Darcie, T.E., Palais, J.C., Kaminow, I.P. “Optical Communication” The Electrical Engineering Handbook Ed. Richard C. Dorf Boca Raton: CRC Press LLC, 2000 71 Optical Communication 71.1 Lightwave Technology for Video Transmission Video Formats and Applications?Intensity Modulation?Noise Limitations?Linearity Requirements?Laser Linearity?Clipping?External Modulation?Miscellaneous Impairments?Summary 71.2 Long Distance Fiber Optic Communications Fiber?Modulator?Light Source?Source Coupler?Isolator?Connectors and Splices?Optical Amplifier?Repeater?Photodetector?Receiver?Other Components?System Considerations?Error Rates andSignal-to- Noise Ratio?System Design 71.3 Photonic Networks Data Links?Token Ring: FDDI, FFOL?Active Star Networks: Ethernet, Datakit ò ?New Approaches to Optical Networks 71.1 Lightwave Technology for Video Transmission T. E. Darcie Lightwave technology has revolutionized the transmission of analog and, in particular, video information. Because the light output intensity from a semiconductor laser is linearly proportional to the injected current, and the current generated in a photodetector is linearly proportional to the incident optical intensity, analog information is transmitted as modulation of the optical intensity. The lightwave system is analogous to a linear electrical link, where current or voltage translates linearly into optical intensity. High-speed semiconductor lasers and photodetectors enable intensity-modulation bandwidths greater than 10 GHz. Hence, a wide variety of radio frequency (RF) and microwave applications have been developed [Darcie, 1990]. Converting microwaves into intensity-modulated (IM) light allows the use of optical fiber for transmission in place of bulky inflexible coaxial cable or microwave waveguide. Since the fiber attenuation is 0.2–0.4 dB/km, compared with several decibels per meter for waveguide, entirely new applications and architectures are possible. In addition, the signal is confined tightly to the core of single-mode fiber, where it is immune to electromagnetic interference, cross talk, or spectral regulatory control. To achieve these advantages, several limitations must be overcome. The conversion of current to light intensity must be linear. Several nonlinear mechanisms must be avoided by proper laser design or by the use of various linearization techniques. Also, because the photon energy is much larger than in microwave systems, the signal fidelity is limited by quantum or shot noise. This section describes the basic technology for the transmission of various video formats. We begin by describing the most common video formats and defining transmission requirements for each. Sources of noise, including shot noise, relative intensity noise (RIN), and receiver noise are then quantified. Limitations imposed by source nonlinearity, for both direct modulation of the laser bias current and external modulation using an interferometric LiNbO 3 modulator, are compared. Finally, several other impairments caused by fiber non- linearity or fiber dispersion are discussed. T.E. Darcie AT&T Bell Laboratories Joseph C. Palais Arizona State University Ivan P. Kaminow AT&TBell Laboratories ? 2000 by CRC Press LLC Video Formats and Applications Each video format represents a compromise between transmission bandwidth and robustness or immunity to impairment. With the exception of emerging digital formats, each is also an entrenched standard that often reflects the inefficiencies of outdated technology. FM Video Frequency-modulated (FM) video has served for decades as the basis for satellite video transmission [Pratt and Bostian, 1986], where high signal-to-noise ratios (SNRs) are difficult to achieve. Video information with a bandwidth of B v = 4.2 MHz is used to FM modulate an RF carrier. The resulting channel bandwidth B is given by B ; Df pp + 2f m (71.1) where Df pp is the frequency deviation (22.5 MHz) and f m is the audio subcarrier frequency (6.8 MHz). As a result of this bandwidth expansion to typically 36 MHz, a high SNR can be obtained for the baseband video bandwidth B v even if the received carrier-to-noise ratio (CNR) over the FM bandwidth B is small. The SNR is given by (71.2) where W is a weighting factor (13.8 dB) that accounts for the way the eye responds to noise in the video bandwidth, and PE is a pre-emphasis factor (0–5 dB) that is gained by emphasizing the high-frequency video components to improve the performance of the FM modulator. High-quality video (SNR = 55 dB) requires a CNR of only 16 dB. This is achieved easily in a lightwave transmission system. Applications for lightwave FM video transmission include links to satellite transmission facilities, transport of video between cable television company head-ends (super-trunking), and perhaps delivery of video to subscribers over large fiber distribution networks [Way et al., 1988; Olshansky et al., 1988]. AM-VSB Video The video format of choice, both for broadcast and cable television distribution, is AM-VSB. Each channel consists of an RF carrier that is amplitude modulated (AM) by video information. Single-sideband vestigial (VSB) filtering is used to minimize the bandwidth of the modulated spectrum. The resultant RF spectrum is dominated by the remaining RF carrier, which is reduced by typically 5.6 dB by the AM, and contains relatively low-level signal information, including audio and color subcarriers. An AM-VSB channel requires a bandwidth of only 6 MHz, but CNRs must be at least 50 dB. For cable distribution, many channels are frequency-division multiplexed (FDM), separated nominally by 6 MHz (8 MHz in Europe), over the bandwidth supported by the coaxial cable. A typical 60-channel cable system operates between 55.25 and 439.25 MHz. Given the large dynamic range required to transmit both the remaining RF carrier and the low-level sidebands, transmission of this multichannel spectrum is a challenge for lightwave technology. The need for such systems in cable television distribution systems has motivated the development of suitable high-performance lasers. Before the availability of lightwave AM-VSB systems, cable systems used long (up to 20 km) trunks of coaxial cable with dozens of cascaded electronic amplifiers to overcome cable loss. Accumu- lations of distortion and noise, as well as inherent reliability problems with long cascades, were serious limi- tations. Fiber AM-VSB trunk systems can replace the long coaxial trunks so that head-end quality video can be delivered deep within the distribution network [Chiddix et al., 1990]. Inexpensive coaxial cable extends from the optical receivers at the ends of the fiber trunks to each home. Architectures in which the number of electronic amplifiers between each receiver and any home is approximately three or fewer offer a good compromise between cost and performance. The short spans of coaxial cable support bandwidths approaching 1 GHz, two SNR CNR W PE=+ ? è ? ? ? ÷ é ? ê ê ù ? ú ú ++10 3 2 log B B f B v pp v D ? 2000 by CRC Press LLC or three times the bandwidth of the outdated long coaxial cable trunks. With fewer active components, reliability is improved. The cost of the lightwave components can be small compared to the overall system cost. These compelling technical and economic advantages resulted in the immediate demand for lightwave AM-VSB systems. Compressed Digital Video The next generation of video formats will be the product of compressed digital video (CDV) technology [Netravali and Haskel, 1988]. For years digital “NTSC-like” video required a bit rate of approximately 100 Mbps. CDV technology can reduce the required bit rate to less than 5 Mbps. This compression requires complex digital signal processing and large-scale circuit integration, but advances in chip and microprocessor design have made inexpensive implementation of the compression algorithms feasible. Various levels of compression complexity can be used, depending on the ultimate bit rate and quality required. Each degree of complexity removes different types of redundancy from the video image. The image is broken into blocks of pixels, typically 8 ′ 8. By comparing different blocks and transmitting only the differences (DPCM), factors of 2 reduction in bit rate can be obtained. No degradation of quality need result. Much of the information within each block is imperceptible to the viewer. Vector quantization (VQ) or discrete-cosine transform (DCT) techniques can be used to eliminate bits corresponding to these imperceptible details. This intraframe coding can result in a factor of 20 reduction in the bit rate, although the evaluation of image quality becomes subjective. Finally, stationary images or moving objects need not require constant retransmission of every detail. Motion compression techniques have been developed to eliminate these interframe redundancies. Combinations of these techniques have resulted in coders that convert NTSC-like video (100 Mbps uncom- pressed) into a few megabits per second and HDTV images (1 Gbps uncompressed) into less than 20 Mbps. CDV can be transmitted using time-division multi- plexing (TDM) and digital lightwave systems or by using each channel to modulate an RF carrier and transmitting using analog lightwave systems. There are numerous applications for both alternatives. TDM systems for CDV are no different from any other digital transmission sys- tem and will not be discussed further. Using RF techniques offers an additional level of RF compression, wherein advanced multilevel modulation formats are used to maximize the number of bits per hertz of bandwidth [Feher, 1987]. Quadrature-amplitude modulation (QAM) is one example of multilevel digital- to-RF conversion. For example, 64-QAM uses 8 ampli- tude and 8 phase levels and requires only 1 Hz for 5 bits of information. As the number of levels, hence the num- ber of bits per hertz, increases, the CNR of the channel must increase to maintain error-free transmission. A 64- QAM channel requires a CNR of approximately 30 dB. A synopsis of the bandwidth and CNR requirements for FM, AM-VSB, and CDV is shown in Fig. 71.1. AM- VSB requires high CNR but low bandwidth. FM is the opposite. Digital video can occupy a wide area, depending on the degree of digital and RF compression. The com- bination of CDV and QAM offers the possibility of squeezing a high-quality video channel into 1 MHz of bandwidth, with a required CNR of 30 dB. This drastic improvement over AM-VSB or FM could have tremen- dous impact on future video transmission systems. FIGURE 71.1Bandwidth versus carrier-to-noise ratio (CNR) required for AM-VSB, FM, and digital video. Increasingly complex digital compression techniques reduce the bit rate required for NTSC-like video from 100 Mbps to less than 5 Mbps. Bandwidth efficient RF techniques like QAM minimize the bandwidth required for each bit rate but require greater CNRs. ? 