by Oleg Artamonov
01/16/2005 | 03:55 PM
Unfortunately, one of the primary tests of our methodology – measurements of the stability of the voltages – was far from visual since we used different load patterns for almost every tested unit, and we couldn’t discuss and compare the results of different PSUs without constantly referring to the peculiarities of the employed patterns. In other words, the results of each PSU used to come with a heap of conditions and reservations, so although a comparison could be made (otherwise our testing wouldn’t make any sense at all), these reservations, alas, made it more difficult to directly compare the numbers we got and the diagrams we constructed.
This article is an introduction of our new methodology of testing power supply units that is going to replace the older voltage-stability-measurement method. The new methodology is going to provide a visual and at the same time very accurate and objective result, suitable both for comparing different PSUs in numbers as well as “by the eye”, i.e. by the look of the resulting diagrams. Our new methodology involves a construction of the so-called cross-load characteristic of the PSU as developed and introduced by our colleagues from the news source ITC Online, but we’ve improved it to make it more visual and informative.
In this article I will also cover various aspects of PSUs used in computers so that those of you who are not well versed in the circuit design of switch-mode power supplies could understand where the various parameters of PSUs come from and what they mean. If you know how switch-mode power supplies are built and how they work, you can skip the first two sections of this article to get right to the description of our test equipment and test methods.
As you know, an electrical power supply is a device that solves the problems of adjusting, controlling or stabilizing the input electrical power; its acts as an interface between a power source and a load.
The simplest and most widespread control method is the consumption of the excess power in the control device, i.e. the power is trivially dissipated as heat. Such power supplies are referred to as linear.
The schematic above describes an example of such a device: a linear voltage regulator. The voltage in the power grid (220 volts) is stepped down by the transformer T1 to the necessary level and is then rectified by the diode bridge D1. Evidently, the rectified voltage must always be higher than the output voltage of the regulator. In other words, some surplus power is required due to the very principle of operation of the linear regulator. In our example, this surplus power is dissipated as heat on the transistor Q1 which is controlled by a certain circuit U1 in such a way as to output the required voltage Vout.
This design has two significant drawbacks. First, the low frequency of the alternating current in the power grid (50 or 60 hertz depending on the country you live in) calls for a big and heavy step-down transformer – a 200-300W transformer would weigh several kilograms. Second, the voltage on the transformer’s output should always be higher than the sum of the regulator’s output voltage and the minimal voltage drop on the control transistor. It means the transistor generally has to dissipate quite an amount of excessive power, which is going to negatively affect the efficiency factor of the whole device.
To solve these problems, the so-called switch-mode voltage regulators were invented which control power without dissipating it right in the control device. You can imagine a simplest sample of such a device as an ordinary switch (a transistor can be used as one) attached in series with the load. In this design the average current flowing through the load depends not only on the resistance of the load and the supply voltage, but also on the frequency of the switching – the higher this frequency, the higher the current is. Thus, changing the switching frequency we can control the average current through the load, and ideally the power won’t dissipate on the switch itself, since the switch has just two states: fully open or fully close. In the former case, the voltage drop on the switch equals zero; in the latter case the current equals zero and the dissipated power (which is the product of current and voltage) is zero, too. Of course, we don’t have such an idealistic picture in reality due to two reasons. First, when transistors are used as switches, a minor voltage drop occurs on them anyway even when they are closed. Second, the process of switching doesn’t happen instantly. These losses are side events, however, and they are much smaller than the excessive power dissipated on the control device of a linear regulator.
If we were to compare the numbers, the efficiency factor of a typical linear regulator would be 25-50 percent whereas that of a typical switch-mode regulator may exceed 90 percent.
Besides, if we put the switch of a switch-mode voltage regulator before the step-down transformer (as you understand, it doesn’t really matter if we control the input or the output voltage of the transformer, since they are both closely interrelated), we will get an opportunity to set up the frequency of the transformer irrespective of the frequency of electric current in the power grid. What for? The dimensions of a power transformer are inversely proportional to its operational frequency, and thus switch-mode regulators can use step-down transformers of just tiny dimensions (if we compare them to their linear analogs). Here are some numbers for you: the ratio of the PSU’s output power to its volume is about 4-5 watts per sq. inch with switch-mode power supplies working at 50kHz and only 0.3-1 watts per sq. inch with linear regulators. Moreover, the power density of a switch-mode power supply can be up to 75 watts per sq. inch which linear power sources cannot achieve even with water-based cooling (the numbers quoted are taken from the book by Irving M. Gottlieb titled “Power Supplies, Switching Regulators, Inverters and Converters”).
Besides that, this realization of a switching regulator is less dependent on the value and frequency of the input voltage – the step-down transformer is more sensitive to these factors, but we can control the voltage and the frequency in any way we like by attaching a switch in line before the transformer. That’s why switch-mode regulators are absolutely indifferent to how many hertz you have in your wall power outlet (50 or 60 in various countries) and can usually tolerate a deviation of 20 percent from the standard voltage in the power grid. And that’s also why switch-mode power supplies suppress noise from the power grid well enough, while linear supplies can let some noise seep through into the load.
Besides the transformer, the use of high frequencies enables a considerable (tenfold) reduction of the capacitance and, accordingly, of the dimensions of the smoothing capacitors (C1 and C2 in the schematic above). This reduction is, however, accompanied with two problems: not all electrolytic capacitors are capable of working normally at such a frequency, and, whatever you do, it is technically hard to get a pulsation swing of less than 20 millivolts on the output of a switch-mode PSU. With linear power supplies, the level of pulsation on the output can be reduced to 5 millivolts and lower.
Evidently, a transformer working at a frequency of several kilohertz is a source of noise not only in its own load, but also in the power grid and in the radio-frequency range. That’s why the PSU should be well screened (a steel case is necessary for high-watt units) and the output filter should be paid much attention to (against the common opinion, this filter not so much protects the PSU from external noise as protects other devices from the noise generated by the PSU itself). Linear power supplies, as I mentioned above, are more sensitive to external interference, but don’t generate any noise themselves, so they don’t call for any special measures to protect the neighboring devices.
