JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 29, NO. 22, NOVEMBER 15, 2011 3483

The Dynamic Gain Modulation Performanceof Adjustable Gain-Clamped Semiconductor

Optical Amplifiers (AGC-SOA)Lin Liu, Student Member, IEEE, Craig Michie, Anthony E. Kelly, and Ivan Andonovic, Fellow, IEEE

Abstract—The growth in demand for high bandwidth serviceshas stimulated the deployment of Passive Optical Networks(PONs), directly to the home or to the kerb. In many cases,particularly extended reach PONs which may cover distances of100 km or more [1], there is the need for low cost reach extensiontechnologies. Semiconductor Optical Amplifiers (SOAs) have akey role in this context, particularly because upstream traffic iscommonly carried at 1.3 m. Upstream traffic in a PON (fromthe Optical Network Unit, ONU to the Optical Line Terminal,OLT) is normally Time Division Multiplexed (TDM) with a widevariation in path loss arising from differences in transmissiondistances and splitting losses. The bursty nature of this trafficcombined with a wide dynamic range of signal strength ( ��

dBm to �� dBm—the difference between a very close ONUwith a small split ratio and a distant ONU with a high splitratio), places severe demands on the burst mode receiver at theOLT. Conventional fibre amplifiers cannot adjust their gain withpacket to packet variations due to their response time. Similarly,conventional SOAs suffer loss of linearity if their bias current andhence gain is rapidly reduced. The paper reports on an adjustablegain-clamped semiconductor optical amplifier (AGC-SOA) de-signed to maximize the output saturated power while adjustinggain to regulate power differences between packets without lossof linearity. Theoretical modeling predicts that this device is ableto modulate gain at nanosecond rates. The analysis is validatedexperimentally.

Index Terms—Gain modulation, Psat, semiconductor opticalamplifier.

I. INTRODUCTION

T HE growing penetration of home DSL connectionsdemonstrates clearly that demand for high bandwidth

services. Realistic predictions estimate annual traffic growthfrom 34% to 50% [1], [2]. Analysis shows that to sustain thislevel of growth economically requires cost reduction in theprice per unit of bandwidth that has hitherto been unachievableby technological developments alone. The solution is thereforeto reduce the amount of equipment (interfaces between nodes)

Manuscript received March 11, 2011; revised September 05, 2011; acceptedOctober 03, 2011. Date of publication October 13, 2011; date of current versionNovember 23, 2011.

L. Liu, C. Michie, and I. Andonovic are with the Department of Electronicand Electrical Engineering, University of Strathclyde, G1 1XW Glasgow, U.K.(e-mail: i.andonovic@eee.strath.ac.uk).

A. E. Kelly is with the Department of Electronics and Electrical En-gineering, University of Glasgow, G12 8LT Glasgow, U.K. (e-mail:Anthony.Kelly@glasgow.ac.uk).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/JLT.2011.2171669

Fig. 1. Schematic of PON with AGC-SOA pre-amplifier.

in the network. These drivers have stimulated interest in ac-cess solutions based upon Passive Optical Networks (PONs)with high splitting ratios [3]. Central, in particularfor extended reach PONs which may cover distances of 100km or more [4]–[7], is the need for low cost optical amplifiertechnologies: Semiconductor Optical Amplifiers (SOAs) have aclear role to play in this context. The wavelength plan for future10 Gbit/s (XG-PON) [8] uses wavelengths from 1260 nm up to1620 nm and alternative optical amplifier technologies do notfunction efficiently across this band.

Upstream traffic in a PON (from the Optical Network Unit,ONU to the Optical Line Terminal, OLT) is normally time di-vision multiplexed (TDM) with a wide variation in path lossarising from differences in transmission distances and splittinglosses. The bursty nature of this traffic combined with a widedynamic range of signal strength ( dBm to dBm—thedifference between a very close ONU with a small split ratio anda distant ONU with a high split ratio), places severe demands onthe burst mode receiver at the OLT. In particular, the design oftrans-impedance and limiting amplifiers and Avalanche PhotoDiodes (APDs) is complicated by the need to cater for this widedynamic range. As the upstream data rate increases, and longerreach PONs are deployed, this issue will become even morecritical.

