Suppose, the core of the choke L1 was saturated at the beginning and the choke let pass the current freely. At time t=0 the voltage on the output of the main regulator changed for negative, and the voltage on the choke is 4v (V1-Vc = 10v - 6v = 4v) and remains such during all the time of the passing of the negative pulse, i.e. 10 microseconds. This switches the choke from the saturated state into high-inductance unsaturated one in which it inhibits the current flow, as any inductance does. So, when at time t=10µsec, there appears positive +10v voltage on the output of the main regulator, the current doesn’t flow through the choke and the voltage V3 on the regulator’s output remains zero. But now the 10v voltage is applied to the choke, driving it back to saturation, but in 4 microseconds (this time depends on the length of the negative-polarity impulse and on the voltage difference on the choke: 10ms*4v/10v = 4ms). Thus, at time t3=14µsec the choke becomes saturated and starts to pass the current through, and the output voltage grows up to 10v. In 6 microseconds more, at time t4=20µsec, the voltage V1 on the output of the main regulator drops to ‑10v, so the output voltage V3 goes down to zero.

Thus, without the choke L1 the load received pulses of 10µsec duration and a period of 20 microseconds, but with this choke the period remains the same and the duration is diminished to 6 microseconds. Easy to calculate, this leads to a voltage drop from the earlier 5v to 3v. The dissipated power is small here – when unsaturated the choke doesn’t practically conduct the current; in the saturated state, the voltage drop is close to zero, so there’s no significant loss. The power consumption from the source Vc is only determined by the material of the choke’s core and the number of coils, so it can be made small irrespective of what current the regulator should give out to the load. In practice, the voltage that determines the length of the pulses on the output of the auxiliary regulator (and, accordingly, its output voltage) is not fixed, but has feedback with the regulator’s output, so the output voltage can be kept the same almost irrespective of the load.

Thus, we have a voltage regulator with a high efficiency factor (and thus not requiring forced cooling), which easily works in pair with the main regulator and takes little space. The choke L1 is rather small, and the rest of the elements are not even worth mentioning.

Until very recently such regulators were typically used on the output +3.3v rail of ATX power supplies. This voltage was derived from the 5-volt winding of the power transformer in the above-described way. Yet it is clear that nothing prevents us from regulating the rest of the voltage likewise. We can just wind the power transformer in such a way as to make its output voltages 2-3 volts above the required levels and then put a magnetic amplifier on each of its outputs that would step the voltage down as necessary. Thus, we have a two-step regulation of the PSU’s output voltages at a relatively low cost, and this PSU is more tolerant to variations in the load than ordinary units with group regulation of the voltages.

Well, considering that there are only three critical output voltages in the computer PSU, among which one (+3.3v) is already regulated with a similar circuit, the problem appears most simple. It is only necessary to regulate one more voltage with a magnetic amplifier and to send a feedback signal from the remaining voltage to the main PWM controller as it can handle the task of regulating one voltage with ease. The negative output voltages can be regulated by ordinary low-power linear regulators – load currents are small there, so they won’t heat the PSU up much.

We haven’t yet tested such units on our site – they are rather scanty in market and cost quite a big sum of money, but for our today’s tests we’ve got only one unit (BeQuiet P4-450W) designed classically, while the products from OCZ and Antec are designed as described above – with dedicated auxiliary regulators on their outputs.

But first I want to present you a small table with the basic load characteristics of the units I am about to test: