10 Gb/s MMF Transmissions over any (loss-limited ...

Sep. 12-14, 2000 UMBCIEEE802.3ae-Equalizer 1.

Fow-Sen Choa

Department of Computer Science and Electrical Engineering, The University of Maryland Baltimore County, Baltimore, MD 21250, USA 410-455-3272

(o), 410-455-6500 (fax), [email protected]

10 Gb/s MULTIMODE FIBER TRANSMISSIONS OVER ANY (LOSS-LIMTIED) DISTANCES USING

ADAPTIVE EQUALIZATION TECHNIQUES


10 Gb/s Ethernets qThe speed of Ethernets increases nearly 10 times for every 3 years It is 1Gb/s now and in the next 1-2 years, it will be focused on the 10Gb/s (meet with SONET)

qAmong the 5 objectives of 10Gb/s GE in the standards meeting (1. 100 m installed multimode fiber, 2. 300m on new multimode fiber, 3. 2 km on single mode fiber, 4. 10 km on single mode fiber, and 5. 40 km single mode fibers) the only one without solution is about the already installed multimode fibers(MMFs).

qThe main source of bit errors in a MMF transmission system is the inter-symbol interference (ISI) caused by the differential mode dispersion (DMD) generated multi-path effects


Multipath Delay Bin Outputs

0

5

10

15

20

25

85 86 87 88 89 90 91 92

Time(ns)

Am

plitu

de(m

V)

0

10

20

30

40

50

84 85 86 87 88 89 90 91

Time(ns)

Ampl

itude

(mV)

0

5

10

15

20

85 86 87 88 89 90 91 92

Time(ns)

Am

plitu

de(m

V)

A single pulse is transmitted through a 1.5 km long MMF at different launching offsets using a 1.55 um gain switched DFB laser. The fundamental mode is at around the time 87 ns. Other higher order modes are followed.


Approaches to Resolve The Multi-Path/ISI Problems

qMulti-fiber parallel transmissions 10x 1Gb/s.

qMulti-Wavelength WDM approaches 4x2.5Gb/s or 2x5 Gb/s.

qMulti-level modulation approaches.

qSubcarrier modulation approaches.

qSingle channel equalization approaches.


1

2

5

10

-40 -35 -30 -25 -20

DFB, 200m M MF, angle=0DFB, back-backDFB, 500m M MF, angle=0.003 °DFB, 500m M MF, angle=0DFB, 500m M MF, angle=0.006 °

Receiving Power(dBm)

Lo

g(B

ER

)

-10 0

-8 0

-6 0

-4 0

-2 0

0

1530 15 40 1550 1560 1 570

CH.1 CH.2 CH.3 CH.4

6.8nm8.4nm7.5nm

W avelength(nm)

Po

we

r(d

Bm

)

Examples (WDM Approaches)

SMF---MMF coupling

MMF---SMF coupling

LD-1

LD-4

LD-3

LD-2MMF

Filter RX

4X1

Attenuator

5

10

-35 -30 -25 -20

Ch.3, After 200m MM FCh.3, Back-to-BackCh.4, Back-to-BackCh.4, After 200m MM F

Power (dBm)

-log

(BE

R)


Equalization Principle

Timing and Decision circuits

τ1

τ2

Delay bins

DetectorPreAmp.

τ1W1

+

τ2W2

MMF

- -

Fundamental


Equalization Architecture

Multi-mode fiberchannelData sender

Clockrecovery

Integratingcircuit

cp

+

_

d(n)= r(n) t

w

e(n)

u w1 w2 wM

D

u(n)

u(n)

D Dt

w

y(n) in training,or rISI(n) in regular

transmission

u(n)

t

w : regular transmission

: trainingDecision level = D


Transmission Stages of the Adaptive Equalizer

nTraining Periods

— A training data stream is transmitted

— A replica of the training data is stored at the receiver end as the input

vector u(n) of the adaptive equalizer

— Desired value d(n) is the detected signal r(n) before decision

— Weight w(n) is updated adaptively (LMS) to approach the impulse

response h(t) of the MMF transmission channel

n Regular transmissions

—The equalizer works with the fixed weights w*(nT)

rISI(n) = w'*T(nT)û(n-1)

