Electrical-to-ical conversion of OFDM 802.11g/a signals by direct current modulation of semiconductor ical amplifiers Francesco Vacondio, Marco Michele Sisto, Walid Mathlouthi, Leslie Ann Rusch and Sophie LaRochelle Centre d ique, photonique et laser (COPL), Department of Electrical and Computer Engineering, Université Laval, Québec, Canada. Abstract We report experimental results showing direct current modulation of a commercial semiconductor ical amplifier (SOA) for electrical to ical conversion of a 64- QAM OFDM analog signal at 54 Mbit/s on 2.5 and 5 GHz carriers. This is believed to be the highest frequency yet experimentally reported for RF signal transmission using a directly modulated semiconductor ical amplifier. reception (b). Optical versions of the RF signals arriving via fiber from the central office to the mobile (downlink) are amplified by the SOA at the base station before photodetection; the photocurrent drives the antenna with the RF downlink signal. Index Terms Optical modulation, microwave communication, ical fiber communication, radio-overfiber (RoF), semiconductor ical amplifiers. I. TRODUCTION The main challenges in radio-over-fiber (RoF) transparent ical links at the physical layer, especially for wireless access networks [1], are the electro-ical (E/O) and the o-electrical (O/E) conversions at the antenna sites. In order to reduce antenna site complexity, and therefore infrastructure cost, O/E and E/O conversions have to be as low cost and efficient as possible, especially for multiple antenna systems, where the cost of the conversion has to be multiplied by the number of antenna elements. To address this problem, we investigate the possibility of using a single SOA as a dual-purpose device (ical amplifier and modulator) in RoF base stations. The SOA could be used for amplification in the downlink, and as an external modulator for a distributed ical source in the uplink [2]. We focus on the popular IEEE 802.11g/a standard that uses orthogonal frequency division multiplexing (OFDM) signals at a maximum transmission rate of 54 Mbit/s. Results show that it is possible to achieve a quality of service of 10-6 BER in the uplink at both 2.5 GHz and 5 GHz using a deeply saturated SOA. However, the saturation level imposes a trade-off: deeper saturation gives less distortion (as measured via error vector magnitude) but less received ical power (for a fixed ical input power). II. THE SOA-BASED BASE STATION Fig.1 presents the principles of operation of the proposed RoF base station during transmission (a) and Fig. 1. Schematic diagram of RoF base station for the downlink (a) and uplink (b) When the antenna is receiving (uplink), the signal coming from the antenna is amplified and summed to a DC current (via a bias tee) to control the injection current of the SOA. In this case a continuous wave (CW) ical signal arriving from the central office will be modulated by current modulation of the SOA. The ideal characteristics of the SOA for the two functions (amplifier and modulator) are disparate: for amplification, an SOA with a long gain recovery time is preferable [3], while for modulation, especially at high frequencies, an SOA capable of fast recovery of the gain is required. Our SOA has a recovery time around 350 psec. In this paper we will focus on the uplink; studying the performance of the SOA as a modulator for analog signals such as OFDM signals on 2.5 and 5 GHz carriers in order to investigate the feasibility of the proposed base station and to understand the constraints imposed on the modulation function by the SOA.
III. EXPERIMENTAL SETUP Fig. 2 shows the experimental setup of the analog OFDM signal conversion. The output of a laser diode at 1562 nm, where the SOA amplification gain is maximum, is injected into the SOA (Optospeed 1550MRI) through a variable attenuator and a polarization controller (our SOA is a polarization sensitive amplifier). The SOA bias current is supplied through a bias-tee in order to sum a DC term and an OFDM modulated radio signal, generated by a vector signal generator (Agilent E-4440A). The signal generator impedance is matched to the biasing circuit by a triple stub, in order to ensure that available signal power from the generator is transferred to the SOA chip in the most efficient manner. Impedance matching is extremely helpful in our case because of the complex impedance presented by the SOA, affected not only by the SOA chip itself, but also by the packaging and the biasing circuit. Moreover, we can employ low loss narrowband impedance matching, thus further improving signal transfer to the SOA, because of the narrowband nature of the OFDM signals (less than 20 MHz around either a 2.5 GHz or 5 GHz carrier). P OUT P P Fig. 2. Experimental Setup (LD: laser diode, VOA: variable ical attenuator, PC: polarization controller, IM: impedance matching, ISO: isolator, BPF: band-pass filter, : photodiode, TIA: trans-impedance amplifier). Gain Gain [db] 20 15 10 5-20 -15-10 -5 0 P [dbm] 82.5 20 55150 325 500-20 27.5 0 Gain [db] 0 240 280 320 360 400 440 480 Fig. 3. a) Gain saturation curve of the SOA biased at 460 ma b) Gain vs. current curve of the SOA for the interesting values of currents for the application; in the inset the same curve is plotted