital metering, use a meter that offers three full digits (up to '999' rather than '199'), and if possible with auto-ranging. There are many choices, so it's up to you to decide how much you want to spend and what accuracy you need. Again, Meters, Multipliers & Shunts gives some worked examples that you may find helpful.
This is where things can get ugly. The front panel is the most important part of the supply, because it has voltage and current controls, on/ off switches (mains and DC), maybe a series-parallel switch, meters, and of course the output connectors (typically combination banana sockets/ binding posts). Of course, you'll also add LEDs for power on, current limit and thermal overload. Everything on the front panel has to be accessible for construction or maintenance, and that invariably means a maze of wiring. The front panel has leads for AC mains, DC outputs, all LEDs and pots, and this all adds up (surprisingly quickly). Maintaining a common supply for all LEDs (e.g. anode to the positive auxiliary supply) means that many of the LEDs can share the same anode voltage, which can save wiring. However, this does not apply to the current limit LEDs in a dual version of the Figure 6.1 circuit, because the two supplies must be kept fully independent until the series-parallel switching.
The internals must contain your power transformer(s), rectifier(s) and filter caps, along with the main heatsink(s) for the output transistors. The latter will have input, output and control wiring, as well as connections for the thermistors and fan(s). At the very least, each output module (assuming a dual supply) will have at least six wires. Then there's the regulator control board(s). You'll have one for each supply (assuming the Figure 7.1 dual supply arrangement), plus a thermal controller board to monitor the heatsink temperature.
It's all too easy to get wiring wrong, and you need a very disciplined approach to ensure that you don't make any wiring errors. Avoid the temptation to try to fit all the control boards to the front panel. It may reduce the wiring needed, but makes servicing a nightmare if the various parts of the supply can't be accessed and tested without having to disconnect wires from boards. Whatever sized enclosure you were thinking of using, if it doesn't have lots of free space then it's too small.
Make sure that all connections can be accessed without having to remove boards to get to the underside. Use pins, wire loops, or any other suitable technique so that all wires can be disconnected from the top (or visible) side of the boards. Avoid plugs and sockets - all connections (especially the really important ones) should be soldered, with the wiring arranged so that if you ever need to remove a board to replace something, the wiring is bound with cable ties so that each wire lines up with the appropriate connection point. Along similar lines, if at all possible, when building the boards (most commonly on Veroboard), keep connections along one edge of the board. This will mean adding jumpers on the Veroboard, but that's far better than having wires all over the board itself. Not only does it mean that wiring is simpler, but it also makes mistakes less likely.
Trimpots are a fact of life for any power supply. Voltages and currents need to be set, and meters calibrated. Thermal sensing also has to be calibrated, so nearly all power supplies will have numerous trimpots - you simply cannot rely on fixed value resistors to provide the proper conditions for anything. If you were to build the Figure 6.1 circuit as a dual supply, with the thermal protection and metering, you'll have at least nine trimpots to set everything up correctly. This is pretty much normal for power supplies, but some may have more!
Make sure that important parts of the supply are easily separated from the rest (and the chassis). For example, the heatsink assembly should be made so it can be removed, and all transistors can be accessed without having to dismantle the entire module. One design I've seen has the main filter caps directly in front of the output transistors, so they cannot be removed without removing the filter caps (or the transistors) from the circuit board. The location of the caps is such that you simply cannot access the transistor mounting screws once the assembly is completed. I strongly recommend that you avoid any similar errors. Having to remove (and/ or desolder) components or boards to gain access to any part of the supply makes it a nightmare to work on later. Consider that it may be in operation for 20 years or more before it needs servicing, and by then you will probably have forgotten many of the 'finer points' of the circuit. After that long, you may not even have the schematic any more, so make sure that you include one inside the case!
While the basics of a power supply aren't overly complicated, there will always be far more wiring than with any typical audio project. This is unavoidable unless you increase the overall cost even further by making your own PCBs. While doing so means a more professional product, there's no guarantee that you'll get the design right first time, and having to make modifications can be very time consuming. If an error has been made on a PCB layout, it can be difficult to diagnose and locate the error so it can be fixed. In general, it's likely to be much easier to hard-wire the final output section. Because of the high currents involved (which may be present for hours at a time), a normal PCB doesn't offer low enough resistance or high enough current capacity unless you use very wide tracks (I'd suggest a minimum of 5mm tracks for 5A, but even that is marginal for continuous duty).
