Test pattern characteristics
Pete Anslow, Nortel Networks
IEEE P802.3ba, New Orleans, January 2009
1
Introduction
It has been proposed to add Pattern 1 and Pattern 2 defined in Table
52-20 to the list of test patterns available in 40 and 100 GbE. (See
comments 562, 563). The current list of test patterns is:
•
•
•
•
Scrambled idle
PRBS31 per lane
TBD Short pattern (expected to be PRBS9)
Square wave - 8 ones followed by 8 zeros per lane
This contribution analyses the characteristics of Pattern 1 and Pattern 2
for baseline wander and variation in clock content and compares them
to PRBS31 and scrambled idle.
2
Baseline wander
In order to be able to compare the effect of various encoding proposals on the
low frequency and clock content characteristics of the signals, it is useful to
define a low frequency content metric (Baseline Wander) and also a clock
content metric (Clock Content).
Since it is highly likely that these signals will be AC coupled at some point, the
baseline wander has been analysed by calculating the amount of offset due to
AC coupling.
In order to make use of recent analysis done within OIF this is the same metric
as was used in the OIF white paper:
http://www.oiforum.com/public/documents/OIF_WP_CEI_Short_Stress_Patterns.pdf
This is defined as:
Baseline Wander is the instantaneous offset (in %) in the signal generated by
AC coupling at the bit rate / 10,000.
3
Baseline Wander of entire PRBS31 pattern
PRBS31
Baseline wander generated by AC coupling at bit rate / 10,000
14%
11.6%
12%
10%
8%
6%
4%
2%
0%
-2%
-4%
4
-2.9%
Baseline Wander of Pattern 1
Pattern 1
Baseline Wander generated by AC coupling at Bit Rate / 10,000
14%
12%
10%
8%
6%
4%
1.5%
2%
0%
-2%
-1.5%
-4%
5
Baseline Wander of Pattern 2
Pattern 2
Baseline Wander generated by AC coupling at Bit Rate / 10,000
14%
12%
10%
8%
6%
4%
1.7%
2%
0%
-2%
-1.7%
-4%
6
Baseline wander PDFs
1.E+00
10,000 year 64B/66B
random limit
Pattern 2
PRBS9
1.E-02
2^31-1 PRBS
1.E-04
Probability of occurence
10,000 year 64B/66B
random limit
Pattern 1
1.E-06
1.E-08
1.E-10
Idle 64B/66B
(red)
1.E-12
Idle 100GbE
Serial
(dashed
green)
Random data
(black)
1.E-14
1.E-16
1.E-18
1.E-20
Fit to
random
64B/66B
1.E-22
1.E-24
1.E-26
-12% -10%
-8%
-6%
-4%
-2%
0%
2%
4%
6%
8%
10%
Baseline wander generated by AC coupling at Bit Rate / 10,000
7
12%
Clock content
Since many of the links that 40 and 100 GbE traverse require the clock
to be extracted from the lane data at the receiver, a second metric to
assess the time variation of the clock content is required. The function
used in the OIF white paper (called “clock wander” there and re-named
to “clock content” in this contribution) is:
Create a function which is a 1 for a transition and a 0 for no transition
and then filter the resulting sequence with a corner frequency of
baud/1667 (6 MHz for 10.3125 GBd)
Filter with corner
frequency
baud/1667
Data
Stream
8
1 Bit
Delay
XOR
Output
Clock content of entire PRBS31 pattern
PRBS31
Clock Content
1
0.9
0.8
0.7
0.620
0.6
0.5
0.4
0.3
0.2
0.1
0
9
0.219
Clock content of Pattern 1
Pattern 1
Clock Content
1
0.9
0.8
0.7
0.577
0.6
0.5
0.4
0.3
0.2
0.1
0
10
0.368
Clock content of Pattern 2
Pattern 2
Clock Content
1
0.9
0.8
0.7
0.6
0.550
0.5
0.4
0.3
0.2
0.1
0
11
0.289
Clock content PDFs
10,000 year random
64B/66B limit
1
0.01
10,000 year random
64B/66B limit
Pattern 2
Probability of ocurrence
0.0001
2^31-1 PRBS
1E-06
Random
data
1E-08
1E-10
1E-12
Idle
64B/66B
Pattern 1
1E-14
Idle
100GbE
Serial
1E-16
PRBS9
1E-18
1E-20
Fit to
random
64B/66B
1E-22
1E-24
Fit to
random
data
1E-26
0.2
0.3
0.4
0.5
Clock content
12
0.6
0.7
0.8
Conclusions
Pattern 2 has slightly more Baseline Wander (BW) than Pattern 1.
However, even the BW of Pattern 2 is approximately a factor of 2 less
stressful than the 2 seconds of scrambled idle shown on the PDF chart
and factor of 6 less stressful than PRBS31.
The Clock Content (CC) of Pattern 2 is significantly more stressful than
Pattern 1 (0.289 Vs. 0.386), although the two seconds of scrambled idle
is slightly more stressful again. PRBS31 (with a minimum CC of 0.219)
is the most stressful pattern and the only one to exceed the two
seconds of scrambled idle.
For BW neither Pattern 1 or Pattern 2 are stressful enough to ensure
that real data or idle will work satisfactorily.
For CC Pattern 1 is not stressful enough. Scrambled idle is a better
pattern to use than Pattern 2 because while they are similar on the low
clock side, Pattern 2 does not contain realistic high clock portions.
PRBS31 is the only pattern providing overstress on the low clock side
compared to scrambled idle.
13
Thanks!
Pete Anslow,
Nortel Networks
With thanks to John Ewen for providing the Pattern 1 and Pattern 2 bit sequences
14