|
| Elliott Sound Products | Regulators Part II |
Main IndexArticles Index Apart from their use within equipment (which is the main topic here), regulated supplies are very handy pieces of test gear. Ideally, a test supply will have a voltage range sufficient to handle everything from logic circuits up to power amplifiers, preamps, and any other electronic circuits that are either faulty, or have just been built. The inclusion of current limiting is especially handy, as you can set the limit low enough that it won't cause any damage (preferably less than 100mA). By increasing the voltage slowly, any fault will cause the current to rise very quickly after you reach a voltage that causes the fault to manifest itself. A very common requirement for power supplies is for battery charging, for lead-acid, nickel-cadmium, lithium or metal hydride cells and batteries. I suggest that you also read Bench Supplies - Buy Or Build?, as there are some relevant points made, along with some more circuits to consider.
A good supply is ideal for charging batteries, as you can set the maximum voltage and current independently. If the battery (or cell) is fully discharged, you limit the charge current to a safe value, so the supply's output voltage will fall, rising as the battery charges. Once charged to a reasonable level, the voltage will remain stable and the current will fall as the battery approaches full charge. For example, to charge a Li-Ion cell, you'd set the open-circuit voltage to 4.2V, and the current to perhaps 1/10 C (i.e. one-tenth of the cell's capacity, say 250mA for a 2,500mA/h cell).
If testing audio equipment, you can verify that the circuit draws an appropriate amount of current (this depends on the circuitry), and doesn't misbehave when a suitable operating voltage is reached. Output voltage (for dual supply circuits) should all be close to zero volts, and if you have at least a couple of amps available, even low-volume tests can be done with power amps connected to a dummy load or a speaker. You can also test power supplies! A 15V regulator should show low output until the input reaches around 17V, and after that the output should not increase beyond 15V. If it does, you know you've made a mistake before it has the opportunity to damage the equipment with which it will be used.
I have a number of different power supplies, and one or more get used almost every time I test a new project or circuits shown in articles. They range from fixed ±12V switchmode (in series for 24V), a fixed 5V switchmode supply, a variable 0 to ±25V supply with current limiting, and a Variac adjustable supply that can provide isolated AC up to 50V, and unregulated DC up to ±25V. The one that gets used depends on what I'm testing. When all else fails, I have another (external Variac controlled) supply that can give up to ±90V at 10A or more. Nothing gets attached to that until it's been verified as being fully functional with one of the others!
Most people's preference for high current is a switchmode supply, but they come with limited voltage ranges (the most common are 5, 12, 24 and 48V). Some have a trimpot to let you control the output voltage over a limited range, others don't. If low noise is a requirement, then you can use a switching supply followed by a linear regulator, and the circuits shown in this article can all be provided with a DC input from a suitable SMPS. Whether you can get the input voltage you need is another matter altogether!
Power supply units (aka PSUs) are everywhere, from large and imposing laboratory units down to 'plug packs' (aka 'wall warts'). They can be regulated or unregulated, but most small switchmode types are regulated, while older (transformer-based) supplies generally were not. When metering is included (with or without a connected computer), these are also known as SMUs - source measure units. For some general ideas for bench power supplies, see Bench Power Supplies - Buy Or Build? Note that it's an article, and the circuits are not part of a construction project.
Most of the time, when anyone mentions voltage regulators we think of IC based solutions. These have been with us for a long time, and they are perfect for most preamps and other relatively low-voltage (5-15V), low-current (less than 1-2A) applications. However, there's also a need for regulators that can provide higher current, higher voltage, or a combination of both. An example is Project 221, which is intended to allow you to run a low-power tweeter amplifier from the main supply of a bigger amplifier.
The circuitry used in the project is deliberately very simple, and it has minimal protection because its output isn't exposed to the outside world (an inherently hostile place for electronics). There's often a need for a regulator that can supply voltages in the range of 50 to 100V (sometimes more), and at relatively high currents. Even with a linear regulator, there's no set limit to the current you can get, but ultimately it comes down to cost. You might have a suitable transformer and other parts, and be understandably reluctant to buy (or try to build) a switchmode supply that can deliver (say) 80V at 10A or more. This is a serious undertaking, but a fairly simple linear regulator may be possible from parts you already have.
This article is based on linear regulators (no switchmode designs), which have the distinct advantage of being (electrically) quiet, something that only very heavily engineered switching regulators can achieve. However, all linear regulators are inefficient, and dissipate significant power at high output current. I'm only going to show series regulators (as opposed to shunt types), which use a transistor in series with the incoming supply and the output. I'll also only show only positive regulators, as negative types (if required) simply use PNP in place of NPN transistors (or negative versions of IC regulators where applicable), and input voltages, diodes and polarised capacitors are reversed.
The basic regulator will be intended to provide a nominal 24V output, at up to 5A or so for most of the design ideas shown, but a few are variable. Output current can be increased by using either higher current series-pass transistors, or using two or more in parallel. For the purpose of the ideas described, the series-pass transistor (TIP35) is assumed to have a gain (hFE) of 45, as that's around the figure you'll normally get (the datasheet says it can range from 15 to 75 with 15A collector current, and the simulator model assumes hFE to be 55). I've also assumed the driver transistor (mostly a BD139) to have an hFE of 75. In reality, these figures will vary, but if you design for the 'worst-case' you may end up with a design that needs too much current. If you design for the 'best case', the design may fail to provide the required current. The TIP35/36 devices are rated for 125W at 25°C.
For the examples that follow, a loaded input voltage of 30V is assumed (which will typically rise to ~35V with no load), and the output voltage is nominally 24V. It will be lower with simple circuits because they have no feedback to correct the output voltage. Output current can range from 1A up to 10A, with the intention of keeping the dissipation in the series pass transistor below 60W if possible.
With a few changes, any of the circuits shown can be driven from an input voltage up to 100V (using a TIP35C), and provide an output voltage of 5V to 90V. Output current depends on the number of output transistors used and the required current. Most of the basic circuits shown are rated for up to 5A with a nominal 24V output, but that's an arbitrary limit I set to make the circuits comparable to each other.
You won't hear many people saying this, but in most applications, regulation isn't necessary. We use regulated supplies for preamps because they provide a nice, low-ripple supply that will never subject the opamps to any voltage above the maximum allowable. However, it doesn't matter if the voltage is ±12V, ±15V or slowly varying between the two! The opamps for an audio circuit really don't care about the actual voltage, nor if it's different between positive and negative. We expect them to be the same, but it doesn't matter. In some cases (particularly for power supplies), the opamp supply voltages may be radically different. The same applies to many other supplies, other than those being used for precision test and measurement circuits.
Accordingly, it's likely that even the simple circuits described will be more than satisfactory for many applications, where you need fairly high voltage and more current than you can get from a 3-terminal regulator IC. Current limiting can be useful, but it's not always essential, it makes the design more complex, and it's more likely to misbehave under some conditions. Sometimes, all you need is an overcurrent trip (an 'electronic fuse') which is far less stressful to implement.