Adjusting the gain of an optical amplifier positioned at theOLT (Fig. 1), in order to regulate path losses and the signalstrength at a packet level substantially alleviates many of theabove issues in long reach PONs. Signal strengths will be pre-sented at the OLT at an optimized power level thus reducingthe receiver dynamic range requirement and thus overall com-plexity. To be of value, the timescale of the gain adjustmentshould take place at least within the guard band of the packet/frame transmission. For 10 Gbit/s PON systems this implies thatthe amplifier be able to regulate its gain, without loss of ,within a 26 ns timescale, (64 bits at 2.5 Gbit/s on the upstream

0733-8724/$26.00 © 2011 IEEE

3484 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 29, NO. 22, NOVEMBER 15, 2011

direction) [9]. Potentially, the dynamic control of the optical am-plifier can be implemented through the standard GPON protocolwhich provides knowledge of distance to a transmitting node.Alternatively, a rapid control circuit based on high speed mea-surements of signal strength may be appropriate.

Erbium Doped Fibre Amplifiers (EDFAs) operate reliably athigh data rates in a multi-wavelength environment, because ofits high output saturation powers and slow gain dynamics. How-ever, these same dynamics cause undesirable gain transientswhen operated within a packet switched or bursty environment[10]. Furthermore, being fibre based, EDFAs have no path tosolutions that demand a high degree of integration. Conversely,SOAs have a small form factor and offer the potential for di-rect integration with other functions on a common InP plat-form [11], [12] and more importantly their temporal dynamicsare better suited for use in switched environments. While SOAshave been demonstrated as effective amplifier solutions in multi-wavelength long distance transmission applications [12]–[14],their operation in this regime requires careful management ofsignal power levels, so as to maintain operation within theirlinear region. This additional degree of management is not desir-able, especially in PONs where bursty signals can operate overa wide dynamic range. Gain clamping has been shown to lin-earize SOA performance over a wider range of input powers,which in turn significantly reduces crosstalk between data bits(TDM crosstalk) at high transmission rates [14]. However inthese cases, gain clamping fixes the gain and as a consequencedoes not provide a solution to dynamic gain adjustment. Con-sequently the current, indirect solution is to combine a linearamplifier with a variable optical attenuator at the input whichseverely degrades the overall system noise figure. This tech-nique has also been applied to cascaded Praseodymium DopedOptical Fibre Amplifiers (PDFAs) specifically for PON applica-tions. However the modulation timescales were limited to 100sof nanoseconds [15].

The paper reports a semiconductor optical amplifier topologywhich has the unique capability to provide variable gain andmaintain linear operation through gain clamping over a wide(40 dB) dynamic range, without compromising the saturableoutput power of the device [11]. A key advantage of thisapproach is that there are no mechanical tuning elements andhence the gain can be adjusted via direct electrical controlat ns timescales. While the operation of this device has beenpresented previously for the static gain case [11], its behaviourunder dynamic gain modulation conditions is not well under-stood. Here theoretical modelling, supported by experimentalanalysis is used to validate, that the geometry can provide gainadjustment and stabilization within a nanosecond timeframe.

Fig. 2 illustrates conceptually the design of the AdjustableGain Clamped SOA (AGC-SOA) [11]. The architecture com-prises two active (gain) regions defining a data path through thesignal SOA (SOA1) and a laser cavity containing SOA1 and acontrol SOA (SOA2). SOA1 amplifies light in the signal path.The lasing mode derives gain from both SOA1 and SOA2. Thecomposite gain provided by both SOAs regulates the conditionfor the onset of lasing. This in turn defines the carrier concen-tration (gain) of the signal SOA. Hence, by controlling the driveto SOA2, the gain imparted by SOA1 can be adjusted. SOA1 is

Fig. 2. Counter propagating ring laser AGC-SOA implementation.

Fig. 3. � variation as a function of SOA gain.

continually operated at full current and therefore the AGC-SOAallows signals to be amplified by SOA1 at a clamped gain whichis varied by SOA2. This maximizes the saturation output powerthereby maintaining an extended linear regime [11].

The key advantage that the AGC-SOA offers over other op-tical amplifiers is that it enables the gain to be adjusted directlythrough the drive current to the clamping SOA without the dra-matic loss of . Hence, linear operation is maintained overa wider range of input signals. In standard SOAs it is possibleto adjust the gain by altering the drive current however, as isdemonstrated in the experimental measurement shown in Fig. 3,this leads to a dramatic loss in .

