|
|
| Elliott Sound Products | Pre-Regulator Techniques |
Main Index
Articles Index
Pre-regulation (or preregulation) circuits have been a common requirement in power supplies for many years. There are two reasons - either to reduce the ripple present on the output, or to minimise the power dissipation of the regulator. This reduces heat generation (in the regulator) and may improve regulation a little because there's less voltage change at the input. There are countless different circuits, but they follow the same general themes - linear, tap-switching, phase-cut and switchmode. The last three can be implemented in many different ways. Linear pre-regulators are usually fairly similar because there's a limited number of options.
The first alternative is to use a linear pre-regulator, with an output voltage that's just high enough to ensure that the regulator remains in control of the output. This has the advantage that the regulator circuit itself is already supplied with a waveform that's essentially free of ripple, ensuring a very low noise output. However, the dissipation in the pre-regulator can be very high - even for a relatively low power circuit.
The simplest form of 'high efficiency' pre-regulation to use two or more voltage taps on the transformer, with the appropriate output voltage being taken from the transformer depending on the set output voltage. Tap-switching (as this is called) is fairly simple to implement, but usually requires a custom transformer. This makes it suitable for manufacturers, but it's far less attractive for DIY unless the constructor is willing to use two multi-tapped transformers, assuming local availability and a dual supply with positive and negative output voltages. You may even have a suitable transformer available in the 'junk box'.
In many early high efficiency power supplies, an AC 'phase-cut' scheme was common. By turning on the AC at that part of the waveform where the peak AC voltage was just above the voltage needed at the output, the voltage across the regulator was kept to a minimum, thus improving efficiency. These systems commonly used thyristors (aka SCRs or silicon controlled rectifiers), which are readily available in high current versions. The very spiky nature of the waveform could create both acoustic and electrical noise. TRIACs were also common, and there was a commercial audio power amplifier design that used this technique.
Modern high power supplies use a switchmode supply at the front end, either with direct conversion from the AC mains, or a low voltage switchmode regulator following the power transformer. These can have high efficiency, and where very high power is expected the AC side may use active power factor correction (PFC) to ensure that the mains waveform is as close to a sinewave as possible. This creates a complex design overall, but is capable of very good results.
For this discussion, we'll look at a supply that can provide up to 50V DC at up to 5A. Although diagrams will only describe a single (positive) supply, the same principles apply for a dual supply with both positive and negative outputs. The primary difference is that for a dual supply, voltage, current and total power dissipation are doubled. Of course this only applies when both polarities are supplying the same voltage and current (a dual tracking power supply). Only the pre-regulator is considered here - the regulator is a separate entity, and is shown as a 'block', similar to a 3-terminal IC regulator.
The drawings below are examples, and in each case shows one way that a particular pre-regulator can be configured. There are as many possibilities as there are designers, and it would not be possible to include a sample of each. A web search for a particular pre-regulator design will often turn up some good examples, along with the usual irrelevant links and some examples that should state that the method shown should be avoided, but someone will still think it's a good idea.
Regardless of the technique used, the regulator circuit (either discrete, IC or hybrid) must always have enough voltage across it to allow proper regulation. That includes the most negative part of the ripple waveform. If a regulator needs 5V differential (input to output), the unregulated (or pre-regulated) voltage must always be at least 5V greater than the output voltage. If there's 3V of ripple, then the most negative part of that voltage still needs to be 5V greater than the input. The most positive part of the ripple waveform will therefore be at 8V above the output.
If the voltage differential isn't great enough, there will be ripple 'breakthrough', and some of it will be visible at the output terminal(s). That means that the average voltage (and therefore the average power dissipated in the regulator) must be a little higher than expected. With 3V peak-peak of ripple, the required average DC voltage is increased by 1.5 volts. It doesn't sound like much, but it increases the power demands on the regulator. With 5A output, the power dissipation is increased by 7.5 watts, so total dissipation (including the required 5V absolute minimum) is 32.5 watts. That's a significant increase from the 25W dissipated if the pre-regulated voltage has no ripple.
Depending on the type of rectification used (normal diodes, SCRs) and other factors, the transformer may also need to handle a higher 'apparent power (VA or volt-amps). A standard bridge rectifier imposes a VA rating that's around 1.8 times the actual power delivered. That means that if you expect the output power (including losses) to be 250W, you need a 450VA transformer if the full output load is maintained for any length of time (longer than a few minutes). A smaller transformer can only be used if you include thermal sensing on the transformer as well as the heatsinks, so the supply will shut down if it starts to overheat. Failure to include this safeguard could lead to a transformer failure.
With any high power bench supply, one of the issues that you will always face is the transistor SOA (safe operating area) limits. Datasheets usually provide this in graphical form, and operating beyond the second breakdown limits (even briefly) can result in instantaneous failure. This must be addressed in a final design, and the details are included below (this circuitry will be in the regulator, not the pre-regulator). Remember that if a regulator transistor fails, it will do so short circuit, so the full supply voltage will be presented to the device being tested. This may lead to the destruction of the DUT (device under test).