Because transistor gain is always a bit of a lottery, I ran some gain tests on a number of TIP35C transistors. With a collector current of 216mA (averaged), the gain was 36. Increasing the current to 400mA, the average gain was 42, rising to 46 at just under 1A. The lowest gain measured was 25 at 150mA, and the highest was 55 at 1.1A. That is in keeping with my expectations, so the simulated circuits shown below will work as shown. Of course, in any batch of transistors there can be 'outliers' that have higher or lower gain than anticipated (one had a gain of only 30 with 20mA base current), and a final design should account for that.
In all examples shown, low value resistors (< 1Ω) are wirewound types, typically 5W ceramic types. All low values are shown in mΩ, so for example, 100mΩ is 0.1Ω. In most circuits, you'll see a reference to the 'series-pass' transistor. This provides the output current, and (for feedback regulators) its base current is controlled continuously to ensure that the selected voltage is delivered, regardless of output current (up to the maximum allowable).
In any regulator (voltage or current), a stable reference voltage is required. It doesn't matter if the output is a fixed or variable voltage, a reference is still necessary. For most simple supplies, a zener diode is the easiest and cheapest, but it is not the most accurate. A zener diode's voltage is dependent on its temperature, with the exception of 5.6V zeners. There are two thermal effects that cancel each other with a 5.6V zener diode, but this doesn't work with other voltages. If you need high stability, zener voltages between 5.1V and 6.8V are pretty good, but this degree of accuracy isn't always needed. See AN008 - How to Use Zener Diodes in the ESP application notes section for more detailed analysis. IC regulators (including adjustable reference 'diodes') use a bandgap reference, typically 1.25V or 2.5V.
The supplies shown use a zener diode, and its current will vary from a maximum at no load to a minimum at full load, because the series-pass transistor(s) need base current from the zener stabilised reference. Sometimes, you may need to use (for example) a pair of 12V zeners in series, rather than a single 24V zener (the same applies to other zener voltages as well). Zener diodes should always be operated with 10% to 50% of rated current (4mA to 20mA maximum for 24V, 1W zeners). Operating at more than 50% of current rating causes zeners to run hot, and they're difficult to cool effectively.
This is easily overlooked, especially when it all appears to be so straightforward. It's a simple job to work out the maximum current for any zener diode, knowing the maximum dissipation and the voltage. For example, a 12V, 1W zener can handle a maximum current of 83.3mA ...
IZ = PZ / VZ
Using this, you can determine that a 12V, 1W zener should carry between 8mA (10%) and 40mA (50%) at no load and full load. The zener current is at its maximum with no load because no current is drawn by the series-pass transistor (includes Darlington and/ or paralleled transistors). When current is drawn by the load, the series-pass transistor's base current increases, leaving less current for the zener diode. If the current falls below 5%, the regulation may be adversely affected.
If you don't provide at least 5% of the rated zener current, the voltage may be lower than expected. Most zener diode datasheets state the test current, which is usually between 5% and 20% of the maximum. Likewise, many datasheets also state that the maximum current is about 10% less than the figure given by the formula shown above. The test current is usually stated, and that's usually a good value to aim for. As noted though, the current varies, so you have to find a 'happy medium' (ideally between 10% and 50% of the maximum). This can be extended to 5% to 50% if you can't manage to keep the current above the 10% value without exceeding the maximum. Meanwhile, you have to allow enough current to drive the series-pass transistor(s).
While it possible to operate a zener at its maximum power rating, it's definitely not recommended. Even at 50%, the diode will run fairly hot, as the only heatsink it has access to is the copper track of a PCB, or other component leads when a PCB isn't used. My test has always been to discover if I can keep holding a component without shouting "rudeword" and dropping it or letting go. This applies to pretty much everything with the exception of ceramic wirewound resistors. Even then, excess heat is likely to cause damage to PCB materials or other parts nearby (especially electrolytic capacitors). It's not uncommon to see burnt patches on a PCB beneath wirewound resistors, and sometimes the solder pads and/ or tracks will de-laminate (separate from the fibreglass).
Rather than a zener diode, you can also use a precision voltage reference, such as the TL431. These can be used with a pair of resistors or a resistor and a trimpot to get a very accurate and stable reference. The maximum allowable voltage is 36V, and the maximum current is 100mA ... but not at the same time. For the TO-92 version, maximum dissipation is 770mW, but it would be unwise to operate the IC at more than 500mW, and preferably less. My suggestion would be around 250mW, so at (for example) 24V, the operating current will only be 10mA. For high output current, a very high gain output stage is needed for the series pass transistor(s) and their driver transistor. MOSFETs are tempting, but come with caveats - see Section 12 - Using MOSFETs.
The general idea for a simple regulator is shown in Figure 2.1. While this will work, it's less than ideal, so we need to add a few parts to improve performance. If the output current doesn't need to be more than about an amp or so it will do the job, but it is quickly found wanting if you need any more. Because there's only a single transistor, R1 has to be able to supply enough base current for Q1 and provide the current for the zener diodes. Even for 1A output at 24V (nominal) with a 30V DC input, R1 has to supply a minimum of 50mA, 28mA 'reserve' current for the zener diodes and 22mA for the base of Q1. With no load, the total current is passed through the zeners. The problems get worse if more current is needed.
Note the diode connected across the series-pass transistor. That's there so that if (when) the supply is connected to a voltage source (such as a battery) but isn't powered on, the diode passes voltage back to the input. By including this, the transistor can never be reverse-biased which can lead to failure. It also bypasses voltage spikes (from inductive loads, motors, etc.) around the transistor. This should be included in any power supply, even if it's not exposed to the outside world.
Figure 2.1 - Basic Regulator Circuit
The simple circuit shown has disadvantages, as you'd expect. The zener current is higher than it should be (so two 12V zeners are used in series) and it varies too much depending on the load. Regulation is mediocre, and there's no protection. If the output is shorted it will supply as much current as it can, leading to almost instant failure of the series-pass transistor. Because there's no driver transistor, the base current that needs to be provided varies widely. We must provide enough current to accommodate the 'typical' hFE, which as stated in the intro we'll take as 45. That means you need 22mA base current, so R1 has to be around 180Ω (30V input), and rated for at least 1W. The zener current will be 61mA with no load (35V input), and around 22mA at full load. It's also necessary to allow for a higher than expected input voltage with no load. If it comes from a transformer, bridge rectifier and filter cap, it will rise to about 35V, and this is the most likely voltage source.
Figure 2.2 - Improved Basic Regulator Circuit
By adding a driver transistor, we lose a bit of output voltage (around 0.7V), but the circuit is far more attractive overall. The pair of zeners can be replaced by a single 24V zener, and by splitting the feed resistance into two (2 x 470Ω) we can add a capacitor to ground. This attenuates ripple for a cleaner output. A larger capacitor reduces noise better. While it's often seen, adding a capacitor in parallel with a zener diode is close to useless because their dynamic resistance is very low, so the cap doesn't achieve anything useful. In Figure 2.2 (D2, marked 'Optional') is used to offset the emitter-base voltage on one of the transistors, or you can use two to get closer to 24V output.