Talking about the pros and cons of the two types of power supplies, I should acknowledge that the switching variety needs much more complex (and more expensive!) electronics than the linear type. Switch-mode power supplies only have a price advantage when they are high-powered, so that the price is mostly determined by the cost of the power transformer and the necessary heatsinks. In this case linear power supplies with their large dimensions and low efficiency are at a disadvantage. Well, the components becoming cheaper, switch-mode power supplies are intruding on the domain of low-powered linear power supplies. For example, 10-15W switch-mode power supplies are no rarity today, but a few years ago the advantages of linear power sources were conspicuous at such small wattages.
In applications that demand small dimensions, however, switching power sources are beyond competition. It is simply impossible to get the same power density from a linear power source as from a switching one.
Today all power supplies employed in computers are of the switching variety because the same wattage density and efficiency at reasonable dimensions and heat generation are not achievable with linear power sources of the same wattage. For example, the wattage density of a typical ATX power supply is 2-5 watts per sq. inch (depending on its output power), and its efficiency factor is not less than 68 percent at the highest load.
The flowchart above represents the design of a typical computer power supply, while the picture below shows you a typical variant of placement of the components in a real PSU (I use a Macropower MP-300AR as an example here – the majority of other models would look much the same):
The 220 volts (or whatever there is in your country) from the wall outlet pass through a dual or triple filter which protects the other devices attached to the power grid from the noise generated by the PSU. After this filter the voltage comes to the rectifier D1 and then to the optional Power Factor Correction circuit (which appears more frequently in newer power supply models). We’ll discuss shortly what this correction is and what it does. Right now I want to linger round the filter since it involves a couple of questions often asked by the users.
This is a schematic of a classic dual filter used in the majority of power supplies. As you probably know, there can be two kinds of interference: differential-mode interference when the interference current is flowing in opposite directions in the wires, and common-mode interference when the interference current is flowing into one direction only. We can also say that differential-mode interference is interference between two hot wires, while common-mode interference occurs between a hot wire and a ground wire.
Interference of the first type is quite easily suppressed in this schematic with the chokes Ld and the capacitor Cx. The resistance of the chokes is too high for high-frequency noise to pass, while the resistance of the capacitor is, on the contrary, too low. It’s worse with common-mode interference. The choke Lc suppresses it somewhat – this throttle’s coils are wound in such a way as to produce a high resistance to this kind of interference, but that’s not enough and the two capacitors Cy are installed to effectively get rid of common-mode interference. The point where these two capacitors are attached to each other is connected to the case of the PSU – and to ground if possible.
It is about these capacitors that the users often ask. Clearly, if your computer’s case isn’t grounded, it will have half the voltage of the power grid, i.e. 110 volts, due to these capacitors. You would feel the current tickling you if you grasped any grounded item (for example, the heating radiator) with one hand and touched the computer’s case with the other hand. The capacitance of these capacitors is very small, however, so the maximum flowing current is negligible and is not dangerous to you in any way. It is somewhat dangerous to various peripheral devices. For example, if you don’t disconnect your ungrounded computer from the wall outlet before attaching an LPT printer to it, those 110 volts may find themselves on the signal pins of the printer’s LPT connector, resulting in damage to the LPT port of the printer or computer. You don’t have to ground everything up, though. It would suffice to have the cases of all devices properly connected electrically, for example through attaching them to one surge protector with three-wire sockets – the devices will be connected through the “ground” wire of the sockets, and you won’t be running the risk of damaging any port. Nothing also threatens “hot-plug” ports (like FireWire or USB) as the pins in these ports are designed in such a way as to lock the “ground” first.
Another question is about the probability of a disruption of one of these capacitors – if this happens, the full 220 volts will be on the computer’s case. I can calm you down here: such circuits make use of special high-voltage self-repair Y-class capacitors with double insulation that are specifically intended for circuits where a disruption of a capacitor is not allowable for safety considerations.
The only situation where it would be really necessary to ground the computer is when your computer creates noise that affects the surrounding equipment (an FM receiver or a TV-set), since, as I mentioned above, common-mode interference cannot be fully avoided without grounding. External surge protectors are helpless here – their schematic is fully analogous to the above-described one, so they don’t work without grounding, either. If you’ve got three-wire electric wiring in your house (with a ground wire), you should just use appropriate power cords, but if you’ve got two-wire wiring, you should contact qualified electricians. Manual grounding is not only unsafe (for example, an egregious mistake is often committed by inexperienced users who attach the computer’s “ground” to the neutral wire in the socket, which is absolutely unacceptable), but may not bring the effect expected since grounding should have as low as possible resistance to be efficient at suppressing electromagnetic interference.
Near the power filter in the PSU there’s usually a safety fuse and varistors (nonlinear resistors whose resistance diminishes abruptly if the threshold voltage is exceeded) attached in parallel to the capacitors Cy. There’s a common misunderstanding about this fuse: some people claim it protects the PSU from breaking down. That’s not true. The safety fuse of a switch-mode power supply melts only after the switching transistors of this PSU fail. In other words, it protects the power grid from the consequences of the PSU’s failure, rather than otherwise. There’s a widespread misunderstanding about the varistors, too. Some people think they can protect the PSU if the voltage in the power grid is high above the norm. That’s again not true since the varistors can only consume rather short-term voltage surges that are provoked by a nearby stroke of lightning, for example. And if you need protection against long-term voltage surges that can occur as air wires close (which is typical for rural areas) or as a mistake of electricians (which are humans and hence can err), you should consider specialized devices for which the manufacturer explicitly declares such protection, for example APC Line-R regulators and others of that kind. Once again – a computer power supply doesn’t have any protection against long-term voltage surges, and it just breaks down if there’s no external protection device.