In the cases discussed, it is assumed that power transistors will be mounted directly to the heatsink, with no electrical insulator present. This minimises the thermal resistance from case to heatsink, but it will always be a non-zero value. The best you can hope for is probably around 0.1°C/W, but that's not easy to achieve in practice. The use of silicone 'thermal' pads is so unwise that I dare not even make mention of them 
, but they exist, and some people still think they're a good idea. While fine for low power applications (up to around 10W continuous) they are ok, but for serious power they are grossly inadequate.
Unfortunately, direct mounting almost always means that the heatsinks are 'hot' (as in electrically 'live'), and they must be insulated from the chassis and great care is needed to ensure that a short to chassis is rendered as close to impossible as you can make it. This isn't necessarily as hard as it sounds, but it does demand a design that's different from the way heatsinks are normally used. As an example, I've included the photo below of a dual live heatsink, which is held together with pieces of acrylic. All screws are countersunk well below the surface, and tape will be applied before mounting to provide a proper electrical barrier. Mounting to the chassis is simple - three holes are drilled through the acrylic, and tapped to take 4mm metal thread screws.
Figure 1 - Dual Live Heatsink, With Fan And Acrylic Separators
The arrangement shown lends itself very well to this application, and one heatsink is for the positive supply, and the other for the negative supply. This is being prepared for an up and coming project that's designed to provide an affordable dual supply, with voltage up to ±25V and load current up to 2A (either or both supplies). Almost all circuitry will be attached to the heatsink, other than the voltage setting pots, current limiting and primary power supply (transformer, bridge rectifier and filter capacitors).
Although the fan is rather puny and the heatsinks aren't overly large (the tunnel is 80mm square, and 160mm long), this heatsink should be good to dissipate up to 50W each side (100W total) fairly easily. This is far more than I'll need, but there's no such thing as a heatsink that's too big. Note that it is absolutely essential that the fan blows air into the tunnel, because fans that suck, really suck! There is a huge difference in performance, and this is covered in detail in the ESP Heatsinks article.
This is the simplest to implement, not counting the thermal management provisions that are essential. For our hypothetical supply, it will require an unregulated voltage of at least 62V DC. If you were to use it with the full 5A output at (say) 5V DC output, the pre-regulator will dissipate at least 260W, with the regulator dissipating a further 25W (assuming a regulator voltage differential of 5V). That is a great deal of heat to dispose of, and attempting it without forced air cooling (a fan) is unrealistic. It can be done, but the heatsink would have to be massive, and the cost of that alone would almost certainly exceed the cost of the power supply itself. This is just silly unless there is an absolute requirement for total acoustic silence, which is rarely the case for a lab/ bench supply.
As the output voltage is increased, pre-regulator dissipation is reduced, until at the very upper limit, it should be passing almost the full unregulated voltage to the regulator. This may mean that output noise (100-120Hz hum or buzz) also increases, because there's no pre-regulation to reduce ripple. This can be countered with a higher unregulated voltage of course, but that increases losses even more. As noted, the greatest advantage is simplicity, but much of that tends to go away when you have to add thermal management circuitry.
Normally, the fan will not be running, and that will be the case for (probably) most of the tests that are typically performed. However, as the heatsink temperature increases, the transistors or MOSFETs used in the pre-regulator become prone to failure due to excessive die temperatures. As soon as the heatsink gets above around 30°C or so, the fan should come on (it can be variable speed), and if the heatsink temperature continues to increase, the supply should be turned off automatically. If these precautions aren't taken, your test load and the supply are liable to be seriously damaged.
While it's potentially the quietest (electrical noise), a linear pre-regulator is the least efficient method. However, this doesn't mean that it shouldn't be considered, especially for lower powers. For supplies providing ±25V or so at current up to 2A, it's limitations are minimised and the loss of efficiency isn't such a big deal. Worst case dissipation may be up to 70W (140W for a dual supply), but that's only under full load at very low output voltages. In 'normal' use (whatever that is ), dissipation will be somewhat less, and in many cases it will only be a few watts when testing preamps or even power amps at low power. It's a technique that isn't dead just yet, and will likely continue for many years to come.
Perhaps one of its greatest advantages is that if built well with good heatsinking, it will outlive most of the people who choose to build it. The parts won't disappear any time soon, and service (if ever required) is usually straightforward if through-hole parts are used throughout. There's no need for SMD parts, because the circuit is so simple. This cannot necessarily be said for some of the alternatives, and especially for switchmode circuits. However, this only applies if a more pragmatic approach is taken, reducing the voltage to ±25V at a maximum of around 2A.
A linear tracking regulator is almost silent, both acoustically and electrically. However, they are also very inefficient, so they need large heatsinks to dissipate the considerable heat that can be generated in a high-power supply. This is not only very wasteful of energy (you pay for the heat generated because of the current drawn from the mains), but also increases the size and cost of the supply.
