Eindhoven University of Technology
MASTER
Automatic gain control in the STOLAS network
Bastiaans, B.L.G.
Award date:
2003
Link to publication
Disclaimer
This document contains a student thesis (bachelor's or master's), as authored by a student at Eindhoven University of Technology. Student theses are made available in the TU/e repository upon obtaining the required degree. The grade received is not published on the document as presented in the repository. The required complexity or quality of research of student theses may vary by program, and the required minimum study period may vary in duration.
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
---ruLe tethnisthe universiteit eindhoven
Automatic gain control in the STOLAS network
Bart Bastiaans
Automatic gain control in the STOLAS network
Bart Bastiaans
Master of Science Thesis
12 August 2002 - 12 May 2003
Final Version Tuesday 20 May 2003
Supervisors:
dr. ir. J.G.L. Jennen
o
Lucent Technologies Nederland Bell LaboratoriesAdvanced Technologies
dr. ir. H. de Waardt
prof. ir. G.D. Khoe
TU/e
Eindhoven University of TechnologyFaculty of Electrical Engineering Electro-Optical Telecommunications
The Faculty of Electrical Engineering of Eindhoven University of Technology disclaims all responsibility for the contents of traineeship and graduation reports.
Preface
This thesis describes the graduation project of my Electrical Engineering study at the Eindhoven University of Technology (TUE). I performed this project within the Bell Labs Advanced Technologies Group of Lucent Technologies in Hilversum.
This thesis gives a description of the design process of an automatic gain control that can be used within the STOLAS project.
Readers who are interested in using the automatic gain control in combination with the Alcatel wavelength converter can find guidelines on tuning the system in section 5.2, page 35. To clarify the used symbols and acronyms, a list of symbols and acronyms is added at page 43 and page 44.
I would like to thank my supervisors, dr.ir. J.G.L. Jennen and dr.ir. H. de Waardt for guiding me through this graduation project and the colleaguesofthe Bell Labs Advanced Technologies Group for their support and coffee.
I am very grateful to my parents for giving me the opportunity to study and always supporting me as much as possible. Last, but not least I would like to thank my girlfriend, Caroline, for the numerous discussions on structuring the project and encouraging me to always pose a higher goal.
Abstract
This work is part of the STOLAS project. The STOLAS project aims to develop an all-optical packet- switched network. The wavelength converter in this project has a small dynamic range of the average optical input power where adequate conversion is obtained. This thesis concentrates on the design process of a gain control system that increases the dynamic range ofthe wavelength converter.
The design process is described based on literature, simulations and characterisation of the automatic gain control system.
The automatic gain control uses two semiconductor optical amplifiers and an electronic proportional- integral control circuit.
This automatic gain control system can be used to increase the input power range of the wavelength converter. The settling time of the control circuit is about 150 ns.
A combination of the automatic gain control and the wavelength converter has been demonstrated.
Data within a 10 dB average input power range can be converted to a new wavelength error-free using the combination of the automatic gain control and the wavelength converter.
Tflble of contents
Preface
Abstract
Table of contents
1 Introduction
ii
1
1
1.1 STOLAS project 1
1.2 Label swapper 2
1.3 Wavelength converter 2
1.4 Automatic gain control 3
1.5 Outline of this thesis 4
2 Literature 5
2.1 Single-stage approach 5
2.2 Multi-stage approach 6
2.3 Conclusions 8
3 Simulations 9
3.1 Theory 9
3.2 Requirements 9
3.3 VPI model 10
3.4 Results 15
3.5 Conclusions 22
4 Design 23
4.1 Layout 23
4.2 Power consideration 25
4.3 Characterisation 25
4.4 Issues 32
4.5 Conclusions 34
5 Demonstration 35
5.1 Requirements 35
5.2 Alcatel wavelength converter 35
5.3 Results 38
5.4 Issues 40
5.5 Conclusions 40
6 Conclusions 41
List of symbols
Table of figures
Bibliography
44 45 48
Appendix A Wavelength converter 49
A.1 Theory 49
A.2 Requirements 50
A.3 Alcatel wavelength converter 53
Appendix B VPI models 56
B.1 Control circuit 56
Appendix C Control circuit 60
C.1 Schematics 60
C.2 Component list 64
C.3Datasheet Samsung SOA 1 66
C.4 Datasheet Burr-Brown OPA 620 68
1 Introduction
In the last years, the demand for wide-band and high-speed transmission systems such as high-speed Internet service and multimedia service has increased rapidly. In order to have sufficient bandwidth for the broadband applications, concept called wavelength division multiplexing (WOM) is developed.
WDM can enhance the system capacity by using different wavelengths within a single fibre core.
Today internet protocol (IP), packet-based data is carried over WDM using asynchronous transfer mode (ATM) and synchronous digital hierarchy (SOH) as the intennediate layers.
Recently a novel approach called multi-protocol lambda switching (MPAS) is proposed for directly transporting IP-over-WDM. This results in more efficient and cost-effective networks because the intermediate SOH and ATM layers are avoided. In the MPAS protocol, optical packets are routed by switching them to other wavelengths (label swapping) at the transmitters in the edge node and in the intermediate network core nodes [7].
This chapter shortly describes the STOLAS project and the wavelength converter that is used to swap the data to another wavelength. The small dynamic range of input powers into the wavelength converter leads to designing an automatic gain control to increase the dynamic range of the wavelength converter. The design of this automatic gain control is the subject of this thesis.
The chapter is ended with an overview of this thesis.
1.1 STOLAS project
STOLAS, which stands for Switching Technologies for Optically Labelled Signals, is a research project that is funded by the European Union as a part of the Fifth Framework Programme "Information Society Technologies" (1ST). The project is a co-operation of different European universities, research centres and companies.
The main objective of the STOLAS project is to improve the throughput of packet-switched networks by using novel optical routing techniques based on stacked optical labels. In addition to optically labelling by assigning a particular wavelength to it (MPAS), a next-level label can be attached by using a particular modulation scheme. In the STOLAS system, the packet data is modulated on the intensity (intensity modulation, 1M) of the optical carrier. The label data is modulated on the frequency (frequency shift keying, FSK) of the optical carrier, which is orthogonal to the intensity modulated packet data.