But let’s get back to the design and operation of the PSU. After the power factor correction unit (if none is present, then directly from the diode bridge) the rectified voltage arrives to the smoothing capacitors C1 and C2 and then to the switch (it is typically made of two transistors) that controls the power transformer T1. The switch in a computer PSU typically works at a frequency of 60 kHz.
A computer PSU outputs up to six voltages (+12v, +5v, +3.3v, -5v, -12v, and +5v of the standby mode), so six voltage regulators are necessary as an ideal. In practice, however, it is not possible to pack even two independent high-power regulators (say, for the +5v and +3.3v power rails) into the cramped space of a power supply without raising its price to astronomic heights. That’s why all modern PSUs use a single switching regulator (frankly speaking, there’s one more regulator – the source of the +5v standby voltage is an independent low-power regulator; its low power (about 10 watts) allows for an easy implementation, though).
So, all output voltages, save for the +5v of the standby mode, are taken from the same transformer T1 (there are only two voltages shown on the flowchart for the sake of simplicity). Note that in all modern PSUs the switches are controlled through pulse-width modulation (the width of impulses changes, while their frequency is constant) rather than frequency modulation (when the switching frequency changes). The wider the impulse, the more power is pumped up into the transformer each cycle and the higher the output voltage is.
But if you just take the feedback signal from one of the output voltages, the PSU will regulate this signal only. Suppose it is the +5v rail. Then, as the load on this rail becomes higher, the voltage goes down. The pulse-width modulation (PWM) controller will increase the width of the impulses, lifting the voltage up back to the norm – and increasing the rest of the voltages as well. Several solutions are employed to avoid this effect.
First, the feedback signal is taken from the two most loaded output lines (+12v and +5v) through a dividing resistor. Thus, the regulation of each of these voltages independently worsens, but the PSU’s regulator now reacts to changes of the load on both voltages, i.e. the PSU works normally at different distributions of the load between these two power rails.
Second, the third heavy-current rail, +3.3v, has its own auxiliary regulator in the majority of PSUs – the so-called saturated-choke circuit. Regulators of that type feature a relatively high efficiency and a good regulation coefficient, although they are not of the switching variety. The +3.3v voltage comes from the same coil of the transformer as the +5v – the surplus 1.7 volts are damped on the choke. Well, there are PSUs where the manufacturer saves on the cost of the auxiliary regulator and just winds additional coils on the power transformer for the +3.3v voltage. Since the regulator doesn’t have a feedback circuit for this voltage, its stability in such PSUs is rather poor.
Third, the feeble-current rails, i.e. -12v and -5v, are sometimes equipped with ordinary linear regulators. The low efficiency factor of such regulators doesn’t practically affect the overall efficiency of the PSU due to the low load currents on these rails. Well, frankly speaking, the -5v voltage is only regulated this way – it is generated from the -12v with the help of a linear regulator for the sake of economy. And since modern PSUs don’t need this voltage at all, linear regulators have left computer power supplies altogether.
Fourth, the output voltages all pass through different coils of the so-called group regulation choke L1. Suppose the consumption on the +5v line has increased. The PWM controller reacts by making the impulses wider and the +5v voltage returns to the norm, but the other voltages also grow up, although their loads have remained the same. Yes, the above-described regulation methods are employed for them, yet it is the +5v voltage that is paid the most attention to. But the group regulation choke does the following: when the current in one coil goes up, the voltage increase in the other coils due to this current is subtracted from the corresponding output voltages. Thus, in our case, the current increase in the +5v coil leads to negative voltages in the coils corresponding to +12v and +3.3v – and these voltages won’t grow up as much as they would do without a group regulation choke.
Thanks to the above-described methods, the power supply ensures an acceptable regulation of all output voltages in a wide range of loads. Still, this regulation is far from perfect and gives rise to the most common problem of computer PSUs – irregular distribution of the output voltages. If the load of the PSU is unevenly distributed across its power rails (for example, the system consumes a high current on +5v, and a low current on +12v, which is typical for many systems with top-end Athlon XP models), the regulator cannot hold the voltages within the necessary limits. The more loaded rails bottom out, while the less loaded rails have too high voltages. That’s also why it is impossible to control independently the output voltages of the PSU: their ratio is strictly determined by the parameters of the power transformer and the group regulation choke, while the pulse-width modulation can only increase or reduce all the voltages at once.
Recently, an interesting solution has been implemented in expensive PSUs (for example, in models from OCZ Technology or Antec): auxiliary regulators on saturated chokes are installed not only on the +3.3v rail, but also on the +12v and +5v rails. This ensures a very good (as computer power supplies go) regulation coefficient of all output voltages and also allows to adjust each voltage independently from the others by changing the parameters of its own auxiliary regulator. Once again, this design is a prerogative of expensive PSU models while middle-range products have the above-described dependence of their output voltages on the load on each of the power rails.
After the group regulation choke, there are electrolytic high-capacitance capacitors (denoted as C3-C6 in the figure above) and filtering chokes at the output of the PSU: they must smooth out the pulsation of the output voltage at the frequency of the PWM controller and the power transformer. Although there is a group regulation choke, discrete filter chokes are still necessary. Due to their low inductance, they are good at suppressing high-frequency noise, which comes through the group regulation choke that has a rather high inductance.
Thus, the two inherent problems of any computer PSU are the dependence of each output voltage on the load on each of the power rails, not only on its own rail, and the pulsation at the frequency of the PWM regulator (usually 60 kHz) at the output of the PSU.
Of course, the manufacturers of PSUs, especially of low-end products, come up with their own “cost-effective innovations”, which it would take too long to enumerate. First of all, the ratings of the components suffer: for example, instead of the diode assemblages at the power transformer’s output, they can put not only assemblages rated for a smaller current than written on the PSU’s label, but even discrete diodes (high-quality units use only assemblages which consist of two diodes in one casing). This often leads to the PSU’s failing after a few minutes under the full load – the more so as the manufacturers often save on the dimensions of the heatsinks these diodes are mounted on.