Figure 2 - Linear Tracking Pre-Regulator
In the above, C1 is 10,000µF (10mF), and is the main smoothing capacitor. It's supplied from the output of the rectifier. Q1 and Q2 form a current source. This provides base current to the series pass Darlington pair (Q3 and Q4). Q4 may consist of two or more paralleled devices if the dissipation is high. The zener diode (ZD1) ensures that the input voltage to the regulator (which may be an IC or discrete) is at least 4.5V greater than the output voltage. If the regulator requires a higher differential voltage, you simply use a higher voltage zener diode. By default, the output from the pre-regulator is fairly well smoothed and contains little ripple because its reference is the regulated output (via ZD1). D1 ensures that the pre-regulator and regulator are not subjected to reverse voltages if a DC source is applied to the output (which can and does happen). Values are not provided for the pot (VR1) or R3 because they are dependent on the regulator topology.
One of the more challenging aspects of any linear design is the transistor SOA (safe operating area). For example, the TIP35/36 devices are low cost and ideal in this role, but there are several things that must be considered. The first is power rating (125W), but that's tempered when you look at the temperature derating curve (power vs. case temperature), the maximum TJ (junction temperature), Rth j-case (thermal resistance, junction to case), and the SOA curves. It should be apparent that with Rth j-case at 1°C/W, if the device is dissipating 70W, the junction must be at ambient (25°C) plus TJ - a total of 95°C. This assumes perfect mating between the case and heatsink, and that the heatsink remains at no more than 25°C.
This is clearly not possible. The maximum allowable junction temperature is 150°C with a case temperature of 25°C, so with 70W dissipation the case temperature cannot exceed 80°C (this is easily calculated, or can be done using graph paper). At 150°C. the die cannot dissipate any additional power, and at a case temperature of 25°C it can handle 125W (which raises the die temperature to 150°C). Note that this only addresses temperature, not SOA! The SOA curve shows that if there's 35V across the device, the maximum current is 2A - that's a maximum of 70W at 25°C. If the voltage or current is increased beyond that, there is a likelihood of second breakdown, an almost instantaneous device failure mechanism. These limits are reduced at higher temperatures!
Despite the apparent simplicity of a linear pre-regulator, there's a great deal of design work involved to ensure that reliability is not compromised. This is why it's so important to examine datasheets, minimise all thermal resistances possible, and usually be prepared to use more parts than you originally thought you'd need. However, this really is the simplest - as soon as more 'advanced' techniques are used, the design challenges are only increased.
If the idea of a linear pre-regulator were to be used for the hypothetical supply (50V at 5A, worst case dissipation of around 300W), the SOA requirements would mean that you'd need a minimum of ten TIP35/36 transistors for each polarity (600mA maximum with 60V across the transistor). This is obviously not a smart way to build a very high power supply. 250W (500W dual) isn't a huge supply by any stretch of the imagination, so alternatives are essential.
Without tap switching, a 50V, 5A supply needs a minimum input voltage of around 55V, so if you were to expect 5A at 1V DC output, the dissipation will be 270W. This assumes that the mains voltage remains at the nominal value, either 230V or 120V. In reality, we need to allow for both high and low mains voltages, so the unregulated voltage should be at least 10% higher than nominal to allow for lower than normal mains voltage. 55V becomes 61V near enough. Dissipation is increased to 300W.
Using tap switching, the transformer has multiple windings (or a single winding with multiple taps), and higher efficiency is available than a regulator that's always supplied with the highest voltage provided from the transformer, rectifier and filter capacitor. For example, for voltages up to 12V, the unregulated DC voltage will typically be at least 18V (average value, requiring an AC voltage of 15V RMS), and it always has to be high enough to ensure that the minimum voltage (based on the amount of ripple) remains above the regulator's dropout voltage (where it can no longer regulate). This varies from around 3V or so, up to 5V or more, depending on the topology of the regulator itself.
With an output voltage of (say) 1V at 5A, the regulator dissipates around 90W. As the output voltage is increased, the transformer tap is automatically selected to provide the required voltage range. There is still a lot of heat to disperse, but it's much lower than a simple regulator supplied with the full secondary voltage at all times.
When the user selects an output voltage of 12V DC or greater, the transformer tapping point is increased so there's more voltage at the regulator's input. For our example, it might rise to 39V DC (AC output from the transformer of 30V RMS), and at full current (5A) with an output voltage of 16V, the regulator dissipates 115W. With a three tap system, the final tap will be selected when the output voltage is set for 34V or above. At 34V with 5A output, the regulator has an input voltage of perhaps 60V, and dissipates 130W.
Note that dissipation is always highest at the low end of any tapped supply voltage. If the regulator is operated with 50V output at 5A, dissipation is around 50W. It will generally be a bit lower at just below the switching voltage for lower voltages, but you must always design for the worst case. You must also allow for a short circuit at the output, and this can be very challenging indeed. Instantaneous power dissipation may exceed 300W, and a heatsink with high thermal mass is needed to absorb such 'transient' events without localised temperature rise. Transistor SOA protection should be included to protect the regulator transistors, and this can be challenging (to put it mildly).
