Switching Power Harmonics
When considering harmonics generated from processor clocks, we usually expect problem frequencies up to about the tenth harmonic. But when working with switching power devices, harmonics go much higher. Let's take a look at the issue, along with possible solutions.
Using Fourier techniques, we find that emission contributions from processor clocks start falling off at the frequency determined by 1/pi*tr — this is the second corner frequency you would see on a Bode plot. Thus, a one nsec risetime will be expected to produce problem emission frequencies up to about 300MHz.
But when dealing with power harmonics, this simple guideline is not valid. Switching edge rates are much lower, but the switching currents and voltages are much higher. A typical power switch will have about 100 nsec rise time, but we may be switching hundreds of volts and tens of amps. Thus, the time rate of change of voltage and current in a slow power switcher may approach that of lower level logic circuits. In one remarkable case, we found a 50kHz switching power supply was producing excessive Class B emissions at 300MHz — that's the 6,000th harmonic!
So we are faced with the task of controlling emissions from the low-order harmonics all the way up to the UHF band. And things will get worse if the risetimes continue to fall and power supply designers strongly resist any attempt to slow the edge rates.
Power switching topologies vary widely, but they all share two common problems — switching current surges on the power rails and output switching power harmonics.
All power switching applications put a heavy load of switching currents at the fundamental frequency and harmonics. We need to do a good job of filtering at the power source side — this means not only the fundamental and lower harmonics, but also the higher order harmonics.
The problem at the output divides into two general cases. The first case is in a power supply, where the switched power is usually passed through a transformer, rectified and converted to DC. In this case, the issue is largely that of blocking high frequency capacitive paths coupling out to the load. The second is where pulsed power (e.g. PWM, or pulse width modulation) is transmitted to a load, often to a motor located remotely from the output drive. In this case, inductive effects are also significant.
Following are the possibilities, noting that not all will be palatable or even implementable in any given case.
1. Slow the switching edge rate. This one creates the greatest heartburn for the power supply designer, as it eats into the conversion efficiency — the faster the switch, the less time spent in the power consuming state and the hotter the switch runs. Nevertheless, if your switch can tolerate a little higher power consumption, this is an exceedingly effective way to control high frequency harmonics. The fix is to slow the gate drive, usually with an RC filter — since gate drive currents and voltages are low, there are no special requirements on the filter components.
2. Filter the power rails. Filtering on the power side prevents switching harmonics from backing out the power input line. But do be aware that there is little isolation between power source and the load — when the switch (or switches) are on, there is virtually a direct connection between power side (voltage or common) and the output lines, especially at the higher frequencies.
Everyone is aware of the need to filter the switching power frequencies, but that is only a start. The key is to keep a very low impedance in the input power loop (which would include the filter capacitors and the switches). This means the capacitors need to be functional at the highest problem frequencies, and that the current path impedance be kept to an absolute minimum. Generally, this means that using traces between the filters and the switch is unacceptable — you need to have near planar paths to keep the inductance (and ground resistance) to a minimum. By the same token, the current loop area must be kept as low as possible. A particular problem with ground impedance is that it creates common mode noise, rendering differential mode filtering ineffective.
This raises the question of whether you can insert series inductors to slow the switching current spike. If you have multiple switching outputs, as you would with a three-phase drive, you might consider using one filter inductor at the source, rather than three to six at the load. Note, that if you do decide to try this one, beware that the inductive kick may put an overvoltage across the gate — output protection snubbers won't help — you need to put a diode snubber across the inductor.
3. Intercept the output coupling path. This applies largely to the high voltage switching node, a textbook case in switching power supplies (Figure 1). The primary paths are capacitive coupling to the heat sink and to the secondary of any transformer. Particularly troublesome is using the enclosure as a heat spreader — noise from the switcher capacitively couples noise directly into the enclosure, and this is tough to deal with. The coupling path must be intercepted, either by reducing capacitance by positioning or with the use of Faraday shields (again, a text book case). Faraday shields for switching transistors are available commercially. Faraday shields for transformers are also available, but not in high favor — inserting a thin metal shield between primary and secondary gives the transformer houses heartburn. One approach we read about interposed an additional winding layer between primary and secondary, reportedely with good results. Whichever choice you make, make sure you don't close the loop of the shield, lest you create a shorted winding.
If you can't block the currents this way, your last resort is to provide a low impedance high frequency path from the secondary back to input power common, typically with a small capacitor.
4. Intercept the drive current harmonics. This problem applies largely to the power drive lines in a PWM application (Figure 2), where the switched currents go out to a remote load, harmonics and all. The problem is that while the low frequency currents go through the motor windings as intended, the high frequency currents capacitively couple from the windings to the motor housing, which is grounded. The result is a power cable that radiates ground currents that become a conducted emission problem.
There are two ways to handle this condition. The best choice is to block the high frequency components from going out the drive cable. A distant second choice is to ensure that the currents that out the cable return on the cable, not by a sneak ground path.
Blocking the currents generally requires some inductance in the drive lines. The good news is that inductance in this path does not reduce power efficiency. The bad news is that the inductors have to have the necessary current capacity — this pretty much eliminates the use of toroidal ferrites. You may also have to cope with inductive kick.
FIGURE 1. Capacitive coupling paths from switching node to heat sink and secondary.
A possible alternate is to note that with most PWM drives, you are only switching one IGBT at a time. Thus, it is possible to use a common mode choke around all the lines, making ferrites a viable possibility.
Whichever you do, make sure your components are working at the frequency range you are trying to kill. Wound inductors resonate at very low frequencies, rendering them useless at the higher frequencies.
FIGURE 2. Noise currents from output drive to motor housing.
Also note that you can't use shunt capacitors at the output drive — these are low impedance devices, and capacitors just increase the current through the IGBT without providing any substantial filtering.
If you can't block the high frequency currents at the drive, your only remaining option is to provide a low impedance path from the load back to drive common, not a trivial task. Usually, this involves a shielded cable and a capacitive connection from the cable ground back to drive common.
When working EMI problems in a power switching application, you have a couple of tasks. First is to ensure the power input to the drive is well filtered, at all problem frequencies, keeping the impedance in power current paths to an absolute minimum. Second is to block high frequency currents on the output side from getting out the load.