2000 by CRC Press LLC Intensity Modulation As mentioned in the introduction, the light output from the laser should be linearly proportional to the injected current. The laser is prebiased to an average output intensity L 0 . Many video channels are combined electron- ically, and the total RF signal is added directly to the laser current. The optical modulation depth (m) is defined as the ratio of the peak modulation L 0 for one channel, divided by L 0 . For 60-channel AM-VSB systems, m is typically near 4%. The laser (optical carrier) is modulated by the sum of the video channels that are combined to form the total RF signal spectrum. The resultant optical spectrum contains sidebands from the IM superimposed on unintentional frequency modulation, or chirp, that generally accompanies IM. This complex optical spectrum must by understood if certain subtle impairments are to be avoided. A photodetector converts the incident optical power into current. Broadband InGaAs photodetectors with responsivities (R 0 ) of nearly 1.0 A/W and bandwidths greater than 10 GHz are available. The detector generates a dc current corresponding to the average received optical power L r and the complete RF modulation spectrum that was applied at the transmitter. An ac-coupled electronic preamplifier is used to remove the dc component and boost the signal to usable levels. Noise Limitations The definition of CNR deserves clarification. Depending on the video format and RF modulation technique, the RF power spectrum of the modulated RF carrier varies widely. For AM-VSB video the remaining carrier is the dominant feature in the spectrum. It is thereby convenient to define the CNR as the ratio of the power remaining in the carrier to the integrated noise power in a 4-MHz bandwidth centered on the carrier frequency. For FM or digitally modulated carriers, the original carrier is not generally visible in the RF spectrum. It is then necessary to define the CNR as the ratio of the integrated signal power within the channel bandwidth to the integrated noise power. Shot Noise Shot noise is a consequence of the statistical nature of the photodetection process. It results in a noise power spectral density, or electrical noise power per unit bandwidth (dBm/Hz) that is proportional to the received photocurrent I r (= R 0 L r ). The total shot noise power in a bandwidth B is given by N s = 2eI r B (71.3) where e is the electronic charge. With small m, the detected signal current is a small fraction of the total received current. The root mean square (rms) signal power for one channel is (71.4) The total shot noise power then limits the CNR (P s /N s ) to a level referred to as the quantum limit. Received powers near 1 mW are required if CNRs greater than 50 dB are to be achieved for 40- to 80-channel AM-VSB systems. Receiver Noise Receiver noise is generated by the electronic amplifier used to boost the detected photocurrent to usable levels. The easiest receiver to build consists of a pin photodiode connected directly to a low-noise 50- to 75-W amplifier, as shown in Fig. 71.2(a). The effective input current noise density, (n), for this simple receiver is given by (71.5) PmI sr = 1 2 2 () n kTF R L 2 4 = ? 2000 by CRC Press LLC where k is the Boltzmann constant, T is the absolute temperature, F is the noise figure of the amplifier, and R L is the input impedance. For a 50-W input impedance and F = 2, n = 20 pA/ . A variety of more complicated receiver designs can reduce the noise current appreciably [Kasper, 1988]. The example shown in Fig. 71.2(b) uses a high-speed FET. R L can be increased to maximize the voltage developed by the signal current at the FET input. Input capacitance becomes a limitation by shunting high-frequency com- ponents of signal current. High-frequency signals are then reduced with respect to the noise generated in the FET, resulting in poor high- frequency performance. Various impedance matching techniques have been proposed to maximize the CNR for specific frequency ranges. Relative Intensity Noise Relative intensity noise (RIN) can originate from the laser or from reflections and Rayleigh backscatter in the fiber. In the laser, RIN is caused by spontaneous emission in the active layer. Spontaneous emission drives random fluctuations in the number of photons in the laser which appear as a random modulation of the output inten- sity, with frequency components extending to tens of gigahertz. The noise power spectral density from RIN is I 2 r RIN, where RIN is expressed in decibels per hertz. RIN is also caused by component reflections and double-Rayleigh backscatter in the fiber, by a process called multipath interference. Twice-reflected signals arriving at the detector can interfere coherently with the unre- flected signal. Depending on the modulated optical spectrum of the laser, this interference results in noise that can be significant [Darcie et al., 1991]. The CNR, including all noise sources discussed, is given by (71.6) All sources of intensity noise are combined into RIN. Increasing m improves the CNR but increases the impairment caused by nonlinearity, as discussed in the following subsection. The optimum operating value for m is then a balance between noise and distortion. Figure 71.3 shows the noise contributions from shot noise, receiver noise, and RIN. For FM or digital systems, the low CNR values required allow operation with small received optical powers. Receiver noise is then generally the limiting factor. Much larger received powers are required if AM-VSB noise requirements are to be met. Although detecting more optical power helps to overcome shot and receiver noise, the ratio of signal to RIN remains constant. RIN can be dominant in high-CNR systems, when the received power is large. AM-VSB systems require special care to minimize all sources of RIN. The dominant noise source is then shot noise, with receiver noise and RIN combining to limit CNRs to within a few decibels of the quantum limit. Linearity Requirements Source linearity limits the depth of modulation that can be applied. Linearity, in this case, refers to the linearity of the current-to-light-intensity (I-L) conversion in the laser or voltage-to-light (V-L) transmission for an external modulator. Numerous nonlinear mechanisms must be considered for direct modulation, and no existing external modulator has a linear transfer function. A Taylor-series expansion of the I-L or V-L characteristic, centered at the bias point, results in linear, quadratic, cubic, and higher-order terms. The linear term describes the efficiency with which the applied signal is converted FIGURE 71.2 Receivers for broadband analog lightwave systems. Coupling a pin to a low-noise amplifier (a) is simple, but improved performance can be obtained using designs like the pin FET (b). C t is the undesirable input capacitance. Hz CNR RIN ] = ++ mI Bn eI I r r r 22 22 22 [ ? 