While it may seem like a minor quibble, I strongly recommend that you use an IEC socket for the mains. In my long experience with test equipment and other gear, there's not much that's quite as annoying as a fixed mains lead. Rather than just unplugging the IEC plug from the back if it needs to be moved, you may have to trace a fixed lead back to its mains outlet, then disentangle it from other leads for the rest of your test bench gear. Depending on just how much gear you have, that can actually be a bigger challenge (and pain in the backside) that you think when it's first installed and plugged in. A minor point, but one worth remembering. Very few test fixtures that I've built have fixed mains leads, and I maintain a good collection of IEC mains leads!
There's one remaining challenge. To test the various parts of your supply before it's fully wired, you need ... a power supply. The chances of getting everything right first time aren't good, so if you don't have a power supply, you will have to devise a way to check that the various sections work properly without the risk of smoke if something isn't right. You may be able to use 'safety' resistors in series with the main supply to limit the damage if there's a wiring error, or (if you have one) use a Variac and a current monitor (see Project 139 or Project 139A so you can test for excessive current as the voltage is increased. Many parts of the supply won't work properly at reduced voltage, so there is always a risk. Testing and calibrating power supplies is not a trivial task, so you'll have a lot to do to get it completed.
While I've only described the basic supply here, many commercial supplies include a 5V output (usually rated for around 3A), and a few include a ±12V supply as well. Because you never know just how the supply will be configured in future use, these will both be fully isolated. Once you tie the ground (or common) connections together internally, that limits what you can do with the supplies. As already noted, you can never anticipate what you'll use a supply for when it's first built, and it would be unwise to assume anything in advance.
This means at least one, but possibly two additional transformers, plus the rectifiers, filters and regulators. You also need more space on the front panel for the connections. Most commercial supplies do not provide metering for any auxiliary supplies, and the circuitry doesn't need to be anything especially fancy. A couple of P05-Mini boards can be used, one for a single +5V output, and the other for ±12V.
Compared to the cost of the rest of the supply, these can be added for (almost) peanuts, with the possible exception of the transformers. Alternatively, they can be built as a separate unit, which does have some distinct advantages. Predictably, I have one of these as well as those on my workbench, and while it doesn't get used a lot, it's invaluable when I do need an extra supply that's isolated from all the others. It's also small enough that I can take it from the workshop to my office, where I also perform some testing and development work. Indeed, that's where it is at the moment.
There are precautions that should be followed with any variable power supply. Unless there is a switch that disconnects the DC (or reduces the output to zero), the supply should never be powered on with your load connected. Most circuits have to go through 'startup' phases (capacitors charging, zener voltages stabilising, etc.) before the output will be stable. If your load is connected, it may be subjected to a dangerous voltage, and current limiting may not be enough to prevent damage. Indeed, until all internal circuitry has the required operating voltages, there may not even be any current limiting!
With the Figure 7.1 circuit, once the supply is powered on and working, reducing the voltage to zero with the switch will work. However, during 'startup' (after mains power is applied), this may not be the case! Nothing should be connected to the output when the mains switch is turned on, because the output can be unpredictable. This has been confirmed by simulation - even with the switch turned off, the output rises to over 4V momentarily when power is applied. The Figure 6.1 circuit should be better in this respect, but it's still best not to have your load connected when the mains is turned on.
The power should be turned on, voltage reduced to zero while you make connections, and then the voltage can be set for the desired level. If testing something for the first time, use a low current limiting threshold to minimise damage if there's a fault in the DUT. If you need a current-limited supply, the voltage should be set such that the current limit is reached, but not beyond. For example, it you wanted to ensure a current of 1A through a 10 ohm load, the voltage only needs to be set for an open circuit voltage of around 12V. A higher voltage setting only increases the risk to your load if something goes wrong.
Setting a low voltage (just sufficient for the task) does not reduce the dissipation in the series pass transistors. The only reason is to ensure that the output capacitor(s) can't charge to 25V, then be discharged through the load. This would almost certainly guarantee that the instantaneous current will be much higher than the threshold set. This isn't only advice for the circuits shown here - it applies to all power supplies unless the operating instructions indicate otherwise. Most will advise against connecting anything until the voltage and maximum current are set before you connect the load.