Figure 3 - Simple 3-Stage Tap Switching
A simple tap switching circuit is shown above. Voltages referred to are loaded to 5A, and assume a power transformer of not less than 500VA. The regulator gets an input voltage of around 19V as long as the output voltage is less than 12V. Above that, the zener diode (ZD1) passes enough current to turn on Q1, which in turn operates the relay (RL1). The relay contacts disconnect the low voltage winding and connects to the next tapping (30V AC), so the regulator's input voltage is increased to 44V (~40V loaded). The regulator can then provide a regulated output of up to 28V DC. If the output voltage is increased further, RL2 operates, connecting the 45V AC tap, giving an unregulated voltage of around 63V (~60V loaded). Without the tap switching, dissipation in the regulator will be much higher than desirable with low output voltages, particularly if the current is high.
The relay contacts are labelled 'NO' and 'NC', meaning normally open and normally closed respectively. The 'normal' state is when the relay is not energised, so the 'NO' contacts will be open (no connection). The relay contacts must be able to handle the full voltage and current, as determined by the design of the power supply. This is usually easy to achieve, and relays have very low resistance when the contacts are closed. You must ensure that there is no possibility of relay contacts for separate voltages to short a winding (this is not possible in the Figure 3 circuit).
ZD2 and ZD4 protect the relay switching transistors from excessive base current with high output voltages. If a pair of comparators are used instead of zener diodes and transistors, power dissipation is reduced and the tap switching voltages will be more accurate. This does add complexity of course, but the cost difference is negligible. The simple scheme shown will certainly work, but switching thresholds are not very precise.
BR2 along with the separate winding provides a low voltage (~12V DC loaded) output permanently to operate the relays, regardless of the selected AC voltage from the transformer. This is best provided by a separate winding, and the output would ideally be regulated for comparators and relay coils. Comparators give better (more predictable) voltage sensing, which gives greater precision and lower power consumption.
If you drive a short circuit and attempt to increase the output voltage, it cannot rise due to current limiting, so the higher voltage taps can't be selected. Although I've shown relay switching, it can also be done using SCRs (silicon controlled rectifiers, aka thyristors), TRIACs or even MOSFET relays. Regardless of the switching technique, the results are much the same. 'Solid state' switching may be thought to be preferable, but is more involved, has higher losses than relays and requires more complex circuitry.
Of course, there's no reason not to include a linear tracking pre-regulator with tap switching, but this will still be subject to the same constraints as a linear pre-regulator if you happen to have set the highest output voltage and there's a sudden short circuit in the load (or just the test leads). The tap will be dropped to the lowest setting almost instantly, but there's still a large filter capacitor, charged to the maximum unregulated voltage! This will cause grief whether linear tracking pre-regulators are included or not, and it must be catered for because it will happen.
The overall efficiency of a tap switching systems is improved with more transformer taps. It's also possible to use different voltage windings that are switched in a sequence that allows (say) three windings to provide five different output voltages from the transformer. You might have a pair of 18V windings and a single 9V winding, switched so that you can have AC voltages of 9, 18, 27, 36 and 45V AC. While this obviously improves efficiency, it also means a complex logic matrix to control the switches. Use of a microcontroller will simplify this task of course, but the relay contact arrangement will be fairly convoluted. The transformer will be a custom design, unless you use multiple smaller transformers.
The regulator design must be sufficiently rugged to ensure that it doesn't fail if shorted while supplying the maximum output voltage, and this will happen, either accidentally or due to a failure in the test circuit. This particular issue refuses to go away, regardless of the technique used for pre-regulation, and failure to provide appropriate protection circuitry will result in a blown-up power supply.
A common approach to pre-regulation in early power supplies was a 'phase-cut' circuit, somewhat similar to an incandescent lamp dimmer. These were popular because they could allow the unregulated voltage to remain just high enough to ensure that the following linear regulator could provide good regulation, without any ripple breakthrough.
However, most of these supplies used SCRs (silicon controlled rectifiers, aka thyristors). The biggest problem was/ is the turn-on speed of the SCRs - they go into conduction very quickly, and that means that they invariably cause some high frequency noise. Because they can only be turned on, they were (in lamp dimmer parlance) leading edge 'dimmers', so most of the AC half-cycle would pass before the SCR(s) turned on. GTO (gate turn-off) thyristors became available later, but they were never used in any 'phase-cut' pre-regulator circuit I've seen.
The rapid turn-on also causes most transformers to growl, so they made acoustic as well as electronic noise. One alternative to the 'traditional' SCR phase-cut pre-regulator is to use a MOSFET switch. This means that it can turn off when the voltage is high enough, so it operates like a trailing-edge dimmer. This is somewhat quieter than an SCR version, and with the MOSFETs one can obtain today, it's also more efficient. However, that doesn't mean that high frequency noise is eliminated.