2000 by CRC Press LLC to linear intensity modulation. The quadratic term results in second-order distortion, the cubic produces third- order distortion, and so on. Requirements on linearity can be derived by considering the number and spectral distribution of the distortion products generated by the nonlinear mixing between carriers in the multichannel signal. Second- order nonlinearity results in sum and difference (f i ± f j ) mixing products for every combination of the two channels. This results in as many as 50 second-order products within a single channel, in a 60-channel AM- VSB system with the standard U.S. frequency plan. Similarly, for third-order distortion, products result from mixing among all combinations of three channels. However, since the number of combinations of three channels is much larger than for two, up to 1130 third-order products can interfere with one channel. The cable industry defines the composite second-order (CSO) distortion as the ratio of the carrier to the largest group of second- order products within each channel. For third-order distortion, the composite triple beat (CTB) is the ratio of the carrier to the total accumulation of third-order distortion at the carrier frequency in each channel. The actual impairment from these distortion products depends on the spectrum of each RF channel and on the exact frequency plan used. A typical 42-channel AM-VSB frequency plan, with carrier frequencies shown as the vertical bars on Fig. 71.4, results in the distributions of second- and third-order products shown in Fig. 71.4(a) and (b), respectively. Since the remaining carrier is the dominant feature in the spectrum of each channel, the distortion products are dominated by the mixing between these carriers. Because high-quality video requires that the CSO is –60 dBc (dB relative to the carrier), each sum or difference product must be less than –73 dBc. Likewise, for the CTB to be less than 60 dB, each product must be less than approximately –90dB. FIGURE 71.3 Current noise densities from receivers, RIN, and shot noise as a function of total received photocurrent. Receiver noise is dominant in FM or some digital systems where the total received power is small. The solid line for receiver noise represents the noise current for a typical 50-W low-noise amplifier. More sophisticated receiver designs could reduce the noise to the levels shown approximately by the dotted lines. RIN and shot noise are more important in AM-VSB systems. FIGURE 71.4 Second-order (a) and third-order (b) distortion products for 42-channel AM-VSB system. The maximum number of second-order products occurs at the lowest frequency channel, where 30 products contribute to the CSO. The maximum number of third-order products occurs near the center channel, where 530 products contribute to the CTB. ? 2000 by CRC Press LLC e- FM or CDV systems have much less restrictive linearity requirements, because of the reduced sensitivity to impairment. Distortion products must be counted, as with the AM-VSB example described previously, but each product is no longer dominated by the remaining carrier. Because the carrier is suppressed entirely by the modulation, each product is distributed over more than the bandwidth of each channel. The impairment resulting from the superposition of many uncorrelated distortion products resembles noise. Quantities analo- gous to the CSO and CTB can be defined for these systems. Laser Linearity Several factors limit the light-versus-current (L-I) linearity of directly modulated lasers. Early work on laser dynamics led to a complete understanding of resonance-enhanced distortion (RD). RD arises from the same carrier-photon interaction within the laser that is responsible for the relaxation-oscillation resonance. The second-harmonic distortion (2f i ) and two-tone third-order distortion (2f i – f j ) for a typical 1.3-mm wavelength directly modulated semiconductor laser are shown in Fig. 71.5 [Darcie et al., 1986]. Both distortions are small at low frequencies but rise to maxima at half the relaxation resonance frequency. AM-VSB systems are feasible only within the low-frequency window. FM or uncompressed digital systems require enough band- width per channel that multichannel systems must operate in the region of large RD. Fortunately, the CNR requir ments allow for the increased distortion. The large second-order RD can be avoided entirely by operating within a one-octave frequency band (e.g., 2–4 GHz), such that all second-order products are out of band. Within the frequency range between 50 and 500 MHz, nonlinear gain and loss, intervalence-band absorption, and, more importantly, spatial-hole burning (SHB) and carrier leakage can all be significant. Carrier leakage prevents all of the current injected in the laser bond wire from entering the active layer. This leakage must be reduced to immeasurable levels for AM-VSB applications. SHB results from the nonuniform distribution of optical power along the length of the laser. In DFB lasers, because of the grating feedback, the longitudinal distribution of optical power can be highly nonuniform. This results in distortion [Takemoto et al., 1990] that can add to or cancel other distortion, making it, in some cases, a desirable effect. Clipping Even if all nonlinear processes were eliminated, the allowable modulation would be limited by the fact that the minimum output power is zero. Typical operating conditions with, for example, 60 channels, each with an average modulation depth (m) near 4%, result in a peak modulation of 240%. Although improbable, modu- lations of more than 100% result in clipping. The effects of clipping were first approximated by Saleh [1989], who calculated the modulation level at which the total power contained in all orders of distortion became appreciable. Even for perfectly linear lasers, the modulation depth is bounded to values beyond which all orders of distortion increase rapidly. Assuming that half the total power in all orders of distortion generated by clipping is distributed evenly over each of N channels, clipping results in a carrier-to-interference ratio (CIR) given by (71.7) FIGURE 71.5Resonance distortion for directly modulated laser with resonance frequency of 7 GHz. Both the second-harmonic 2f i and two-tone third-order 2f i ± f j distortion peak near half the res- onance frequency and are small at low frequency. Also shown is the same third-order distortion for an external modulator biased at the point of zero second-order distortion. CIR= + 2 16 2 3 12 2 p m m m () / e ? 2000 by CRC Press LLC where the rms modulation index m is m = m (71.8) External Modulation Laser-diode-pumped YAG lasers with low RIN and output powers greater than 200 mW have been developed recently. Combined with linearized external LiNbO 3 modulators, these lasers have become high-performance competitors to directly modulated lasers. YAG lasers with external modulation offer a considerable increase in launched power, and the low RIN of the YAG laser translates into a slight CNR improvement. The most challenging technical hurdle is to develop a linear low-loss optical intensity modulator. Low-loss LiNbO 3 Mach–Zehnder modulators are available with insertion losses less than 3 dB, modulation bandwidths greater than a few gigahertz, and switching voltages near 5 V. The output intensity of these modulators is a sinusoidal function of the bias voltage. By prebiasing to 50% transmission, modulation applied to the Mach–Zehnder results in the most linear intensity modulation. This bias point, which corresponds to the point of inflection in the sinusoidal transfer function, produces zero second-order distortion. Unfortunately, the corresponding third-order distortion is approximately 30 dB worse than a typical directly modulated DFB laser, at low frequencies. This comparison is shown on Fig. 71.5. For high-frequency applications where RD is important, external modulators can offer improved linearity. A means of linearizing the third-order nonlinearity is essential for AM-VSB applications. Various linearization techniques have been explored. The two most popular approaches are feedforward and predistortion. Feedforward requires that a portion of the modulated output signal be detected and compared to the original applied voltage signal to provide an error signal. This error signal is then used to modulate a second laser, which is combined with the first laser such that the total instantaneous intensity of the two lasers is a replica of the applied voltage. In principle, this technique is capable of linearizing any order of distortion and correcting RIN from the laser. Predistortion requires less circuit complexity than feedforward. A carefully designed nonlinear circuit is placed before the nonlinear modulator, such that the combined transfer function of the predistorter-modulator is linear. Various nonlinear electronic devices or circuits can act as second- or third-order predistorters. Difficulties include matching the frequency dependence of the predistorter with that of the modulator, hence achieving good linearity over a wide frequency range. Numerous circuit designs can provide reductions in third-order distortion by 15 dB. Miscellaneous Impairments Laser chirp can cause problems with direct laser modulation. Chirp is modulation of the laser frequency caused by modulation of the refractive index of the laser cavity in response to current modulation. The interaction of chirp and chromatic dispersion in the fiber can cause unacceptable CSO levels for AM-VSB systems as short as a few kilometers. Dispersion converts the FM into IM, which mixes with the signal IM to produce second- order distortion [Phillips et al., 1991]. These systems must operate at wavelengths corresponding to low fiber dispersion, or corrective measures must be taken. Chirp also causes problems with any optical component that has a transmission that is a function of optical frequency. This can occur if two optical reflections conspire to form a weak interferometer or in an erbium- doped fiber amplifier (EDFA) that has a frequency-dependent gain [Kuo and Bergmann, 1991]. Once again, the chirp is converted to IM, which mixes with the signal IM to form second-order distortion. Although externally modulated systems are immune to chirp-related problems, fiber nonlinearity, in the form of stimulated Brillouin scattering (SBS), places a limit on the launched power. SBS, in which light is scattered from acoustic phonons in the fiber, causes a rapid decrease in CNR for launched powers greater than approximately 10 mW [Mao et al., 1991]. Since the SBS process requires high optical powers within a narrow optical spectral width (20 MHz), it is a problem only in low-chirp externally modulated systems. Chirp in DFB systems broadens the optical spectrum so that SBS is unimportant. N2 ? 