ieee 802.3ap proposal to use pr-4 signaling for 10gbase-kr...

Michael AltmannIEEE 802.3ap Task Force - 21-Jan-05

IEEE 802.3ap

Proposal to use PR-4 Signaling

for 10Gbase-KR links

IEEE 802.3ap Task ForceJan’05

Michael Altmann, Fulvio Spagna, Jun Wang


Slide 2

Objectives

� Propose a PR-4 based signaling solution for 10Gbase-KR signaling on backplanes

� Allows leveraging signaling techniques & experience from the storage industry

� PR4 signaling can acceptably equalize typical BP channels� Simulation results will be shown


Slide 3

Agenda

� Layer Model� Proposal Overview� Link Model� PR-4 Signaling� AN and Link training� PR-4 simulation results

� Adaption simulations

� MMSE-optimized eye diagrams


Slide 4

IEEE802.3ap Layer Model


Slide 5

Overview

� Use PR-4 signaling for 10Gb serial BP channels� PR-4 has spectrum & implementation advantages

� Requires pre-coding to limit error propagation

� Use existing 802.3 clauses for other functions:� PCS (clause 49)

� PMA (clause 51)

� MDIO (clause 45)

� Auto-Negotiation (clause 28)� Possible signaling modifications as in brink_01_1104.pdf or

thaler_01_1104.pdf)

� Programmable Tx and Rx equalization optimized with training


Slide 6

PR-4 Signaling - Background

� Used extensively in disk drive industry� Equalized systems use non-unity target polynomial

0 0.5 1 1.5 2 2.5 3

x 1010

0

10

20

30

40

50

60

70

Line code=nrz

Freq [Hz]

Spe

ctra

l Pow

er [d

bmW

/Hz]

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x 10-10

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2Filtered Eye Diag

Time [s]

[V]

0 0.5 1 1.5 2 2.5 3

x 1010

0

10

20

30

40

50

60

70

Line code=pr4

Freq [Hz]

Spe

ctra

l Pow

er [d

bmW

/Hz]

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x 10-10

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2Filtered Eye Diag

Time [s]

[V]

PR4: Y=(1-D2)XNRZ: Y=X


Slide 7

PR-4 Signaling - Background

� Multi-level encoding� 2-level NRZ becomes 3-level PR-4 encoded data at Rx� Adds coding redundancy which alters the spectrum

� Nulls at DC and fbit/2� True Nyquist channel possible for minimum noise� Encoding mechanism is combination of Tx Eq, channel, Rx

AFE and Eq

� ‘Allowable’ ISI reduces equalization boost from NRZ� See ‘altmann_02_0904.pdf’

� Zero crossing between +1, -1 pulses is an excellent timing indicator� Enables baud-rate Rx design� Reduces sampler complexity by 50%

� PR-4 requirements� Requires both Tx and Rx equalization

� Tx pre-coding to limit error propagation


Slide 8

PR-4 Signaling

� Data encoding requirements� Tx pre-coding not explicitly required, however it eliminates error

prpoagation from incorrectly decoded bits

� Data Decoding options� Can use bit-by-bit or optional MLSE (ex. Viterbi) detection

� Receiver can operate at half-rate, as data stream is intrinsically interleaved

� Both decoding methods requires a pre-coder to eliminate error propagation.

� PR-4 Pre-coder


Slide 9

PR-4 Signaling vs NRZ

� Controlled ISI creates 3-level eye diagram� In BW-limited channels, eye height reduces slower than NRZ� Jitter is higher due to direct +1 →→→→ -1 transitions

0 20 40 60 80 100 120 140 160 180 200-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5Data & clock response: f0=7.5GHz

Filtered NRZ Eye: ωωωω0=7.5GHz Filtered PR-4 Eye: ωωωω0=7.5GHz

0 20 40 60 80 100 120 140 160 180 200-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5Data pattern & clock response: fr=750m


Slide 10

Link Model (from brink_01_1104.pdf)


Slide 11

Auto-Negotiation & Eq Training

� AN and link training required� AN will exchange PHY capabilities and determine bit-rate

� Follow method similar to thaler_01_1104.pdf or brink_01_1104.pdf

� Timing Acquisition� Initial timing acquisition using ¼-rate signal: +1 +1 -1 -1

(approx a clk at PR4 spectral peak) before equalization� Rx signal is approximate sine wave� Final timing acquisition using simple (TBD) known data

patterns

� Equalizer training� Initial equalizer training can occur using (TBD) training

sequences (altmann_02_1105.pdf)� Optimal (TBD) training sequence will be defined for faster