At high gains, where the SOA is highly inverted, thevalue is at its highest. However, in this region, adjusting thesmall signal gain through bias current has a dramatic effect onthe value. In the example depicted in Fig. 3, at high gainsthe changes with gain at a rate of dBm/dB- i.e., forevery dB that the gain is reduced, the value drops by 3 dBm.As the drive current is further reduced, the drop in with gainis weaker at dBm/dB. However, by the time that this pointhas been reached the value is already significantly compro-mised (5 dBm compared to the high gain value of 10 dBm). TheAGC-SOA enables gain modulation to be achieved without thisdramatic loss of value.

Fig. 4 depicts the gain of AGC-SOA as a function of outputpower for a set of different clamping currents ranging from0 mA to 200 mA. The values are constant over the range ofclamping currents despite significant gain reduction ( dB).

II. MODEL OF AGC-SOA

Several numerical models have been developed to investi-gate the characteristics of both conventional SOAs and gain-clamped SOAs using either an external or internal lasing mode.However the underlying mechanism of gain clamping achieved

LIU et al.: DYNAMIC GAIN MODULATION PERFORMANCE OF AGC-SOA 3485

Fig. 4. AGC-SOA gain as a function of output power at different clampingcurrents.

Fig. 5. Schematic of the simulation model for an AGC-SOA.

by adjusting wideband amplified spontaneous emission (ASE)power still remains largely unknown. In this study, the widebandsteady-state SOA model [16] is adapted to form a ring cavity;the evolution of travelling ASE power and spectrum within thering cavity, important for gain clamping, is then characterized.The gain, Noise Figure (NF), maximum output power at gainsaturation of an AGC-SOA under different clamping cur-rents are also studied. Based on this model, the timescale foradjusting and stabilizing the gain, crucial for dynamic packetequalization, is evaluated.

As shown in Fig. 5, ASE circulating within the AGC-SOAtravels in clockwise and counter-clockwise directions, however,the isolator in the ring cavity ensures that ASE travelling in theclockwise direction is not amplified. Thus, the counter-clock-wise ASE generated by both SOAs accounts for gain clamping.In this model, the two SOAs are simulated as independent mod-ules using different sets of material parameters summarized in

TABLE IDEVICE PARAMETERS USED IN AGC-SOA SIMULATION

Table I. (typical bulk SOA parameters from [16], [17]). BothSOAs generate ASE in the forward and backward directions inthe active regions. In each SOA, the ASE profile extends over1300 nm–1650 nm and is partitioned into discrete frequencybands.

The numerical model for the whole system of Fig. 5 isachieved using iterative circulations. In the first iteration, theASE in both directions of SOA2 is calculated assuming no ASEpower is coupled in. Then the ASE generated by SOA2 travelsin both clockwise and counter-clockwise order towards SOA1.Under this boundary condition, the ASE originated from SOA2together with the one generated by SOA1 is amplified by SOA1as it travels through, however only the backward ASE inside theSOA1 is input to SOA2. For any successive iterations, the ASEfrom SOA1 couples into SOA2 before SOA2 generates ASE.When ASE travels inside the ring cavity, the facet reflectivityand coupling loss of both SOAs, the insertion loss for the iso-lator and WDM couples are taken into account. The round triptime is ns viz. the fibre length is about 0.5 m. Therefore,fibre loss and dispersion are neglected. The iterative procedureis terminated when the maximum difference of the ASE powersat each discrete frequency band between successive iterationsis less than the desired tolerance. The numerical model isimplemented using Matlab. In the present case an amplifiercentred around 1550 nm was studied. Although it is acceptedthat in a deployed PON, the operation would be at 1300 nm,evaluation at 1550 nm was a compromise based on access tocomponents for experimental characterization. Further sincethe object of the work was to investigate the dynamic behaviorthe extrapolation to 1300 nm is assumed to be valid.

The characteristics of AGC-SOA were studied from the ini-tial state when no input signal is introduced. The ASE power

3486 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 29, NO. 22, NOVEMBER 15, 2011

Fig. 6. Counter-clockwise ASE power after N loops, Cross: SOA2 output ASEpower; Circle: SOA1 output ASE power.

after every counter-clockwise ASE round trip is recorded. InFig. 6, the counter-clockwise output ASE powers from SOA2and SOA1 are displayed after each loop transit.