2000 by CRC Press LLC Summary A wide range of applications for transmission of video signals over optical fiber has been made possible by refinements in lightwave technology. Numerous technology options are available for each application, each with advantages or disadvantages that must be considered in context with specific system requirements. Evo- lution of these video systems continues to be driven by development of new and improved photonic devices. Defining Terms Chirp: Modulation of the optical frequency that occurs when a laser is intensity modulated. Composite second order (CSO): Ratio of the power in the second-order distortion products to power in the carrier in a cable television channel. Composite triple beat (CTB): Same as CSO but for third-order distortion. Direct modulation: Modulation of the optical intensity output from a semiconductor diode laser by direct modulation of the bias current. Erbium-doped fiber amplifier: Fiber doped with erbium that provides optical gain at wavelengths near 1.55 mm when pumped optically at 0.98 or 1.48 mm. External modulation: Modulation of the optical intensity using an optical intensity modulator to modulate a constant power (cw) laser. Fiber dispersion: Characteristic of optical fiber by which the propagation velocity depends on the optical wavelength. Fiber nonlinearity: Properties of optical fibers by which the propagation velocity, or other characteristic, depends on the optical intensity. Lightwave technology: Technology based on the use of optical signals and optical fiber for the transmission of information. Linear: Said of any device for which the output is linearly proportional to the input. Noise figure: Ratio of the output signal-to-noise ratio (SNR) to the input SNR in an amplifier. Rayleigh backscatter: Optical power that is scattered in the backwards direction by microscopic inhomoge- neities in the composition of optical fibers. Relative intensity noise: Noise resulting from undesirable fluctuations of the optical power detected in an optical communication system. Shot noise: Noise generated by the statistical nature of current flowing through a semiconductor p-n junction or photodetector. Related Topics 42.1 Lightwave Waveguides ? 69.1 Modulation and Demodulation ? 73.6 Data Compression References T.E. Darcie, “Subcarrier multiplexing for lightwave networks and video distribution systems,” IEEE J. Selected Areas in Communications, vol. 8, p. 1240, 1990. T. Pratt and C.W. Bostian, Satellite Communications, New York: Wiley, 1986. W. Way, C. Zah. C. Caneau, S. Menmocal, F. Favire, F. Shokoochi, N. Cheung, and T.P. Lee, “Multichannel FM video transmission using traveling wave amplifiers for subscriber distribution,” Electron. Lett., vol. 24, p. 1370, 1988. R. Olshansky, V. Lanzisera, and P. Hill, “Design and performance of wideband subcarrier multiplexed lightwave systems,” in Proc. ECOC ’88, Brighton, U.K., Sept. 1988, pp. 143–146. J.A. Chiddix, H. Laor, D.M. Pangrac, L.D. Williamson, and R.W. Wolfe, “AM video on fiber in CATV systems, need and implementation,” IEEE J. Selected Areas in Communications, vol. 8, p. 1229, 1990. A.N. Netravali and B.G. Haskel, Digital Pictures, New York: Plenum Press, 1988. K. Feher, Ed., Advanced Digital Communications, Englewood Cliffs, N.J.: Prentice-Hall, 1987. ? 2000 by CRC Press LLC B.L. Kasper, “Receiver design,” in Optical Fiber Telecommunications II, S.E. Miller and I.P. Kaminow, Eds., San Diego: Academic Press, 1988. T.E. Darcie, G.E. Bodeep, and A.A.M. Saleh, “Fiber-reflection-induced impairments in lightwave AM-VSB CATV systems,” IEEE J. Lightwave Technol., vol. 9, no. 8, pp. 991–995, Aug. 1991. T.E. Darcie, R.S. Tucker, and G.J. Sullivan, “Intermodulation and harmonic distortion in IaGaAsP lasers,” Electron. Lett., vol. 21, 665–666, erratum; vol 22, p. 619, 1986. A. Takemoto, H. Watanabe, Y. Nakajima, Y. Sakakibara, S. Kakimoto, U. Yamashita, T. Hatta, and Y. Miyake, “Distributed feedback laser diode and module for CATV systems,” IEEE J. Selected Areas in Communi- cations, vol. 8, 1359, 1990. A.A.M. Saleh, “Fundamental limit on number of channels in subcarrier mulitplexed lightwave CATV systems,” Electron. Lett., vol. 25, no. 12, pp. 776–777, 1989. M.R. Phillips, T.E. Darcie, D. Marcuse, G.E. Bodeep, and N.J. Frigo, “Nonlinear distortion generated by dis- persive transmission of chirped intensity-modulated signals,” IEEE Photonics Technol. Lett., vol. 3, no. 5, pp. 481–483, 1991. C.Y. Kuo and E.E. Bergmann, “Erbium-doped fiber amplifier second-order distortion in analog links and electronic compensation,” IEEE Photonics Technol. Lett., vol. 3, p. 829, 1991. X.P. Mao, G.E. Bodeep, R.W. Tkach, A.R. Chraplyvy, T.E. Darcie, and R.M. Derosier, “Brillouin scattering in lightwave AM-VSB CATV transmission systems,” IEEE Photonics Technol. Lett., vol. 4, no. 3, pp. 287–289, 1991. Further Information National Cable Television Association (NCTA), Proceedings from Technical Sessions, annual meetings, 1724 Massachusetts Ave. NW, Washington D.C., 20036, 1969. Society of Cable Television Engineers (SCTE), Proceeding from Technical Sessions, biennial meetings, Exton Commons, Exton, Penn. T.E. Darcie, “Subcarrier multiplexing for lightwave multiple-access lightwave networks,” J. Lightwave Technol., vol. LT-5, pp. 1103–1110, Aug. 1987. T.E. Darcie and G.E. Bodeep, “Lightwave subcarrier CATV transmission systems,” IEEE Trans. Microwave Theory and Technol., vol. 38, no. 5, pp. 534–533, May 1990. IEEE J. Lightwave Technol., Special Issue on “Broadband Analog Video Transmission Over Fibers,” to be published Jan./Feb. 1993. 71.2 Long Distance Fiber Optic Communications Joseph C. Palais When the first laser was demonstrated in 1960, numerous applications of this magnificent new tool were anticipated. Some predicted that laser beams would transmit messages through the air at high data rates between distant stations. Although laser beams can indeed travel through the atmosphere, too many problems prevent this scheme from becoming practical. Included in the objections are the need for line-of-sight paths and the unpredictability of transmission through an atmosphere where weather variations randomly change path losses. Guided paths using optical fibers offer the only practical means of optical transmission over long distances. Long-distance fiber systems tend to have the following operational characteristics: They are more than 10 km long, transmit digital signals (rather than analog), and operate at data rates above a few tens of megabits per second. This section primarily describes systems in this category. Figure 71.6 illustrates the basic structure of a generalized long-distance fiber optic link. Each of the compo- nents will be described in the following paragraphs. A useful figure of merit for these systems is the product of the system data rate and its length. This figure of merit is the well-known rate-length product. The bandwidth of the transmitting and receiving circuits (including the light source and photodetector) limits the achievable system data rate. The bandwidth of the ? 2000 by CRC Press LLC fiber decreases with its length, so that the fiber itself limits the rate-length product. The losses in the system, including those in the fiber, also limit the path length. Systems are bandwidth limited if the rate-length figure is determined by bandwidth restraints and loss limited if determined by attenuation. The first efficient fiber appeared in 1970, having a loss of 20 dB/km. Just 7 years later the first large-scale application, a link between two telephone exchanges in Chicago, was constructed. By this time the loss had been reduced to around 3 dB/km. The digital technology used could accommodate a rate of 45 Mbps over an unrepeatered length of 10 km and a total length of over 60 km with repeaters. The unrepeatered rate-length product for this initial system was a modest 0.5 Gbps ′ km. As fiber technology advanced, this figure steadily increased. Unrepeatered rate-length products have improved to 500 Gbps ′ km (e.g., 8 Gbps over a path of 60 km) and beyond. Allowing repeaters and/or optical amplifiers increases the net rate-length product considerably. Values beyond 70 Tbps ′ km (70,000 Gbps ′ km) are achievable with optical amplifiers. This latter figure allows construction of a transmission system operating at 5 Gbps over a 14,000-km path. The longest terrestrial paths are across the Atlantic and Pacific oceans, distances of about 6,000 and 9,000 km, respectively. Fibers are capable of spanning these distances with high-capacity links. Fiber All fibers used for long-distance communications are made of silica glass and allow only a single mode of propagation. The silica is doped with other materials to produce the required refractive index variations for the fiber core and cladding. The important fiber characteristics that limit system performance are its loss and its bandwidth. The loss limits the length of the link and the bandwidth limits the data rate. Figure 71.7 shows the loss characteristics of single-mode silica fibers at the wavelengths of lowest attenuation. As indicated in the figure, there are three possible windows of operation. In the first window (around 820 nm), the loss is typically 3 dB/km. This is too high for long systems. In the second window (near 1300 nm), the loss is about 0.5 dB/km. In addition, the fiber bandwidth is quite high because of low pulse dispersion at this wavelength. The second window is a reasonable operating wavelength for high-capacity, long-distance systems. At 1550 nm (the third window) the loss is lowest, about 0.25 dB/km. This characteristic makes 1550 nm the optimum choice for the very longest links. Dispersion refers to the spreading of a pulse as it travels along a single-mode fiber. It is due to material and waveguide effects. This spreading creates intersymbol interference if allowed to exceed about 70% of the original pulse width, causing receiver errors. The dispersion factor M is usually given in units of picoseconds of pulse spread per nanometer of spectral width of the light source and per kilometer of length of fiber. FIGURE 71.6Long-distance fiber communication system. ? 