ieee-sa - working group - extension of statistical multimode fiber channel model...

www-g.eng.cam.ac.uk/photonic_commsPhotonic Communications Research

Extension of Statistical Multimode Fiber Channel Model to Electronic Dispersion

Compensation

J. D. Ingham, R. V. Penty and I. H. WhiteUniversity of Cambridge, Department of Engineering

Trumpington Street, Cambridge, CB2 1PZ, United [email protected]


Talk outline

• Brief review of statistical MMF channel model

• Extension of model to EDC• Model verification for EDC

• Types of EDC studied

• Link yield as function of • EDC type• Fiber launch• Link length

• Conclusions


Brief Review of Statistical MMF Model

see Cambridge channel modeling talk for detail


Multimode Fiber Numerical Modeling Input:Fibre profileProfile / position of launch beam

Calculate fibre modeprofiles /group indices

Calculate overlap integrals,impulse resp.

Output:bandwidths, CPRDMD, loss

0 10 20 30 40 501.47

1.48

1.49

radial distance / µm

refra

ctiv

e in

dex

Unoptimumregion

Optimised profile

0 10 20 30

22

26

30

High-order fibre modes

Mode group

Del

ay ti

me

(ns/

km)

0

200

400

600

1000

0 5

Ban

dwid

th /

(MH

z.km

)

ExperimentTheoryOFL-experimentOFL-theory

• Laser Launch conditions determine Mode Power Distribution (MPD) amongst fiber modes

Dispersive low-order fibre modes

1.5 800

10 15 20 25Offset / µm

• Fibre bandwidth determined by propagation characteristics and the distribution of power amongst fiber modes


Steps in the GbE statistical model

• Generate a set of representative fiber index profiles

• Calculate OFL frequency response and bandwidth

• Calculate impulse and frequency responses at beam offsets ranging across the entire fiber core radius

• Calculate DMD from assessment of the impulse responses at each offset

• Compare DMD to “worst-case” value, e.g. 5% of installed fibers have DMD > 2 ns/km for 62.5-µm MMF at 1300 nm

• Convert to set of “worst-case” fibers by scaling frequency responses according to ratio of DMD to “worst-case” DMD


Statistical MMF Bandwidth Simulations

• 81 fiber profiles chosen

• DMD scaled to 2ns/km for each profile (worst case 5% population)

• Bandwidths calculated using mode profile and group indices

• Statistical scaling allows yield calculation

M. Webster, L. Raddatz, I. H. White, D. G. Cunningham, “A statistical analysis of conditioned launch for Gigabit Ethernet links using multimode fiber,” Journal of Lightwave Technology, vol. 17, no. 9, pp. 1532-1541, 1999


Initial results for 10GbE EDC


• At GbE, the statistical model is used to extract MMF bandwidths.

• In an unequalized link, there is a strong inverse correlation between ISI penalty and bandwidth

• With EDC, knowledge of the bandwidth is not sufficient - the entire MMF link frequency response determines link performance

• Use calculated frequency response to determine link response to PRBS

• Noise can also be incorporated into the BER calculation determine BERs and penalties

• Easy to add different EDCs in the Rx (time or frequency domain)

Considerations relating to the extension of model to EDC

-10

-8

-6

-4

-2

0

0 1 2 3 4 5 6frequency, GHz

rela

tive

resp

onse

/ dB

EDC uses power in passband


Steps in the extended model

BER

MMF frequency responseextracted from GbE modeland scaled to desired linklength – includes chromaticdispersion

f

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09

0

2

4

6

8

10

12

14

16

18

x 10-4

time / ns

optic

al p

ower

/ W

0 0.010.020.030.040.050.060.070.080.09-0.50

0.51

1.5

time / ns

EDC Penalties

Laser output modeled byPRBS waveform which islow-pass filtered to ensureappropriate rise and fall times

PINTIA thermal noisemodeled in conjunctionwith low-pass filter.Other noise sources, e.g. RIN,accounted for at this point


Inclusion of Noise into System Model• Important consideration due to the potential for noise

enhancement effects in some EDC architectures• Noise sources considered: receiver thermal noise & RIN

0 50 100 150 200 250 3000

10

20

30

40

50

60

70

80

90

100850-nm 10-Gb/s MMF link without EDC

0 50 100 150 200 250 3000

10

20

30

40

50

60

70

80

90

100

link length / m

perc

enta

ge o

f wor

st-c

ase

links

w

ith p

enal

ty b

elow

4 d

B

850-nm 10-Gb/s MMF link without EDC

No RINNo EDC

RIN of -130 dB / Hz

• Additional penalties due to RIN found to be in line with 10 GbE spreadsheet calculations

link length / m


Types of EDC

Decision-feedback equalizer (DFE)Feedforward equalizer (FFE)For each of the two EDC types, we have considered:• Unconstrained-complexity equalization to determine upper bounds on performance • Constrained-complexity equalization to determine what may be realistically expected of an implementation.