The STOLAS project will focus on the development of some essential components for optical packet routing networks:
• Tuneable lasers for wavelength labelling which can be wavelength-switched very fast;
• Wavelength converters for optical label swapping on wavelength and frequency;
• Optical-label-controlled cross connect and add/drop nodes;
• 2R-multiwavelength regenerator;
• Management and control of the optical labelling and routing functions;
• Validation of the optical routing techniques in a small-scale test bed.
The STOLAS project is structured in four technical areas of work (work packages, WPs). A fifth package is named for the interworking and the project management tasks. The five work packages are:
Chapter1Introduction
•
WPO project management;•
WP1 system and network architecture;•
WP2 optical components;•
WP3 network elements;•
WP4 system integration and trial.1.2 Label swapper
1.2 Label swapper
Within WP2, research is performed on the label swapping for the STOLAS system. Figure 1.1 shows the configuration of the STOLAS label swapper. Both the intensity modulated payload and the orthogonal (FSK) modulated header are processed separately. The tap coupler in the optical path splits a part of the signal for the label processing. The label-processing unit recovers the header signal and a new header label is FSK-modulated on the fast-tuneable laser wave. The label-processing unit also tunes the wavelength of the laser.
A"
FSKlIM
Figure 1.1Circuit scheme ofanetwork node for IP label swapping [7].
The 1M information on the original signal is modulated on the new FSK-modulated wave using a Mach- Zehnder interferometer (MZI) wavelength converter. This converter is not sensitive to phase information of the old signal. Therefore, the 1M payload information of the incoming signal is transferred to the new wavelength but the frequency-modulated information is stripped.
The delay-line is introduced to allow the label detection and processing unit to detect the labels and tune the laser to the right wavelength before the payload data arrives at the wavelength converter. The wavelength converter will be integrated on an optical chip obtaining a cost-efficient solution for optical label processing [7]. The automatic gain control (AGC) is used to amplify the payload information into the wavelength converter and the average power is controlled to reach an optimal input power for the WC. The MZI-WC has a very small dynamic range (as shown is section 1.3) and therefore a gain control system should be used.
1.3 Wavelength converter
The basic principle of an MZ-based wavelength converter consists of a Mach-Zehnder interferometer with a semiconductor optical amplifier (SOA) in each branch as shown in the outlined section in Figure 1.1. The input signal can be coupled into the MZI-WC from the same side as the laser wave (co propagating) or from the opposite direction (counter propagating). The STOLAS-WC is used as a co propagating converter. Without an incoming signal, the laser wave is equally split in both branches, amplified and combined in the second splitter again by constructive interference.
An incoming signal in one of the branches will unbalance the MZI. The branches get different phase shifts due to self-phase modulation (SPM). The optical intensity changes the refractive index of the active region of the SOA and this results in a phase shift. Agrawal [1] describes the phase shift in an SOA:
Chapter1Introduction 1.4 Automatic gain control
(1.1 )
Where Uh [-]is the linewidth enhancement factor and gL [-] is the amplifier gain in linear units. Due to the higher optical intensity in the arm with the incoming signal, this arm will receive a different phase shift. Dependent on the phase difference between the two arms, the waves will interfere constructively or destructively in the second splitter. This will result in a transfer function as shown in Figure 1.2.
Because of the lower gain in one of the arms due to saturation of the amplifiers, the resulting output power of the MZI-SOA will not be zero for a phase difference ofn.
inverting non-inverting
Figure 1.2Transfer function of an MZI-SOA wavelength converter.
From this figure, an inverting and a non-inverting working area can be discriminated, depending on the incoming signal power. Figure 1.2 shows that an 1M input signal at the signal-wavelength will result in a converted signal at the CW-wavelength.
The principle of the wavelength converter is described in detail in Appendix A. The opto-electrical devices (OED) group of the Eindhoven University of Technology (TUE) will perform the actual implementation of this device. However, this research is still in an initial stage and therefore this thesis will focus on a wavelength converter from Alcatel. The Alcatel WC is available at the TUE and has been documented well [9].
1.4 Automatic gain control
Implementation of non-linear, all-optical functions results in tight power constraints. Some kind of power management should be used to compensate for:
• Power fluctuations due to laser ageing;
• Different routes in the network due to protection switching;
• Routing through a meshed network.
The automatic gain control (AGC) described in Figure 1.1 compensates for these power fluctuations and keeps the average input power into the wavelength converter at a constant level.
Chapter1Introduction
1.5 Outline of this thesis
1.5 Outline of this thesis
This thesis is a description of the design and implementation process of the automatic gain control function for the STOLAS project.
Chapter 2
In the late 80s and the beginning of the 90s research has been done on the subject of automatic gain control. This chapter gives a summary of these publications as a basis for the design of the STOLAS AGC-system.
Chapter3
The simulations on the new control circuit and the adaptation of existing models are described in this chapter. These models are based on simulations of semiconductor optical amplifiers (SOAs) and the control loop.
Chapter4
This chapter describes the measurements on the developed AGC-solution. The technical feasibility for implementation in the STOLAS label swapper will also be discussed.
ChapterS
The AGC system will be used in a STOLAS field trial. Guidelines for combining the AGC with the Alcatel wavelength converter will be presented in chapter 5.
Chapter6and 7
In these concluding chapters a review will be given on the design process and the characteristics of the developed AGC. Recommendations will be presented for future research and the use of the AGC solution for other areas of optical switching.
2 Literatu re
Automatic gain control is necessary in the STOLAS system in order to increase the dynamic range of the wavelength converter and to compensate for power fluctuations. Almost any type of optical amplifier, e.g. Raman, EDFA or SOA, can be used. The gain is controlled by a feedback loop that controls the pump source as a function of the optical output power.
The different types of gain control can be divided based on the type of amplifier used and on the method of measuring the reference value. Because the AGC should be integrated in the STOLAS system some boundary conditions should be observed. These conditions are the availability of the components in bulk material, costs, potential for integration in the STOLAS label swapper and the speed of the control circuit.
Because of the potential of on-chip integration, an SOA-based solution is preferred. The fluctuation of the input power to the AGC can be calculated from a power budget analysis. Based on these conditions a model for the AGC can be chosen.
In the late 80s and the beginning of the 90s research has been done on the subject of automatic gain control. This chapter gives a summary of these publications as a basis for the design of the STOLAS AGC-system.
2.1 Single-stage approach
The junction voltage of a SOA is directly related via the quasi-fermi levels to the electron density in the active layer and thus to the photon density in the amplifier. It is difficult to find a control function that controls the bias current of the SOA as a function of the error voltage(VJ - V,ef) because the reference voltage is also dependent on the bias current.