2000 by CRC Press LLC In the range from 1200 to 1600 nm, the dispersion curve for silica can be approximated by the expression (71.9) where l is the operating wavelength, l 0 is the zero dispersion wavelength, and M 0 is the slope at the zero dispersion wavelength. M 0 is approximately 0.095 ps/(nm 2 ′ km). The pulse spread for a path length L, using a light source whose spectral width is Dl, is then Dt = MLDl (71.10) The zero dispersion wavelength, close to 1300 nm for silica fibers, makes this wavelength attractive for high- capacity links. The dispersion at 1550 nm is typically close to 20 ps/(nm ′ km). This is a moderate amount of dispersion. If a proposed 1550-nm system is bandwidth limited because of this spread, several alternatives are available. One solution is to use dispersion-shifted fiber, which is a special fiber with a refractive index profile designed to shift the zero dispersion wavelength from 1300 nm to 1550 nm. Another solution is to transmit soliton pulses, which use the nonlinearity of the fiber to maintain pulse shape during transmission. Figure 71.7 includes relative unrepeatered, unamplified values of rate-length products in the three transmis- sion windows. Because of high loss, the first window can be used only for moderate lengths (around 10 km). Because of high dispersion, the data rates are also limited in this region. In the second window, nearly zero dispersion allows high-rate transmission, but the losses limit the distance that can be covered (typically around 50 km). In the third window, the loss is about half the 1300-nm attenuation so that twice as much distance can be covered. Dispersion-shifted fiber allows the same high rates as does 1300-nm operation. Repeaters and amplifiers extend the useful distance of fiber links well beyond the distances listed here. Modulator A digital electrical signal modulates the light source. The driver circuit must be fast enough to operate at the system bit rate. As bit rates increase into the multigigabit per second range, this becomes increasingly difficult. Modulation can be done in the optical domain at very high speeds. In this case, the modulator follows the laser diode rather than preceding it. External modulation is usually accomplished using integrated-optic structures. FIGURE 71.7 Fiber loss and relative unrepeatered, unamplified rate-length product. M M = ? è ? ? ? ÷ 00 4 4 3 l l l – ? 2000 by CRC Press LLC Light Source Laser diodes or light-emitting diodes (LEDs) supply the optical carrier waves for most fiber links. LEDs cannot operate at speeds in the gigabit range, but laser diodes can. For this reason, laser diodes are normally required for high-rate, long-distance links. Laser diodes can be modulated at frequencies beyond 40 GHz. Laser diodes emitting in the second and third fiber transmission windows are semiconductor heterojunctions made of InGaAsP. The exact emission wavelength is primarily determined by the proportions of the constituent atoms. Output powers are commonly on the order of a few milliwatts. Typical laser diode spectral widths are between 1 and 5 nm when operating in more than one longitudinal mode. Single-mode laser diodes can have spectral widths of just a few tenths of a nanometer. As predicted by Eq. (71.10), narrow-spectral-width emitters minimize pulse spreading. Minimizing pulse spreading increases the fiber bandwidth and its data capacity. Solid-state lasers other than semiconductor laser diodes may be useful in specific applications. Example of such lasers are the Nd:YAG laser and the erbium-doped fiber laser. Source Coupler The light emitted from the diode must be coupled as efficiently as possible into the fiber. Because the beam pattern emitted by a laser diode does not perfectly match the pattern of light propagating in the fiber, there is an inevitable mismatch loss. Good coupler designs, sometimes using miniature lenses, reduce this loss to about 3 dB when feeding a single-mode fiber. Isolator An optical isolator is a one-way transmission path. It allows power flow from the transmitter toward the receiver but blocks power flow in the opposite direction. It is used to protect the laser diode from back reflections, which tend to increase the laser noise. Connectors and Splices Connections between fibers and between the fiber and other components occur at numerous points in a long- distance link. Because there may be many splices in a long system, the loss of each one must be small. Fusion splices with an average loss of no more than 0.05 dB are often specified. Mechanical splices are also suitable. They often involve epoxy for fixing the connection. Connectors are used where remateable connections are required. Good fiber connectors introduce losses of just a few tenths of a decibel. In addition to having low loss, good connectors and splices also minimize back reflections. This is especially important for connections near the transmitter to reduce laser noise. Fusion splices produce little reflection, but mechanical splices and all connectors must be carefully designed to keep reflected power levels low. Reflections occur because of small gaps at the interface between the mated fibers. Successful techniques for reducing reflections include the physical contact connection, where the fiber end faces are polished into hemispheres (rather than flat surfaces) so that the cores of the two mated fibers are in contact with each other. Even better performance is obtained by angling the end faces a few degrees so that reflected light is filtered out of the single propagating mode. Optical Amplifier Many fiber links are loss limited. One cause is the limited power available from the typical laser diode, which (together with the losses in the fiber and the other system components) restricts the length of fiber that can be used. The fiber optic amplifier increases the power level of the signal beam without conversion to the electrical domain. For example, gains of 30 dB are attainable at 1550 nm using the erbium-doped fiber amplifier (EDFA). Quite importantly, the EDFA has a bandwidth of over 20 nm so that several WDM or numerous OFDM channels (described later in this section) can be amplified simultaneously. As indicated in Fig. 71.6, there are a number of possible locations for optical amplifiers in a system. An optical amplifier just following the transmitter increases the optical power traveling down the fiber. Amplifiers ? 2000 by CRC Press LLC along the fiber path continually keep the power levels above the system noise. An amplifier located at the fiber end acts as a receiver preamplifier, enhancing its sensitivity. Many amplifiers can be placed in a fiber network, extending the total path length to thousands of kilometers. Repeater The repeater is a regenerator that detects the optical signal by converting it into electrical form. It then determines the content of the pulse stream and uses this information to generate a new optical signal and launch this improved pulse train into the fiber. The new optical pulse stream is identical to the one originally transmitted. The regenerated pulses are restored to their original shape and power level by the repeater. Many repeaters may be placed in a fiber network, extending the total path length to thousands of kilometers. The advantage of the optical amplifier over the regenerator is its lower cost and improved efficiency. The greater cost of the regenerator arises from the complexity of conversion between the optical and electrical domains. The regenerator does have the advantage of restoring the signal pulse shape, which increases the system bandwidth. This advantage is negated by a system propagating soliton pulses, which do not degrade with propagation. Photodetector This device converts an incoming optical beam into an electrical current. In fiber receivers, the most commonly used photodetectors are semiconductor pin photodiodes and avalanche photodiodes (APD). Important detector characteristics are speed of response, spectral response, internal gain, and noise. Because avalanche photodiodes have internal gain, they are preferred for highly sensitive receivers. Both Ge and InGaAs photodiodes respond in the preferred second and third fiber windows. InGaAs performs better at low signal levels because it has smaller values of dark current (that is, it is less noisy). The current produced by a photodetector in response to incident optical power P is i = GheP/hf (71.11) where G is the detector’s gain, h is its quantum efficiency (close to 0.9 for good photodiodes), h is Planck’s constant (6.63 2 10 –34 J s), e is the magnitude of the charge on an electron (1.6 2 10 –19 ), and f is the optical frequency. For pin photodiodes (G = 1), typical responsivities are on the order of 0.5 mA/mW. Receiver Because of the low power levels expected at the input to the receiver, an electronic amplifier is normally required following the photodetector. The remainder of the receiver includes such electronic elements as band-limiting filters, equalizers, decision-making circuitry, other amplification stages, switching networks, digital-to-analog converters, and output devices (e.g., telephones, video monitors, and computers). Other Components There are a number of other fiber components, not shown in Fig. 71.6, that can be found in some systems. These include passive couplers for tapping off some portion of the beam from the single fiber and wavelength- division multiplexers for coupling different optical carriers onto the transmission fiber. System Considerations Long-distance fiber links carry voice, video, and data information. Messages not already in digital form are first converted to it. A single voice channel is normally transmitted at a rate of 64,000 bits per second. Video requires a much higher rate. The rate could be as much as 90 Mbps or so, but video compression techniques can lower this rate significantly. Fiber systems for the telephone network operate at such high rates that many voice channels can be time-division multiplexed (TDM) onto the same fiber for simultaneous transmission. For example, a fiber operating at a rate of 2.3 Gbps could carry more than 30,000 digitized voice channels. ? 