<−

>−=

Dnrnr

Dnrnrnu

)]()([ if ,0

)]()([ if ,1)(ˆ

ISI

ISI


Nature of the MMF channel

q Impulse response of the MMF channel

q Impulse response of the inverse system

q Length of filter approaches h(t)

Mmin = (τmax - τmin ) / T

The filter approaching h(t) is much shorter than the one approaching h’(t)

)()(0

∑=

+=N

kkk tAth τδ

=′−

=∑

1

0

21)(N

k

fjk

- keAth τπF


Statistics of the Input Tap Vector

n Correlation matrix (R = E{u(n)uT(n)})

Assuming i.i.d. input samples u (n)

n Condition number of the matrix (χχ(R))

Conditioning of R becomes worse as M increases

p

pp2

p2

p2 pp2

p2

p2

M x M

R =

ppM

−−+

=1

)1(1)(Rχ


Performance of the LMS Algorithm

n Maximum step size µµmax for LMS to converge

n µµmax becomes smaller and χχ(R) bigger as M increases

Slowing down the convergence of the LMS algorithm

n Mmin is proportional to the transmission distance z

Longer distance between communication hosts requires

more iterations for LMS algorithm to converge

Mp2

max =µ


Simulation Results: Comparison of Converged Filter weights

w2 has a simpler structure than w1 and can be represented by a few nonzero coefficients

0500 1000

15000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

h(t)

time (ps)0

20 40 60

80-0.5

0

0.5

1

1.5

2

weight index

w1(

n)1

05 10 15

20-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

weight index

w2(

n)


Simulation Results: Comparison of Learning Curves

System ID based approach converges much faster

0 1000 2000 3000 4000 5000 6000-35

-30

-25

-20

-15

-10

-5

0

iteration index (n)

Log

10(J

)

Standard EqualizerSystem ID


Simulation Results (I): Performance of the Proposed Equalization approach

Impulse response of the MMF channel and the converged filter weights

0 500 1000 1500 20000

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

h(t)

time (ps)0 5 10 15 20

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

weight index

w(n

)


Simulation Results (II): Performance of the Proposed Equalization approach

0 10 20 30 40 500

0.5

1

1.5

Received signal waves in NRZ formid

eal

0 10 20 30 40 500

0.5

1

1.5

with

out e

qual

izin

g

0 10 20 30 40 500

0.5

1

1.5

with

equ

aliz

ing

Units in bit


Simulation Results (III): Performance of the Proposed Equalization approach

0 1 2

0.2

0.4

0.6

0.81

No

equa

lizat

ion

Eye-pattern diagrams of the received signal

0 1 2x 10-10

00.2

0.4

0.6

0.81

time (second)

with

equ

aliz

atio

n

x 10-10


Experimental Setup

Transmitter

Receiver and Part of the Equalizer

Equalizer is composed ofMultilink Decision Ckt.,Delay lines, Amplifiers (providing phase reversal).


Experimental Results

(a)

(b)

(c)

a. Transmitted Pattern, b. Before equalization, c. After equalization

Since the delaypath (cables) are long, we send fixed data pattern and use signals in the current frame to cancel multipathcopies in the next frame.


Conclusions (I)

qWe have demonstrated using adaptive equalization techniques to overcome the signal degradation caused by differential modal dispersion in a conventional multimode fiber.

qWith this technique, we can not only obtain MMF-based 10 Gb/s GEs but also upgrade all the already installed MMFs (OC-3 backbones) to higher-speed pipes at nearly any (loss limited) lengths.


Conclusions (II)

qThe good news: The modal diffusion constant in MMF is small and once the initialization process is done no more adaptive processes are required in later transmission. No matter how we change the fiber temperature and stress if the launching condition is not changed, the excited modes will be very stable in the MMF. Ethernet protocol can be boosted after the initializationwithout any modification to accommodate the adaptive process, since there is no more "adaptive equalization“

10 Gb/s MMF Transmissions over any (loss-limited ...

Documents

Transcript of 10 Gb/s MMF Transmissions over any (loss-limited ...