When his card fails Wander Transfer, an engineer wonders how the unit
is going to meet spec when the stimulus exceeds spec
I joined Fujitsu in May of 97. I had 14 years of experience but the
majority of that was in defense electronics designing guidance systems.
I’d been working in telecom for just over a year and had just enough
telecom knowledge to get the job. I had 19 board designs under my belt
that included a wide variety of technologies--I was a Jack-of-all-trades. The one specialty I did have
was test and troubleshooting.
This skill took me into the second stage
of my career – test and integration. I spent five years in integration
and achieved a fair amount of success including the first error-free
system to roll off our production line. How I achieved that would
certainly be a worthy topic for this article, but I prefer to tell you
about another problem I solved, using the same techniques I applied to
produce that first error-free system.
I’d been at Fujitsu about 5 years when our group was converted from
designing line cards to optical line units. My first design was an
OC-48 Line Unit. It was a complex design that not only had the optics,
but also had the SONET pointer processor, the switch fabric, and the
system synchronization circuitry. The project was plagued with
problems. There were 2 huge ASICS and both ended up being respun. One
of them twice.
Enter the manager from the Underworld. He was the only non-American
manager in the hardware group, on assignment from the mothership back
in Japan. He had a reputation as one of their most talented and one of
their toughest. He was a typical workaholic. His wife nagged that he
worked too much so to keep her happy he would wake up at 4:00 AM, log
into the office and work from home until he had to drop his son off at
school. He’d work until 7:00 or 8:00 PM, whatever his wife would
tolerate. Not sure if it was his culture or his wife that made him so
tough on us.
The first year I worked on this OC-48 unit, he gave me a bad
performance review because of the ASIC problems. The ASIC group was
100% responsible for the ASIC and the problem wasn’t even design
related. It was a process problem and the ASIC vendor took
responsibility and did the respin for free. I got a bad review because
my manager said I should have anticipated the process problem and had a
backup plan. You’re kidding me, right? I was supposed to have a backup
plan for a 4 million gate ASIC that was the largest switch fabric
available at the time. After the review, I honestly thought this guy
expected me to pull my short sword and slit my belly right there in his
office. That might have been the easy way out as it would turn out.
Another year passed, another switch fabric ASIC respin and most of
my hair had turned gray by then. The ASIC vendor had finally gotten it
right and my card was in the final stages of qualification testing
(IQT). A problem surface during IQT testing –the synchronization
system. The card failed Wander Transfer. A quick tutorial on Wander
Transfer : Wander is just jitter, but very low frequency. Wander
transfer is the addition of wander when wander is injected into your
The synch circuitry that was used is also used on many Fujitsu
products, so I didn’t think it was the unit. My focus was on the
measurement system. We used the Tektronix SJ300 as our wander source.
The SJ300 can not measure OC-48 rate, so we used the SJ300’s DS1 output
as a reference clock for an OC-48 SONET Analyzer to produce a wandered
source. I measured the source and it did not meet the SONET specs and
neither did the DS1 output from the SJ300. So how was the unit going to
meet spec when the stimulus exceeded spec?. I presented my finding to
IQT and my manager. IQT was willing to pass the unit. My manager
disagreed. He simply told me to go fix it.
I felt there had to be a way to call out these measurement errors
and that’s just what I did. I set up a measurement system that measured
the stimulus as well as the unit under test. By definition, wander
transfer is a comparison between wander input wander and output wander.
After measuring both input and output, I entered the data into a
spreadsheet and set up a formula to subtract input wander from the
output and compare it to the wander pass/fail mask. As we suspected,
the synch circuit passed with flying colors. My card eventually passed
IQT and was released into production. Since then, many of these units
have been sold and reside in the labs of providers such as Verizon and
My measurement system idea was presented at our patent review board.
Fujitsu didn’t file on my idea because we really don’t produce
measurement equipment, but this method has become the preferred method
to measure wander transfer in our North American hardware development
labs. I guess the most important outcome was that my manager was
finally satisfied and for now my short sword and my belly are in their
Author Mark Berry is a Senior Manufacturing Test Engineer at Fujitsu
Network Communications. Mark has both MSEE and BSEE degrees from
Southern Methodist University. Mark holds two patents; 1) Protection
switching and 2) System and Method for Providing a Circuitry Management