The ASE power increases rapidly within the first 3 loops andthen stabilizes. Since the cavity round trip time is ns, sta-bilizing the travelling ASE power in the loop takes between1.7–5.1 ns (several round trips). The ASE spectrum within theclamping mode was examined after every circulation. Fig. 7shows the SOA2 output ASE spectrum at different loop transits.In the first loop, the output ASE power from SOA2 is relativelylow, and the whole spectrum is divided into two parts fallingoutside the C-band due to presence of the WDM coupler in thering cavity.

Initially, the ASE power within the S-band is greater than thatwithin the L-band. However, as the lasing mode becomes estab-lished, the output ASE power develops as predicted in Fig. 6.With the ASE power within S-band decreases significantly, thespectrum becoming sharp and narrow.

As ASE circulation progress, ASE emission in the S-bandis further restrained becoming negligible and ASE within theL-band accounts for gain clamping. This results from the factthat the lasing threshold is lower at longer wavelengths henceonce lasing action is established the shorter wavelength energystates are depleted. The results agree well with experimentalobservation.

Having established the steady state conditions of model op-eration, an input optical signal was introduced after the ASE in-side the AGC-SOA cavity stabilizes. The gain of a 1550 nm CWlight as a function of travelling ASE loop numbers is depictedin Fig. 8. An optical signal power of dBm was injected intoSOA1. The drive current of SOA1 was set at 200 mA, and SOA2at 65 mA. The simulation predicts that the signal gain settleswithin the first ASE loop transit and then remains unchanged;thus after AGC-SOA stabilizes from the initial state (shown inthe evolution depicted in Fig. 7), it takes ns (within one looptime) for the gain to settle.

The variation of the gain with clamping SOA drive currentwas modelled over a range of clamping currents to corroboratethat the model faithfully reproduced the experimental behaviour

Fig. 7. SOA2 ASE spectrum (counter-clockwise). (a) ���� ����� �;(b) ���� ����� ��; (c) ���� ����� �; (d) ���� �����

���.

Fig. 8. CW light (1550 nm, � � dBm) gain as a function of circulating ASEloops after AGC-SOA stabilization.

of the AGC-SOA. The model outputs (Fig. 9), indicate broadlythe model is predicting the trends. There is no significant lossof with gain reduction however the exact values of gainand differ slightly from experimental measurements. Thisdifference most likely derives from small differences betweenthe physical parameters used in the model and the real device.

The DC parametric operation of the AGC-SOA can be esti-mated using the above model. A CW light (1550 nm) was in-troduced into AGC-SOA once the steady state operation was

LIU et al.: DYNAMIC GAIN MODULATION PERFORMANCE OF AGC-SOA 3487

Fig. 9. AGC-SOA gain as a function of output power at different clampingcurrents (modeling result).

Fig. 10. Modeled steady state AGC-SOA performance.

established. The input signal power was then increased steadilyfrom dBm to 20 dBm and the normal performance metricsof gain, maximum output power at gain saturation andnoise figure (NF) were recorded for a given clamping currentcondition; the clamping current is change from 0 mA to 200 mA.The simulation results are presented as a function of clampingcurrent in Fig. 10. The overall trends given by the models are ingood agreement with previously reported experimental charac-terizations [11]. Gain clamping begins at a clamping bias currentof greater than 0 mA. Experimentally this value was observedto be nearer 10 mA before there was sufficient gain within theclamping SOA to overcome loop losses and allow the lasingmode to stabilize. This difference is mainly due to the overesti-mation of ASE noise within the model [16]. Strong clamping isobserved when the control SOA is operated at high gain levels;here the AGC-SOA is driven into attenuation. As the gain isclamped, the NF increases as the gain is clamped but in the main,the value remains relatively constant when the clampingcurrent is mA.

The main thrust of the analysis was to better understand thedynamic behaviour of the AGC-SOA. In order to do this, themodel was run under the following conditions. Firstly stable

Fig. 11. Dynamic gain variation as a function of clamping current.

operation of the AGC-SOA was ensured by running the sim-ulation with only ASE for the first 30 loop iterations and at aSOA2 (clamping SOA) bias of 20 mA. At this point a 0 dBminput signal was introduced and it can be seen that the gain of theAGC-SOA is around 10 dB. The clamping SOA bias current wasthen increased every 10 loops, from 20 mA to 200 mA, in stepsof 20 mA and the gain change observed (Fig. 11). It is clear fromthese simulations that the gain is indeed adjusted and stabilizedwithin 1 or 2 loops of the model iteration. This implies there-fore that the gain can be adjusted within nanosecond timescalesmeeting the requirements for dynamic packet equalization.