Ellis [2] proposed a gain control with a low frequency tone superimposed on the signal, but regarding the modulation scheme used in the STOLAS-project it is not possible to insert an extra tone.
2.1.1 Amplified signal monitor [8]
Tiemeijer [8] reports a SOA module with an amplified signal monitor based on two photodiodes in the module. Figure 2.1 shows the set-up.
Figure 2. 1SOA module with signal monitor [8].
Two high-quality aspherical lenses are used for the chip-to-fibre coupling. Two photodiodes are carefully aligned above the lenses. The differential signal from the two photodiodes is capable of
Chapter2Literature 2.2 Multi-stage approach
VVhen using a feedback loop that drives the bias current of the SOA, the output signal can be stabilised within 2 dB for a 25 dB dynamic range of the input power [8].
The advantage of this approach compared to a signal monitor with couplers is the possibility to monitor the signal without introducing extra loss due to the couplers.
The alignment of the diodes is very difficult and therefore expensive. On-chip integration of the diodes with the SOA is possible. The disadvantage of this system is the dependence of the saturation behaviour on the drive current.
2.2 Multi-stage approach
A multi-stage SOA uses an active area that is divided by two or more separate electrodes. Therefore, it is possible to keep one section at a fixed bias current and the reference junction voltage is only dependent on the optical intensity in the active layer. The actual reference voltage can be used to control the current in the other sections.
2.2.1 In-line detector [6]
Optical detection with a SOA is based on the dependence of the carrier density in the active region on the optical intensity. The amplification mechanism is a result of the stimulated emission of photons in the active region of the amplifier. For increasing optical power, the carrier density decreases, reducing the separation between the quasi-fermi levels and therefore the junction voltage. The junction voltage at a distance z on the longitudinal axis is [6]:
(2.1)
Vlpmono
VqJref VIpSIQ
VVhereT] [-] is a heterojunction ideality factor; ks [.I/K] is the Bolzmann's constant; T [K] is the absolute temperature; q [C] is the electronic charge; NSias[11m3]is the carrier density when no light is injected and L\N is the carrier density change when an optical signal is amplified .. The voltage measured at the junction of a SOA is the average of V<jJ(z)
M
along the active region axis. Figure 2.2 shows a set-up where the electrode is split in two parts of lengthL,[m] and Ls[m].-=======R=e=fe~f=n=ce=========--====s=il=na=1==~
, ,v'l'
(z»,,,
1 - -...-=--::..:;--.:.:--:=,:--- ---.---~..,,-",--••-.- •• '
-+----... _'!"I""'-t-...,f-+-...
zVariation of the voltage along z-axis
_ _~~ i
•
...
~,~~....
...'7... ::~:j::::::::::.:: ...
jIlean value L overL
Mean voltage on each electrode
Lr Ls
Figure 2.2Two-stage SOA for signal detection [6].
The different voltages are the average junction voltages in the reference, signal and total section respectively.
Chapter 2 Literature 2.2 Multi-stage approach
Vrp_ref =
i-
r JVrp(z)dz (2.2)(2.3)
(2.4)
In the case of a single-stage SOA the "reference" and the "signal" section are combined together, the responsivity is about-80VIW at50mA bias current [6]. The voltage variation in the active region is exponential along the longitudinal axis. At the input facet, the voltage variation is very weak. V\lhen the SOA active region is divided in two parts as shown in Figure 2.2, the signal average is only made over the section where the variation is strong. The responsivity can be improved significantly by using this two-section SOA.
Figure 2.3 shows the set-up of the differential detection with a two-section SOA. The bias current is shared into the two parts of the amplifier. V\lhen no light is injected, the voltage on the two electrodes is equal(Vref= Vsignal). In this set-up, the reference voltage is correlated with the signal bias voltage.
er cal
lib,..
I
Current sharingI
~
- outputVref VSlgnal
Ampli~edpow
---+---+
---+
Two-section SOA
Input power
Figure2.3Differential detection with atwo-stageSOA [6].
This two-section differential-detection SOA is less sensitive to temperature drift and bias current change than the one section amplifier.
Rampone [6] shows a three-section SOA with a reference, middle and signal electrode. The middle section acts as a pre-amplifier. The differential signal is defined as (Vsignal- V ref). This results in a differential detector with a good responsivity and a large operating range.
2.2.2 Current-Voltage [3]
Previous research at Lucent Technologies has been performed on automatic gain control by using SOAs [3]. Jansen describes a gain control system with two stages as shown in Figure 2.4. The second SOA is set at a fixed bias current and is used as a reference amplifier. Because the output of the second SOA should be at a constant power level, the reference voltage is determined as the junction voltage at which the desired output power is reached.
Chapter2 Literature
Voltage-<:ontrolled
current source Difference amplifier
2.3 Conclusions
~--~~ --_.
~-
VrefI
I *~M -r>'H -e:'H~* I
Tuneable laser source First SOA Second SOA Optical Receiver
Figure2.4 Two-stage automatic gain control[3].
A control circuit drives the bias current of the first SOA as a function of the difference(V,ef- Vj,2). The bandwidth of this system is about 6.4 kHz and a maximum error for the output power of 0.5 dB for an 8 dB dynamic range. For the speed of the control loop, a trade-off should be made between the time needed for controlling burst-mode data and pattern effects due to interaction of the control loop on the data.
In this set-up, the second SOA is driven at a constant drive current. The first SOA is controlled to output a constant average power into the second SOA. The advantage of this system, with respect to the other concepts discussed in this chapter, is the fixed saturation behaviour of the AGC system because of the constant average optical power in the second SOA.
2.3 Conclusions
Almost any kind of optical amplifier can be used in an AGC-system. The advantage of SOAs, however, is the possible full integration in photonic integrated circuits and therefore enabling a large cost reduction.
The system with the amplified signal monitor [8] is a good concept but it is very expensive to align the diodes above the SOA surface. Integrating the diodes in the same process as the SOAs can be useful. The disadvantage of this concept is the different saturation output power depending on the gain setting of the device.
A two- or three-section SOA [3,6] is the most promising concept for the AGC concept in the STOLAS project because it is a relatively simple, cost-efficient solution.
The control speed of the AGC in the STOLAS system has to be limited between a minimum and a maximum value. It has to be fast enough to follow the average power fluctuations in a burst-mode system but without significant signal distortion. Using the AGC-solution as proposed by Jansen [3], research should be performed on components for a faster control response and the usage of other types of control feedback.