The ratings of the capacitors suffer, too. A reduction of the capacities of the input capacitors worsens the reaction of the power supply to minor sags of the input voltage; a reduction of the capacities of the output capacitors leads to a stronger pulsation at the PSU’s output.
Sometimes this cost cutting can be observed with your own eyes: there are fewer output connectors, and the section of the wires in the power cables is 20AWG rather than the required 18AWG (in the AWG system, a bigger number means a smaller section of the wire). This results in a higher voltage drop on the wires, which leads to stronger pulsations of the voltage right on the power connectors of the consumers. If the load is high, thinner wires become perceptible hot, too.
The filtering chokes are the last to be abandoned – reducing their size doesn’t bring any serious cost advantages, so these chokes are present until the manufacturer decides they are unnecessary at all. A replacement of the chokes with straps leads to a stronger pulsation at the output of the PSU (if these were the output chokes) or to more noise the PSU is issuing into the power grid (if these were the chokes of the input filter).
One of the most memorable ways of making cheaper low-end PSUs was the realization of the standby +5v power source as a blocking generator with an electrolytic capacitor in the feedback circuit. In this scheme, which is a switching power supply based on a blocking generator, the output voltage is determined by the frequency of the impulses, and this frequency in its turn is inversely proportional to the capacitance of the capacitor in the feedback circuit. Cheap capacitors, intended to work at a temperature of 85°C at most, couldn’t stand the hard thermal conditions of the standby unit, typical for cheap PSUs (the standby is always on, while the PSU fan is only working when the computer is running), and after a year and a half of such work, the capacitor began to dry up, and its capacity – to decline. With the reduction of the capacity, the output voltage of the standby source grew up, and this voltage powers up the main regulator of the PSU. So in the end the main regulator just breaks down on your turning the computer on, and this breakdown is accompanied with an output of two- or threefold voltage along all of the power rails. Of course, the computer was just burned out after that – you could even see the burnings on the mainboard’s chips, in the hard disk drive and so on. The CPU and the memory were the only devices to have any chance at all – if their own regulators located on the mainboard endured the trial.
Of course, the manufacturers came to their senses at last and began to install near-infinite film capacitors into the standby circuit instead of electrolytic ones, since small capacity was quite sufficient there. Unfortunately, they had already produced quite a number of such “delayed-action” bombs which were a serious argument in favor of purchase of more expensive and high-quality PSUs which were free from such dubious engineering solutions.
The following kinds of power are generally recognized in alternating current circuits: 1) instantaneous power is the product of the currant by the voltage in the particular moment of time; 2) active power is the power generated on a purely resistive load and is measured watts. Active power fully goes to useful work (heating, mechanical motion) and is usually referred to as consumed power; 3) since a real load usually has inductive and capacitive constituents, reactive power accompanies active power. Reactive power is measured in reactive volt-amperes and is not consumed by the load. Received during one half-cycle of the line voltage, it is fully returned back into the power grid during the next half-cycle, just uselessly loading the power wires. Thus, reactive power is absolutely useless and various corrective devices are employed to oppose it.
Power factor is the ratio of active power to full power, i.e. to the vector sum of the active and reactive power.
A switch-mode power supply without any additional correction circuitry is a high capacitive load – as the schematic I showed you earlier shows, there are two capacitors (of a rather high capacitance) right after the diode bridge D1. On the PSU’s connection to the power grid, the first quarter-wave of the voltage loads the capacitors to 300 with something volts, then the voltage goes down quickly (the second quarter-wave), while the capacitors are more slowly discharging into the load (i.e. into the switching regulator). As the result, when the voltage starts to grow up again (the third quarter-wave), the voltage on the not-fully-discharged capacitors is about 250 volts, and the charge current will be zero while the voltage in the power grid is smaller than that (the rectifier’s diodes are locked by the applied reverse voltage which equals the voltage difference between the capacitors and the power grid). During the last third of the quarter-wave (of course, I give just approximate numbers since in reality they depend on the load and the capacitance of the capacitors) the voltage in the grid is again higher than the voltage in the capacitors, and the charge current starts to flow. The charge stops as soon as the voltage in the grid is again smaller than on the capacitors – in the first half of the fourth quarter-wave. As the result, the PSU is consuming power from the power grid in short pulses, approximately coinciding with the peaks of the sinusoid of the voltage in the power grid:
Power Supply without Power Factor Correction
The green line in the oscillogram above denotes the line voltage, the yellow line marks the current consumed by the PSU from the line. The power factor equals roughly 0.7 here, i.e. about one third of the power goes to heat up the cables, without doing any useful work. Home users shouldn’t bother much about this number, as home electric meters only measure the active power, but a low power factor may become a problem for large offices and rooms where there are many computers running at the same time, because the electric wiring and the accompanying equipment should be made according to the full power. In other words, it should be one third higher (at power factor = 0.7) than it would be if the PSU didn’t consume reactive power. A low power factor may also affect the choice of an Uninterruptible Power Supply (UPS) since they are limited by the full rather than active power, too.
That’s why Power Factor Correction devices are becoming ever more popular. The simplest and most widespread device like that performs passive power factor correction. This device is an ordinary choke of a rather high inductance, attached to the circuit in series with the power supply.
Power Supply with passive Power Factor Correction
This oscillogram shows that a passive PFC device smoothes out the pulsation of the electric current somewhat, stretching it out in time, but the inductance of a choke which can be packed into a PSU cannot seriously affect the power factor, so the power factor of PSUs with passive PFC is about 0.75.
