A DYNAMIC MULTI-WAVELENGTH SIMULINK MODEL FOR EDFA

Stephen Pinter Jean Jiang Xavier Fernando

Ryerson University Ryerson University Ryerson University
[email protected] [email protected] [email protected]

Abstract is important to compensate for this non-uniform gain spec-

trum.
Erbium doped fiber amplifier (EDFA) is an imperative Currently, the simulation tools available to investigate
element in DWDM networks. This all-optical amplifier en- EDFA dynamics consist of OASIX and the photonic trans-
ables simultaneous amplification of multiple wavelengths. mission design suite (PTDS) [2]. These software programs
Nevertheless, uneven gain across the usable wavelength re- allow for simulations of EDFA models that are mostly static.
gion remains a concern. Understanding and modelling the However, the Simulink model developed in [2] provides for
dynamic characteristics of an EDFA is an important step to- a dynamic EDFA model with the ability to modify the input
wards achieving a flat gain spectrum. signal power and more importantly the input pump power.
A Simulink model for investigating EDFA dynamics has Also, Simulink is a well-known simulation tool that is less
been developed by Novak and Gieske (2002). Following specialized than PTDS and OASIX, therefore being readily
their work, we have developed an enhanced version of the available. Last year, the EDFA Simulink model produced in
EDFA Simulink model with more capabilities. Previous re- [2] was reproduced by Jean Jiang in [3].
sults were verified for the given doped fiber length with- In this project we reverse-engineered the model in [3].
out noise. In addition, the forward Amplified Spontaneous Jean Jiang provided her MATLAB and Simulink files as a
Emission (ASE) noise is added to the model and the op- starting point for the project. The MATLAB code and the
timum length is determined (with and without considering Simulink models were thoroughly analyzed and certain cor-
ASE) by simulation and used in all subsequent stages. A sig- rections made (which will be discussed later). The dynam-
nificant addition to the model is the ability to handle multi- ics of the corrected EDFA model were then verified. Ad-
ple channels using the absorption and emission coefficients ditions made to the EDFA model include: 1) determined
previously obtained experimentally. The resulting model ac- the optimum length to perform simulations, 2) expanded
curately represents EDFA gain dynamics and forward ASE. the model to show the various gains over the entire 1550
Simulation results show a 12 dB gain fluctuation across a nm window, and 3) integrated forward ASE into the EDFA
40 nm window per EDFA. model.

Keywords: EDFA Simulink model, multiple channels, gain

dynamics, ASE integration. 2. OVERVIEW OF EDFA EQUATIONS
An ordinary nonlinear differential equation for studying
EDFA gain dynamics has been derived by Sun et al. [2],
1. INTRODUCTION
and is shown below.
Optical amplifiers are interesting because they provide ∂
a method by which long distance communication over op- N2 = PS (0, t)[1 − eBS N2 −CS ]
∂t
tical fiber can be done. However, when transmitting over N2
long distances, signal attenuation can occur. In this case an + PP (0, t)[1 − eBP N2 −CP ] − (1)
τ
optical amplifier must be used to regenerate the signal. One
of the most widely used amplifiers for long-haul telecom- Equation (1) is the key equation for studying gain dynam-
munication applications are erbium doped fiber amplifiers ics in an EDFA and is implemented in Simulink as shown
(EDFA) [1]. A major concern with using EDFA’s in wave- in Fig. 1. The co-directional input pump power is PP (0, t)
length division multiplexed (WDM) systems is the non-uni- and the input signal power is PS (0, t). These input powers
form gain over the 1550 nm window (i.e. 1530 nm - 1560 are in photons/second and are related to the power in Watts
nm). Different wavelength signals experience different gains by P̃P,S = PP,S (hν), where ν is the frequency in Hertz and
and therefore experience a different signal to noise ratio. It h is Plank’s constant in units of J/Hz [4]. For our purposes

CCECE 2004 - CCGEI 2004, Niagara Falls, May/mai 2004

0-7803-8253-6/04/$17.00
2004
c IEEE
 Fig. 1. Simulink EDFA module for EDFA dynamics

40
where we must consider the wavelength, the equation is re-
written as P̃P,S = PP,S hc
λ , where c is the speed of light in
m/s and λ is the wavelength in m (note: in [3], this relation-
20

ship was incorrectly implemented and the results of the gain

0
simulations were inaccurate). The output pump and signal
powers are −20
Gain (dB)