A limitation is the noise behaviour of the system. Without filtering, the influence of amplified spontaneous emission (ASE) is significant. Research should be performed on decreasing the influence of noise by applying an EDFA pre-amplifier or using one single gain control for all wavelengths.
3 Simulations
This chapter describes the simulations on the automatic gain control. The purpose of the simulations is to find a set of parameters and their tolerances for which the AGC can be used within the STOLAS label swapper.
The production of the STOLAS label swapper is still in a preliminary phase and therefore an Alcatel wavelength converter will be used instead as was noted in section 1.3. The automatic gain control has to be designed in order to fit the boundary conditions of both the Alcatel-WC and the future STOLAS- WC. The Alcatel WC will be further discussed in section 5.2.
The simulations are performed using the VPI transmission maker software [10]. In this chapter, the theory of the VPI SOA model and a description of the AGC model are presented. The model is simulated for different types of control feedback.
Based on the simulations a type of control feedback will be chosen and guidelines for designing the automatic gain control will be discussed.
3.1 Theory
The amplification process in an SOA is based on the process of a photon stimulating an electron to change energy levels within the semiconductor and, in doing so, releasing energy in the form of a second photon. A set of rate equations describes the temporal and spatial dependence of the carrier density on the photon intensity in the active region [5,10). For the SOA model in the simulations, it is useful to rewrite the rate equations in a way that temporal rate equations are obtained [5).
Equation 3.1 describes the balance between injected carriers and the carriers lost by spontaneous and stimulated emission:
dN I 2 3 ( ) S
- : - - A N - B Ndt qV -CaugN -vaN-No - -9 1+
as
(3.1)Where N [11m3]is the carrier density; I [A] is the amplifier drive current,
q
[C] is the electronic charge;V [m3] is the volume of the active region, equal to Lxwxd; the coefficients A [1Is], B [1/s.m1 and C [1/s.m1 account for defect, bimolecular and Auger recombination, respectively; vg[m/s] is the group velocity; a is the linear material gain coefficient [m2]; a [11m3]is the material gain nonlinearity and S [W/m3]is the photon density within the cavity.
The intensity variation in the active region is the result of stimulated gain plus spontaneous emission noise minus the stimulated absorption.
dS S 2 S
- : V
a
of .(N -
No ) - -+ B .r .
N .fJ - -
dt 9 1+O"S 'fp (3.2)
Where
r [-]
is the confinement factor and13 [-] is spontaneous emission coupling factor and'p[s] is the photon lifetime. Equation 3.2 represents the temporal intensity variation in the active region.3.2 Requirements
Chapter3Simulations 3.3 VPI model
The extinction ratio E[dSI of an 1Moptical signal can be defined as the ratio between the mark power (b1) and the power of a space (bo):
G=10 .
L09(~J
bo'
(3.3)
In the STOLAS project, the header data is FSK modulated on the 1M payload. A high 1M extinction ratio will result in a large difference between the 1M spaces and marks and therefore a good performance of the1Mdetector. A high1Mextinction ratio, however, also means a low power for the1M marks and therefore the FSK data is modulated on this low power, In the STOLAS project, a trade-off has to be made between a high extinction for the 1Mdetector and a low1M extinction ratio for the FSK detector. In the STOLAS project a consensus has been reached on an1Mextinction ratio of 6 dB.
The AGC has to be fast enough to control an incoming burst-mode signal without loosing too much data on controlling the average power, but a high control speed will result in pattem distortion because the data bits are also affected, For a 1.25 GbiUs, PRBS 27_1 modulated data pattern, the lowest frequency component (7 consecutive spaces or marks) will be 1.25/14=89 MHz. In the simulations, a 3 dB bandwidth of 100 MHz will be used as a starting value,
3.3 VPI model
The VPI model of the control circuit is presented in Figure 3.1. This scheme is a simplified illustration of the actual model layout in VPI. A brief description of the model is given here and in the next paragraphs, each of the control loop components is discussed in more detail.
The design of the SOA model is similar to that of the AGC solution in section 2.2, The first and second SOA are referred to as SOA A and SOA B, respectively. Equal parameters are used for SOA A and SOA B values are listed in Table B.2.
Amplifier with slew rate
Filler, 3rd order LP
~.~. ~~
-D- Gam i I
, LJ
§
A
II~i~Wj=kt/q
yL.(' +Re'lbiasIn (NlNi)8
>--1:-<f".--~.>. ,IFork'~'~
18 B
Figure3.1 VPI model
to
simulate theAGesystem behaviour.A continuous wave (CW) laser is externally modulated with a 1.25 GbiUs PRBS 27_1 data pattern and the modulated wave is coupled into SOA A. This SOA is driven by a bias current, which is dependent on the control loop. The optical output of SOA A is coupled into SOA B, which is driven by a constant bias current of 50 mA. The optical output of SOA B is displayed in the time domain. Using equation
Chapter3Simulations 3.3 VPI model
3.7, the bias voltage will be calculated from the carrier density in the active region and the bias current into SOA B.
A reference voltage is subtracted from the junction voltage and led to an operational amplifier (OPAMP). The OPAMP model has a filter, gain section and a slewing rate. The output drives the bias current of SOA A. In the next paragraph, each part of the control loop is described.
By using a reference amplifier with a fixed bias current, it is possible to have all signals, regardless of the input power, at the same saturation power because the control loop adjusts the input power of SOA B to a constant level.
3.3.1 SOA model
The SOA model that is used for the VPI simulations is based on the transmission-line laser model (TLLM) from Lowery [5]. The TLLM discretises both space and time by dividing the cavity into sections, each containing a scattering node, and linking the sections with transmission lines, each with a constant delay (Figure 3.2). The scattering node calculates the gain, absorption, phase change and spontaneous emission in this section using the carrier density and the optical power in the model section (equation 3.2). After each scattering operation, the outgoing fields are passed to the next section and the incoming fields are received from the previous section. In the next iteration the carrier density is recalculated from equation 3.1 using the average photon density in the section, the injection current into the section and the spontaneous recombination rates [10].
Bias current
Figure 3.2Transmission-line laser model;
the laser cavity is divided into sections represented by scattering matrixes [5].