III. EXPERIMENT

Theoretical modelling of the AGC-SOA predicts that the gaincan be modified within a timescale of nanoseconds. In a dy-namic packet equalization scenario, the AGC-SOA must be ableto adjust and stabilize its gain within the period of the guardband of the PON transmission viz. 32 bit periods which repre-sents ns. For 10 Gbit/s PON systems (64 bits at 2.5 Gbit/son the upstream direction) this equates to 26 ns [9]. The modeltherefore predicts that this is feasible.

Central to attaining high speed gain modulation is the capa-bility to directly modulate the gain of the clamping SOA throughdrive current. Reflective SOAs (RSOAs), packaged in a TO can,have been designed to both amplify and modulate signals at datarates in excess of 1 Gbit/s [18]. The topology of the AGC-SOAwas therefore modified to include an RSOA as shown in Fig. 12topologically equivalent to the geometry shown in Fig. 2. Sig-nals to be amplified are input through the P (pass) port on theWDM coupler. C-band signals are passed through to the SOA.At the output of the SOA, the amplified C-band signal is passedthrough the WDM coupler on the output. Out of band signals,ASE from the S-band and L-band signals are amplified by theRSOA and used to form a counter propagating clamping mode.

The RSOA was driven with a data stream between two gainextremes viz. high gain and low gain. In the RSOA high gainstate, the gain of the AGC-SOA is highly clamped, i.e., it hasvery low gain. Conversely, when the RSOA is turned to a lowgain state, the gain of the AGC-SOA is high. The transition timebetween the two states represents the minimum time required to

3488 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 29, NO. 22, NOVEMBER 15, 2011

Fig. 12. RSOA regulated AGC-SOA.

Fig. 13. Directly modulated SOA ASE (blue), driving signal (red).

Fig. 14. Directly modulated seed signal.

adjust and settle the gain of the AGC-SOA. Fig. 13 shows thevariation in ASE signal emitted by the in-line signal SOA inresponse to directly modulating the RSOA gain. The transitionbetween a high and a low state is in response to the signal.This shows clearly that this transition time is ns.

The experiment was repeated with a steady state signal of0 dBm presented at the input to the AGC-SOA, Fig. 14 showsthat the amplitude of the seed signal has been directly modi-fied through changing the gain of the AGC-SOA. Again it isclear that the transitions between gain states are ns (2 ns perdivision timescale). The experimental characterization and themodeling are therefore in broad agreement and indicate that thegain of the AGC-SOA can be adjusted and stabilized within the

Fig. 15. Gain equalization experiment.

Fig. 16. Eye diagram of a gain modulated data signal.

26 ns timescale that would be required for packet equalizationin PONs.

High speed gain modulation of the AGC-SOA on a data signalwas demonstrated using the experimental arrangement shownin Fig. 15. Data from an Agilent N4903A J-BERT Bit ErrorRate (BER) test set was used to modulate a CWDM transmittermodule (Finisar FDB 1027) at a data rate of 1.25 Gbit/s. Thesignal amplitude was then regulated using a combination of anErbium Doped Fibre Amplifier (EDFA) and a variable opticalattenuator (VOA). The output of the VOA was then input to theAGC-SOA. The gain of the AGC-SOA was regulated by ap-plying a square wave to the RSOA via an ETS3869 laser driver,which provides control over the bias current and the datamodulation power. The output from the AGC-SOA was detectedusing a pin photoreceiver and displayed on an Agilent DigitalCommunications Analyzer (DCA). All instruments were syn-chronized from a common clock source so that modulation anddata signals could be observed at the same time.

An eye diagram of the recovered data signal is shown below inFig. 16. Indicating that the output from the SOA was presentedwith a signal level that is sufficiently high to produce a degree ofpatterning. The gain of the AGC-SOA is then rapidly reduced to

LIU et al.: DYNAMIC GAIN MODULATION PERFORMANCE OF AGC-SOA 3489

the point where the onward transmitted signal is much reducedbut is substantially cleaner. The transition from the high gainstate to the low gain state takes place over two bit periods i.e.,just over 2 ns.