The Figure 2.2 circuit is easily capable of 5A output with a 30V input. Zener current is well within the desirable limit, and even with no feedback, the regulation is acceptable. It's not precision, but nor are most of the circuits shown in this article. They are best described as 'utilitarian', in that they will do the job 'well enough' for most applications. If you need precision, you won't get it from simple discrete circuits.
The two regulators so far are very basic, having no form of protection, and no way to adjust the output voltage to be closer to the desired 24V. This is because they lack feedback, which is essential for reasonable performance. Feedback is also used to provide good overload protection, but that will come later. The Figure 2.2 circuit is capable of reasonably good regulation, although the output voltage is only about 22V with a 5A load. The output of both of these simple regulators can be boosted a little, by adding a diode (or two) in series with the zener. The forward voltage of the diode(s) helps to offset the base-emitter voltage of the transistors.
Note that for all of these simple regulators, I've only shown a single TIP35 power transistor. In most cases, at least two should be used (in parallel, with emitter resistors) to keep the temperature down to something 'sensible'. The emitter resistors can also be used for current sensing, and an additional resistor isn't needed. If there are two transistors in parallel, the emitter resistance should be double the value shown, and sensing taken from both resistors as shown next.
Figure 2.3 - Improved Basic Regulator Circuit - Parallel Output Transistors
The above should be used in most cases, but only a single transistor (and current sense resistor) are shown in the other circuits for clarity. It's important to sum the two voltages dropped across R4A and R4B, because the transistors will not be matched, and one will supply more current than the other. The effective current limit resistance is 135mΩ, which will bias 'on' a current limit transistor at around 4.8 - 5.2A total output.
One of the first things that regulators that interface with the 'outside world' need is current limiting. It comes with caveats though, especially if the output is shorted (which will happen). Figure 3.1 shows the general principle, which has been around almost for as long as discrete regulators. It's very basic and just uses diodes. When the combined base-emitter voltage and that across the sense resistor (R4) exceeds the voltage drop of the diodes (about 2.6V), the diodes shunt the base current from Q2 (the driver) to the output. As a current limiter it's best described as "better than nothing", as it lacks any pretense at precision. However, it might just save the series-pass transistor(s) from failure, provided the fault is transient.
In reality, it's almost impossible to apply a direct short across anything, because there are always connectors and wiring forming part of the circuit. The total resistance depends on many factors, but it's 'traditional' to always design for the worst case. In fact, the transformer, bridge rectifier and internal wiring also add to the total series resistance, but in general it would be unwise to assume more than 100mΩ (0.1Ω) of external resistance.
Figure 3.1 - Improved Basic Regulator With Diode Current Limit
As shown, the simulator tells me that current limiting starts at 5A, and with a shorted output the current is 6.5A. A better scheme is shown next. R4 is the current sense resistor, and if the voltage across it exceeds 0.65V, Q3 will conduct, and it will bypass base current from Q2 to maintain the set current. The advantage is that the current limiter has gain, so it is more accurate than the Figure 3.1 circuit. With 0.1Ω (100mΩ) for R4, current limiting starts at about 5.5A, with the final current into a short-circuit limited to about 6A. This still isn't a precision limiter, but it's a lot better than a string of diodes.
Figure 3.2 - Improved Basic Regulator With Variable Current Limit
A simple transistor current limiter will often rely on a resistor value (for R4) that's unobtainable. The solution is to add a low-value pot (VR1) so the current can be adjusted. This can be used with any of the following circuits, and it lets you set the current with reasonable accuracy. Because the single current-sense transistor has limited gain, expect the current to vary by up to 300mA or more from the onset of limiting to a shorted output. This isn't a problem, as the limiting is intended only to provide some protection for the series-pass transistors, and it's not intended to be a precision circuit.
One thing that may appear strange is the use of an NPN transistor for limiting. It doesn't look like it, but both the base and collector are positive with respect to the emitter, so it must be NPN. In some of the other circuits shown below, the transistor is PNP, and the base and collector are negative with respect to the emitter. This can get confusing, but it depends on how the current limit circuit is configured. Make sure that you follow the drawings thoroughly to ensure that you understand when (and why) an NPN or PNP limiter transistor is used.
The problem with all simple limiters is that Q1 will dissipate up to 175W (35V across the transistor at 5A), far more than a TIP35 can handle under short-circuit conditions. It will usually be less in reality, because the incoming DC supply will never have perfect regulation because it has some internal resistance (transformer windings, diode resistance and wire resistance). Even if these add up to 0.5Ω, Q1 will still be subjected to a dissipation of almost 160W, and it will still fail. Simple limiters require that the series-pass transistors can dissipate the maximum power, with particular attention paid to safe operating area. See The Elephant In The Room for details.
Figure 3.3 - Improved Basic Regulator With Foldback Current Limit
The answer to this is a technique known as foldback current limiting. As the voltage across the series-pass transistor increases, the allowable current is reduced. With the arrangement shown above, the circuit can only provide 1.6A into a short-circuit, while still being able to provide 5A at full voltage. The addition of just one resistor (R6) means that as the output voltage falls, Q3 gets additional base current through R6, turning it on harder and reducing the available output current.
The highest power in Q1 is 60W at an output current of about 3A and an output voltage of 9V. The general characteristics for foldback limiting are shown in the following graph. This is for the circuit shown above, and the trends are similar with most foldback regulators. The short circuit current is determined by R4, R5 and R6, and they are interactive. If any one of these resistors is changed, the limiting characteristic is modified. There is some leeway with R6 without seriously affecting the maximum current, but not very much.
Figure 3.4 - Foldback Current Limiting Voltage, Current And Power
As you can see, as the current increases, the voltage remains steady until the maximum (4.8A) is reached. This causes the output voltage to fall, which allows more current through R6, turning Q3 on harder. With the output shorted, the maximum current is 1.6A, and Q1's dissipation is 46W. Worst-case dissipation is 61W, with an output voltage of 9.3V and a current of 3A. All foldback limiters have a hidden 'gotcha', in that the circuit may not power up normally with anything close to full load. Foldback limiting is a form of positive feedback, and like all positive feedback systems it can be unstable under some conditions.
Figure 3.5 - Foldback Current Limiting (Traditional View)
Figure 3.4 shows the 'traditional' way that foldback current limiting is shown on a graph. A 'regular' current limiter simply provides constant current at any voltage once it's active, but the foldback limiter reduces the current as the load impedance falls. With simple limiting, if the regulator's input voltage is 30V and the output is shorted, it will deliver 5A, resulting in a regulator dissipation of 150W. With a foldback limiter, the maximum current with a shorted output is 1.5A, so the regulator dissipates only 45W. The lower the output voltage (with intermediate load currents), the lower the output current. As you can see, with an output voltage of 10V, the basic limiter still provides 5A output, where the foldback limiter reduces that to about 3.1A. You can work out the dissipation for each limiter type easily, and a foldback limiter always has lower power dissipation in the series pass device(s).