Not only the dimensions, but also the influence of the choke on the operation of the PSU does not permit to use a choke of a greater inductance. A high inductance attached in series with the power supply worsens the dynamic characteristics of the latter, i.e. its reaction to quick changes of the load as well as to sudden surges in the power grid.
The PFC choke can also suppress interference, but only low-frequency one. Due to its high inductance, it lets high-frequency noise pass through.
Thus, the role of passive PFC is ambiguous. It does very little to improve the power factor, but worsens the dynamic characteristics of the PSU. So, when choosing between two PSUs – with and without passive PFC, you should base your choice on other factors, rather than on the presence/absence of passive PFC.
Unlike a passive PFC, an active PFC device is yet another switching power source, which increases the voltage. An active PFC is attached between the power grid and the main regulator, providing a constant voltage of about 380-400v on the input of the latter. Unlike the main switching regulator, an active PFC device is designed in such a way that it doesn’t require a smoothed-out voltage on its input, so it doesn’t require capacitors. Thus, the active PFC switching power source doesn’t put a capacitive load on the circuit and, accordingly, has a power factor close to 1.
Power Supply with active Power Factor Correction
As you see, the shape of the current consumed by a PSU with active PFC differs but little from the consumption of an ordinary resistive load – the resulting power factor of such a PSU may be 0.95-0.98 at full load. The power factor diminishes at smaller loads, to 0.7-0.75 at the minimum, i.e. to the level of units with passive PFC. Still, peak values of the consumption current are much smaller with active-PFC PSUs than with any other type of power supply, even under small loads.
The graph below shows you the experimentally measured dependence of the power factor on the load for three PSUs: without PFC, with passive PFC and with active PFC.
Besides providing a power factor closest to the ideal, active PFC improves the operation of the PSU, unlike passive PFC. Firstly, it additionally regulates the input voltage of the PSU’s main regulator. The unit becomes less sensitive to voltage sags in the power grid, and it is easy to design PSUs with a universal 110-230v power input that doesn’t require manual switching. Secondly, active PFC improves the reaction of the PSU to short-term (fractions of a second) sags in the ac voltage. In such moments the PSU works thanks to the power of the capacitors C1 and C2 of the high-voltage rectifier, and this power is proportional to the squared voltage in them. As I mentioned above, this voltage is 400 volts with active PFC instead of the ordinary 310 volts, so the efficiency of the capacitors improves by more than a half.
In fact, active PFC has two drawbacks only. First, like any other complication of the design, it reduces the reliability of the power supply unit. Second, the efficiency of the PFC device is not 100 percent, so it contributes to the heating-up of the PSU. Still, the advantages of active PFC usually counterweigh these drawbacks.
So, if you need a PSU with power factor correction, you should first consider models with active PFC. They alone provide a really good power factor, also improving other characteristics of the power supply. From the home user’s standpoint, PSUs with active PFC may come in handy for owners of low-power UPSes. Suppose you have a 500VA UPS – 50VA consumed by your LCD monitor and 450VA are left to your system case – and you want to upgrade the latter to the up-to-date level, knowing that a serious configuration may consume up to 300 watts at maximum from the PSU. In this case a PSU with a power factor of 0.7 and an efficiency factor of 80 percent (these are typical numbers for a good PSU) would give us a full power consumption of 300/(0.75*0.8) = 500VA, while the same PSU with a power factor of 0.95 would consume 300/(0.95*0.8) = 395VA. As you see, you have to upgrade your UPS for a power supply without PFC, since the current UPS won’t handle the load, while an active-PFC power supply will even give you a small reserve of 55VA. Well, we should also take into account the fact that the voltage on the output of cheap UPSes has a trapezoidal rather than sinusoidal shape – but the PSU with active PFC would still have an advantage, only the absolute numbers would be different.
Finishing this section of the article, I want to warn you against confusing two terms: power factor and efficiency factor. These denote two quite different things. Efficiency factor equals the ratio of the output power of the PSU to the active power it consumes from the power grid. Power factor is the ratio of the active power consumed from the power grid to the full power consumed from the same grid. The PFC circuitry in the PSU affects the amount of the consumed active power only indirectly – because the PSU itself consumes some power, plus the input voltage of the main regulator changes. The main purpose of PFC is a reduction of the consumption of reactive power by the PSU, but reactive power is not accounted for when calculating the efficiency factor. So, there’s no direct relation between power factor and efficiency factor.
The basis of our testbed for checking out power supply units is a semi-automatic device capable of setting any required load on the +5v, +12v, +3.3v and +5v standby power rails of the tested PSU, measuring the corresponding output voltages.
The hardware part of the device is based on a 4-channel Maxim MX7226 DAC to whose outputs power sources are attached. The power sources are made on operational amplifiers LM324D and high-power field transistors IRFP064N installed on heatsinks with forced air cooling.
Each transistor has a peak dissipated power of 200W, and since we use three such transistors in each of the heaviest-load rails (+5v and 12v), the testbed allows testing any existing ATX-compliant PSU up to the most powerful ones. Even considering the reduction of the allowable dissipated power of the transistors at high temperatures, the allowable load power on each rail is no less than 400 watts.
To measure the load currents and the output voltages of the tested PSU we use two 4-channel Maxim MX7824 ADCs – one converter is responsible for the currents, another for the voltages.
The testbed is fully controlled from the computer through the LPT port, from its powering-up to performing various tests, registering and processing the results. We wrote a special utility that allows to manually select the load currents independently for each power rail and to perform some non-standard tests (for example, to build the cross-load characteristic as described below) – fully automatically.
Besides the main testbed we use two auxiliary test devices. The first of them is a generator of rectangular pulses with a frequency that can be discretely varied from 60Hz to 40kHz:
The generator is attached to the tested PSU as a load – a switch allows to select the rail (+12v or +3.3v). In any case, the generator creates a load current of about 1.3 amperes. This allows to estimate how well the tested PSU reacts to relatively powerful load pulses of rectangular shape that follow at frequencies from tens of hertz to tens of kilohertz.
