The Adventure of the Singing Chassis

Dwight Bues, Contributing Writer

During my career as a Hardware Engineer in the 1990s, a former employer decided to produce a rack-mounted, fan-cooled, 21-slot VME chassis with a front-mounted, touch panel display. The engine to power this beast was a 1,500W WESTCOR power supply providing the following voltages: +5V @ 150A, +/-15V @ 15A, and +24V @ 10A to power the fans.

The power supply was modular, so that modules could be strapped together to provide more current. Five modules were strapped together to provide the +5V @ 150A. Four of the five were connected with a bus bar, but the fifth module was jumpered with 10 AWG bus wire to the others - this particular asymmetry may have been a contributing factor to the instability we encountered.

One AWG welding cable was used to connect the power supply to the backplane bus bars. Positive and negative cables were between 12 and 15 inches long. Twisted pair 20 AWG wire ran from the power supply sense lead input about 20 to 25 inches to the backplane bus bars (the length of these wires was due to wire-harnessing convenience).

The first two programs (of about 40 in all) that used this chassis were successful. But a follow-on program used new CPU cards that had a much greater CMOS content. CMOS ICs generally have lower quiescent current, but the overall supply currents can be greater and be more dynamic.
The chassis, when fully populated, would “sing” at about 12-15 kHz. At first, we thought it was just annoying. When we blew the first CPU card, we realized it was more than just an annoyance. The AC ripple on the +5V bus was actually exceeding the VME spec by about a volt. The “death toll” on this program was about 8 CPU cards ($8K each) and 12 power supplies ($800 each) for a total of $150-200K, including rework labor.

We traced the problem to a revision in which the Mechanical Designer had moved the studs to the opposite end of the backplane bus bars, adding about 10 inches to each supply cable. That change, along with the presence of additional CMOS CPU cards, meant the singing was no longer merely a random occurrence, but always present.

The only practical solution that presented itself, was to shorten the circulating current “loop” by terminating the sense leads right on the power supply output. This removed about 30 inches from the loop. A voltmeter verified only a 0.1V drop from the power supply output to the backplane slot with the shortened sense leads, well within the VME spec.

This was an expensive lesson to learn, but that’s why an Electrical Engineer needs to carefully review the Mechanical Design of any system.

About the Author: Dwight is a Georgia Tech Computer Engineer with 27 years experience in Computer Hardware, Software, and Systems and Interface Design.

Get Published in Sherlock Ohms!
Have you applied your deductive reasoning and technical prowess to troubleshoot and solve an engineering mystery that even the fictional Sherlock would find most perplexing? Tell us about it in 600 words and we’ll pay you $100 if we publish your case.
Email Karen Field at: [email protected] .