Designing your own power supplies: Ripple and filters
Since the overwhelming majority said "yes" when i asked if you wanted to learn how to design your own SMPS, i'll start writing some useful stuff from time to time. Today i'll be talking about the various ripple mechanisms in a SMPS supply. This is going to get long probably, but it'll be full of useful info so read on. Ever wondered why, while the switching frequency is in the 10s, sometimes even 100s of kHz, the output L-C filter has a cutoff in the 100s of Hz? Well, here you go.
In a SMPS we have many types of ripple.
Now i'm sure you realize that those poor caps take quite a beating.
Note that i omitted mains frequency ripple from the list. In a regulated SMPS the controller cancels that out. This is how they get by with so small primary capacitors. However, a regulated SMPS isn't the ideal power supply for all types of equipment... Audio amplifiers for example do better with unregulated rails. If your SMPS is unregulated you have to worry about mains frequency ripple too, and the load ripple becomes several times larger! And there we go, we need lots of caps again...
Filters.
While all SMPS books give formulas for filter calculation, there is something that i haven't seen mentioned too much, and that becomes very apparent in simulation. That is, filter overshoot when transitioning from a heavy to a light load. Most dangerous when you're going from max load to zero. An inductor is a device that opposes changes in current. That's actually why SMPS supplies work in the first place. When current demand is suddenly removed, the inductor has to release the energy it had been storing to deliver that current. And this translates to a sudden increase in output voltage. Most books probably assume that we're doing fixed output power supplies and there is some sort of overvoltage eating circuitry at the output, such as a zener diode. Even so, that must be sized properly, and if you hit it all that often, efficiency suffers. This can be mitigated by having a proper filter to begin with.
Generally speaking, the bigger the inductor, the higher the overshoot. Now, with a bigger inductor, we would need less caps since there is less ripple current left, isn't it? Well, in theory. In practice, taming overshoot means we would need bigger caps. Fortunately this isn't an issue in PC SMPS as they always cheap out on inductors. However, the smaller the inductor, the greater the minimum load required... are we running in circles already?
But why is minimum load so important? Well, the ringing of the output filter:
The first one is a non-issue generally. The second however is a big problem. The fact is that in an effort to get good transient response (which btw is only essential in some specialty applications), many design engineers do not leave enough phase margin so this ringing could cause control loop oscillation. It's also unfortunate that at the transition between continuous current mode and discontinuous current mode in the inductor, the maths change a lot. And control theory is hard enough as it is.
Now, i propose a different approach. Let the sucker ring all it wants and adjust compensation components to suit. But this is for another topic... Anyway. Simulation can be a great aid to people like us who have no access to expensive equipment like network analyzers. You cannot compensate a control loop without a Bode plot of the frequency response, and by plugging real-world measured data in a simulation, you can get *very* close to what the real thing would do. You have to design on the conservative side anyway, since real-world parts have real-world tolerances, so overcompensating in the simulation is perfectly adequate.
I once read about a power supply design that worked fine as prototype but when they received the first batch from the factory it oscillated. It turned out that one little Chinese worker forgot to put one of the layers of tape on the transformer. This changed the response of the entire supply... IMO this screams of bad design. If one missing layer of tape threw it into oscillation, i can only imagine what real-world aging capacitors would do to it.
That's it for now, stay tuned.
In a SMPS we have many types of ripple.
- Ripple from the switching frequency (this is always present)
- Ripple from the load (especially true for PC and LCD power supplies where the loads are other switching power supplies)
- Ripple from whatever frequency the controller is correcting at (which may or may not be the load frequency!)
- Ripple from ringing of the output filter at low loads (hence the need for those pesky minimum load resistors)
Now i'm sure you realize that those poor caps take quite a beating.
Note that i omitted mains frequency ripple from the list. In a regulated SMPS the controller cancels that out. This is how they get by with so small primary capacitors. However, a regulated SMPS isn't the ideal power supply for all types of equipment... Audio amplifiers for example do better with unregulated rails. If your SMPS is unregulated you have to worry about mains frequency ripple too, and the load ripple becomes several times larger! And there we go, we need lots of caps again...
Filters.
While all SMPS books give formulas for filter calculation, there is something that i haven't seen mentioned too much, and that becomes very apparent in simulation. That is, filter overshoot when transitioning from a heavy to a light load. Most dangerous when you're going from max load to zero. An inductor is a device that opposes changes in current. That's actually why SMPS supplies work in the first place. When current demand is suddenly removed, the inductor has to release the energy it had been storing to deliver that current. And this translates to a sudden increase in output voltage. Most books probably assume that we're doing fixed output power supplies and there is some sort of overvoltage eating circuitry at the output, such as a zener diode. Even so, that must be sized properly, and if you hit it all that often, efficiency suffers. This can be mitigated by having a proper filter to begin with.
Generally speaking, the bigger the inductor, the higher the overshoot. Now, with a bigger inductor, we would need less caps since there is less ripple current left, isn't it? Well, in theory. In practice, taming overshoot means we would need bigger caps. Fortunately this isn't an issue in PC SMPS as they always cheap out on inductors. However, the smaller the inductor, the greater the minimum load required... are we running in circles already?
But why is minimum load so important? Well, the ringing of the output filter:
- Appears as unwanted ripple at the output
- Can confuse the control loop and cause it to oscillate
The first one is a non-issue generally. The second however is a big problem. The fact is that in an effort to get good transient response (which btw is only essential in some specialty applications), many design engineers do not leave enough phase margin so this ringing could cause control loop oscillation. It's also unfortunate that at the transition between continuous current mode and discontinuous current mode in the inductor, the maths change a lot. And control theory is hard enough as it is.
Now, i propose a different approach. Let the sucker ring all it wants and adjust compensation components to suit. But this is for another topic... Anyway. Simulation can be a great aid to people like us who have no access to expensive equipment like network analyzers. You cannot compensate a control loop without a Bode plot of the frequency response, and by plugging real-world measured data in a simulation, you can get *very* close to what the real thing would do. You have to design on the conservative side anyway, since real-world parts have real-world tolerances, so overcompensating in the simulation is perfectly adequate.
I once read about a power supply design that worked fine as prototype but when they received the first batch from the factory it oscillated. It turned out that one little Chinese worker forgot to put one of the layers of tape on the transformer. This changed the response of the entire supply... IMO this screams of bad design. If one missing layer of tape threw it into oscillation, i can only imagine what real-world aging capacitors would do to it.
That's it for now, stay tuned.