IV. CONCLUSION

This paper has presented a theoretical analysis of an AGC-SOA. Both the steady state operation and the case where thegain of the AGC-SOA is dynamically modulated are addressed.The DC parametric behaviour is shown to agree well with exper-imental measurements. An adaptation of the models has shownthat in principle, the gain of the AGC-SOA can be regulateddynamically to respond within the guard band of packet basedPON transmissions. Gain settling times within the order of 2 nsare predicted. This analysis was shown to agree well with exper-imental evaluation of the AGC-SOA behaviour which demon-strated gain modulation and settling within two bit periods of a1.25 Gbit/s data signal.

A key motivation for this study was to better understand thedynamic behaviour of the AGC-SOA with a view to establishinga means for dynamically modulating the gain of the AGC-SOAin response to changes in packet amplitude. For this to be effec-tive, the gain should be adjusted and stabilized within a timespanof less than 20 ns. Both the theoretical analysis and the experi-mental investigation indicate that it is possible.

ACKNOWLEDGMENT

Strathclyde University gratefully acknowledges fundsreceived from the British Council under the Prime Minis-ters Initiative II, Research Cooperation. This work was alsosupported in part by the Glasgow Research Partnership inEngineering (GRPe).

REFERENCES

[1] R. P. Davey and D. B. Payne, “The future of optical transmission in ac-cess and metro networks—An operator’s view,” in Proc. ECOC, 2005,vol. 5, pp. 53–56.

[2] Cisco Systems, Cisco Visual Networking Index: Forecast and Method-ology, 2010–2015 Jun. 2011 [Online]. Available: http://www.cisco.com/en/US/solutions/collateral/ns341/ns525/ns537/ns705/ns827/white_paper_c11-481360.pdf

[3] S. Chatzi and I. Tomkos, “Techno-economic study of high-splittingratio PONs and comparison with conventional FTTH-PONs/FTTH-P2P/FTTB and FTTC deployments,” presented at the Optical FiberCommun., Los Angeles, CA, 2011, Paper JWA15.

[4] R. P. Davey, D. B. Grossman, M. Rasztovits-Wiech, D. B. Payne, D.Nesset, A. E. Kelly, A. Rafel, S. Appathurai, and S.-H. Yang, “Long-reach passive optical networks,” J. Lightw Technol., vol. 27, no. 3, pp.273–291, Feb. 2009.

[5] J. R. Stern, “Optical wideband subscriber loops and local area networksin the UK,” in Proc. ICC, Amsterdam, The Netherlands, May 1984, pp.884–887.

[6] L. Spiekman, D. Piehler, P. Iannone, K. Reichmann, and H. Lee,“Semiconductor optical amplifiers for FTTx,” in Proc. ICTON, 2007,pp. 48–50.

[7] R. P. Davey, P. Healey, I. Hope, P. Watkinson, D. B. Payne, O. Marmur,J. Ruhmann, and Y. Zuiderveld, “DWDM reach extension of a GPONto 135 km,” J. Lightw Technol., vol. 24, no. 1, pp. 29–31, Jan. 2006.

[8] F. J. Effenberger, “The XG-PON system: Cost effective 10 Gb/s ac-cess,” J. Lightw Technol., vol. 29, no. 4, pp. 403–409, Feb. 2011.

[9] ITU-T Recommendation G.987.2, 10-Gigabit-Capable Passive OpticalNetworks (XG-PON): Physical Media Dependent (PMD) Layer Spec-ification Jan. 2010.

[10] A. K. Srivastava, Y. Sun, J. L. Zyskind, and J. W. Sulhoff, “EDFAtransient response to channel loss in WDM transmission system,” IEEEPhoton. Technol. Lett., vol. 9, no. 3, pp. 386–388, Mar. 1997.

[11] C. Michie, A. E. Kelly, I. Armstrong, I. Andonovic, and C. Tombling,“An adjustable gain-clamped semiconductor optical amplifier (AGC-SOA),” J. Lightw Technol., vol. 25, no. 6, pp. 1466–1473, Jun. 2007.