PR4 adaption

� After link training, equalizers continuously update with active traffic


Slide 12

PR-4 Equalizer Adaption Simulations

� Simulated results for channels equalized to PR-4 target polynomial

� Initial Eq estimate provided from channel response� Substitute for training sequence adaption

� Simulation conditions� Data Pattern: PRBS15, 100kbit sim time

� NEXT/FEXT: off

� Tx Pulse shaping: 4th Order Bessel Filtering

� Rx Noise Filtering: None

� Tx Eq: 8 tap FIR, 10b coeffs

� Rx Eq: 5 tap DFE, 10b coeffs


Slide 13

PR-4 Equalization Simulations – Tyco Case 5

0 500 1000-2

-1

0

1

2Tx EQ coeffs

0 500 1000-0.5

0

0.5

1

1.5Rx EQ coeffs

0 500 1000-0.06

-0.04

-0.02

0Normalized Sampling Phase

0 500 1000

0.5

0

0.5

Sampler Eye Diagram

0 500 1000-0.5

0

0.5Sampler Error

-2 0 2 4 6-1

0

1

2Tx EQ final coeffs

0 2 4 6-0.5

0

0.5

1Rx DFE final coeffs

Channel model: tyco: Case5_DS_13_10_T_D13_L6


Slide 14

PR-4 Adaptive Eqn Sims – Intel T32

0 500 1000-5

0

5Tx EQ coeffs

0 500 1000-1

-0.5

0

0.5Rx EQ coeffs

0 500 100-0.1

0

0.1Normalized Sampling Phase

0 500 1000

0.5

0

0.5

Sampler Eye Diagram

0 500 1000-0.5

0

0.5Sampler Error

-2 0 2 4 6-5

0

5Tx EQ final coeffs

0 2 4 6-1

-0.5

0

0.5Rx DFE final coeffs

Channel model: Intel T32


Slide 15

PR-4 Equalization Simulations – Eye Diagrams

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

-1

-0.5

0

0.5

1

X: 0.7188Y: 0.2558

X: 0.5045Y: 0.4376

X: 0.3007Y: 0.27

X: 0.5056Y: 0.08217

Tyco Case5 PR4 equalized Eye

Time [UI]

Sam

pler

Vol

tage

[V]

'-1' Lower'-1' Upper'0' Lower'0' Upper'+1' Lower'+1' Upper

Data: PRBS15Tx Eq: 7T FIRRx Eq: 5T DFETR: Data-Directed Sign-Sign

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

-1

-0.5

0

0.5

1

X: 0.3468Y: 0.2386

X: 0.501Y: 0.3983

X: 0.6518Y: 0.26

X: 0.5053Y: 0.1146

Intel T32 PR4 equalized Eye

Time [UI]

Data:PRBS15Tx Eq: 7T FIRRx Eq: 5T DFETR: Data-Directed Sign-Sign

0.355

0.205

Eye Height

[V]

0.691 0.418 Tyco Case #5

0.205 0.154Intel T32

Max NEXT/FEXT

[Vp-p]

Eye width[UI]

Channel

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-0.02

-0.015

-0.01

-0.005

0

0.005

0.01

0.015

0.02X: 0.8438Y: 0.01975

Worst-Case NEXT/FEXT at sampler vs phase

Time [UI]

Sam

pler

Vol

tage

[V]

X: 0.4688Y: -0.01872


Slide 16

PR-4 Simulation Summary

� Simulation conditions:� Noise filter: 2nd order linear filter ωωωω0=6GHz & 10GHz

� NEXT/FEXT & noise: None

� Fixed coefficients, post-MMSE adaption

� Linear tap values (no quantization)


Slide 17


Intel B12 Intel M12 Intel T12 Intel B8


Slide 18


Intel M8 Tyco Case #1 Tyco Case #5


Slide 19

PR-4 Simulation Summary – Eye Parameters

0.506 0.315 Case #1

Intel Channels (Peters)

0.691 0.375 Case #2

Tyco Channels (D’Ambrosia)

0.288 0.261 Intel M8

0.343 0.264 Intel B8

0.205 0.154Intel T12

0.341 0.277Intel M12

0.461 0.314Intel B12

Eye Height[V]

Eye width[UI]

Channel


Slide 20

Summary

� Presented a PR-4 signaling proposal for 10Gbase-KR� Leverages existing clauses for PCA, PMA and MDIO� Leverages known PR-4 implementation advantages� Includes Tx (FIR) and Rx (DFE) adaptive equalization� Successful adaption simulations shown� Successfully equalizes a range of Tyco and Intel

(ATCA) backplanes

ieee 802.3ap proposal to use pr-4 signaling for 10gbase-kr...

Documents