in db and for a larger current range. Two values of P are considered: -30dBm (solid) and -10 dbm (dashed) The input ical power P is held constant while the SOA current is modulated by the data signal. The SOA gain is modulated via the current signal, transferring the modulation to the output ical power P OUT [4-5]. The modulated output passes through an isolator to prevent harmful reflections, a 0.24 nm band-pass filter to select the laser at 1562nm, and a second variable ical attenuator for controlling the power on the photodiode P before the signal analyzer (Agilent E4440A). As a reference, some of the most important characteristics of our SOA are shown in Fig. 3. The gain saturation can be seen in Fig. 3-a, and the variation of the gain with the SOA injection current in Fig. 3-b. The inset shows the gain in db, while the main plot is a zoom of the inset in linear scale, in the zone where the SOA has significant gain. IV. LEARITY AND NOISE PERFORMANCES OF THE SOA MODULATOR Since we are dealing with analog complex signals, the linearity of the modulator is the first issue to be investigated. Ours being a narrowband (sub-octave) application, we are only interested in odd-order distortion products, as the even ones can be easily filtered electronically at the receiver. Furthermore, SOA add noise. Thus, in order to identify an imum operating point we will focus on three parameters: the power of the fundamental component, the 3rd order inter-modulation (ID3) products and the carrier to noise ratio. Recall that when the SOA is saturated the noise level decreases because the amplified spontaneous emission (ASE) level decreases [6]. Fig. 4-a shows the result of the inter-modulation distortion measurements. For these measurements the vector signal generator produced two tones with 1 MHz spacing around 2.5 GHz, and we measured the ratio between the powers of the received tones versus the ID3 products expressed in dbc, i.e. power relative to the carrier (fundamental). The vector signal generator was calibrated to assure accurate measurements; ID3 could be measured to levels as low as -80dBc. We varied the bias point and the level of saturation of the SOA (by varying i DC and ical power P ) for a fixed modulation power of +8dBm per tone (+11 dbm total). The power of the fundamental component increases with the saturation level because of the greater power on the photodiode; at the same time the ID3 increases because we are in the saturation regime. The shape of the curves is affected by the slope of the curve in Fig. 3-b (linear gain versus the bias current): the fundamental increases with the bias level, and the non-linearity is higher when the linear gain slope is rapidly changing with bias current. 2
Note that CNR is given in db/hz, since the noise is integrated over a 1 Hz bandwidth. In Fig. 4-b the vector signal generator produced one tone at 2.5 GHz, and we measured the carrier-to-noise ratio (CNR) on the signal analyzer varying the SOA bias point, and for various ical input powers. We see in Fig. 4-b the CNR increases monotonically to some maximum value at moderate pump currents. For high pump bias current the CNR stays fairly constant, although small degradation is visible at low saturation. As the SOA pump current increases the desired signal (the fundamental) increases to a maximum. Once reaching moderate pump current (300 ma), the principal noise source is ASE and its beating with the input signal. This noise decreases with deeper saturation, hence the improvement in CNR with increasing pump current and saturation. Power on fundamental [dbm] -28-33.4-38.8-44.2-49.6-30 -35-40 -45-50 -55 240 280 320 360 400 440 480-55 ID3 [dbc] 115 105 100 95 90 85 80 240 280 320 360 400 440 480 75 Fig. 4. a) Power of the fundamental and relative 3rd order inter-modulation (ID3) distortion and b) CNR versus bias current i DC for different ical input powers to the SOA P = -5 dbm (Δ), -10dBm (x), -15 dbm (o), -20 dbm ( ) and -25 dbm ( ). Since we have low non-linearity (even for RF powers as high as +8dBm per tone, the ID3 is always below -42 dbc), and an almost flat CNR, for the next sections we choose the working point i DC = 460mA, in order to maximize SOA gain. CNR [db/hz] power P (adjusting VOA1), corresponding to the ical power available from the remote, distributed CW laser source. Fig. 5 shows the bit-error rate results of our analog OFDM signal E/O conversion: dotted lines correspond to the BER without forward error correction (FEC); solid lines correspond to the BER after correction, using the IEEE 802.11 standard FEC with a hard-decision Viterbi decoder. In Fig. 5 the vector signal generator provided an OFDM signal on a 2.5 GHz carrier. As P increases the SOA becomes saturated, and we can see that the impact of amplifier saturation on the E/O conversion is positive, with decreased BER for greater ical powers. BER 10 0 10-1 10-2 10-3 Increasing P P = -14dBm, with FEC 10-4 P = -14dBm, no FEC P = -10dBm, with FEC 10-5 P = -10dBm, no FEC P = -6dBm, with FEC P = -6dBm, no FEC 10-6 -20-15 -10-5 0 5 [dbm] P = -10 dbm Fig. 5. BER performances for the O/E conversion on a 2.5 GHz carrier for fixed ical power on the and increasing saturation levels P = -14(o), -10( ), -6( Δ ) dbm. In Fig. 6 we plot the error vector