2000 by CRC Press LLC Several optical carriers can simultaneously propagate along the same fiber. Such wavelength-division mul- tiplexed (WDM) links further increase the capacity of the system. Systems using two or three optical carriers are common. Adding more than a few channels (8 or so) puts strong constraints on the multiplexers and light sources. In long systems wideband optical amplifiers are preferred over regenerators for WDM systems because a single amplifier can boost all the individual carriers simultaneously while separate regenerators are needed for each carrier wavelength. Total cable capacity is also increased by placing numerous fibers inside the cable. This is a cost-effective strategy when installing long fiber cables. The added cost of the extra fibers is small compared to the costs of actually deploying the cable itself. Fiber counts above 100 are practical. Multifiber cables can have enormous total data capacities. Still further capacity is possible using optical frequency-division multiplexing (OFDM). In this scheme, many optical carriers very closely spaced in wavelength (maybe a few tenths of a nanometer) operate as independent channels. Hundreds of channels can be visualized in each of the two low-loss fiber windows. Systems of this type require coherent detection receivers to separate the closely spaced carriers. Error Rates and Signal-to-Noise Ratio The signal-to-noise ratio is a measure of signal quality. It determines the error rate in a digital network. At the receiver, it is given by (71.12) where P is the received optical power, r is the detector’s unamplified responsivity, G is the detector gain if an APD is used, n accounts for the excess noise of the APD (usually between 2 and 3), B is the receiver’s bandwidth, k is Boltzmann’s constant (k = 1.38 ′ 10 –23 J/K), e is the magnitude of the charge on an electron (1.6 ′ 10 –19 coulomb), T is the receiver’s temperature in degrees kelvin, I D is the detector’s dark current, and R L is the resistance of the load resistor that follows the photodetector. The first term in the denominator of Eq. (71.12) is caused by shot noise and the second term is attributed to thermal noise in the receiver. If the shot noise term dominates (and the APD excess loss and dark current are negligible), the system is shot-noise limited. In this case the probability of error has an upper bound given by: P e = e –n s (71.13) where n s is the average number of photoelectrons generated by the signal during a single bit interval when a binary 1 is received. An error rate of 10 –9 or better requires about 21 photoelectrons per bit. Shot noise depends on the optical signal level. Because the power level is normally low at the end of a long-distance system, the shot noise is small compared to the thermal noise. Avalanche photodiodes increase the shot noise compared to the thermal noise. With APD receivers, ideal shot-noise limited operation can be approached but (because of the APD excess noise and limited gain) not reached. If the thermal noise dominates, the error probability is given by P e = 0.5 – 0.5 erf (0.354 ) (71.14) where erf is the error function. An error rate of 10 –9 requires a signal-to-noise ratio of nearly 22 dB. System Design A major part of fiber system design involves the power budget and the bandwidth budget. The next few paragraphs describe these calculations. In a fiber system, component losses (or gains) are normally given in decibels. The decibel is defined by dB = 10 log P 2 /P 1 (71.15) S N GPR GeRBI P kTB L n LD = ++ () () r r 2 24 SN ? 2000 by CRC Press LLC where P 2 and P 1 are the output and input powers of the component. The decibel describes relative power levels. Similarly, dBm and dBm describe absolute power levels. They are given by dBm = 10 log P (71.16) where P is in milliwatts and dBm = 10 log P (71.17) where P is in microwatts. Power budget calculations are illustrated in Table 71.1 for a system that includes an ampli- fier. A specific numerical example is found in the last two columns. The receiver sensitivity in dBm is subtracted from the power available from the light source in dBm. This difference is the loss budget (in decibels) for the system. All the system losses and gains are added together (keeping in mind that the losses are negative and the amplifier gains are positive). If the losses are more than the gains (as is usual), the system loss dB SL will be a negative number. The loss margin is the sum of the loss budget and the system loss. It must be positive for the system to meet the receiver sensitivity requirements. The system loss margin must be specified to account for component aging and other possible system degradations. A 6- dB margin was found for the system illustrated in the table. The fiber in the table has a total loss of 24 dB. If its attenuation is 0.25 dB/km, the total length of fiber allowed would be 24/0.25 = 96 km. In addition to providing sufficient power to the receiver, the system must also satisfy the bandwidth require- ments imposed by the rate at which data are transmitted. A convenient method of accounting for the bandwidth is to combine the rise times of the various system components and compare the result with the rise time needed for the given data rate and pulse coding scheme. The system rise time is given in terms of the data rate by the expression t = 0.7/R NRZ (71.18) for non-return-to-zero (NRZ) pulse codes and t = 0.35/R RZ (71.19) for return-to-zero (RZ) codes. An example of bandwidth budget calculations appears in Table 71.2. The calculations are based on the accumulated rise times of the various system components. The system in Table 71.2 runs at 500 Mbps with NRZ coding for a 100-km length of fiber. Equation (71.18) yields a required system rise time no more than 1.4 ns. The transmitter is assumed to have a rise time of 0.8 ns. The receiver rise time, taken as 1 ns in the table, is a combination of the photodetec- tor’s rise time and that of the receiver’s electronics. The fiber’s rise time was calculated for a single-mode fiber operating at a wavelength of 1550 nm. Equation (71.9) shows that M = 18 ps/(nm ′ km) at 1550 nm. The light source was assumed to have a spectral width of 0.2 nm. Then, the pulse dispersion calculated from Eq. (71.10) yields a pulse spread of 0.36 ns. Because the fiber’s rise time is close to its pulse spread, this value is placed in the table. TABLE 71.1Power Budget Calculations Source power dBm s 3 Receiver sensitivity dBm r –30 Loss budget: dBm s – dBm r dB LB 33 Component efficiencies Connectors dB c –5 Splices dB s –2 Source coupling loss dB cl –5 Fiber loss dB f –24 Isolator insertion loss dB i –1 Amplifier gain dB a 10 Total system loss dB c + dB s + dB cl + dB f + dB i + dB a dB SL –27 Loss margin: dB LB + dB SL dB LM 6 TABLE 71.2Bandwidth Budget Calculations a Transmitter t t 0.8 Fiber t f 0.36 Receiver t r 1 System total: t s 1.33 System required t 1.4 a All quantities in the table are rise time values in nanoseconds. ttt fr 222 ++ ? 2000 by CRC Press LLC The total system rise time is the square root of the sum of the squares of the transmitter, fiber, and receiver rise times. That is: (71.20) In this example, the system meets the bandwidth requirements by providing a rise time of only 1.33 ns, where as much as 1.4 ns would have been sufficient. Defining Terms Coherent detection: The signal beam is mixed with a locally generated laser beam at the receiver. This results in improved receiver sensitivity and in improved receiver discrimination between closely spaced carriers. Material dispersion: Wavelength dependence of the pulse velocity. It is caused by the refractive index variation with wavelength of glass. Quantum efficiency: A photodiode’s conversion efficiency from incident photons to generated free charges. Single-mode fiber (SMF): A fiber that can support only a single mode of propagation. Spectral width: The range of wavelengths emitted by a light source. Related Topics 42.2 Optical Fibers and Cables ? 