Second, we use an ordinary shunt on powerful wire resistors with a total resistance of about 0.61 Ohms to make oscillograms of the power consumed by the PSU and of the supplied AC voltage.
When testing a PSU, the probes of a digital two-channel oscilloscope are attached to this board. One channel is drawing the oscillogram of the AC voltage and another – the oscillogram of the power consumption of the PSU. Then, these oscillograms are processed with a special utility which calculates the parameters we’re interested in: the active, reactive and full powers consumed and the power and efficiency factors.
To take oscillograms we use a digital dual-channel “virtual” oscilloscope M221 from the Slovak company ETC (we call it “virtual” because this oscilloscope is a board installed into the computer and, unlike ordinary oscilloscopes, cannot work without a computer since it doesn’t have its own hardware tools for controlling and displaying information). The analog part of the oscilloscope has a pass band of 100MHz; its maximum speed of digitization of a random signal is 20 million samples per second and it has a sensitivity range from 50 mV/del to 10 V/del. Besides measuring the efficiency and power factors of the tested PSU, the oscilloscope is employed to estimate the amplitude, shape and frequency content of the pulsation of the PSU’s output voltages.
For a quick evaluation of the currents and voltages during the tests, and also for periodical tests of other measurement equipment, we use a Uni-Trend UT70D multimeter which accurately measures currents and voltages, also of non-sinusoidal shape, which is very important when testing PSUs without power factor correction. Many measurement devices, without the “TrueRMS” label, cannot produce adequate measurements of alternating currents and voltages whose shape is other than sinusoid.
A digital thermometer Fluke 54 Series II with thermocouples 80PK-1 and 80PK-3A (the model names are quoted from the Fluke catalogue) measures the temperature inside the PSU in our tests. Unfortunately, our remote infrared digital thermometer was inaccurate on shiny metal surfaces (like the aluminum heatsinks of computer power supplies), so we had to use a thermocouple thermometer instead.
An optical tachometer Velleman DTO2234 measures the speeds of PSU fans. The PSU remains closed at that (i.e. we don’t interfere with its regular thermal conditions) – we only stick a thin stripe of reflective material on one of the fan’s blades.
And the last touch: to feed the same AC voltage to all PSUs, irrespective of its daytime fluctuations, and to test PSUs at a higher or smaller AC voltage, they are attached to the power grid through a laboratory auto-transformer Wesley TDGC2-2000 with an acceptable load power up to 2,000 watts and an adjustable voltage range of 0-250 volts.
The first and foremost test for each power supply is the construction of the so-called cross-load characteristic. As I told you in the theoretical part of this article, each output voltage of the PSU depends on the load on the corresponding power rail, but also on the loads on the rest of the rails.
The ATX standard describes the maximum possible deviations of the output voltages from their defaults: 5 percent for all positive voltages (+12v, +5v and +3.3v) and 10 percent for all negative voltages (-5v and -12v; modern PSUs have only the latter of them, though). The cross-load characteristic (CLC) of a PSU is the set of possible combinations of the loads when none of the output voltages goes out of the permissible range.
The CLC is represented on a plane where the X axis shows the load on the +12v power rail, and the Y axis – the total load on the +5v and +3.3v rails. When constructing the CLC graph, the testbed is automatically changing the load on these rails stepping 5 watts, and if the output voltages of the PSU all fit into the required ranges, a dot is placed on the plane. The color of the dot – from green to red – reflects the deflection of the voltages from the standard. Our testbed controls three main output voltages, so three diagrams are produced for each PSU, in which the same area is colored with different colors that reflect the stability of each of the voltages. Below you see the cross-load characteristics of a Macropower MP-360AR Ver.2 PSU, colored for the +12v voltage (we will publish animated pictures in our reviews that will show all three voltages one by one; the currently displayed voltage will be indicated in the top right corner of the diagram, above the color scale).
Each dot of this diagram corresponds to one measurement step. For the sake of convenience the dots where the voltages are out of the acceptable limits are colored gray and have a smaller size – it helps the tester to follow the progress of the test in real time. After the measurements are made, the data are processed with bilinear interpolation. Thus, instead of discrete dots we get a colored area with sharp edges for better reading:
So what does this diagram say? The tested PSU excellently handles the load on the +12v rail – it is capable of outputting the necessary voltages when there’s the maximum load on this rail and only 5 watts on the +5v rail (5 watts is a typical starting value in our measurements; for high-power PSUs, which are unstable at so small loads, we start with 15 or 25 watts).
The straight vertical borderline in the bottom right part of the diagram says that the PSU has got to the power limit of the +12v rail (for the given PSU, it is 300 watts) and the testbed didn’t increase the load current further to avoid damaging the PSU. The vertical line changes into a slanting one in the top right corner of the diagram – this is where the testbed reached the maximum allowable power of the PSU (340 watts with this model) and had to reduce the load on the +12v to increase the load on the +5v, while still keeping the PSU out of danger.
Then, in the top part of the diagram, the slanting line becomes perfectly flat. This is where the testbed reached the maximum allowable load on the +5v rail and didn’t proceed further for safety reasons, although the voltages the PSU yielded were within the norm.
And lastly, in the top left corner of the diagram we see an irregular slanting line which is evidently not a power limit, since the load on the +12v is too low for this area. But this line is explained by the red color of the diagram: when the load was high on the +5v, but low on the +12v rail, the +12v voltage deviated more than 5 percent out of the norm, thus setting up the border of the CLC diagram.
Thus, the cross-load characteristic tells us that the given PSU outputs quite stable voltages and can yield the declared wattage, but it would be preferable for modern systems where the CPU and the graphics card are supplied from the +12v power rail, since this PSU behaves better when there’s a high load on the +12v rather than on the +5v.