While the drawing shows a sharp transition from voltage to current regulation, this isn't the case with simple limiting circuits. In most cases, you'll see the voltage sag noticeably as the maximum preset current is approached, and for a 2.5A limiter this may start to be measurable from perhaps 2.3A onwards. Beyond the preset current limit, simple limiters will also allow the current to increase with decreasing load resistance. A precision current limit isn't usually required, and even the most basic arrangement will be sufficient to prevent disasters if everything is designed to handle the worst case.
Figure 3.6 - E-Fuse Protected Basic Regulator
There is another way to provide protection, and this one is (close to) bulletproof. An SCR (T1 for 'thyristor 1') is triggered if the current exceeds a preset maximum. Once it's triggered, the SCR shorts out the zener diode, and reduces output voltage and current to zero. It's reset by turning the power off and on again, or you can use a pushbutton in parallel with the SCR. It will cease conduction when it's shorted out, because there is no holding current. The nice part of this is that if the fault is still present, the SCR will be triggered again as soon as you release the pushbutton, and there is no way to force the regulator to provide more than around 6.4A. The extra capacitor (C3) is necessary to allow the regulator to charge C2. Note that R1 and R2 should be 1W if you use this arrangement, as they will dissipate just under 0.5W when T1 is triggered.
Note the connection of the 'Reset' switch. I have seen similar circuits where the switch is in series with the SCR, but that means that if the switch is open there is no protection! By having the switch in parallel, provided the fault has been cleared, output voltage is restored when the switch is released. If the fault is still present, the SCR will be re-triggered the instant that the switch is opened, so protection is never compromised. There are many things that have to be properly thought through with circuitry, and just putting a switch in the wrong place can lead to failure.
For many regulators, this arrangement can be the saviour of the series-pass transistor. While R4 does reduce the regulation (the output will fall by 0.5V from no-load to full load), this is rarely an issue with a simple design. R4 can be repositioned so it (and Q3 with associated resistors) comes before the regulator itself. The position doesn't matter, as the extra current for the regulator is minimal (only about 10mA with a 30V input). There's a solution for everything, even if it's not immediately obvious. There's also another way (as always), and it's far from obvious.
Figure 3.7 - 'Lossless' Current Detector
The reed switch shown above has the advantage that there is very little resistance in the circuit (I used 1mm wire, and heavier gauge wire would be used for higher current), but there is a (very) small delay because it's a mechanical contact. With the switch I tested, it requires 32 ampere-turns (2.3A, 14 turns) to operate, and it can be configured for almost any current you like. Anything over 32A would be a challenge though, as that implies less than one turn. Positioning the coil along the body of the switch provides some minor control over the trip current. Also of interest is just how fast the reed switch is. With only a small over-current (about 2.5A), it operates in 250µs - and yes, you did read that correctly. With a higher current it just gets faster, and I measured 200µs with 3A. That's not as fast as you'd normally expect from semiconductor circuitry, but it is still fast enough to protect the series-pass transistor.
Figure 3.8 - Reed Switch E-Fuse Protected Basic Regulator
The implementation is shown in Figure 3.7, and the trip current is set by the number of turns. Since all reed switches will be a little different, you'll need to test the coil and switch combination to work out the number of turns for the preset current. My test switch pictured above has 14 turns, and will trip reliably with 2.3A. If the winding is reduced to 7 turns, the trip current is 4.6A. Should the output be shorted, the instantaneous current from C2 will be very high, so operation should be close to instantaneous. If it's only used as an e-fuse, the exact current probably doesn't matter too much, as it's there for protection, not for precision current limiting.
When you have both voltage and current regulators (any form of current limiting), it's usual that the current regulator is 'dominant'. In other words, when the current limiter is active, it controls the output voltage and overrides the voltage setting. This usually (but by no means always) results in a stable circuit, because the two regulators cannot fight for control. By making the current control dominant, the preset current will be delivered whenever the load demands more, regardless of the voltage setting. The latter is automatically altered to maintain the preset current, unless the load current is less than the limit. Then (and only then) is the voltage control active.
The next set of drawings show feedback regulators, which have better regulation than the simple versions shown above. Feedback is used to ensure that any change in the output voltage is compensated by means of an error amplifier. This term explains what it does - if there's an error, the error amp makes the necessary compensation to restore the voltage to its preset value. All IC regulators contain an error amplifier plus comprehensive protection schemes. These include current limiting and thermal protection that turns the IC regulator off if the temperature rises beyond the preset limit (typically a die temperature of around 125°C).
Figure 4.1 - Basic Feedback Regulator
The above circuit used to be the mainstay of regulators before the advent of IC-based versions. I used it as a 48V phantom power supply in Project 93, but configured for much lower current. The feedback is via R5 and R6 to the base of Q3. If the output voltage falls, Q3 turns off just enough to restore equilibrium, and R4 (which can be installed for current sensing) has no effect on the output voltage because the feedback is taken from after the resistor. It can be used with foldback limiting (Figure 3.2), or an e-fuse arrangement as shown in Figure 3.4. Foldback limiting has to be set up carefully, because there are two feedback networks, one negative (to maintain the set voltage) and one positive (to provide foldback). The two will fight each other for control.
R7 may look out of place, but it's intended to stabilise the current through the zener, ensuring better regulation. By taking it from after the regulator, there is no injected noise (mainly ripple). Even at 5A output, the ripple is attenuated by over 40dB. The output voltage remains within 100mV of the target value from no load to 5A. If you need a more accurate voltage setting, either R5 or R6 can be replaced with a trimpot in series with a fixed resistor, allowing the voltage to be set precisely.
Far better performance can be obtained by using an opamp in place of Q3, but that comes with limitations. Most are rated for a supply voltage of no more than 36V, so high voltage regulators cannot be realised with readily available opamps. Despite the number of so-called 'super' regulators used to power preamps and the like, a circuit such as that shown is perfectly acceptable in most cases. The circuit can also be used as a 'pre-regulator', allowing preamp circuits to be powered from the main power amp supply, with the discrete regulator followed by an IC version. This will provide almost infinite power supply ripple rejection.
Figure 4.2 - Opamp Based Feedback Regulator
The circuit in Figure 4.2 uses an 'ideal' opamp (available in the simulator I use), and as such it's close to perfect. The current limiter comes in at 2.6A, and reduces the reference voltage to maintain the preset current. For true precision, the current limit circuit would also use an opamp, as that provides much higher gain than the two transistors, and therefore has much better control of the output current. Since this article is primarily about 'simple' regulators, adding another opamp is out of scope. Be warned that when you do include an opamp or additional gain stages, there's always the likelihood that the circuit will become unstable, and it's necessary to include compensation capacitors to roll off the gain at high frequencies, where oscillation is likely to occur. Bear in mind that the 'ideal' opamp can provide as much base current as is needed by the series-pass devices, where a real may be unable to do so.