BP N2 −CP
PP (L, t) = PP (0, t)e
PS (L, t) = PS (0, t)eBS N2 −CS
−40
(2)
In equation (2), quantities B and C characterize the physical −60

EDFA and are given by

−80
1520nm ≤ λs ≤ 1570nm
α+β
[BP , BS ] =
4.3429ρA −100
0 5 10 15 20 25 30 35 40 45 50

αL length (m)
[CP , CS ] = (3)
4.3429
In this equation, α is the absorption coefficient in dB/m,
Fig. 2. Dependance of gain on amplifier length in the range
β is the emission coefficient in dB/m and the scale factor
of wavelength’s indicated
1/4.3429 converts decibels to log base e. This is important
because the gain produced by the EDFA model is in base e
and must be multiplied by 4.3429 to convert it to decibels.
The multiple input channels are shown in Fig. 1 by the is proportional to the length of the fiber and had to be re-
numbers beside the block connecting lines. In this case calculated for each fiber length. Fig. 2 shows the gain as a
there are 26 signal wavelengths and 1 pump wavelength. function of amplifier length in the range of 0 m to 50 m. It is
The EDFA module in Fig. 1 is called from the main Simulink clear that the gain varies significantly over wavelength, and
model where the input signal power is -30 dBm and the in- that there are two distinct peaks, one around 12 m and the
put pump power is 17.8 dBm. other around 30 m. The wavelengths that reach maximum at
12 m on Fig. 2 correspond to the wavelengths on the emis-
sion and absorption spectra (Fig. 3) where the peak occurs.
2.1. Optimum Length
The effect of the emission/absorption peak is of key interest
Before performing simulations, the optimum length (Lopt ) so the optimum length is chosen as 12 m. This length is op-
of the EDFA model was determined. The parameter [CP , CS ] timal only in the sense that the signal gain of the amplifier
 38

36

34

32

30

Gain (dB)
28

26

24

22

20

18
1520 1525 1530 1535 1540 1545 1550 1555 1560 1565 1570

wavelength (nm)

Fig. 3. Emission and absorption cross-section coefficients Fig. 4. Gain versus wavelength for a length of 12 m, λP of
[6] 980 nm, pump power of 18 dBm and signal power of -30
dBm
is maximized. When ASE is included, the optimum length
is determined by maximizing the signal gain in the presence gain versus wavelength. Using a length of 12 m, the small
of ASE, which is also a function of the length [5]. The ASE signal gain over 1520 nm ≤ λS ≤ 1570 nm is plotted as
dependance on length will be discussed in section 4. shown in Fig. 4. An input signal power of -30 dBm is used
because a large signal would drive the EDFA into saturation
– causing the difference in the gains at different wavelengths
3. HANDLING MULTIPLE
to be small. The ratio between the absorption and emission
WAVELENGTHS cross-sections at a particular wavelength is critical in deter-
mining the gain for the amplifier [5]. The shape of the gain
Multiple signal wavelengths are handled by allowing
in Fig. 4 is highly dependant on the parameters α and β,
BS and CS in equation (3) to be multidimensional [2]. The
particularly the peak value. So, different variations in gain
parameters that determine the wavelength dependency of
can be produced when using different cross-sectional data.
BS and CS are α and β, the emission and absorption cross-
section coefficients, respectively. For this project, the emis-
sion and absorption coefficients used were those from Fig. 4. AMPLIFIED SPONTANEOUS
3. The input signal power PS (0, t) is also wavelength de- EMISSION
pendent as shown by the formula P̃S = PS hc λ . Therefore
PS (0, t) was also a multidimensional quantity and had to be The forward ASE is now considered for the EDFA model.
calculated for each wavelength. The forward ASE power is given in [5] by
It is important to note that simulating one wavelength
at a time will give different results than simulating all the PASE = 2nsp hν∆ν(G − 1) (4)
wavelengths simultaneously. EDFA’s are designed to work 1
where, nsp = (5)
in the nonlinear regime, so properties like linear superpo- 1 − βαPP αβSS
sition don’t hold. This is because when there are several
channels in an EDFA there is an effect called gain stealing. The ASE power is in Watts, G is the gain, ∆ν and ν refer to
How much of the energy each of the channels takes from the wavelength deviation of the ASE power around λ, h is
the pump depends on the details of the emission and ab- Plank’s constant, and nsp is the population-inversion factor
sorption spectra. The Simulink model in [2] uses only one which is dimensionless. In an EDFA, complete inversion
channel along with the pump. For this project the signal was can only be obtained when being pumped at 980 nm; at 980
expanded to 26 wavelength channels. This provided an ad- nm βP = 0 and therefore nsp = 1.
equate representation of gain versus wavelength, however, The ASE power versus length relationship is shown in
more channels can be used to get a more accurate represen- Fig. 5. In this figure it is clear that the ASE power builds up
tation. as the length of the fiber increases. This is an expected result
So, an important relationship for EDFA’s is found, i.e. because as spontaneously emitted photons travel down the
 −5
9