The transmission-lines represent the waveguide propagation delay between two sections. The transmission-lines all have an equal delay, which is called the time step. The time step is equal to the inverse of the sampling rate that is predefined by the global parameter SampleModeBandwidth in VPI.
The choice of SampleModeBandwidth also sets the number of sections used (and with this the accuracy) in the model [10]. As the computation time is approximately proportional to the square of the SampleModeBandwidth, a trade-of should be made between computation time and accuracy of the model. The VPI manual [10] therefore recommends a SampleRateBandwidth of 2560 GHz for a cavity lengthof350 j.1m. The TLLM then uses M=12 sections.
Gain
The TLLM used infinite-impulse response filters to represent the spectral dependence of the spontaneous and stimulated emission. These filters can be approximated by a Lorenzian response if the bandwidth of the filter is significantly less than the simulation bandwidth (SampleModeBandwidth).
Figure 3.3 shows the position of the filters within the signal path. The dependence of the gain on the carrier density is logarithmic in the model:
Chapter3Simulations
Gstim(N,S)=
exp[ar'ln(~),-S-'/}'L]
No 1+oSVVhere /}.L =LIM is the section length [m]. The spectral density of the ASE noise is:
3.3 VPI model
(3.4)
(3.5)
VVhere G [dB] is the power gain across the active region; hv [J] is the photon energy; nsp [-] is the population inversion parameter; Uint [11m] is the internal loss and the factor gI(g - aint) accounts for the population inversion due to the internal loss in the section.
Input Field
Spontaneous Emission Filter
Figure3.3Signal flow through the stimulated and spontaneous emission filters [10].
The incoming and outgoing fields are first filtered before entering or leaving a model section. As can be seen from Figure 3.3, the stimulated emission filter only operates on the fraction of the optical field generated by optical gain. If the gain due to stimulated emission is zero, the optical field passes through the section unchanged and unfiltered. At the full width half maximum (FWHM) points, the power gain in dB will fall to half the peak gain in dB, which is different from the common definition of the FWHM being the point at which the gain has fallen to half the linear peak gain (-3 dB point).
Model I/O
The TLLM is a time-domain model, so all the outputs are time series. The model has a forward- and backward optical field and an average-carrier-density output. The junction voltage
vi
can be calculated from this average carrier density [10].(3.6)
VVhere "1 [-] is the heterojunction ideality factor (usually around 2); k is Bolzmann's constant (1.38x 10'23J/K); T is the junction temperature (298 K); q is the electronic charge (1.602.10'19 C);
N[1/m1 is the average carrier density across the junction and Ni [11m3]is the intrinsic carrier density in the active region (1.5x 1016
1/m\ The junction voltage cannot be measured directly because of a series resistance Rsof about 2
n.
The junction voltage in the model is therefore represented by:Chapter3Simulations 3.3 VPI model
(3.7)
Transient
In the simulations, a transient time should be discarded before the signal is introduced, because the amplifier needs some time to reach a steady state. The model starts with an initial carrier density defined by the parameter InitialCarrierDensity and it takes about 7 ns to reach a steady carrier density.
Parameters
In the actual AGC implementation, two Samsung bi-directional amplifiers (Appendix C, Table C.5) will be used. In the remainder of this report, these amplifiers will be referred to as SOA 1 and SOA 2.
A large effort has been made to map the parameters of the simulation-amplifiers on the Samsung amplifiers. Because Samsung does not issue physical parameters and measuring these parameters is not possible, the process of mapping the simulation-parameters is very difficult. Therefore, the simulations can only be used to show trends for different parameter settings of the control circuit.
By assuming a logarithmic dependence of the gain coefficient on the carrier density (equation 3.4) the model does not describe the decrease of the gain coefficient for high carrier densities that can be observer in real SOAs.
The characteristic transfer functions of the model SOA are presented in Appendix B and the model parameters are presented in Table B.2. The complete datasheet for the Samsung SOA 1 is presented in appendix C.3.
3.3.2 Control circuit
The control circuit is divided in an ideal block that subtracts a reference voltage from the junction voltage of SOA B and a block that represents the electrical amplifier. This amplifier block has a filter to model the spectral dependence of an amplifier and decrease the control speed for reasons addressed in section 3.2. The gain section models the control transfer and a slewing rate block represents a finite rising speed of an operational amplifier.
Reference voltage
Figure 3.4 shows a simulation of the junction voltage versus the SOA output power and for a bias current of 50, 60 and 70 mA respectively. A higher bias current increases the number of carriers that are inserted into the active region whereas a higher output power results in a lower junction voltage due to a larger carrier recombination rate.
Chapter3Simulations 3.3 VPI model
1112
r-;:::======::;---:::----;:==::::r==---:--i
~109
0( .
oCIl
~108~
~
~
§1.07~
1.06
6 S 10
Output power IdBm)
·2
·6 ·4
·8
1.045' - - - ' _ - - ' _ - ' - _ - - - - ' _ - - - - L _ - - ' - _ - - ' - _ - ' - _ - ' - - _ - - - '
·10
Figure 3.4 Junction voltage versus output power of the SOA for different drive currents.
In the saturation area of SOA B, the junction voltage change per watt input power is larger than for lower powers. A realistic value for the average signal input power into the Alcatel wavelength converter is 0 dBm. Therefore, the AGC needs an average output power of 0 dBm.
Figure 3.4 shows that the responsivity of the SOA at 0 dBm output power is the highest for a bias current of 50 mA. The reference voltage for the simulations is chosen from Figure 3A. With Ibias
=
50 mA and Pout
=
0 dBm this results in V,el=
1060 mV.Operational amplifier
In the AGC an OPAMP from Burr-Brown, type OPA 620, will be used. The datasheet is presented in appendix CA. The error voltage Vj - V,el is amplified by the control transfer function and the output drives the bias current of SOA A. The simulations model the slewing rate and frequency dependence of this OPAMP. Different types of feedback (proportional, integral and derivative) are evaluated.
Slewing rate
An operational amplifier has a slewing rate. The slewing rate S, [VIs] is defined as the maximum voltage swing per second for the maximum input sweep.
S = t1V
r
M
(3.8)The OPA 620 has a slewing rate of 250V/lJ.s. The VPI model for the slewing rate is given in appendix B.1.2.
Bandwidth
An OPAMP is frequency dependent. The speed of the control loop should be limited to prevent control of the single data bits. In section 3.2 it has been discussed that this results in a filter bandwidth that should be around 89 MHz. In the simulations, an infinite response low-pass filter of order 3 is used with a bandwidth of 100 MHz.
