Vote No on PRBS31! (and the trouble with journalism these days)
Ransom Stephens- November 6, 2012After posting my last rant, “PRBS31: Slower, costlier, worse,” I decided that I should heed the call of journalism and seek out the other side to the argument. I set out across that land in search of PRBS31 test pattern defenders (i.e., I sent four emails, each one to someone in the Test & Measurement industry).
I posed the question in a fair and balanced way (“What idiots would use PRBS31 when it's so damn long?) and, after my death-defying trek from the Great Northwest, through Silicon Valley, up (and down) the Rocky Mountains, and across the Chestnut Ridge of New York, (i.e., after waiting a couple of days for their replies), I got my answers.
Despite my comprehensive investigation, I couldn’t find a single person who would defend the PRBS31 test pattern. (It should be noted that, with a sample of four, the statistical uncertainty in that statement is 50% and, since I only asked people in the T&M industry, the sample has a 100% bias, making the total uncertainty 112% = sqrt[100^2 + 50^2] (you can’t get this kind of analysis from MSNBS or Faux News)).
Objective visionary from LeCroy (I’m not willing to attach “Teledyne” to LeCroy until I get my complementary Waterpik), Dr. Alan Blankman said, “We are in complete agreement with you regarding the unwieldy nature of PRBS31.”
In a clandestine meeting in a parking lot near The Garden of the Gods, Rick Eads of Agilent Technologies whispered, “Engineers want to test with patterns that are close to what their random traffic looks like. This makes intuitive sense to them. Nevertheless, it is my belief that they do not fully comprehend the practical concerns of how the algorithms work that are used by the T&M industry to evaluate and to decompose jitter.”
That’s all they said. No one could find a reason to defend the PRBS31. No one.
Okay, I’m lying.
(Don’t pretend like you didn’t see this coming, it’s politics season. Wait, admitting a lie is impolitic? That explains my humiliating defeat in the Petaluma City Council election of ’06 – didn’t get a single vote. I didn’t even vote for me.)
In the even-handed tradition of engineering, the experts did in fact offer explanations for the use of PRBS31.
Defender of justice in the beaver-damming northwest, Pavel Zivny from Tektronix pointed out that “PRBS31 covers all of the events we care about. If we have a system that fails on a run-length longer than 31 UI, we don’t care because there is no pain in re-sending one packet a week.” He concluded that “PRBS31 is the complete test with 0 assumptions.”
Rick Eads added that custom patterns, like OIF-CEI’s Jitter Tolerance Test Pattern, “require memory to be allocated in the IC to hold the pattern. This adds cost and vendors are thus reluctant to do it. LFSRs [linear feedback shift registers], on the other hand, are cheap to implement and thus lend themselves to PRBS type patterns.”
By virtue of volume, it makes cents to pay more for test equipment – in both capital costs and test engineer sweat – than to increase component cost. Since the PRBS31 is generated by a simple hardware shift-register it costs little and enables BIST (built in system test). The alternative that I endorsed (and still endorse!), OIF-CEI’s “CID Jitter Tolerance Pattern” would take up memory and resources.
Despite the truthiness of these “facts,” I insist that there is a better candidate for all-purpose test pattern than PRBS420!
Linear shift register (from the Ransom's Notes "Art" studio)
Surely you (or maybe the engineer in the cube next to yours (yeah, get her/him to do it)) can derive a polynomial for a shift register that generates a much shorter pattern than the PRBS31 and has all of its wonderful features.