The original designer of this power supply should have been promoted into management before he/she could design any more disasters.
Inheriting someone else's bad design can sometimes be a real Frankenstein's nightmare.
My bad dream was a switching power supply for a legacy telecom shelf that provided +12V, +5V, and -5V onto a backplane. It was not our design; our Mighty Management had purchased the product line from a defunct manufacturer and expected us to continue support.
From the minute I saw this piece of garbage, I concluded that the designer should have been promoted into management before he/she could design any more disasters.
Units kept coming in from the field with burned PCBs from overheating zener diodes. These reverse-biased diodes were paired in series directly across the +12V rail, perhaps in an attempt to clamp surges. Not a bad idea, but their 5 percent tolerance range included the nominal operating voltage! It was a no-brainer to ECO a change to a slightly higher voltage rating that excluded the operating voltage over all tolerances.
But that was not the worst of it: The regulation was on the +5V only and was very unstable. The test fixture for these power supply cards had individually switchable loads that all had to be turned on together for light loading or full loading. Any combination of light load vs full load would cause stability problems. For example, when the 12V was fully loaded and the +5V left lightly loaded, the thing would "motorboat," which means the output voltages would rise and fall at a few tens of Hz. Unacceptable.
It appeared that the designer had assumed that the shelf loading was constant among the three voltages based on other cards loading each supply equally. This was not really the case, some cards loaded the -5V more than others, some the 12V more than others, so it was not possible to predict the actual load on each voltage in a randomly populated shelf.
I did not have much experience with stabilizing power supply feedback loops, but I did get enough information from on-line app notes that the existing control loop did not allow for the ESR of the output capacitors (4 in parallel). Changing the feedback compensation stabilized the operation, no more motorboating on the test fixture under any combination of no load, light load, or full load on the three voltages. Further tests with a switchable electronic load again under all load combinations showed good transient stability in all cases.
I have since left that... uhm, company, but I do keep in touch with the production manager. He tells me that not a single one of the improved power supplies has ever come back from the field.
The Frankenstein's Fix has just come to an end. Stay tuned to read the submissions and see what kind of difficult job of judging we have ahead of us! Submission details and full contest rules here.