Optical filtering
The signal will receive a large amount of broadband ASE noise within the AGC. The oscilloscope in the VPI model is implemented as a photo detector. The photo detector introduces quantum noise to the signal and this noise is filtered out using an internal rectangular low-pass filter with a cut-off frequency of 1.0 GHz.
Chapter3Simulations
3.4 Results
3.4 Resuffs
The next paragraph discusses simulations with a slow varying input power in order to evaluate the static response of the automatic gain control system. Secondly, a modulated signal is used to evaluate the effect of the bandwidth of the control loop and the extinction ratio of the modulated signal.
3.4.1 Static response
This paragraph describes three different types of control feedback. First, the proportional control is simulated. Secondly, a proportional-derivative control is evaluated in order to increase the damping of the control signal. Finally, an integral factor is added to eliminate the steady state error. The consecutive values of the input power are: -30, -20 and -10 dBm.
P-control
The P-control AGC follows the approach as proposed by Jansen [3]. The proportional transfer function Dpis:
(3.9)
The parameter settings of the Bulk_SOA model are shown in Table B.2 and the control-loop settings are presented in Table 3.1.
Table3.1Parameters of P-control simulation of of theAGewithout data.
System Symbol Parameter Value Unit
Global TimeWindow 51.2 ns
SampleModeBandwidth 640 GHz
SOAB I Bias current 50 mA
Control loop Vre, Reference Voltage 1060 mV
I~W
Bandwidth loop filter 100 IMHZ Proportional gain 10 AN Slewing rate 5.108 'IV/sFigure 3.5 shows different signals of the P-control AGC. The drive current plot shows the effect of the transient time of the SOAs. T
=
0 s, the initial carrier density is very low (1.0x 10241/m\
and from equation 3.6 the junction voltage is also low resulting in a negative error voltageVj -Vref.Therefore, the drive current is zero in the first 7 ns.Chapter3Simulations 3.4 Results
a.
PowerIdem] Input signal -9
-15
-20
-25
b.
uj-uref [mV]
o1f---fHHf--I'-'---l
-30 ",E:::==~....J!~ """'!-2L - _ - ' - _ - - - - ' - _ - - - ' ' - - _....
o 100 Time Ins] 20: 0 50 100 Time Ins] 20:
100Time Ins] 205
c.
driue current [rnA]
20
-1 1 : : = = = = = = = = = - 2 0LoL.. ....L ----'J
o 100 Time Ins) 20: 0 10
d.
Power[dem] output SOA 2 33r - : - - - -10
Figure 3.5 Proportional control signals;
(a) input signal; (b) error voltage; (c) drive current of SOA 1;(d) output power of SOA 2.
After the transient time, the AGC reaches a steady control situation for the -30 dBm input power. The drive current of SOA 1 is 30 mA and using equation 3.9 and K=10 the error voltage is 3 mV.
After 69 ns, the input power into SOA 1 changes to -20 dBm and this results in an instantaneously power step at the output of SOA 2. The increased photon density in SOA 2 saturates this amplifier and therefore the junction voltage decreases due to the lower carrier density. Because of the saturation effect, the output power of the AGC has become lower than the desired value and the drive current is increased to reach a steady output power. The drive current stabilises at 22.5 mA and the error voltage is 2.25 mV.
The outputofthe SOA 2 for a -30 dBm input has more noise than for the -20 dBm input. Because the automatic gain control has to amplify the -30 dBm more, also a larger noise contribution will be added.
Different input powers result in different error voltages and consequently different output powers for the AGC. Figure 3.5.d shows indeed a 5 dB difference in the final value between the -30 and -20 dBm input power.
For t=137 ns the input power increases to -10 dBm and again a step in the output power occurs. For this input power the control signal oscillates. The proportional gain is too high for this input power.
PD-control
In order to improve stability, a proportional-derivative (PO) control is introduced with a transfer function DpD :
(3.10)
\l\lhere "Cd [s] is the derivative time constant. The derivative feedback increases the damping and improves the stability of the system. Table 3.2 shows the parameters of this simulation, which are the
Chapter3Simulations 3.4Resuffs
same as the parameters in the proportional control simulation except for the extra derivative component.
Table3.2 Parameters of PO-control simulation of of the AGC without data.
System Symbol Parameter Value Unit
Global TimeWindow 51.2 ns
SampleModeBandwidth 640 GHz
SOAB I Bias current 50 mA
Control loop Vref Reference voltage 1060 mV
BW Bandwidth loop filter 100 MHz
K Proportional gain 10 AN
'rd Derivative time constant 10.10-9 s
S, Slewing rate 5.108 Vis
The control signals in Figure 3.6 show more damping compared to the P-control response. A PD- control still has the disadvantage of a nonzero steady-state error.
b.
vj.vref[mV]
2
3.4 r - - - , 3
Ou- ----'- ....u
100 Time Ins] 205 0 100 Time Ins] 205 -25
-20 -15
a.
PowerIdBm] Input signal -9
-30
1::::===----l
----tI o100 Time Ins) 205 -10
c.
drive current [rnA]
20
-1 t:::::=========:I:l.20 ....L... ---lJ
o 100 Time Ins] 205 0
10
d.
Power[clBm] output SOA 2 33 . - - - , 1 0
Figure 3.6 PO-control signals;
(a) input signal; (b) error voltage; (c) drive currentof SOA 1;(d) output powerof SOA 2.
Figure 3.7 shows the individual contributions of the proportional and derivative components to the total control signal of the AGC. The derivative component increases the transient speed of the control signal and its contribution decreases for the steady state.
Chapter3Simulations 3.4 Results
Proportional and derivative contribution to the drive curent [rnA]
Proportional' d/dx • sum' oH--f+-"'>O'C__--...:....---...,.=...---r+-...,...,f-";C""'7":::..",,,.-.----j
3Sr - - - ,
30
10 20
.10
-20_ W - l L - - " ' - - - ' - - - - ' - - - ' - - - ' - - - ' - - - ' - - - ' - - - ' - - - . - J
o 20 40 60 80 100 120 140 160 180 200
TimeInsl
Figure 3.7 Proportional and derivative contributions to the PO-control signal.
