comparison of nrz, pr-2, and pr-4 signaling
TRANSCRIPT
Comparison of NRZ, PR-2, and PR-4 signaling
Presented by: Rob Brink
Contributors: Pervez Aziz
Qasim Chaudry
Adam Healey
Greg Sheets
IEEE P802.3ap Task Force2 March 15, 2005 (r1.0)
Scope and Purpose
� Operation over electrical backplanes at 10.3125Gb/s is investigated using NRZ, PR2, and PR4 signaling.
� A common equalizer architecture is used in all cases.
� Estimated BER, as well as voltage and timing margin at 1E-12, is reported.
IEEE P802.3ap Task Force3 March 15, 2005 (r1.0)
Agenda
� Simulator Overview
• Link Model
• Transmitter Model
• Receiver Model
• Equalization Strategy
� Test Cases
� Sample Results
� Results Summary
IEEE P802.3ap Task Force5 March 15, 2005 (r1.0)
Link Model
RXPackage
2
4
50
50
1
3
TXPackage
1
3
2
4
50
50Backplane
2
4
1
3
TP1 TP4
TX Package Model
Backplane Model
RX Package Model
Mellitz “Capacitor-Like” Model
As described in “Test Cases” section…
Mellitz “Capacitor-Like” Model
IEEE P802.3ap Task Force6 March 15, 2005 (r1.0)
Crosstalk (1/2)
� For each crosstalk aggressor…
• The response to a PRBS-15 pattern (with an additional trailing “0”) is computed.
• This response is sampled at baud-spaced intervals at 16 offsets from 0 to (15/16)T in T/16 steps.
• At each offset, the amplitude distribution of sampled response is computed.
• The aggressor amplitude distribution is defined as the average of the amplitude distributions computed at each sample offset.
IEEE P802.3ap Task Force7 March 15, 2005 (r1.0)
Crosstalk (2/2)
� The overall crosstalk distribution is defined as the convolution of the individual aggressor distributions.
� The effect of crosstalk on the eye is modeled as the convolution of the overall crosstalk distribution and amplitude distribution of the “thru” path at each sample phase.
� This methodology is has been previously described by Moore:
• http://ieee802.org/3/ap/public/channel_adhoc/moore_c1_0305.pdf
� Computed RMS and peak-peak crosstalk will be reported.
IEEE P802.3ap Task Force8 March 15, 2005 (r1.0)
Transmitter Model
� Transmitter differential output voltage fixed at 800mVp-p.
� Transmit filter is Gaussian.• Rise Time (20-80%): 24 ps
� Transmitter output jitter• Duty Cycle Distortion: 0.05 UIp-p (even-odd)
• Deterministic Jitter: 0.10 UIp-p (sinusoidal)
• Random Jitter: 0.15 UIp-p (at 1E-15), 9.4mUIrms
� Parameters defined at package model input and do notinclude package parasitics.
• The impact of the package model is investigated in the next slide.
IEEE P802.3ap Task Force9 March 15, 2005 (r1.0)
Transmitter Output at TP1
TXPackage
1
3
2
4
50
50
TP1
50
50
800 mV
p-p800 m
Vp-p
0.70 UIp-p at 1E-15
IEEE P802.3ap Task Force10 March 15, 2005 (r1.0)
Receiver Model
� Receiver modeled as a single pole at 75% fbaud.
• Noise Bandwidth (Bn): 11.4 GHz
• Noise Figure: 18 dB
• sqrt( 4kTRBn ): 1.08 mVrms
� Receiver jitter:
• Random Jitter: 0.15 UIp-p (at 1E-15), 9.4 mUIrms
� No gain stages have been included in the receive path.
�∞
=0
2)( dffHBn
IEEE P802.3ap Task Force11 March 15, 2005 (r1.0)
Equalization Strategy
� Transmitter Finite Impulse Response filter.
• 3 taps, T-spaced with “infinite” tap weight resolution.
� Receiver Decision Feedback Equalizer
• 5 taps, unconstrained with “infinite” tap weight resolution.
� Sequential adaptation
• Transmit FIR is adapted first, then the DFE.
� Sample phase chosen to minimize mean-squared error.
• T/32 resolution
IEEE P802.3ap Task Force13 March 15, 2005 (r1.0)
Test Patterns
� Equalizer trained with PN-11 pattern with a trailing “0”.
� Equalizer settings are then frozen.