43.2 Amplifiers References E. E. Basch, Ed., Optical-Fiber Transmission, Indianapolis: Howard W. Sams & Co., 1987. C. C. Chaffee, The Rewiring of America, San Diego: Academic Press, 1988. M. J. F. Digonnet, Rare Earth Doped Fiber Lasers and Amplifiers, New York: Marcel Dekker, 1993. R. J. Hoss, Fiber Optic Communications Design Handbook, Englewood Cliffs, N.J.: Prentice-Hall, 1990. L. B. Jeunhomme, Single-Mode Fiber Optics, 2nd ed., New York: Marcel Dekker, 1990. N. Kashima, Passive Optical Components for Optical Fiber Transmission, Norwood, Mass.: Artech House, 1995. G. Keiser, Optical Fiber Communications, 2nd ed., New York: McGraw-Hill, 1991. R. H. Kingston, Optical Sources, Detectors and Systems, New York: Academic Press, 1995. J. C. Palais, Fiber Optic Communications, 3rd ed., Englewood Cliffs, N.J.: Prentice-Hall, 1992. S. Shimada, Coherent Lightwave Communications Technology, New York: Chapman and Hall, 1994. A. Yariv, Optical Electronics, 4th ed., Philadelphia: Saunders College Publishing, 1991. Further Information Continuing information on the latest advances in long-distance fiber communications can be obtained from several professional society journals and several trade magazines including: IEEE Journal of Lightwave Technol- ogy, IEEE Photonics Technology Letters, Lightwave, and Laser Focus World. 71.3 Photonic Networks Ivan P. Kaminow Lightwave technology has been developed and widely utilized for local and long-distance transmission in the public telephone network (see Section 71.2 and Miller and Kaminow, 1988) and in modern CATV (cable TV) networks (Section 71.1). Computer communications have been provided utilizing copper transmission lines in private local-area networks (LAN) that cover short distances, L < 10 km, and involve low data throughputs S < 10 Mb/s. The throughput is defined as S = NB tttt stfr =++ 22 2 ? 2000 by CRC Press LLC with N the number of simultaneous interconnections and B the communication bit-rate per user. At the higher ranges of the bit-rate-distance product, above BL = (10 Mb/s) (10 km) = 100 Mb/s·km or, equivalently, higher ranges of the bit-rate-delay product M = BD = (BL)(n/c), where D is the propagation delay, c/n is the (group) velocity of bits on the transmission line, and c is the velocity of light, optical fiber may be preferable to copper. For optical fibers, with refractive index n = 1.5, the delay is n/c = 5 ′ 10 –9 s/m = 5 ms/km. Thus, with BL = 100 Mb/s·km, M is 500 bits, i.e., there are 500 bits in transit on the transmission line between transmitter and receiver. As M gets larger, the performance of copper transmission lines—twisted pairs or coax—becomes unsatis- factory because of attenuation and pulse dispersion. The economic break-even value for M, where the added cost of lightwave technology is justified, though not precise, is in the neighborhood of 500 bits. In this section, we will cover aspects of lightwave data networks that utilize the lightwave technology discussed in Section 71.2 and some of the multiple-access methods for LANs discussed in Section 66.3. The latter section touches on commercial LAN standards that utilize optical data links for point-to-point transmission between nodes, often with multimode fiber. Here, we will first discuss some of the recent optical LAN standards and then briefly mention proposed approaches to photonic networks with terabit-per-second throughput and gigabit-per- second user access, and the novel optical components that are needed to realize this high performance. When such networks connect users separated by L ~ 1000s of kilometers, M ~ 10s of megabits may be in transit, requiring new approaches to congestion control for multiple access. Data Links A data link consists of a transmitter (T) that converts electrical pulses to optical pulses (E/O) and sends the optical pulses on an optical fiber to a receiver (R) which converts the optical pulses back to electrical pulses (O/E). The transmitter may use a light-emitting diode (LED) or a laser diode (LD) as the optical source. The LED is cheaper but has lower output power into the fiber (~10 mW vs. ~1 mW), lower modulation bandwidths (~100 Mb/s vs. ~1 Gb/s), and wider optical spectrum, leading to chromatic dispersion due to the variation of optical velocity in the fiber with wavelength. Pulse dispersion limits BL when the pulse spreading approaches a bit period. The receiver may employ a PIN (positive-intrinsic-negative) or APD (avalanche photodiode) photodetector. The former is cheaper and easier to bias but has poorer sensitivity by about 5 dB. The sensitivity of a good PIN receiver is about –50 dBm at 100 Mb/s and –35 dBm at 1 Gb/s for a bit-error-rate (BER) of 10 –9 . Optical devices operating at a wavelength of ~0.87 mm use gallium-aluminum-arsenide materials and are less expensive than those operating at 1.3 or 1.5 mm and using indium-gallium-arsenide-phosphide materials. However, for 1.5 mm, the fiber attenuation and, for 1.3 mm, the chromatic dispersion is much less than for 0.87 mm. The fiber joining transmitter and receiver may be multimode or single mode. Multimode (with a typical core diameter of 62.5 mm) is cheaper, but since each mode travels at a different optical velocity, the modal dispersion further limits BL. The LED data links generally employ multimode fiber, and the combination of chromatic and modal dispersion limits BL to values below ~1 Gb/s·km. For an LD, single-mode fiber (core diameter ~10 mm) data link, BL of ~100 Gb/s·km is possible. Optical data links are employed to connect electronic components of a LAN when copper is no longer feasible. However, because of its lower cost and the fact that it is often already installed, clever tricks are now being used to extend the utility of copper. Token Ring: FDDI, FFOL Figure 71.8 illustrates a ring network. The real topology may be a good deal more irregular than a circle, depending on the accessibility of stations. In its usual application, which uses a token-ring protocol for media access, an electronic repeater that operates at the aggregate network rate is required at each station. A token—e.g., a “1” or a “0” bit, or a token packet—is propagated in one direction from station to station. When a station has a packet to send to another station, it adds the address of the receiving station in a header ? 2000 by CRC Press LLC and holds the combined packet in a buffer. The sending station reads the tokens as they go by until it receives an empty token, a “0.” It then converts the “0” to a “1,” a busy token, and appends the packet. Intermediate stations repeat the bits in the packet and also “listen” for their own addresses. If a station recognizes its address in the packet header, it copies the packet. When the packet returns to the sender, it serves as an acknowledgment, and the sender removes it from the ring, after converting the token back to “0.” Commercial token rings use wire interconnections or optical data links to join stations at rates in the 10- Mb/s range. Actual network use is less than 10 Mb/s because of the time it takes an empty token to pass around the ring. This transit-time delay increases linearly with the number of stations. It includes propagation delay between stations and processing delay at each station, which must examine the header of every packet before repeating the bits to the next station. A token-ring architecture is not especially attractive for a high-speed optical network (where S ~ B is above 1 Gb/s) because of the cost of high-speed repeater optoelectronics at each station and the packet-processing delay. In addition, at high bit rates, the packet time may be much shorter than the propagation time around the ring, unless a packet contains an impractically large number of bits. Efficient use of the ring with short packets may call for multiple tokens, which can lead to complex protocols. Increasing the number of bits per packet increases the packet time but places added burden on the high-speed buffer at each station. Reliability—if one station is disabled or if the fiber breaks—is a problem in both fiber and wire rings. To address these reliability problems, a double-ring optical network can provide for bypass of defective stations and loop back around a fiber break. Each station has two inputs and two outputs connected to two rings that operate in opposite directions; this, of course, increases the cost. The fiber distributed data interface (FDDI) [Ross, 1986, 1989] is a standard proposed by the American National Standard Institute for a 100-Mb/s double-ring, time-division multiplexed (TDM) LAN that uses 1.3- mm multimode fiber and LED (or single-mode fiber and laser diode) data links between stations. This LAN is designed to provide both backbone services that interconnect lower speed LANs and back-end services that interconnect mainframe computers, mass storage systems, and other high-speed peripherals. FDDI provides datagram packet service with up to about 4,500 data bytes per frame. It employs 4B/5B coding so that the clock rate is 125 MHz for maximum S of 100 Mb/s. The FDDI network is designed to operate with low-cost components that were commercially available in 1986. The standard can provide both packet-switched and circuit-switched services. As many as 1000 stations can be connected, with a maximum of 2 km between stations and a maximum perimeter of 200 km. An FDDI follow on LAN (FFOL) will operate with laser and single-mode fiber data links at bit rates corresponding to the synchronous optical network (SONET) or synchronous digital hierarchy (SDH ) standards FIGURE 71.8 Undirectional ring network. R and T represent the receiver and transmitter functions, respectively. ? 