PID-control
The problem of the steady-state error can be solved using a proportional-integral-derivative (PIO) control. The main advantage of combining the PO-control with an extra integral control is the reduction of the steady state error Vj -Vref .The control transfer function OPID is:
0P/O(s)= K·(1+_1_+S'dJ
S'I
(3.11 )
VVhere 'd [s] and 'I [s] are the derivative and the integral time constant, respectively. The proportional feedback increases the speed of the response; the integral component eliminates the steady-state error whereas the derivative part can damp the dynamic response.
Table 3.3shows the parameters of the PIO simulation. The input signal is alternated from -30to-20 and -10 dBm to evaluate the effect of a step in average power on the AGC behaviour.
Table 3.3 Parameters of PIO-control simulation of of the A GC without data
System Symbol Parameter Value Unit
Global TimeWindow 102.4 ns
SampleModeBandwidth 640 GHz
SOAB I Bias current 50 mA
Control loop Vref Reference voltage 1060 mV
BW Bandwidth loop filter 100 MHz
K Proportional gain 10 AN
'td Derivative time constant 10.10-$ s
'tl Integral time constant 0.5,10·g s
S, Slewing rate 5.108 VIs
Chapter3Simulations
a.
Power [dBm] Input signal -19.7
-25
b.
uj-uref [mY]
4.3 . - - - ,
2
oI---"'h'h_--tf\-tv---!
-2
3.4 Results
-30.2 -5
0 100 TimeIns] 205 0 100 TimeIns) 205
C. d.
driue current [rnA) Power (dBml output SOA 2
48 7
40
0 30
20 -10
10
-1 ·25
0 100 TimeIns) 205 0 100 TimeIns] 205
Figure 3.8 PID-control signals;
(a) input signal; (b) error voltage; (c) drive current of SOA 1;(d) output power of SOA 2.
The effect of the integral component on the AGC response is as expected. The error voltage in Figure 3.8b fades out resulting in a zero error voltage of the AGC control signal. Figure 3.9 shows the proportional, integral and derivative contributions to the total control current. The integral component slows the response but it also eliminates the error voltage, as the proportional line decreases to zero.
The derivative component is only present at the points where the input power changes.
Chapter3Simulations 3.4 Results
Proportional, integral, derivative contribution to the drive current [rnA]
51 , - - - ,
200 180 Proportional·
dldx·
Int·
sum·
160 140 120
60 100
Time [ns) 60
40 20
okt--t...::'i---f-==~---=,.r::=:::==~----d~?=~----j 20
40
-20
Figure 3.9 Proportional, derivative and integral contributions to the total control signal.
3.4.2 Data response
In the previous paragraph, a steady-state analysis was made. The influence of the control loop on an intensity modulated signal will now be simulated. Regarding the speed of the control loop, a trade-off should be made between the time it takes for the AGe system to control the input signal and the pattern degradation due to equalisation of the different signal bits by the control loop.
The derivative component in the PID-control is very sensitive to the fast data pattern. Therefore, it is decided to use a proportional-integral (PI) control because the faster transient response does not compensate for the bad steady-state performance.
The absense of the derivative component results in less damping of the control loop. Therefore, the proportional gain should be lowered for the PI-feedback compared to the PID-control to decrease the oscillations.
The influence of data on the PI-control loop will be simulated in this section. The data speed is 1.25 GbiUs, because this is one of the bit-rates that will be used in the STOLAS system.
It is important to achieve a steady-state control and a fast reaction on burst-mode data. Equation 3.12 shows the new PI transfer function.
Dp,(s)= K .
(1
+_1_J
sr, (3.12)
The parameters of this PI simulation are listed in Table 3.4. The input data, with an average power of -17 dBm and an extinction ratio of 6 dB, is coupled into SOA A after 80 ns.
Chapter3Simulations 3.4 Results
Table3.4Parameters of PI-control simulation of of the AGe with1.25Gbit/s data modulation.
System Symbol Parameter Value Unit
NRZ coder BitrateDefault 1.25 GHz
Type PRBS 27_1
Risetime Risetime data 200 ps
Control K Proportional gain 10 AN
~I Integral time constant 0.5.10-9 s BW Bandwidth loop fitter 100 MHz
MZ Modulator E Extinction ratio 6.0 dB
Figure 3.10 shows the resulting signals within the control loop. The simulation starts with a very low input power. Figure 3.10.d shows that the drive current has not yet reached a stable value when the data is entering the AGC. When the separate contributions of the proportional and derivative components to the total control signal are observed, the integral contribution is still rising. This is a result of the start-up effect of the SOA model (paragraph 3.3.1, transient).
After 69 ns, a signal with a relative small extinction ratio is coupled into SOA A. This situation illustrates that the control loop is still too fast. The proportional component shows a large reaction on the data whereas the integral component keeps integrating and does not reach a steady value. This unstable drive current results in an average power that is not constant. A control loop bandwidth of 100 MHz is therefore still too high.
Unfortunately, because the VPI licence expired, it is not possible to simulate the AGC with a smaller bandwidth.
100 Time [ns] 205
b.
uj.uref
·5 a.u. pe·3j
4.3 , - - - ,
l.----'--"==---'- ....J..I.10.7 l . - ~ __'_'
100 Time [nsl 205 0
·40
·60
a.
Power [dBm] Input signal
·16
-t---....
Ita,
-50
d.
Power [dBm] outputSOA2 13
Proportional ·100
Integr'l PH
·132 '--- ....1....::::--_--:---"
100 Time (nsl 205 0 100 Time [nsl 205
c.
P and I component [rnA]
o
20
45 , - - - ,
-23 '-- ~ ____!J
o
Chapter3Simulations
3.5 Conclusions
3.5 Conclusions
The simulations show that a proportional-integral control system is the most suitable for the STOLAS automatic gain control. The integral component is necessary to eliminate the steady state error.
Design rules for the values of the proportional gain, integral time constant and bandwidth of the control loop, resulted from the simulations.
A derivative component is not used in the feedback. The derivative component follows the fluctuations on the junction voltage due to the high-speed data and therefore the control signal becomes dependent on the data pattern, which is undesired.
Proportional gain
The proportional gain should not be limited. A high proportional gain results in increasing the overshoot and the chance of oscillation. A proportional gain of 10 is used for the simulations because a higher value increases the oscillations of the control signal.
Bandwidth filter
The bandwidth of the filtering has to be significantly lower than the lowest component in the data pattern. A PRBS 27_1 data pattern at 1.25 GHz has a lowest frequency component (7 consecutive spaces or marks) of 1.25/ 14=89 MHz. Simulations showed that good results are obtained for a control loop bandwidth of 1/1Oth of the lowest frequency component in the data. The control should therefore be around 9 Mhz.