[12] K. Wakao, H. Soda, and Y. Kotaki, “Semiconductor optical active de-vices for photonic networks,” FUJITSU Sci. Tech. J., vol. 35, no. 1, pp.100–106, Jul. 1999.

[13] G. Onishchukov, V. Lokhnygin, A. Shipulin, and P. Riedel, “10 Gbit/stransmission over 1500 km with semiconductor optical amplifiers,”Electron. Lett., vol. 34, no. 16, pp. 1597–1598, Aug. 1998.

[14] L. H. Spiekman, J. M. Wiesenfeld, A. H. Gnauck, L. D. Garrett, G. N.Van Den Hoven, T. Van Dongen, M. J. H. Sander-Jochem, and J. J. M.Binsma, “Transmission of 8 DWDM channels at 20 GB/s over 160 kmof standard fiber using a cascade of semiconductor optical amplifiers,”IEEE Photon. Technol. Lett., vol. 12, no. 6, pp. 717–719, Jun. 2000.

[15] M. Fujiwara, K. Suzuki, K. Hara, T. Imai, K. Taguchi, H. Ishii, N.Yoshimoto, and H. Hadama, “Burst-mode compound optical amplifierwith automatic level control circuit that offers enhanced setting flexi-bility in a 10 Gb/s-class PON,” in Proc. ECOC, 2010, pp. 1–3.

[16] M. J. Connelly, “Wideband semiconductor optical amplifier steady-state numerical model,” IEEE J. Quantum Electron., vol. 37, no. 3, pp.439–447, Mar. 2001.

[17] S. L. Chuang, Physics of Optoelectronic Devices. New York: WileyInterscience, 1995.

[18] D. C. Kim, B. S. Choi, H. S. Kim, K. S. Kim, O. K. Kwon, and D. K.Oh, “2.5 Gbps operation of RSOA for low cost WDM-PON sources,”in Proc. ECOC, 2009, pp. 1–2.

Lin Liu (S’11) was born in Zhangjiagang, China, in 1983. He received theB.S. degree in telecommunication engineering and M.S. degree in optical en-gineering from Nanjing University of Posts and Telecommunications, China, in2005 and 2008 respectively. He is currently working toward the Ph.D. degreeat the Centre for Intelligent Dynamic Communications (CIDCOM), Departmentof Electronic and Electrical Engineering, University of Strathclyde, Strathclyde,U.K.

His current research interests include modelling of semiconductor optoelec-tronics devices and passive optical networks.

Craig Michie, biography not available at the time of publication.

Anthony E. Kelly, biography not available at the time of publication.

Ivan Andonovic (SM’97–F’11) received the B.Sc. (Hons.) degree in electronicand electrical engineering from the University of Strathclyde, Strathclyde, U.K.,in 1978, and the Ph.D. degree in lithium niobate waveguide devices in conjunc-tion with Glasgow University, Glasgow, U.K., in 1982.

He is presently the Head of the Centre for Intelligent Dynamic Commu-nications Systems (CIDCOM), Department of Electronic and Electrical En-gineering, Strathclyde University. He joined the University of Strathclyde in1985 following several years at Barr and Stroud, where he was responsible forthe design, manufacture, and test of optical guided wave devices. He has helda two-year Royal Society Industrial Fellowship, in collaboration with BritishTelecommunications (BT) Labs, during which he investigated novel approachesto broadband networking. He has edited two books and authored/coauthoredsix chapters in books and over 300 journal and conference papers. He has beenchairman on the IET Professional Group E13, held a BT Short Term Fellow-ship, Visiting Scientist status with the Communications Research Laboratoriesof Japan, and Visiting Professor with the City University of Hong Kong andPrinceton University. He is Topical Editor for the IEEE TRANSACTIONS ON

COMMUNICATIONS and was Technical Programme Co-Chair for the recent IEEEInternational Conference in Communications. He was cofounder, Director, andChief Technology Officer of Kamelian Ltd., a high-growth technology start-upfocussing on the design and manufacture of advanced semiconductor devices.He was also a member of flagship Scottish Enterprise (government agency foreconomic growth) establishment team of the Intermediary Technology Institutes(ITIs), aimed at bridging the gap between basic research and company growth.His research interests center on the development of broadband networks, accessand home networking and wireless sensor systems.

Prof. Andonovic is a Fellow of the IET and a member of the Optical Societyof America.