The easy answer to the opamp conundrum (for high voltages) is to make it discrete as well, but the circuit becomes much more complex. In the majority of cases where you'd use a discrete regulator it's simply not necessary, but feel free to experiment if you want to. It will make voltage regulation better, but stability issues are always waiting to pounce. You will be able to get the voltage change (no load to full load) down to less than 1mV with a suitable opamp, but that's rarely important. It's a different matter if it's a lab supply where very accurate voltages are essential, but 'general purpose' regulators don't need to be that precise.
For any regulator, it's important to ensure that there is enough input-output differential to ensure there's no ripple 'breakthrough'. All regulators require some 'headroom', the difference between the input voltage and the output voltage. I've used an example of 5V in the examples, but that's often cutting it fine, especially if there's ripple on the incoming supply. This is where the selection of the transformer, bridge rectifier and filter capacitance is very important. If you get it wrong, the regulator will not perform as expected.
The transformer is at the heart of any power supply. To ensure that its regulation is adequate and to ensure it won't overheat, it needs to have a higher rating than you may expect. Capacitor input filters impose a heavy load on a transformer, so if the average output voltage is 30V and the current is 5A, that's 150 watts. However, transformers are rated in VA (volt-amps), and the VA rating should be not less than (around) 1.7 times the DC power. That means a 225VA transformer. Power supplies are used differently from (for example) audio amplifiers, and are often expected to provide full current for extended periods. To get a reliable 30V average DC voltage, the transformer will normally have at least a 25V secondary. The voltage will be close to 35V with no load, and AC current is double the DC current. A 25V transformer delivering 5A DC after rectification and filtering needs to deliver 10A AC (RMS), which is 250VA. Note that it doesn't matter if the DC output voltage is 5V or 25V, if the output is 5A then the transformer is still delivering 250VA. For safety, you'd use a 300VA transformer, as that's a standard size.
The circuits shown so far can function with an input-output differential of less than 3V, provided the average voltage remains at 30V or more. This means that the ripple's negative voltage can extend down to 27V (a total of ~8V peak-peak with an average of 30V) and the circuit will maintain regulation without ripple breakthrough. The next thing is to work out the DC supply for the regulator, which often produces a few mental shock-waves when you start to add up the pieces. For this, I'll assume the full-load ripple to be 4V P-P ...
The required capacitance for a given load current and ripple voltage is determined (approximately) by the formula ...
C = ( IL / ΔV ) × k × 1,000 µF ... where
IL = Load current
ΔV = peak-peak ripple voltage
k = 6 for 120Hz or 7 for 100Hz ripple frequency
Since I will always use 100Hz ripple frequency (50Hz mains), this can be checked easily, so ...
IL = 5A, ripple = 4V p-p, therefore C = 8,750µF (use 10,000µF)
This is well within expectations, and with a 25V transformer the average voltage (simulated) is just under 30V, with 2.9V p-p ripple. However, we've not considered the transformer's regulation yet, and it has a big influence on the final outcome. Transformers never provide the same regulation with a rectifier and capacitor load as they do with a resistive load (see Linear Power Supply Design for more on this topic). To be able to get 24V output at 5A means the transformer will have to provide an output current of close to 22A peak, or a bit over 9A RMS. We know that the voltage will sag under load, so we will probably need an output of 30V RMS to ensure that the voltage doesn't collapse too far. That means a 300VA transformer. It's only just big enough, but it will work at full load. Note that the RMS current from the transformer is almost double the DC current, something that isn't always appreciated or accounted for.
Of course, you may not need the full 5A output continuously so a smaller transformer (lower VA rating) may be suitable. This is entirely dependent on the application, something I can't predict. This is all part of the design process, and you need all of the information. Many people ask questions on forum sites with the bare minimum (and often not even that) and expect others to help them with a solution. It can't be done - all of the info needs to be available, and there's quite a bit of design work involved just to determine the transformer and filter capacitor requirements.
If you need a higher voltage, it's just a matter of increasing the zener voltage for the simple regulators, or changing the feedback resistors for the Figure 4.1 circuit. Ideally, the zener voltage for this will be around half the desired output voltage, so you'd use a 24V zener diode for a 48V supply. The series-pass transistor (Q1) will be happy with anything up to 100V input, provided you use the TIP35C. However, if you increase the voltage, you also increase the chances of failure if only a single transistor is used. My recommendation would be that if you double the voltage (from 24V to 48V) the output current should be halved (from 5A to 2.5A). It's important to always be aware of SOA - See #8.
Be particularly careful if the input voltage is much greater than the output voltage. While it's certainly possible to have 100V input and 5V output, it would not be sensible. Even with 1A of output current, Q1 will dissipate 95W (until it fails, which it will), and it's hard to get that much heat out of the transistor and into a heatsink. A heatsink that can dissipate 95W and remain at a sensible temperature is going to be a very substantial piece of aluminium - you're looking at a heatsink with a thermal resistance of 0.27°C/W for a temperature rise of 25°C (50°C heatsink temperature). The maximum allowable DC current through a TIP35C with 95V across it is only 100mA, limited by second-breakdown. No-one ever said that this was easy, other than someone who's never done it.
You can use higher voltage transistors if you really do need to reduce a high voltage to the required voltage, but you must consider the safe operating area (see below). There are many considerations, and it's not just about the transistor voltage rating. All resistors will dissipate more power too, and in general, regulating high voltages can be particularly challenging.
Figure 6.1 - 48V Feedback Regulator
As an example, Figure 4.1 is easily modified to provide 48V output at 2A. The circuit can supply more, but in the interests of minimal change, 2A is realistic. There are a few resistor changes, and Q3 is changed for a higher voltage version (The BC546 is rated for 80V) and the zener voltage increased to 24V. Even without any adjustments, the output voltage (simulated) is 46V, well within the limits set for phantom microphone power for example.
For other voltages (and currents) it's a matter of selecting the component values to ensure sufficient base current for the series-pass devices, a stable zener current, and transistors that are within their safe operating area at all times. I didn't include current limiting, but an e-fuse circuit would be useful if there's any chance of the output being shorted. As you can see, the topology isn't changed at all, and with suitable high-voltage transistors a circuit such as this can regulate almost any voltage you like.
High voltage regulators were very uncommon with valve (vacuum tube) amplifiers, but there are some valves that are particularly fussy about the screen voltage. Transmitting valves (for RF work) have been used for audio, with one I'm familiar with being the 6146B. With a 750V plate supply, failure was assured if the screen was operated at more than 200V, and the only way to ensure reliability was a regulator. When these were built, no transistors were available that could handle the voltage, so it used a zener-controlled valve regulator. It worked well enough, but today there are many transistors that would be a lot better.