• FFE is modeled by an N-tap transversal filter which tap coefficients according to the MMSE criterion. • DFE is modeled by an N-tap MMSE transversal filter followed by M bits of decision feedback


FFE bound with next-generation MMF

0 100 200 300 400 500 6000

1

2

3

4

5

6

7

8

910

link length / m

optic

al IS

I pow

er p

enal

ty /

dB

Conventional Receiver

FFE

• 10 Gb/s at 850 nm over 50-µm 2000 MHz.km MMF• IBM (OFC 2003): 7-tap transversal filter reduced ISI penalty by 8.8 dB

from 11 dB to 2.2 dB over 600 m of next-generation fiber• Simulations agree with IBM results

P. Pepeljugoski et al. “Improved performance of 10 Gb/s multimode fiber optic links using equalization,” Proc. OFC 2003, vol. 2, pp. 472-474


1300-nm 62.5 µm 500-MHz.km MMF – Model Parameter SetTransmitter Transmit pulse rise time 0.2 UI Laser center wavelength 1300 nm Laser RMS spectral width 0.1 nm Laser RIN -130 dB / Hz Fiber Launch Launch spot 7 µm FWHM Gaussian Offset launch conditions 20 µm +/- 3 µm Fiber Properties Fiber DMD for scaling process 2 ns/km Fiber chromatic dispersion 10 ps/(nm.km) Fiber length 30 m to 300 m Receiver Properties Photodetector responsivity 0.9 A/W TIA input-referred noise current 1.1 µA TIA bandwidth 7.5 GHz


50 100 150 200 250 3000

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

link length / m

optic

al IS

I pen

alty

/ dB

50 100 150 200 250 3000

1020304050

607080

90100

link length / mpe

rcen

tage

of w

orst

-cas

e lin

ks

with

ISI p

enal

ty b

elow

4 d

B Installed base yield / %

99

98

100

97

1300nm 10G Transmission Without EDC – Using 81 Fiber Model with 20 µm ± 3 µ m offset launch (ie 243 links)

• RH Graph shows pass% of “81 fibre” links with <4dB ISI penalty• Since these represent worst 5% of installed fibers, it is possible to extrapolate to installed fiber link failure for 4dB penalty• Installed fiber base yield (%) = (95+100-pass yield/20)• Valid for pass % > 50% (ie installed base yield >97.5%)


IEEE 802.3z MBI field test: yields for FFE and DFE bounds

50 100 150 200 250 3000

10

20

30

40

50

60

70

80

90

100

link length / m

perc

enta

ge o

f lin

ks w

ith IS

I pe

nalty

bel

ow th

resh

old


4dB

6dB

50 100 150 200 250 3000

10

20

30

40

50

60

70

80

90

100

perc

enta

ge o

f lin

ks w

ith IS

I pe

nalty

bel

ow th

resh

old

link length / m

4dB

6dB

DFE bound

50 100 150 200 250 3000

10

20

30

40

50

60

70

80

90

100

perc

enta

ge o

f lin

ks w

ith IS

I pe

nalty

bel

ow th

resh

old

link length / m

4dB

6dB

FFE bound

• ROFL (1.0dB)