There are several power supply designs that use a microcontroller to manage the functions, but be very wary of anything (DIY or commercial) that requires you to 'program' the voltage or current using a keypad. The use of low-tech conventional pots means that you can increase the voltage (or current) with the twist of a knob, and quickly reduce the voltage if any anomalies are seen. Trying to do this using push buttons is usually impossible, and much damage may be caused simply because you couldn't reduce the voltage quickly enough at the first sign or trouble. The 'high-tech' look and feel of a programmable power supply may be appealing, but it's impractical for anything other than laboratory tests, where the equipment being powered is a known quantity from the outset.
If all of the above hasn't frightened you away from the idea of building your own supply, I strongly suggest that you start with something fairly simple (such as Project 44). I know that DIY is about doing it yourself, but that should hold true only when it makes sense. As discussed earlier, I did build a ±0 to 25V, 2A supply with fully variable current limiting, thermal cutout and a dual speed fan. It's been in fairly consistent use for around 30 years (at the time of writing), and has never let me down. However, it's a complex circuit, and isn't really suitable for amateur construction. Rather annoyingly, the circuit diagram cannot be found, and it's not an easy circuit to 'reverse engineer'. With seventeen transistors, five opamps, two 12V regulator ICs, five trimpots as well as the expected bunch of resistors, diodes, filter caps, switches, meters and voltage/ current setting pots, it's not something I would recommend - even if I did have a complete circuit for it. The cost would be considered unacceptable to most constructors who may not need it all that often anyway.
The simple circuit shown above (Figure 7.1) is not bad. It's not as good as the one I built, but it's certainly acceptable for normal test-bench work. It does have the advantage that it can limit at a lower current than mine (~50mA is my minimum), and that's useful for sensitive circuitry. More importantly, it's simple enough to build even on Veroboard, with the current limiting circuits wired directly to the switch and voltage setting pots. This leaves only the basic circuit on Veroboard, which should be fairly straightforward. Overall, the Figure 6.1 circuit is better, but the switching for series parallel operation needs to be done with great care.
Perhaps surprisingly (or perhaps not), current sensing is generally far more difficult that it seems at first. It's pretty easy if you use a simple switched resistor scheme, but making it adjustable is not so straightforward. There are specialised ICs that are designed for this exact application, but most are SMD only, and they're not inexpensive - especially if they are only available in a pack of five. This is very common with SMD parts. Of course, this is just the sensing part - it's still necessary to get current regulation. As already noted, at the transition point (from voltage to current regulation), there are two separate regulators, both trying to impose their will on the output. Without a great deal of design time, the result is often oscillation (either transient or continuous).
The main idea of this article is to show you some of the options available. Ideally, most DIY constructors want something that does the job, is reliable, and doesn't cost a small fortune to build. If it can use parts you already have available, then that's even better. If you do have to buy the parts, you want to be reasonably sure that the circuit you choose is up to the task. As already noted, the circuits I've shown had to be adapted to ensure reliability (especially with low output voltage and high current). Failure to provide protective measures (current limiting, fan and over-temperature cutoff) will result in a circuit that not only lets you down, but may blow up the circuit you're testing as well.
When you look at the cost of the components needed, you'll discover very quickly that they add up to a fairly scary figure. Just the transformer(s) will be expensive, and while many of the parts are cheap enough, that doesn't apply to the filter capacitors or the heatsinks. You also have to provide a case and other hardware, and that will require significant machining to accommodate meters, fans, connectors, etc. It's very doubtful that you'll spend less than the equivalent of AU$400 in your currency of choice, even if you have many of the smaller parts in stock. I've seen a 0-30V, 3A dual supply for as little as AU$325 on-line, and it's highly doubtful that you can build one for less unless you have almost everything needed in your 'junk box'.
This should not under any circumstances be seen as a construction article! It is intended only to demonstrate that building even a modest bench supply is not a trivial exercise, and that there are considerations that you may not have thought much about. Some of the designs you'll find elsewhere on the Net are not well designed, and fail to provide adequate safety margins for the series-pass transistor (in particular) and most have no warnings about transistor SOA, thermal failure or any of the things that can go awry. As this article has shown, there are many things that can go wrong, especially if any part of the supply is underrated for the abuse it will get in normal use.
Main Index
Articles Index