Often, the only thing you need to do to get more current is to use paralleled series-pass transistors, and you may also need to upgrade the driver transistor as well. Current up to 20A or so is usually not especially difficult, but the power transformer, bridge rectifier and filter caps become a serious (financial) problem if 20A or more is needed on a continuous basis. You'll also be looking for a pretty serious heatsink, depending on the load's duty-cycle. For momentary current up to 20A (less than ~100ms) you often don't need to do very much, but if the current is required for more than a few seconds you're probably better off with a switchmode supply. Again, it's important to be aware of SOA - See #8.
If you need lots of current at a relatively low voltage, a switchmode supply followed by a linear regulator will usually work well. The SMPS will be regulated, so you don't have to consider transformer regulation or other losses within a 'linear' supply. A voltage differential of 5V will normally be quite sufficient, and the regulator is greatly simplified.
Figure 7.1 - 24V @ 10A Regulator
The principles aren't changed one bit. We need an extra output transistor to handle the current, and that in turn requires a bigger driver transistor (Q2) and error amplifier (Q3). Due to the higher current through the circuit, we must ensure that everything is well within its limits for a long and happy life. Each output transistor has its own emitter resistor to force current sharing, but if current limiting or an e-fuse were needed, all three should be monitored, using summing resistors as shown in Figure 2.3. The higher the resistor value, the better, but we still need to keep the voltage drop to less than 0.5V, so 100mΩ resistors would be preferred.
Because the two base current feed resistors (R1 and R2) are a lower value, the bypass capacitor should be increased to ensure good ripple rejection. 220µF is ideal, and maintains much the same performance as we had with the lower-current version. While the circuit was simulated with a fixed 30V input, in reality it will likely be 35V at no load, and a 500VA transformer would be needed to maintain a voltage of not less than 30V (including ripple) at the input. Add to this the need for at least 20,000µF filter caps, a very good heatsink, plus the components. Adding current limiting would make it more complex of course.
In this case, the 'elephant' is SOA - safe operating area. There are three different parts to an SOA graph, the bonding wire limit (before it acts as a fuse), the thermal limit (how much heat can be removed from the junction) and second breakdown. Thermal and bonding wire limits are easy to deal with, but as shown below, second (or secondary) breakdown becomes an issue once the collector-emitter voltage exceeds 30V. While you may think that SOA only applies to the power transistors, it applies to every transistor in the circuit. The driver transistor is the next most at-risk, but it's unusual to see an SOA curve in datasheets for smaller devices (the BD139/140 are rated for 1.5A maximum, with a dissipation of 8W). It's always better to err on the side of making the driver transistor bigger (e.g. TIP41) than smaller, but you also need to consider the hFE at the expected collector current.
I've only shown the TIP35/36C graphs, as the 'A' and 'B' versions are identical, other than a lower maximum voltage (some suppliers only stock the 'C' versions). One of the reasons I recommend these transistors is that they are very rugged, and they are low-cost. The graph below was adapted from that shown in the Motorola datasheet, but it applies whoever makes the transistors. The essence of the graph is unchanged, but I made mods to the graph to make it easier to read.
Figure 8.1 - TIP35C, 36C Safe Operating Area
The second breakdown area is where things can get out of hand very quickly. The phenomenon is caused by 'hot-spotting' on the silicon die. If there's any difference between the temperature of one section versus another (which will always be the case), one small section will be a little hotter than the rest. This increases gain in that area, and also reduces the base-emitter forward voltage. The hot section then becomes hotter because it carries more current. This cycle continues until the transistor fails, which can happen very quickly. You can see that the SOA changes with time, so for DC it means lower voltage and/ or current than for momentary pulses. The shortest pulse shown is 300µs. For a regulator, we are primarily interested in the DC conditions unless we know (for certain) that short pulses are the normal load for the regulator in use.
For example, at a collector voltage of 30A, the maximum current is 4A, a dissipation of 125W (the full rating for the transistor). Increase the voltage to 40V and the current is only 2A (80W). A further increase to 50V, and current is only 1A (50W). At 100V, the current is reduced to 100mA (a mere 10W). You ignore the SOA of any transistor at your peril, because failure is never a matter of 'if', but 'when'! The transient ratings mean that you can get more current at higher voltages, provided the time is short. With 40V collector-emitter, you can get 4.5A if the 'event' is over in 300µs, so charging an output capacitor (for example) won't usually kill the transistor - provided the current is limited to remain within the SOA. The SOA topic is discussed in detail in the article Semiconductor Safe Operating Area, but with an emphasis on power amplifier designs.
In all of this, there is still another 'gotcha'! Note that the figures shown are all for a case temperature of 25°C. Maintaining this in use is generally impossible, so the maxima all have to be derated at elevated temperatures. For the TIP3x devices, the dissipated power is derated by 1W/°C (from the universally accepted 25°C), so at a case temperature of 50°C, the maximum power is reduced by 25W (maximum allowable dissipation is therefore 100W, not 125W). At a case temperature of 150°C, no power may be dissipated at all. The die (or junction) will also be at a temperature of 150°C, and any additional power will increase the junction temperature to the point of failure (150°C is the maximum allowable). Most bipolar transistors are the same in this respect, but some MOSFETs can tolerate up to 175°C junction temperature.
Failure to accommodate the SOA vs. temperature curves is a major reason for failure, and few people consider the thermal resistance from case to heatsink. A transistor dissipating 50W can easily have the case temperature a full 50°C above the heatsink temperature (1°C/W). See The Design of Heatsinks for a very detailed discussion of how to apply a heatsink, knowing the power to be dissipated, transistor specifications, etc. Heatsinks only seem simple, but there's a lot needed to get it right. Using the right thermal interface material (aka 'TIM') is essential to minimise the thermal resistance, which can mean success or failure of the end result.
There are countless power supply schematics on the Net, and a majority of them underestimate the power dissipation, and give nary a thought to SOA. Perhaps surprisingly, many of these circuits will work with most typical loads, but unfortunately, if you have a power supply, it will be used 'inappropriately' at some stage. This is the nature of a power supply, you never know what it may be expected to drive in advance, and it's not until you have one that you'll come up with 'exciting' ways to use it. Yes, I am speaking from personal experience, with at least 40 years using power supplies in ways I didn't envisage when I built my first unit. Fixed (internal) supplies have one major benefit - you know exactly what they need to drive, and they're generally in the same chassis.
A semi-discrete design can be engineered to have excellent performance, and an example is shown below. The error amplifier is now an opamp driving a transistor, so it has a vast amount of gain for high accuracy and very good ripple rejection. The extra complications are not particularly DIY friendly, as there's a lot of extra parts. Of greater concern is stability. No-one wants a power supply that thinks it's an RF transmitter with some loads, and stability needs to be verified at every possible combination of output voltage and current. In any high-gain circuit, ensuring complete freedom from oscillation can be surprisingly difficult, and power supplies are no different.