For the comparison’s sake, here’s the CLC diagram of a cheaper PSU, the L&C LC-B300ATX model with a declared wattage of 300W. The diagram is again built for the +12v power rail only:
This diagram obviously differs from the previous one. The bottom line of the CLC is not horizontal anymore, but is going up from left to right. The red color of the diagram there indicates that this is not because the +5v voltage is out of the normal range (it often happens when there’s a high load on the +12v), but because the +12v power rail bottoms out. Then, there’s no horizontal “ceiling” on top of the diagram; its top point corresponds to a load of about 150 watts on the +5v rail, which means you cannot get the promised 180 watts from this power supply in practice at any combination of the loads. Then, although the declared wattage on the +5v and +3.3v rails is higher with this PSU than with the MP-360AR (180 watts against 130 watts), you can see that the slanting line in the top left corner of the MP-360AR’s diagram starts at a load power of more than 80 watts on the +5v rail. With the LC-B300, however, this slanting line starts at abut 50 watts. This means that although the LC-B300 boasts a higher declared wattage on the +5v rail, you can more often get a much higher real wattage on this rail from the Macropower unit.
Attentive readers may have noted that if we put the two diagrams in one, in the same scale, the CLC of the Macropower unit will be more stretched along the +12v axis than the CLC of the unit from L&C. This comes as these two units comply with two different versions of the ATX/ATX12V Power Supply standard which described different preferable distributions of the loads among the PSU’s power rails. The following figure shows you the CLCs which PSUs of different production years should have had, according to Intel (who is the originator of all the family of ATX standards):
You can see that the ATX standard originally supposed a big consumption from the +5v and 3.3v rails, and the whole stuffing of the computer did feed on these two rails, while the +12v was only loaded by the mechanics of the optical and hard disk drives.
Things were changing, however. As central processors became power-hungrier, their supply from the +5v poised a whole lot of problems before mainboard developers: firstly, it was already clear then that the growth of the power consumption of CPUs would continue further and would raise the problem of transferring such high currents to the mainboard – the standard connectors might not hold. Secondly, the mainboard’s power connector would have to be either crammed in somewhere near the CPU’s voltage regulator module or placed farther, but it would be difficult to wire a high-current bus from this connector to the VRM.
Taking these things into consideration, Intel proposed the ATX12V standard, according to which the CPU should be supplied by the +12v rail. Evidently, at the same consumption, it means a current 2.4 times smaller. But as the main ATX connector had only one +12v wire, they had to introduce an additional 4-pin ATX12V connector, which solved two problems at once: 1) the connector’s contacts wouldn’t burn because of high load currents and 2) the mainboard’s PCB design was simplified as it was easier to find a place for a small 4-pin connector near the VRM than for a big 20-pin one.
Unfortunately, AMD didn’t support Intel’s initiative, and many owners of Socket A mainboards, about 20-25 percent of which still don’t have an ATX12V connector, have suffered from the problems Intel was talking about four years ago. As soon as there appeared high-powered processors for the Socket A platform, users began to report of burned-out pins in the PSU’s connectors and a strong misbalance of its output voltages (as the above-shown CLC diagrams indicate, even cheaper PSUs are better at handling the load on the +12v rail)…
In fact, there is only one technical drawback from the introduction of ATX12V – the efficiency of the CPU’s voltage regulator has diminished somewhat since the efficiency of any switch-mode converter diminishes when there’s a bigger difference between the input and the output voltages. This, however, was well compensated by the increased efficiency of the PSU proper. Like for the mainboard developers, the decision to orient on the +12v power rail as the main source of power has simplified the construction of the PSU for the designers.
You can see it in the diagrams that the versions of ATX12V up to 1.2 inclusive differ from the original ATX standard in a higher allowable consumption on the +12v power rail. Version 1.3 of the standard brought more serious changes: for the first time in the evolution of computer PSUs the required allowable load on the +5v rail was decreased, while the allowable load on the +12v rail was increased even higher. In fact, computer PSUs began to adapt to the more modern systems in which fewer consumers remain on the +5v rail (processors have long been feeding on the +12v, and graphics cards are now following the suit). Unlike earlier models, an ATX12V version 1.3-compliant power supply didn’t have to maintain stable voltages when there’s a high load on the +5v rail and a small on the +12v rail.
The last version of the standard for today is ATX12V 2.0. Easy to see, the allowable load on the +5v rail has been decreased even more, and now it is only 130 watts. Meanwhile, the allowable power load on the +12v has grown up further. Besides that, ATX12V 2.0-compliant PSUs acquired a 24-pin power connector for the mainboard instead of the older 20-pin one. Four years ago the older connector was found to be insufficient for power-supplying the CPU, and they invented ATX12V, but now the maximum allowable current in this connector is not enough to give juice to new graphics cards with the PCI Express interface. ATX12V-compliant PSUs also have two sources of the +12v voltage, which are actually the same source inside the PSU, but with independent over-current protection circuits – according to the safety requirements of the IEC-60950 standard, currents over 20amps are not allowable on the +12v rail, so this power rail has to be split in two. When there’s no need to comply with this standard, the manufacturers can just leave out the appropriate circuitry and an ATX12V 2.0 PSU with, say, 10- and 15-ampere currents on its +12v rail, can be considered as a PSU with one +12v rail with a max current of 25amps.
So, returning to the above-discussed PSUs, we can say that the MP-360AR Ver.2 complies with the ATX12V 2.0 standard, while the LC-B300 – with ATX12V version 1.2, hence the difference in their CLCs. Well, the problem is not only in a formal compliance with this or that version of the standard. I said above that the LC-B300 cannot provide the declared wattage on the +5v rail, and now let’s compare its CLC diagram with Intel’s recommended CLC for ATX12V 1.2 units:
As you see, this power supply just doesn’t meet the requirements to 300W models in the allowable load on the +5v rail, so we can only regard it as a 300W PSU with a reservation that these watts are not very honest. For comparison, here’s the diagram of the MP-360AR in comparison with the recommended CLC for 350W ATX12V 2.0 units:
As you see, we have almost a perfect coincidence here. I guess there’s no need to comment on the comparative quality of the two PSUs.