• Calculated frequency responses scaled to lengths from 0 to 300 m

• Links failing the 500 MHz km specification are not included


1300-nm 62.5-µm 500-MHz.km MMF - Unequalized & FFE bound

• Optical ISI power penalty vs. link length


FFE bound

50 100 150 200 250 3000

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

link length / m

optic

al IS

I pen

alty

/ dB

50 100 150 200 250 3000

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

link length / m

optic

al IS

I pen

alty

/ dB


81 fiber statistical model: yields for FFE and DFE bounds

50 100 150 200 250 3000

10

20

30

40

50

60

70

80

90

100

link length / m

perc

enta

ge o

f wor

st-c

ase

links

with

IS

I pen

alty

bel

ow 4

dB

CONVENTIONAL RECEIVER

FFE BOUND

Installed base yield / %

99

98

100

50 100 150 200 250 3000

10

20

30

40

50

60

70

80

90

100

DFE BOUND


perc

enta

ge o

f wor

st-c

ase

links

with

IS

I pen

alty

bel

ow 4

dB

link length / m


99

98

100

• All launches offset launch and values quoted for 4dB power penalty

• Note simulations do not take error propagation in DFE into account


50 100 150 200 250 3000

10

20

30

40

50

60

70

80

90

100FFE BOUND

9-TAP FFE


perc

enta

ge o

f wor

st-c

ase

links

with

IS

I pen

alty

bel

ow 4

dB

link length / m

50 100 150 200 250 3000

10

20

30

40

50

60

70

80

90

100

DFE BOUND

(5,3)-TAP DFE

perc

enta

ge o

f wor

st-c

ase

links

with

IS

I pen

alty

bel

ow 4

dB

link length / m



99

98

100


99

98

100

81 fiber statistical model: yields for constrained FFE and DFE for 4dB penalty

• Constraining EDC implementation results in reduced achievable link length


DFE BOUND

Yields for constrained FFE and DFE for 6dB penalty

50 100 150 200 250 3000

10

20

30

40

50

60

70

80

90

100

link length / m


(5,3)-TAP DFE

perc

enta

ge o

f wor

st-c

ase

links

w

ith IS

I pen

alty

bel

ow 6

dB Installed base yield / %

100

99

98

50 100 150 200 250 3000

10

20

30

40

50

60

70

80

90

100

link length / m


FFE BOUND100

99

98

perc

enta

ge o

f wor

st-c

ase

links

w

ith IS

I pen

alty

bel

ow 6

dB Installed base yield / %

9-TAP FFE

• Allowing ISI penalty to rise to 6dB allows greater link lengths


EDC Results Summary

Yield at 300 m for 4dB ISI penalty (installed base yield)

Yield at 300 m for 6dB ISI penalty (installed base yield)

ROFL 1.0 dB MBI fibers

Unequalized 10 % 16 % FFE bound 55 % 74 % DFE bound 77 % 100 %

Offset Launch 81 fiber statistical

Unequalized 9% (-) 26% (-) FFE bound 62% (98.1%) 86% (99.3%)

N=9 tap FFE 52% (97.6%) 74% (98.7%) DFE bound 87% (99.3%) 100% (100%)

(N,M) = (5,3) tap DFE 62% (98.1%) 87% (99.4%)

• Note: MBI fibre link yield not representative of installed based yield


Modelling Caveats

• 2 ns/km DMD for 5% worst-case is now dated – should we revisit?

• Model does not include error propagation in EDC

• Known to be a problem, particularly in DFE

• Model does not take into account any noise statistic modification in the EDC chip

• May give rise to additional penalties

• Precise implementation of EDC will cause variations in the link yields predicted

• Should look upon the yield figures calculated to date as an upper bound


Conclusions• GbE statistical “worst-case” model reviewed and extended to

10G EDC, including noise effects• DMD scaling technique appears reasonable • General agreement with published experiments

• Extended model allows different EDC approaches to be easily investigated

• FFE and DFE implementations appear valid

• Preliminary indication of benefit of EDC for over 62.5-µm MMF at 1300 nm is achievable link length increase by a factor ~ 2 for 99% link yield

• Precise EDC yield curves will be implementation specific


Future work• Work with the Ethernet standards body to develop a

statistical fiber model for EDC-enabled 10GbE links

• Once the model is agreed, provide a set of corner-case impulse responses for input into simulations to enable estimation of theoretical performance of different classes of EDC

• Help out in the development of a “worst-case” spreadsheet based on 10GbE spreadsheet and power-budget model for EDC-enabled links

Acknowledgment• Cambridge would like to acknowledge the advice / input of

Phyworks Ltd


References

• M. Webster, L. Raddatz, I. H. White, D. G. Cunningham, “A statistical analysis of conditioned launch for Gigabit Ethernet links using multimode fiber,” Journal of Lightwave Technology, vol. 17, no. 9, pp. 1532-1541, 1999.

• D. G. Cunningham, W. G. Lane, Gigabit Ethernet Networking. Indianapolis, IN: Macmillan Technical Publishing, 1999.

• P. Pepeljugoski, J. Schaub, J. Tierno, J. Kash, S. Gowda, B. Wilson, H. Wu, A. Hajimiri, “Improved performance of 10 Gb/s multimode fiber optic links using equalization,” Proc. OFC 2003, vol. 2, pp. 472-474.

• E. A. Lee, D. G. Messerschmitt, Digital Communication, second edition. Norwell, MA: Kluwer Academic Publishers, 1994.

• S. Quershi, “Adaptive equalization,” IEEE Communications Magazine, March 1982, pp. 9-16.

ieee-sa - working group - extension of statistical multimode fiber channel model...

Documents