Figure 9.1 - Semi-Discrete Regulator
As you can see, the opamp needs its own power supply (±12V), and there are two capacitors to ensure stability. You may wonder where the reference voltage is, as it's shown using a zener diode for the other designs. The reference is the -12V supply! This circuit is adapted from Bench Power Supplies - Buy Or Build?, a discussion as to whether one should consider building a variable bench supply or not. It's been changed so that only the highest current range is included, and voltage adjustment has been set up to allow it to be trimmed with the preset. The circuit was originally devised by John Linsley-Hood and was published in 1975. Although the circuitry is rather dated, it will still perform very well. C3 and C4 are included to slow down the circuit, and these prevent oscillation. Their inclusion also means that there's overshoot and undershoot when the load is connected or disconnected, and this may not be desirable with some sensitive circuits. Q3 must be mounted on a heatsink, as its dissipation can be up to 2W.
If you only need a fixed voltage, and your requirements are fairly relaxed, this is not the kind of circuit you'd normally use. The article has several other circuits that are worth looking at, but the complexity is fairly high in all cases. Note that the Figure 9.1 circuit is designed to be able to drive full current into a shorted output, so it uses two TIP36 power transistors. They are within the SOA curve at all times, but a fan forced heatsink is essential. I doubt that many readers will find this an attractive proposition.
If you imagine that even better performance is needed (particularly accurate current limiting), then the pain is increased accordingly. When you have two feedback systems (one for voltage, one for current), there is always a point where both are active, and if not worked out properly they may be fighting each other for control. This will lead to instability (oscillation) that will usually be very difficult to suppress successfully, so there may be some combinations of output voltage and current that cannot be used without the supply oscillating. This is unlikely to be high on anyone's wish-list.
Figure 9.2 - LM317 Based Regulator
One arrangement that's very common is a current-boosted LM317 (and/ or LM337). Without external current limiting provided by Q3 and Q4, there is no protection at all, so a short or severe overload at the output will cause the booster transistors to fail. When current limiting is applied at the input side as shown, there may be some ripple on the output when the current-limit circuit is active. The only way to eliminate that problem is to have a separate sensing resistor at the output, but that affects regulation. Note that both emitter resistors for Q1 and Q2 are monitored, as the gain of the transistors will be different (emitter resistors notwithstanding). The ICs have their own internal bandgap reference, using 1.25V.
It's shown here using just a 10k pot to set the output voltage, and the trimpot (VR2) provides the ability to use a standard value linear pot to set the voltage for a variable supply. As shown, the output is adjustable from 0V up to 25V. The requirement for a clean negative supply for the current limiter and voltage pot is a nuisance, but that can be provided by a low-current regulator. There are endless possibilities for voltage regulation, and the circuit needs to be selected based on your needs. A boosted 3-terminal regulator is a good solution when you need particularly good regulation, but without protection it's vulnerable to damage by overload. Zero volts output is possible by using the low-voltage negative supply (as close as possible to -1.25V). This is used for VR1, R2 and the emitter of Q4. The three diodes are important. Without D2, if the output is shorted the IC will be damaged. The other two protect the supply against an external voltage (of either polarity).
This is about as simple as it's possible to make a power supply based on the LM317. If you need alternative current limits, the easiest is to use another sense resistor in series with the input (but after C1 of course). This can use switched values, or the voltage across it can be amplified. The latter is a more complex solution, and isn't shown here. Some example circuits are shown in Bench Power Supplies - Buy Or Build?. The current limiter needs to be fairly fast to provide full protection for the current boost transistors.
While there are some good examples in the LM317 datasheet, most are without explanation, and a few appear to be rather suspect. I cannot vouch of any of the circuits described in the datasheet, as many will not simulate properly (if at all), and others are just basic modifications to the generally accepted circuits. I suggest that if you do decide to use any of the demonstration circuits that you do so with care, and be prepared to encounter difficulties (oscillation can be particularly troublesome).
Note Carefully: In the documentation for various regulators, the input-output differential voltage is quoted. This is 40V for the LM317, and many people seem to think that it's therefore alright to have an input of (say) 60V, provided the output voltage is set for at least 20V. This myth is backed up all over the place, but fails to consider reality. When power is applied and there's a decent sized output capacitor, it's discharged at power-on, so the full 60V is across the regulator. If there's a momentary short at the output, the full 60V is across the regulator.
So, while it's claimed that the input voltage can be greater than the input-output differential voltage, relying on this can lead to failure. You may be able to bypass the IC with a 36V zener diode that can handle the output cap's charge current, but even a momentary short will probably kill the zener, and the output will be at the full unregulated voltage. You won't find many people talking about this, but it's very real. I would never advise anyone to operate any regulator IC with more than it's maximum voltage at the input.
All of the power supply circuits shown are capable only of sourcing current. That means they can provide power to a load, but they cannot sink, which is to accept current from another source. There are laboratory power supplies that can do either, namely provide or accept current. For most testing, this isn't necessary, but a supply with this capability is known as a '2-quadrant' supply if it can source or sink current of one polarity, or a '4-quadrant' unit can source or sink current of either polarity.
A basic 'electronic load' is (usually) a single quadrant current sink. It can absorb current, but cannot supply anything to an external load. These are specialised, and are typically used to test power supplies. It's unlikely that you'll ever need one, as most of the time a suitable resistor bank is the easiest (and you may already have one as a dummy load for amplifiers). If you do happen to need a true electronic load, some modern switchmode types use 'regenerative' capabilities, and can return the absorbed power back into the mains, minimising wasted power. There's a lot involved, and they are definitely not a DIY project.
A supply that can both source (supply) and sink (absorb) power needs a set of transistors for each function. It requires feedback to ensure that its output voltage remains fixed regardless of whether it's sourcing or sinking current, and a dual-polarity current limit so that excessive power won't cause damage. Consider a supply set for 6V, but connected to a 12V car battery. The battery can deliver hundreds of amps (at least for long enough to blow up the PSU), so the supply must be designed to limit the maximum current being absorbed to a safe value. As you can imagine, this involves a great deal of circuitry, and most people will never need one. I do have a current sink - it's called a dummy load, and can be set for 4, 8, 12 or 16Ω. I have never had a need for anything more advanced in my workshop, but I did design one for a company I worked for because there was one type of supply that required 'soak testing' to ensure the voltage never fell below a critical voltage level.
These are specialised supplies, and require significantly more electronics than a 'simple' power supply. An audio power amplifier is a 4-quadrant power supply if it can amplify DC, but the normal transistor complement is nowhere near sufficient to allow it to be used as a power supply. Because these are so specialised, they are mentioned in passing, and details will not be provided here.
However, there is one simple supply that can source and sink current - a shunt regulator (often nothing more than a resistor and a zener diode). Note that I mention this in the interests of completeness, even though it's of no practical use in 99% of cases. More information is available in Voltage & Current Regulators And How To Use Them.