You can think of this arrangement as an 'infinitely variable' tap switcher, because the transformer's output voltage is continuously variable. The unregulated output voltage can be as low as 6V if the control is on the secondary side of the transformer. Many supplies used phase-cut circuits on the primary side, because that reduces the current involved, which in turn lowers the losses in the SCRs or TRIACs (a TRIAC is a bi-directional AC switch). Of course this introduces additional complexity, because the SCRs or TRIACs need isolated control circuitry. There are specialised ICs designed specifically for powering TRIACs (e.g. MOC3020 ... MOC3023), but control circuitry is still required. A zero-crossing detector is necessary so the circuitry can identify the point where the AC waveform passes through zero (and the SCRs or TRIACs turn off).
In the following circuit, the zero-crossing detector is not required as a separate sub-circuit. The switching system doesn't actually identify the zero-crossing, but turns on the MOSFET whenever the AC voltage is below the target voltage. The current limiter uses a 50mΩ resistor (R2), which limits the peak MOSFET current to a bit over 13A. If the peak current is decreased, the MOSFET will be on for longer, and total power dissipation will be increased. The current must be high enough to ensure that the filter cap (C2) can charge to the required voltage under full load. Ultimately, the peak current is also limited by the transformer's winding resistance.
Figure 4 - Phase-Cut Pre-Regulator
The drawing shows a version of a phase-cut pre-regulator that you almost certainly won't find elsewhere. Although simplified, it works well as shown, and needs only a few changes for a practical circuit. The P-Channel MOSFET is turned on as the unfiltered DC waveform falls below the target voltage, and turns off again when the target unregulated voltage is reached. With a 5.1V zener as shown, the regulator's differential voltage is about 5V at any output voltage setting. The opamp comparator requires a 'full time' power supply, or it can't function. As with all phase-cut circuits, the filter capacitor's ripple current can be much higher than normal. This is mitigated (to an extent) by using the current limiter for the MOSFET as shown, but that increases its dissipation. For better overall performance, Q3 is a current sink. This makes the MOSFET drive signal less affected by the instantaneous voltage. R7 and R10 are required so the circuit will start, as without them there is no voltage on the comparator's non-inverting input, and it won't turn on. It only requires a few millivolts to start operating, and the process is self-sustaining after that. R7 also provides the zero crossing signal, although at times the circuit will switch on at other points on the waveform (as seen in the waveforms below).
Using phase-cut circuitry on the secondary (low voltage, high current) side of the transformer was once impractical, but MOSFETs have changed that. They are available with almost scary current ratings, and 'on' resistances so low that power dissipation is minimal. The circuitry needed isn't frighteningly complex, but it's usually a wise move to impose some form of current limiting (other than the transformer's winding resistance) to ensure that the filter capacitor ripple current is manageable. Without current limiting, the filter capacitor can have a very hard (and commensurately short) life. Fortunately, this isn't too difficult to achieve, and requires only a few low-cost parts. One technique that was commonly used in older systems is a filter 'choke' (inductor), but that's a large, heavy and expensive addition. However, it does give good results when implemented properly.
It's rather doubtful that any manufacturer would use this scheme in a new design, but that's not because it's not effective. The biggest issues with all phase-cut systems is poor transformer utilisation and high capacitor ripple current. For a DIY build, and provided that the DIYer is willing to experiment, it can produce good results, but manufacturers will now use a switchmode supply (and most use only the switchmode supply, without any linear regulation to minimise noise). Note that leading edge (SCR or TRIAC) systems are essential if used on the transformer primary, but if the phase-cut is performed on the secondary side, a trailing edge switch is preferred. Note that the supply voltage to the comparator must be no less than the regulator's output voltage to prevent damaging the input circuits of the comparator IC (or opamp). The comparator has some hysteresis built in (provided by R5), which helps to prevent spurious oscillation.
Of the options shown, the MOSFET phase-cut circuit is probably the simplest to implement if you need high efficiency, but it comes at a cost. While the circuit is conceptual (rather than a complete solution), it simulates very well and there's no reason to expect that it won't also work very well. It doesn't need any extras other than a suitable supply voltage for the comparator (typically around 30V DC). Apart from a switchmode circuit, it can have the highest overall efficiency at any voltage or current of any of the techniques.
So, what are the costs? At low to mid voltages, expect filter capacitor ripple current to be up to twice that of a conventional rectifier supplying the same output current. It also suffers from rather poor transformer utilisation (in common with all phase-cut circuits). The power factor at the voltage and current shown below is only 0.327 which means the transformer's VA rating may be up to 800VA for an output of 250W (50V at 5A). You need a much larger transformer than expected to get the current and voltage required. A 'conventional' rectifier and filter cap needs a 450VA transformer for the same output power. The same effects are seen with any phase-cut system - it's not something that's limited to the one shown.