8
−7.5

−10
6

ASE Power (dBm)

ASE power (mW)

5
−12.5

−15
3

2
−17.5

−20
0 1520 1525 1530 1535 1540 1545 1550 1555 1560 1565 1570
0 2 4 6 8 10 12
wavelength (nm)
length (m)

Fig. 5. ASE power versus amplifier length for λS of 1530 Fig. 6. ASE power versus wavelength for λS of 1530 nm
nm and λP of 980 nm and λP of 980 nm for an amplifier length of 4 m

where numerous EDFA parameters can be changed and the

fiber they get amplified and they also stimulate the emission
effect on the output easily observed. A significant appli-
of more photons [5]. From Fig. 5 it is observed that the
cation is in the area of gain flattening where innovative ap-
ASE power is relatively small at lengths around 4 m. In this
proaches to gain flattening can be simulated/tested using our
case the optimum length is chosen to be 4 m compared to
model.
12 m obtained in section 2.1. However, at 4 m the EDFA
gain is reduced as shown in Fig. 2. Essentially, this is one
of the compromises that has to be considered when making Acknowledgements
a selection of amplifier length.
The output spectrum of the ASE is shown in Fig. 6 for The authors would like to thank S. Novak for useful dis-
a signal wavelength of 1530 nm. The ASE spectrum is very cussions regarding the Simulink model.
similar to that of the gain spectrum shown above in Fig. 4.
This is expected because of the relationship in equation (4).
Also, the ASE is present over the whole operational range References
of the EDFA, thereby reducing the overall gain of the sys-
[1] G. Keiser, Optical Fiber Communications, [Third Edition].
tem. The obvious difference between the ASE spectrum and
New York: McGraw-Hill, 2000, pp. 423-447.
the gain spectrum is the output power. In the EDFA usable
range of 1530 nm ≤ λS ≤ 1560 nm, the ASE spectrum [2] S. Novak and R. Gieske, “Simulink Model for EDFA Dynam-
varies from -5.95 dBm to -14.7 dBm. Clearly, ASE is a ics Applied to Gain Modulation,” Journal of Lightwave Tech-
nology, vol. 20, no. 6, pp. 986-992, June 2002.
dominant noise generated in the amplifier.
[3] J. Jiang, “Simulink Model for EDFA Dynamics in a WDM
System,” EE8114 Course Project, Department of Electrical
5. CONCLUSIONS and Computer Engineering, Ryerson University, 2002.
[4] S. Novak and A. Moesle, “Analytic Model for Gain Modu-
This paper presented a method for investigating EDFA lation in EDFAs,” Journal of Lightwave Technology, vol. 20,
dynamics using the tools MATLAB and Simulink. Using no. 6, pp. 975-985, June 2002.
the Simulink model implemented by Novak and Gieske [2], [5] P.C. Becker, N.A. Olsson, and J.R. Simpson, Erbium-Doped
we successfully improved the model to simulate gain spec- Fiber Amplifiers Fundamentals and Technology. San Diego:
trum and forward ASE. The optimum simulation length was Academic Press, 1999.
determined and the model in [2] was expanded to include [6] Ö. Mermer, Ege University, “EDFA Gain Flattening By Using
multiple wavelengths. Forward ASE was also added to the Optical Fiber Grating Techniques,” [course notes], October
model and the simulation results verified with [5]. The re- 2001, http://bornova.ege.edu.tr/∼wwweee/
sulting model accurately represented EDFA gain dynam- docs/seminer.pdf
ics and forward ASE; thereby providing a simulation tool