� Voltage and timing margin is estimated based on PN-15 pattern with a trailing “0”.
• Thru, NEXT, and FEXT channels share the same output amplitude (800 mVppd) and transmit FIR settings.
� Decision threshold set at the mid-point between nominal signal levels, as determined by the 1E-4 contour.
• Reported margins are twice the minimum distance from the sample phase (or threshold) to the BER contour of interest.
IEEE P802.3ap Task Force14 March 15, 2005 (r1.0)
Channels Studied
� All Tyco Electronics channels
• Test Cases 1 through 7
• Note that the FEXT channels were included twice due to connector symmetry.
� Recommended subset of Intel channels
• Test Cases 8, 11, and 12 map to T1, T12, T20
• Test Cases 14, 17, and 18 map to B1, B12, B20
• Test Cases 20 and 24 map to M1, and M20
� Other channels not simulated due to a lack of time…
IEEE P802.3ap Task Force16 March 15, 2005 (r1.0)
Disclaimer
� The following results are based on PRBS-11 pattern and are included for illustrative purposes.
IEEE P802.3ap Task Force17 March 15, 2005 (r1.0)
NRZ Sample Results: Eye at Slicer InputTe
st C
ase
#1Te
st C
ase
#6
IEEE P802.3ap Task Force18 March 15, 2005 (r1.0)
PR2 Sample Results: Eye at Slicer InputTe
st C
ase
#1Te
st C
ase
#6
IEEE P802.3ap Task Force19 March 15, 2005 (r1.0)
PR4 Sample Results: Eye at Slicer InputTe
st C
ase
#1Te
st C
ase
#6
IEEE P802.3ap Task Force21 March 15, 2005 (r1.0)
1.0E-361.0E-341.0E-321.0E-301.0E-281.0E-261.0E-241.0E-221.0E-201.0E-181.0E-161.0E-141.0E-121.0E-101.0E-081.0E-061.0E-041.0E-021.0E+00
0 5 10 15 20 25
Test Case
BE
R
NRZ
PR2
PR4
BER Estimates
IEEE P802.3ap Task Force22 March 15, 2005 (r1.0)
Voltage and Timing Margin at 1E-12
Vertical Margin at 1E-12
0.000
0.005
0.010
0.015
0.020
0.025
0.030
0.035
0.040
0.045
0.050
0 5 10 15 20 25
Test Case
Mar
gin
(V
p-p)
NRZ
PR2
Horizontal Margin at 1E-12
0.000
0.050
0.100
0.150
0.200
0.250
0.300
0 5 10 15 20 25
Test Case
Mar
gin
(U
I p-p)
NRZ
PR2
IEEE P802.3ap Task Force23 March 15, 2005 (r1.0)
Crosstalk Environment
RMS and Peak-Peak (at 1E-12) Crosstalk
0.0000
0.0200
0.0400
0.0600
0.0800
0.1000
0.1200
0.1400
0 5 10 15 20 25
Test Case
Pea
k-P
eak
Cro
ssta
lk (
V)
0.0000
0.0020
0.0040
0.0060
0.0080
0.0100
0.0120
RM
S C
ross
talk
(V
) NRZ (p-p)
PR2 (p-p)
PR4 (p-p)
NRZ (rms)
PR2 (rms)
PR4 (rms)
IEEE P802.3ap Task Force24 March 15, 2005 (r1.0)
Required FFE Boost
FFE Gain at (Symbol Rate)/2
0.0
5.0
10.0
15.0
20.0
25.0
30.0
0 5 10 15 20 25
Test Case
Gai
n (
dB
) NRZ
PR2
PR4
IEEE P802.3ap Task Force25 March 15, 2005 (r1.0)
Conclusions (1/2)
� The target response for PR4 is a poor fit to the channel and therefore higher equalizer complexity is required to achieve acceptable performance.
� NRZ and PR2 both support 1E-12 operation over the Tyco channels.
• In general, PR2 requires considerably less boost to achieve this objective.
• In the majority of test cases studied, NRZ offered superior voltage and timing margin.
IEEE P802.3ap Task Force26 March 15, 2005 (r1.0)
Conclusions (2/2)
� The Intel “T” channels were not supported by any of the signaling schemes studied with the chosen equalizer architecture.
� NRZ signaling may be feasible for select Intel “B” and “M” channels.
• Crosstalk is a significant impairment on these channels.