Generally speaking, it is quite hard to meet Intel’s strict requirements to the cross-load characteristic. There are few units available that can boast a full compliance with the standard, but the egregious violation of the recommendation the LC-B300 is an example of happens rarely, too.
As for the coloring of the CLC diagram, it would be all green in an ideal world. In our reality, it is normal when each voltage, except the relatively stable +3.3v, goes through the entire range from green or yellow-green at one end of the diagram to red at the other end. Sometimes, there’s no green color in the CLC diagram at all – it means the voltage is originally higher than necessary. The worst case, however, is when a voltage goes twice through the entire color range, from red at one end, through green in the middle, and to red at the other end (see the diagram of the LC-B300 above). It means that the voltage sags low at one end of the CLC (clearly, when there’s a small load on the +5v and a high load on the +12v, the only thing the latter rail can do is to sag), but rises high at the other end; in other words, this voltage lacks stability.
To end the CLC-related section of this article, I want to offer you an example of an ideal power supply. I mentioned PSUs from Antec and OCZ with dedicated voltage regulators on each of the primary rails, so here’s the experimentally measured CLC of a PowerStream OCZ-470ADJ unit from OCZ Technology (this is the comprehensive picture with all the three voltages – the animated GIF is changing each 5 seconds):
You can see that the whole CLC area is limited by the maximum allowable load of this PSU. Moreover, none of the voltages comes close to a 5-percent deviation. Alas, such power supplies are rather expensive as yet…
Of course, we don’t end our tests with building the cross-load characteristic. We also check out the stability of each PSU under a constant load from zero to the maximum, stepping 75 watts. Thus, we can see if the PSU can sustain such a load at all.
Then, we measure the temperature of the diode assemblages of the PSU and the rotational speed of its fan under different loads (the fan speed depends on the temperature one way or another in nearly all modern PSUs).
The temperature measurements should be regarded with some skepticism, though. The design of the heatsinks and the placement of the diode assemblages vary between different PSU models, so the temperature measurements are not very accurate. On the other hand, the showings of the thermometer may be quite interesting in critical situations when the PSU is close to dying from overheat (this sometimes happens to cheapest models). I have seen PSUs whose heatsinks were above 100°C hot under full load.
The measurements of the fan speed produce more curious results. Although all manufacturers claim the fan speed to be temperature-dependent, the practical realization of the fan control system varies greatly. As a rule, the starting fan speed of low-end PSUs is about 2000-2200rpm and only grows up by 10-15 percent as the unit heats up. With high-quality models, the starting speed may be just 1000-1400rpm and may double under full load. So, the PSU will always be noisy in the first case, but in the second case owners of average system configurations may hope for noiselessness.
When the PSU is working at its full wattage, we also measure the pulsation swing of the voltages it outputs. According to the standard, the pulsation swing in the range up to 10MHz should not exceed 50mV for the +5v rail and 120mV for the +12v rail. In practice, there can be pulsations of two frequencies at the output of the PSU: about 60 kHz and 100Hz. The former frequency is the result of the operation of the PWM controller of the PSU (which usually works at 60 kHz) and is present more or less in all PSUs. A typical pulsation is shown on the next oscillogram (the +5v rail is marked with green; the +12v rail is marked with yellow):
This is the case when the pulsation on the +5v rail has deviated beyond the permissible 50 mV. You see the classic triangular shape of the pulsations in the oscillogram, but in more expensive PSUs the switching moments are smoothed out by the output chokes.
The second frequency is the double frequency of the power grid (50Hz) that usually passes through to the output because of an insufficient capacity of the high-voltage rectifier’s capacitors or various design flaws. Such pulsations (they are given with a timebase of 4 ms/del) are usually observed in low-end PSUs and are rare in middle-range products. The swing of these pulsations grows proportionally to the load on the PSU and can sometimes go out of the allowable limits at peaks.
We also attach a generator of rectangular pulses to the PSU at 150W load to measure the amplitude of the pulses on the other wire of the PSU, i.e. not on the one the generator is attached to. Thus we check out the PSU’s overall reaction to such pulse loads and, particularly, how well it suppresses noise from each of the devices attached. Due to the voltage surges provoked by switching, the measurements are not very accurate, but sometimes they yield most curious results.
As for measurements of the efficiency factor and the power factor of the PSU, this is the least interesting and important section of our tests. Our experience says these parameters are very close among different PSU models of the same type, and minor differences are not important at all for the absolute majority of users, so we perform such measurements in rare cases only: the power factor is measured for PSUs for which PFC is declared, and the efficiency factor comes along (in fact, the value of the efficiency factor is produced automatically – no additional measurements are necessary) or is measured specifically if we suspect the efficiency factor of a particular PSU goes out of the acceptable limits, but this is a rare thing.
Winding up this article, I want to say some words about what we are not measuring and are not going to measure, although we could. We don’t trust tests that measure the absolute maximum power as outputted by the PSU – when the load on the PSU is being increased till the PSU fails or is saved by its protection circuitry. Such tests produce results that vary between two samples of the same PSU model as well as depending on how exactly the tester loads the PSU, i.e. how the load is distributed across the rails of the PSU. Besides that, the ability of the PSU to maintain certain wattage is not necessary for a normal operation of the computer. What’s important is the ability of the PSU to output voltages and pulsations within the limits described by the standard, but such tests don’t pay attention to this fact. They produce pretty numbers, which have little relation to reality.
So, our new methods of testing computer power supplies allow to examine the behavior of a PSU in more detail as well as to compare visually different PSU models. The cross-load characteristics make the tests more intuitive and they show objectively and without any additional reservations what this or that PSU is like.