Integral time constant
The value of the integral time constant depends on the speed of the control loop and the proportional gain. The integral time constant should be chosen as low as possible to reach the final value as fast as possible, without getting oscillations.
4 Design
Now it is determined that a proportional-integral control is the best solution for the STOLAS automatic gain control, the next step is to design the actual control system. In this chapter, the design-process of the PI-control automatic gain control is described.
Section 4.1 describes the layout of the AGC and the components that are used in the system followed by a calculation of the reference voltage based on the expected optical powers in the system. A detailed characterisation is presented in section 4.3 and section 4.4 describes some issues concerning the present AGC solution.
4.1 Layout
The automatic gain control system is divided in a temperature control and a gain control circuit, each driven by a separate supply voltage source. Figure C.1 shows the schematic of the voltage sources. In the next paragraphs the temperature and gain control circuits are discussed. In appendix C the complete schematics and component lists can be found.
4.1.1 Temperature circuit
Figure C.3 and Figure C.4 show the schematics of the thermo-electric cooler (TEC) circuit for the two SOAs. Each SOA has its own temperature control loop. The thermistor between pin 5 and ground of the SOAs is the sensor of the control loop. Using potmeters R37 and R59, the reference temperature can be adjusted. These potmeters are set at 16.7 kQ for an SOA temperature of 298 K.
4.1.2 Gain control circuit
The gain control circuit is divided in an implementation of the PI transfer function and a protection circuit between the output of the OPAMP and the SOA 1. Figure C.2 (appendix C) presents the complete control circuit. In the next paragraph, the PI and protection circuit are discussed separately.
Proportional-integral circuit
The AGC control system is developed using two Sam sung bi-directional SOAs (Table C.5). The PI- control schematic is presented in Figure 4.1. At the right side, the reference voltage is made using a buffer-OPAMP (U3). The operational amplifier U1 is a differential amplifier that amplifies the error voltageVj -V,e" Assuming R31
=
R34 ,R30=
R36and C5=
Cg the transfer function of this section is:v -
R36 .tV - V )ul - R (1 R C ) \ j ,.,
34 +S 36 9
(4.1)
In section 2.2.2, it is explained that the bandwidth of the control loop has to be limited to prevent a reaction of the control loop on the data. For a 2.5 GbiUs, PRBS 27_1 pattern, the lowest frequency component is 2.5 GbiUs /14=179 MHz. The simulations showed that the control loop bandwidth has to be a factor of 10 lower than the lowest frequency component in the data to eliminate the influence of the data on the control loop. The bandwidth should therefore be lower than 20 MHz.
For the control loop, two OPA620 operational amplifiers from Burr-Brown are used. These amplifiers
Chapter4Design 4.1 Layout
AGNO
U2 OPA620
V antral
+S
Ul OPA620
+S
R32 10k
AGNO
Figure 4.1 Proporlional-integral control schematic.
OPAMPU2adds the integral component to the control function. The transfer function ofU2is:
(4.2)
The total transfer function of the control loop is:
(4.3)
\Nhich has the same shape as the PI-control function in equation 3.12, with K = R36 and " =R3SC10 . R34
The values of the proportional gain and the integral time constant are tuned using the guidelines in section O. The proportional gain is set at K= 12, because a high proportional gain results in a lower bandwidth of U1 . The integral time constant is chosen as low as possible without getting oscillations.
The integral time constant resulted in'tl = 20 ps.
Protection circuit
The next step is to control the bias current into SOA 1 and limit this current at 200 mA to prevent damage to this amplifier. Figure 4.2 shows the protection circuit. The output swing of the OPA620 amplifiers is ±3.5 V.
Rl0 200 L---4I---.._~----+--_+_--+--_-~-~ +S 01
BAS161S0T
R14 SO
R1B SO SOAI
SOAanode . . .---1>---to__---t--~---.--~-~-t---<r____,-__,,-___,
Figure 4.2 Cu"ent limiting circuit.
The resistors in the circuit are implemented with a network of parallel resistors because the maximum power dissipation of one resistor is 0.1 W. The right hand side of the circuit delivers a current of about 150 mA to the node with the SOA anode. Depending on the output V of the control OPAMP. this
Chapter4Design 4.2 Power consideration
current is drained away into the control OPAMP or the OPAMP adds an extra current into SOA 1. In order to tune the control loop in a way that an average optical input power in the range -20 thru -10 dBm is properly controlled, Rl1 , R15, R1 and R2 are not connected. The total resistance of RA becomes 33.33 nand RB25 n. Table 4.1 shows the limits of the SOA 1 current for Vu= ±3.5 V.
Table4.1 Bias current SOA for limits control voltage.
Vu +3.5 V
-3.5 V
4.2 Power consideration
ISOA2
165 rnA
o
rnAExact values for the extinction ratio and the input power range in the STOLAS project are not specified yet. Therefore, realistically expected values are used. In the STOLAS system, an extinction ratio of 6 dB has been chosen (section 3.2).
In the AGC system, a current of50mA drives the second SOA. The drive current of SOA 2 defines the saturation behaviour of the AGC circuit. The worst-case of a 6 dB extinction ratio at the output of the AGC is evaluated.
SOA 2 has a 3 dB saturation power of +3 dBm [3]. The saturation of an SOA can be approximated by:
(4.4)
In order to reduce pattern effects on the data, only 1.5 dB gain saturation will be tolerated on the data marks. From equation 4.4 the 1.5dB saturation power will be calculated as:
(4.5) V\lhere Psat,3dB
=
+3 dBm results in Psat.15dB=
0 dBm. The AGC reference voltage should be tuned to get a mark power of 0 dBm. The average power for this situation can be calculated from equation 4.6:£ +1
Paverage = - _ .Pmark
2·£ (4.6)
V\lhere E is the extinction ratio (6 dB) in linear units. This results in Paverage
=
-2 dBm. The AGC reference voltage is tuned to get an average output power of -2 dBm for an input power of -10 dBm and a wavelength of1550nm.4.3 Characterisation
This section describes the behaviour of the AGC system. First, a slowly varying input power is applied to the AGC, representing the typical power transitions in a burst-mode system. Secondly, the effect of the control circuit on continuous data is characterised and then, a burst-mode modulation is combined with a data pattern.