If you think that you really need a 2-quadrant or 4-quadrant power supply, you could look at the OPA549, but it's rather limited since it's a single IC and has fairly low power dissipation. It's also expensive, but it does include programmable current limiting (set with a resistor or a pot). You could also use an LM3886 IC power amplifier, but the available current is even more limited, and getting the heat out of any IC will always be a challenge. There are several other similar options, but none that I'd really recommend. This is simply because it's not necessary in most cases.
It's commonly believed that MOSFETs don't suffer from second breakdown, and therefore should be 'better'. However, the vast majority of MOSFETs available are designed for switching, not linear operation. They also suffer from a failure mechanism that's remarkably similar to second breakdown, but it's usually spoken of in hushed tones, lest anyone find out about it. Ok, that might be a stretch, but in almost all cases, MOSFETs are optimised for low RDS-On (on resistance) and high switching speeds. The only MOSFETs that are specifically designed for linear operation are lateral types, as used in Project 101. These have a very different set of output characteristics from 'vertical' MOSFETs (e.g. HEXFETs and their ilk), with a high RDS-On and lower transconductance (roughly equivalent to gain).
Many people (including me) have used switching MOSFETs in linear circuits, and with care they will work. Some of the early types were almost suitable due to comparatively high RDS-On compared to the latest and greatest. However, the design of MOSFETs has evolved, and linear operation is no longer something you can rely on. They will often work quite well (I've tested and verified this), but in general they are simply not recommended (and that's the manufacturer's recommendation, not just mine).
When you look at the SOA curves for MOSFETs, you'll see curves for various time limited operation, but nothing for DC. The difference between the allowable voltage and current in a lowly IRF540N shows 10ms, 1ms and 100µs curves, but nothing for DC. Most are the same, and only a few show DC characteristics (mainly older devices that may or may not still be available). You may be able to use a MOSFET if it's significantly derated, but you would need to run extensive (and likely destructive) tests to determine if it will survive in your application.
You can often see just from basic specifications if a MOSFET is likely to work in linear mode. The first clue is a high RDS-On, usually greater than 0.5Ω. An example is the IRF840, rated for 500V at 8A. However, the TO-220 case is terrible for dispersing heat because the tab is small. With 30V drain-source voltage, only 4A is available, or 8A with 15V. These are the same as the TIP35/36, but the latter have a larger case and better heat dispersion. You will never get 120W of heat out a TO-220 package, so the MOSFET must be operated at a lower current (or use multiple devices in parallel). With 100V across the device, an IRF840 can deliver up to 1.25A, and although this is significantly better than the TIP35/36, you'll still be unable to get the heat (125W) out of the TO-220 package if the power is anything other than a transient event.
MOSFETs cannot be used as regulators 'open-loop' (no feedback) either, because the gate-source voltage is highly variable (from one device to the next and with temperature). However, the IRF840 might be a good choice if you need a 400V regulated output at relatively low current (preferably less than 100mA). It will need extensive protection, both to limit the power dissipated, and to ensure that the gate-source insulation cannot be damaged (this requires a 12-15V zener diode).
The RDS-On of a MOSFET operated in linear mode is irrelevant. The power dissipated is the product of drain-source voltage and current. If you imagine that a low RDS-On makes a difference, then you don't understand how linear circuits work (and yes, I have seen this claimed, hence the comment here).
It's easy to see why switchmode supplies have taken over for high current outputs. The entire SMPS will be smaller than just the transformer, and will also cost a great deal less. At the time of writing, I had a quick look on eBay and found (for example) a 24V, 10A SMPS for just over AU$23.00 up to around AU$30.00 or so. It's impossible to compete with this price, and even if they cost twice as much, it's still way cheaper than one you could build using linear techniques. This applies to many different voltages and currents, but the choices are limited. You can get 5V, 12V, 24V and 48V SMPS at various current ratings. 'Traditional' suppliers are more expensive of course, but you'd still be hard-pressed to build a linear supply for less than even the most expensive SMPS.
None of this makes the circuits here redundant or of no use, as it's all about the principles. Supplies such as the one shown above were used regularly before the advent of low-cost switchmode supplies. Early SMPS were both complex and expensive, and having worked with them many years ago, I have first-hand experience with them. Unlike those today, they were always repaired after a failure (which were fairly common), and even the repair process was tricky. Like all SMPS, everything had to be fully functional, or the supply would blow up again when tested. Before I devised some specialised test jigs, technicians used to power-on a 'repaired' supply with a broom-handle, lest the SMPS blow up in their face. This is not made up - it's 100% factual. 
One thing that building a supply can provide is flexibility. If you need (say) 13.8V with a preset current limit, you'll probably pay dearly for that (that describes the requirements for a lead-acid a battery charger). There are many other places where your needs aren't provided for by COTS power supplies, and unless you're an experienced SMPS designer you don't have many choices. Under these conditions you will end up having to use linear supplies, and even more so if high-frequency noise is an issue. In some cases, you can use a COTS supply followed by a linear regulator, which reduces the size, weight and cost, and you get the best of both worlds.
Voltage regulators aren't actually hard to design, but it's important to consider all of the factors. It's not just finding a transistor that can pass the current you need, but finding one (or more than one) that can handle the power, won't be subjected to second breakdown, and has the thermal ratings needed to ensure it can be kept to a reasonable temperature. Thermal derating has to be factored into the design, along with the input voltage variation. While none of this is difficult, there are many pieces to the puzzle, and they all have to fit together.
You also need to factor in the transistor gain at the current it has to carry, as most transistors vary their gain across the current range. Choosing a suitable heatsink can be a challenge as well, and if you don't understand how everything fits together then the end result is a lottery. Failure to keep the transistors within their safe operating area means that the regulator will fail when you push it to its limits, unleashing the full incoming DC upon your circuitry.
Getting very good regulation (both input or 'line' and load) demands more complex circuitry, so it needs to be tested thoroughly to ensure that the regulator doesn't oscillate at any load. This can be difficult if you use a high-gain error amplifier, and it's made worse when current limiting is included. Foldback limiting can be particularly hard to get right, as the voltage and current curves must remain within the safe operating area at all times, compensated for elevated heatsink temperatures of course.
The last thing I want to do is turn people off building their own regulator designs, as you will learn a great deal by doing so. What I do want to do is provide enough information so that your design has some chance of working without failure, hence the details presented here. It's especially important to be aware that extremely good regulation isn't often needed. You do need to be able to provide a voltage that's close to the desired figure, but unless you're working with precision test gear you rarely need perfect regulation.
What you do need is low ripple and some control over the maximum allowable output current. Once you understand that exceptional voltage stability is rarely needed, that makes your job that much easier. Most circuits won't care one bit of the voltage falls by a couple of hundred millivolts from no load to full load, as that's still far better than you'll get from a transformer, bridge and filter capacitor. Some circuits do care though, so a thorough analysis of the regulator's requirements is always necessary.
Main IndexArticles Index