Figure 4.1 - Phase-Cut Pre-Regulator Waveforms
Of all the designs shown here, the MOSFET phase-cut switching version is the only one that requires a waveform to show how it functions. The unregulated voltage and current are shown, for an unregulated output of just over 23V at an output current of 0.8A. The MOSFET current is limited using the circuit shown above, and is about 13A peak. As you can see, when the unregulated voltage falls below the threshold, the MOSFET turns on and remains on, until the voltage exceeds the threshold. You can see the slight 'bump' in the DC waveform if the MOSFET turns on just before the zero crossing. The main power transfer occurs after the zero crossing. The comparator's output is shown in blue, and you can see that it turns on just before the zero crossing, and turns off at the instant the voltage has reached the desired peak value. As output current is increased, the MOSFET turns on for longer, allowing the filter cap to fully charge to the required voltage.
As noted earlier, this is not a scheme you're likely to come across elsewhere. There are certainly MOSFET switching systems published, but most try to operate in exactly the same way as a 'traditional' SCR or TRIAC version, and don't use the simpler scheme shown here. There's nothing to suggest that a more traditional method is 'better', and I suggest that the opposite is true, since the circuit shown above operates primarily as a trailing edge control system, which helps to reduce the capacitor's ripple current.
Care is needed with the MOSFET selection, because they have a defined safe operating area. This is critical when they are operated partially in the linear region (for which few MOSFETs are optimised), and the datasheet must be consulted to ensure that the MOSFET used can handle the combination of voltage and current. The current limiter makes life easier for the filter capacitor, but harder for the MOSFET. Conversely, removing the current limiter makes life easier for the MOSFET, but places more strain on the filter capacitor.
With a switchmode pre-regulator, you retain the normal mains transformer, bridge rectifier and filter capacitor. However, rather than using a linear (or phase-cut) pre-regulator, it will (most commonly) be a 'buck' (step down) switching regulator. There are countless ICs available for this, and it would be rather foolish of me to try to describe a complete circuit (so I won't). Instead, the buck converter is shown as a circuit 'block', with a separate P-Channel MOSFET acting as the switch. The feedback has to ensure that the output voltage is higher than the regulated output, and as before the voltage differential depends on the regulator topology.
This arrangement is capable of high efficiency, so there will be little wasted power. The biggest problem will always be ensuring that the switching noise is not coupled through to the output. For some applications, a bit of high frequency noise isn't a problem, but if you are trying to measure the signal to noise ratio (SNR) of an audio frequency circuit, any high frequency noise can play havoc with your measurements.
Figure 5 - Switchmode Buck Converter Pre-Regulator
The DC voltage from the transformer, bridge and filter cap must be greater than the highest regulated voltage required, because the buck converter will always need some voltage differential (just like the linear regulator). One major advantage is that if you need high current at a low voltage, the switchmode converter applies 'transformation'. Assuming no losses, if the buck converter has an input voltage of 60V, an output voltage of 10V and a current of 5A (50W), its input current will only be 833mA (also 50W). In reality it will be more because no circuit can achieve 100% efficiency. It's reasonable to expect that the input current will be around 1A (60W) representing only 10W 'wasted' power. Even a small heatsink can dispose of that easily, although not all of the power is dissipated in the switching MOSFET - some is also dissipated in the inductor and rectifier diode.
Q2 is a very simple differential amplifier, which ensures that there's about 6 volts across the regulator. If the input voltage decreases due to external loading, Q2 is partially turned off, which provides a lower feedback voltage to the switchmode converter, forcing its output voltage to increase. The converse is also (obviously) true. Because the MOSFET is a high speed switch, dissipation will be low, and is a combination of the turn on/off speed, and the 'on' resistance (RDS on). Inductor dissipation depends on the core losses and AC resistance (which is affected by skin effect, and is higher than its DC resistance). A means of providing short circuit protection or current limiting is necessary for the buck converter.
Today, the trend is to use a switchmode power supply to provide the unregulated voltage. It is actually regulated, but set up so that the SMPS output voltage is high enough for the linear regulator to regulate properly. Hopefully, any residual high frequency noise will also be eliminated, but that can be a lot harder than it sounds. A switchmode supply can be either on the mains side (eliminating the 50/60Hz transformer) or on the secondary side using a simple buck regulator as shown in Figure 4. Using a mains switchmode supply is more efficient, but you then have a great deal of circuitry that's all at mains potential. This is not usually a wise choice for most DIY people, although it can be done if you are skilled in the finer points of 'off-line' (powered directly from the 'line' (mains) voltage) switchmode supplies. I've shown an SMPS with active PFC (power factor correction), but this isn't essential. These are far more complex than 'simple' switchmode supplies.
