Design tips for switching regulators and the LTC3873

A friend of mine was helping his friend with a switching power supply. The fellow was using a Linear Tech chip and was having problems. According to my buddy:

John’s complaint is the data sheet is not complete enough to make the proto work. His claim is the compensation is omitted from the datasheet and phone calls do not get any more than "try it". Now he can’t even get calls returned. So far he has fried several parts. [He’s using the] LTC3873-5. Not sure why this part was chosen. Apparently, this part requires loop compensation on a pin. It is a constant-frequency current-mode device, used in flyback mode with a 7 (!) output transformer and using FET current sense. As I look at the datasheet, the 2 circuits have different compensation values and no mention in the datasheet. His configuration may not even look like the examples.

So I wrote some of my power supply expert pals. One, who definitely wants to remain anonymous, said:

Read about compensation in LT App Notes 19 and 25. Step response method is fine for current mode topologies that have 40% inductor ripple current and enough Cout. If you really want to see loop response; you need an FRA (frequency response analyzer) like a Ridley or Venable.

Yes, compensation discussion is missing from all LT data sheets.   I always wondered how they had the balls to call it Opti-Loop. I think Mystery-Comp would have been a more appropriate moniker.

Then I got really neat response from National Semi crack FAE Jon Dutra. He writes:

Need more info to work on the problem
If he is "Frying parts" he is violating some Abs Max spec,
(Usually that means running close to the max with a poor layout
To help him I Need ……
= Customer phone number and e-mail
= Vin (Range)
= Vout
= Iout min and max
= Customer schematic (Here you see what "Funny stuff" the customer is attempting)
= This part should be nearly impossible to blow up if he is using a current sense resistor
Generally I compensate these types of regs  by the "quick method"…

Lift a pin on the  11.8K resistor and put an RC substitution box from Ith to gnd. (Across the 47 pF). Have a 100 Hz transient load on the output going from 10% load to 100% load. Tune the RC box for critically damped response.

So what a dude Jon is— he is willing to help someone with a competitor’s part. Of course, once Jon knows the details, I am sure he will try to get the guy to use a National Semi part. This reminds me of when I first met Jon back in 1998. He was an FAE at Linear Tech and I was consulting to Hewlett Packard. I was designing the analog front end for an automotive diagnostic tool— a cross between a DVM and an oscilloscope. John helped me out with several parts of the design, dutifully putting in Linear Tech parts. Then I asked him about the multiplexers. He said: “Try Maxim, Harris and Intersil.” I was astonished. I pointed out that they were competitors to Linear Tech. He said: “Yeah, I know, and if Linear Tech had parts like this I would tell you about them, but LT does not make muxes. My job is helping you solve your problems. Once you trust me to do that we can see about using LT parts. That way, the next design you have you will call me in and give me a chance to help you again.” I wish all people and companies were like Jon and Linear Tech. They are reputed to have some of the best FAEs in the business. I am surprised that my friend said his buddy could not get help from an LT apps engineer. I suspect that is because the buddy was trying to get the FAE to do the whole design, and all and FAE can do is get you started in the right direction. Then again manybe the FAE was so good he just expected everybody would know how to compensate a power supply circuit.

Here is a picture of Bob Widlar’s compensation cheat box (click to enlarge). When Bob was playing with decompensated amplifiers or maybe when he had and amp built with kit parts prior to doing the design, he would use this tweak box pull the part into stability. Yeah Bob.