magnitude (EVM) results for various modulation powers against the ical input power. The dash-dotted line corresponds to a BER lower than 10-6 on the coded bits (EVM = -22.5 db, or 7.5%, inferred from the BER measurements of Fig. 5). The figure can be compared to Fig. 3-a, to see that the greater SOA saturation level leads to better performances in terms of EVM (better modulator), and to a loss in ical gain (worse amplifier). V. PERFORMANCE OF SOA MODULATOR WITH OFDM SIGNALS Our figure of merit is the bit error rate (BER) versus the RF power modulating the SOA pump current around its bias point of i DC =460mA. The ical power incident = -10 dbm is held constant by adjusting VOA2, so that variations in available gain from the SOA do not interfere with interpretation of our results. We consider three values for the input ical to the photodetector P 3
EVM [db] -5-10 -15-20 -25-30 -35-20 -18-16 -14-12 -10-8 -6-4 -2 0 P [dbm] = -10dBm = -5dBm = 0dBm = +5dBm Fig. 6. EVM for fixed ical power on the : P = -10 dbm. The points under the dotted line correspond to BER<10-6 with FEC for a 2.5GHz carrier. Interestingly, we observe that performance improves when the SOA is operated in saturation. The 3rd order distortion is low enough that operating in the nonlinear saturation regime does not degrade the OFDM signal quality. which is always the dominating noise component, except in deep saturation. In order to establish the functionality of the SOA as an external modulator for 802.11a (at a carrier frequency of 5 GHz), Fig. 8 shows the comparison of the performance of the conversion with carrier frequency of 2.5 GHz and 5 GHz, where we fixed the saturation level at the amplifier corresponding to the best case in Fig. 5, P = -6 dbm. In Fig. 8 the SOA is saturated (its input power is -6 dbm) and the received ical power is -10 dbm. The total attenuation introduced is then 17dB (the gain of the SOA being 13dB). The penalty to be paid in passing from a 2.5 GHz carrier to a 5 GHz carrier in terms of to achieve BER = 10-5 on the bits after FEC is 14 db. Note that the same BER could be achieved with less RF power and a deeper saturation level, or less RF power and less attenuation. In fact the combination of the free parameters in this application affects the performance, and the tradeoff is between the three of them: ical power at the input of the SOA P, the RF power, and the received power P. 10 0-100 10-1 100 90-105 - -115 BER 10-2 10-3 CNR [db/hz] 80 70 60 50 = -10dBm = -5dBm = 0dBm = +5dBm -120-125 -130-135 -140-145 40-20 -18-16 -14-12 -10-8 -6-4 -2 0-150 P [dbm] Fig. 7. CNR and noise for fixed ical power on the : P = -10 dbm. Dashed line is noise floor given by the sum of shot, thermal and laser relative intensity noise (R). The enhancement in performance can be justified by Fig. 7, where we plot the CNR and the noise floor as a function of the ical input power to the SOA P, for various RF powers. The CNR improves because of the lowering of the noise, as we can argue from the dotted line (for a fixed PRF, the power of the carrier does not change with P ). Since the power on the photodiode is constant ( P =-10dBm), the contribution of shot noise, thermal noise and of laser relative intensity noise (R) is constant. The noise that decreases with the saturation level is then the signal-ase beating that falls at 2.5 GHz, Noise [dbm/hz] 10-4 2.5 GHz, with FEC 10-5 2.5 GHz, no FEC 5 GHz, with FEC 5 GHz, no FEC 10-6 -20-15 -10-5 0 5 10 [dbm] Fig. 8. BER performances for the O/E conversion on a 5 GHz carrier fixed saturation level of the amplifier and increasing received powers P = -10 dbm. VI. CONCLUSIONS We demonstrated that electro-ical conversion of analog signals such as 64-QAM, 54 Mb/s OFDM signals on 2.5 and 5 GHz carriers is possible by direct injection current modulation of an SOA. The SOA can thus be used for amplification in the downlink, and as an external modulator for a distributed ical source in the uplink. REFERENCES [1] A. J. Seeds, Wireless access over ical fibre: from cellular radio to broadband; from UHF to millimetre-waves Proc. LEOS 2002, pp.471-471, Nov. 2002. 4
[2] J. M. Kang, Y. Y. Won, S. H. Lee and S. K. Han, Modulation characteristics of RSOA in hybrid WDM/SCM-PON ical link Proc. OFC 2006, paper JThB68, Mar. 2006. [3] W. Mathlouthi, P. Lemieux, L. A. Rusch, Optimal SOA-based noise reduction schemes for incoherent spectrum-sliced PONs, submitted to ECOC 2006. [4] J. Mørk, A. Mecozzi and G. Eisenstein, The modulation response of a semiconductor laser amplifier, Journal of Selected Topics in Quantum Electronics, vol.5, pp. 851-860, May/June 1999. [5] J. Yao, N. G. Walker, C. Hunter, G. R. Walker, G Sherlock, 5 Gbit/s transmission system using a MQW semiconductor ical amplifier as an amplitude modulator, Proc. IEE Colloqium Sources for Very High Bit Rate Optical Communications Systems, pp. 6/1-6/5, Apr. 1992. [6] M. J. Connelly, Wideband Semiconductor ical amplifier steady-state numerical model, Journal of Quantum Electronics, vol.37, No.3, pp. 439 447, Mar. 2001. [7] T. G. Hodgkinson and R. P. Webb, Carrier-density modulation effects in a travelling wave semiconductor ical amplifier: communication theory analysis and experiment, Journal of Lightwave Technology, vol.9, No. 5, pp. 605 622, May 1991. 5