The most common SMPS uses a rectifier direct from the mains, with a high voltage filter capacitor. This is followed by a switchmode control IC, and one or more MOSFETs to switch the high voltage DC to the transformer. Low power systems (less than 50W or so) will use a flyback converter, while more powerful supplies use full or half bridge drive to the transformer. The output voltage on the secondary side is controlled by pulse width modulation (PWM). The feedback control system must monitor the output voltage from the regulator, as well as its input voltage (from the SMPS), and ensure that there is sufficient voltage differential to maintain regulation. The SMPS requires short circuit protection, which is not shown in the circuit. For additional information about switchmode topologies, see the ESP article Switchmode Power Supply Primer.
Figure 6 - 'Off-Line' Switchmode Pre-Regulator
Because there are so many possibilities and so many variables, the above is presented as a block diagram only. The secondary rectifier has to use either Schottky or ultra-fast diodes, as they will typically be operated at 50kHz or more. The feedback system uses the same differential arrangement as Figure 5, but includes a resistor (R1) to limit the maximum LED current in the optocoupler. There are many things that can (and do) go awry with an SMPS, and every eventuality has to be catered for. The SMPS is shown as a circuit block for this approach, simply because of its overall complexity. The purpose of this article is to provide ideas, not complete circuit diagrams.
Note that apart from the X2 capacitor (C1), there is no mains input filtering, switching or fusing shown. These are all necessary in an operational circuit. Switchmode supplies can provide both conducted (through the mains wiring) and radiated (through the air) radio frequency interference (aka EMI - electro-magnetic interference), and filtering is always necessary to ensure that other equipment isn't compromised. Commercial equipment requires compliance testing, and the necessary filtering is essential to obtain certification. It's generally unlawful for any manufacturer to sell non-compliant equipment.
There is no doubt that a well engineered SMPS can give very good results. As shown, you can also use a smaller main filter capacitor (C2) because the frequency is so much higher than normal mains. This minimises stresses if (when) the supply is shorted, because it discharges much faster than a larger cap. Unfortunately, the difficulty lies in the implementation, as these supplies are very complex. Most of the ICs used are SMD, and should the supply fail 10 years after you build it, the chances of getting replacement parts is not good (especially PFC and SMPS controllers).
Despite the apparent simplicity of the block diagram shown, the reality is that there is nothing even remotely trivial about this technique. You can simplify the final design by not using active PFC, but there are still many serious challenges to overcome. The design of a switchmode transformer is almost a 'black art', and achieving full isolation that complies with relevant safety standards is a feat unto itself. Ultimately, while it's certainly likely to provide the highest efficiency of all the methods discussed, the circuit complexity (and the danger of working with live mains powered circuitry) means that it's very hard to recommend as a DIY project.
When a regulator is providing the maximum possible output voltage, an accidental short (or a failed DUT) can stress the regulator's series-pass transistors well beyond their SOA limits. This can result in instantaneous failure, especially if the current limiting is set to its maximum value. Consider a transistor with 60V across it, attempting to pass 5A. The instantaneous power is 300W, and it takes time for the main filter capacitor to discharge. The more capacitance you use, the worse it is for the transistor(s). While most transistors can handle up to three times their rated power for very short durations, the time needed to discharge a 10,000µF capacitor will exceed the capabilities of simple series pass stages. At the same time, the transformer and rectifier are trying to keep the cap charged! The pre-regulator will reduce its output voltage, but this is never instant - expect at least 10ms, often more.
This is especially true when using a pre-regulator, because that is generally used to limit the dissipation in the regulator. Consequently, the regulator may normally only have to dissipate around 100W (worst case), and usually less. Unless some form of specialised limiting is used (commonly known as V-I limiting in audio power amplifiers), the result will mean an expensive repair, and the supply is out of action until it's fixed. This isn't a trivial undertaking, and some fairly serious design work is necessary to get a V-I limiter that provides complete protection against short circuits. Don't imagine for a moment that it won't happen, because it will - that is pretty much guaranteed!
Figure 7 - TIP35C/36C SOA Curves
The data above is adapted from the Motorola datasheet for the TIP35C/36C (25A, 100V, 125W), and only the 'C' version is shown, since the lower voltage parts are now hard to obtain. Below 30V, the limits are based on power dissipation alone, so at 10V the limit is 12.5A (125W) and at 30V the limit is 4.16A (125W). At any collector to emitter voltage greater than 30V, second breakdown becomes the limiting factor, and woe betides the designer who fails to take it into consideration. Higher current is permissible if the duration of the overload is short enough, so you can get up to 1.75A with a duration of 300µs (87.5W), but that's not sensible for a power supply.
As you can see, if there's 50V across the transistor, its maximum collector current is only 1A (50W vs. 125W). This is the secondary breakdown limit - at a case temperature of 25°C ! As the temperature increases, the SOA limits fall, so maintaining the lowest possible heatsink temperature is obviously critical. The TIP35/36 devices are rated at 125W, but that can only be reached at a VC-E of 30V or less, and at a case temperature of 25°C. This is normal, and you'll see the same trend with any BJT you care to examine. Some are better than others, but all are limited by physics.
Using switching MOSFETs in linear mode is generally considered to be a bad idea (by the manufacturers), and while they don't suffer from second breakdown per sé, they have a remarkably similar failure mode brought about by localised overheating within the silicon die. It's well outside the scope of this article to try to cover this, but it's a very real phenomenon and has caused the death of many a MOSFET. If you look at the vast majority of MOSFET datasheets, you'll see curves for various periods, such as 10ms, 1ms and 100µs. They don't show operation at DC, because they don't handle it well. Switching MOSFETs are designed for switching!
Of course, there's no good reason that you can't use lateral MOSFETs - the same as used for audio amplifiers such as the Project 101 MOSFET power amp. Lateral MOSFETs such as the ECX10N16 (125W, 160V, 8A) have a much greater SOA than bipolar transistors, and the primary limit is simply power dissipation. For example, if the device has a drain-source voltage of 100V, the maximum current is limited to 1.25A, because they are rated for 125W. If the voltage is 50V, current is 2.5A (also 125W). As standard practice, all power ratings are for a case temperature of 25°C. Lateral MOSFETs are much more expensive than BJTs or switching MOSFETs, and are uncommon in regulators or pre-regulators. There are a few MOSFETs (other than lateral types) that are designed for linear operation, but they are hard to find, and usually very expensive.
Everything in electronics ends up being a compromise. We compromise noise for efficiency, and (in many cases) we may compromise efficiency for ease of construction. There is no single 'ideal' solution, so a trade-off is always necessary somewhere. Simple techniques are usually easy to build but inefficient, and as the circuit is improved along the way, it will end up being more complex. With modern SMD (surface mount device) construction, there's little or no cost in terms of PCB real estate, but the final product may not be able to be repaired if it's smothered in tiny SMD parts.
The best design for any given purpose is not necessarily the most efficient or the most expensive, and it may not even require particularly good regulation. The best design is one that suits the purpose, and for DIY, is easy to build and service should that ever be necessary. Highest efficiency almost always means greatest complexity, and that's especially true of switchmode circuits. If you intend to use the supply regularly, it needs to provide the functions you think are essential, and ideally can be modified later to make improvements should they be found necessary.
A bench supply doesn't need 0.01% regulation, because it's invariably used with test leads that will degrade the regulation anyway, even if they are of sufficient gauge to minimise voltage drop. To use test leads that don't affect regulation means that you have to employ remote voltage sensing, so you need five leads for a dual power supply. In all the years I've been using power supplies, I've literally never wished that I had remote sensing capabilities, because most of the time a small voltage variation is of no consequence. It's a different matter if you are performing particularly precise measurements, but if that's the case you need a supply that's been designed for the purpose. DIY usually won't provide the performance needed without considerable effort and expense.
Major manufacturers may spend hundreds (possibly thousands) of hours designing a supply that can be classified as true 'laboratory grade', and few individuals have the time, resources or money to spend on multiple prototypes to arrive at a final design. For example, a small miscalculation when designing a custom power transformer will mean that a new one has to be built. This can add a significant financial burden if you are building a single supply for your own use.
Like the article discussing Bench Power Supplies - Buy Or Build?, this is not intended to show complete and/or tested and verified circuitry. It's a collection of ideas, selected to show common ways to minimise regulator dissipation. Each has been simulated (other than the switchmode versions), and has advantages and disadvantages. The not-so-small issue of protecting the regulator should the voltage be set to maximum and the test leads are shorted has not been addressed with any additional circuitry. If the regulator is a 3-terminal type (unlikely given the voltage and current suggested in the introduction) this should be 'automatic', but for a discrete regulator some form of instantaneous dissipation limiting must be considered.
Bench power supplies are not trivial, and the protection requirements become quite onerous for a supply that can deliver high voltage and current. Since most (or at least a great many) applications today require a dual supply, everything is doubled. I consider that a supply that can deliver up to ±25V at perhaps 2A or so to be a sensible limit for a home made supply. Anything larger becomes very expensive to build, and is much harder to protect against accidents or misuse (deliberate or otherwise).
Many pre-regulator circuits rely on a separate 'always on' supply to power the control circuitry (always on when the power supply is turned on, not 24/7). While the current needed is usually quite low, this adds to the overall circuit complexity, and it's worse for a dual supply. In addition, separate floating supplies may be required for digital meters, and while inexpensive, they too add to the build complexity and final cost. Some people don't care, and simply want to build the best supply possible that suits their needs. If that's your goal, then choose wisely, and be prepared to build several prototypes before you get everything just right.
The most useful reference is shown below, along with the ESP article. The HP circuit is an advanced design (for its time), and uses tap switching to obtain 0-50V at 0-10A output. There are countless circuits on the Net, with some being perfect examples of what not to do, while others are interesting (which should be in quotes for some). Otherwise, there are few other references, because the information available was either far too complex to consider, or had issues that would make a reference less than useful. The references to using MOSFETs in linear mode are worthwhile just for interet's sake, as many people are unaware of the likely problems.
Main Index
Articles Index