2000 by CRC Press LLC of 622 Mb/s and 2.5 Gb/s, possibly with ATM (asynchronous transfer mode) cells. The geographical size will also be increased. Active Star Networks: Ethernet, Datakit ? Carrier sense multiple access with collision detection (CSMA/CD) Ethernet networks operating at 10 Mb/s connect users on a copper bus. The length of the bus must be less than 1/2 the distance light propagates in the time required to transmit a packet frame. Thus, for speeds much greater than 10 Mb/s, where optics might be needed, the length of the bus will be limited unless the maximum frame contains an impractically large number of bits. Further, the number of stations that can be supported by an optical bus is limited by the nature of optical taps, as opposed to electrical taps [Kaminow, 1989]. Finally, the collision detection algorithm does not work well on an optical bus because the intensities of optical packets from two different stations may vary considerably along the bus. Thus, an active electronic star, as shown in Fig. 71.9, with optical data links from users is often employed. The AT&T Datakit [Fraser, 1983] packet switch behaves as a virtual circuit switch (VCS) in that a reliable data path is set up for each session, and packet retransmission because of collisions is not required. Remote stations that may consist of mainframe computers, concentrators that bring together many terminals, or gateways to other networks are connected by 8-Mb/s fiber-optic data links to individual electronic modules at the central node as shown in Fig. 71.10 [Kaminow, 1988]. These modules plug into two electronic buses that are short (about 1 m) compared to a packet propagation length (16 bytes). In the module, packets are formed and stored with a header that contains the source address. When the packet is complete (it has the full number of bytes, or a fixed waiting period for added bytes has passed), the module transmits its binary address on the contention bus while listening for bits transmitted by others. If the module transmits a 1 and hears a 1, it transmits the next address bit. But if it transmits a 0 and hears a 0, it transmits the next bit; and, if it transmits a 0 and hears a 1, it stops transmission, having lost the contention. This process is equivalent to a logical OR operation and assigns the contention to the highest address. The winner transmits the packet on the contention bus in the next time frame. FIGURE 71.9Active star network. Optical-to-electrical (O/E) and electrical-to-optical (E/O) converters must be provided at the star. R and T represent the receiver and transmitter functions, respectively. ? 2000 by CRC Press LLC The switch at the end of the bus replaces the source address with the destination address and transmits the packet on the broadcast bus, where the appropriate module records the destination address and sends the packet to the remote station over the fiber link. Because the switch establishes a correspondence between source and destination at the beginning of a session (as in a circuit switch), source modules need not know the bus position of destination modules. The switch has a directory of positions and terminal names. If we were to go to very high bit rates, the physical bus length might no longer be short compared with a packet, and collisions caused by delays might upset the “perfect scheduling” of packets. Although methods have been proposed [Acampora and Hluchyj, 1984] for overcoming this limitation, the electronic circuit costs and electrical reflections on the bus may limit the effectiveness of a centralized bus at very high data rates. New Approaches to Optical Networks The preceding conventional networks with optical data links replacing copper can improve their throughputs thanks to the increased bandwidth of the transmission medium. However, a revolutionary improvement in throughput to terabit-per-second levels with gigabit-per-second access requires entirely new approaches for the physical connectivity, architecture, and access protocols. We can use much of the photonic technology employed in long-haul lightwave systems to provide physical connectivity, but we also need devices with new functionality to realize proposed architectures, and, conversely, with new component functionality we can dream of new architectures. We can provide connectivity among users in three dimensions: space, time, and optical frequency or wave- length, employing space-division multiplexing (SDM), optical time-division multiplexing (OTDM), and optical frequency-division multiplexing (OFDM) or wavelength-division multiplexing (WDM), respectively. To control the path routing we need optical switches for OTDM and frequency routing technology for OFDM. At present, network architectures and protocols are at the research stage. We mention some of these components and switches in the following paragraphs. More details can be found in the References [Miller and Kaminow, 1988; Special Issue, 1990]. FIGURE 71.10 Datakit a VCS network. Remote stations are connected to the electrical node by 8-Mb/s data links. The length of the bus is much shorter than the propagation length of a packet. ? 2000 by CRC Press LLC A star topology seems most attractive for gigabit-per-sec- ond multiple-access photonic networks [Kaminow, 1989], as shown in Fig. 71.11. Each station has its own transmitter and receiver. For optical TDM, the connectivity can be provided by an N ′ N electrooptic switch and suitable controller, and for optical FDM, the connectivity is provided by a passive N ′ N star coupler. Electrooptic N ′ N switches based on inte- grated titanium-diffused lithium niobate waveguide elements [Korotky and Alferness, 1988] have been demonstrated with N = 16 and operating at B = 2.5 Gb/s for each input. The switch connections can be rearranged in a few nanoseconds. It is estimated that such switches could be interconnected to provide N = 256. Unlike electronic switches, electrooptic switches are transparent to the bit rate, i.e., they can connect any bit stream independent of B. The problems of suitable multiple-access protocols and controls have not yet been fully addressed. The passive N ′ N star coupler [Kaminow, 1989] in Fig. 71.11 has N single-mode fiber inputs and N outputs. In an ideal passive star, a signal incident on any input is divided equally among all the outputs, i.e., the star broadcasts every input to every output. Unlike the OTDM case, each transmit- ter uses a different optical frequency and each receiver must tune to the frequency of the channel intended for it, as illus- trated in Fig. 71.12. Alternatively, the receiver frequencies may be fixed and the transmitters tunable. Thus, the control can be distributed in the terminals. Calculations indicate that such a network can support throughputs of several terabits per second. Current research [Special Issue, 1990] is aimed at devising multiple-access protocols and demonstrating the novel devices needed for optical frequency routing. These include fast tunable lasers and receivers that can cover many channels (switching speeds of ~10 ns at 2.5 Gb/s with ~50 channels appear feasible), optical frequency translators for frequency reuse, and integrated star couplers and integrated optical frequency routers. One challenge in photonic network design is to make them “all-optical,” as nearly as possible, in order to avoid throughput bottlenecks by electronic components and the expense of O/E and E/O conversions. In principle, clear all-optical channels would offer connectivity independent of data-rate and format for a wide variety of applications. However, many physical technology problems remain and new concepts for multple access and congestion control [Special Issue, 1991] suited to large bit-rate-delay (M) systems must be found. FIGURE 71.12An optical FDM network with passive star distribution. Optical transmitter frequencies f 1 … f N are modulated with data at the transmitter and selected by a filter at the receiver. FIGURE 71.11Electrooptic switch network for optical TDM or passive star network for optical FDM. The N ′ N switch or star have N single-mode optical fiber input ports and N optical fiber output ports. As in the active star (Fig. 71.9), two fibers connect a remote station with the star. R x and T x represent the receiver and transmitter functions, respectively, for station x. ? 2000 by CRC Press LLC Related Topic 72.2 Computer Communication Networks References A. S. Acampora and M. G. Hluchyj, “A new local area network architecture using a centralized bus,” IEEE Communications Magazine, vol. 22, no. 8, pp. 12–21, 1984. A. G. Fraser, “Towards a universal data transport system,” IEEE J. Selected Areas in Communications, vol. SAC- 1, no. 5, pp. 803–816, 1983. I. P. Kaminow, “Photonic multiple access networks,” AT&T Technical Journal, vol. 68, no. 2, pp. 61–86, 1989. I. P. Kaminow, “Photonic local networks,” in Optical Fiber Telecommunications, II, New York: Academic Press, 1988, chap. 26. S. K. Korotky and R. C. Alferness, “Waveguide electrooptic devices for optical fiber communication,” in Optical Fiber Telecommunications, II, New York: Academic Press, 1988, chap. 11. S. E. Miller and I. P. Kaminow, Eds., Optical Fiber Telecommunications, II, New York: Academic Press, 1988. F. E. Ross, “FDDI—A tutorial,” IEEE Communications Magazine, vol. 24, no. 5, pp. 10–17, 1986. F. E. Ross, “An overview of FDDI—The fiber distributed data interface,” IEEE J. Selected Areas in Communica- tions, vol. 7, no. 7, pp. 1043–1051, 1989. Special Issue, “Congestion control in high speed networks,” IEEE Communications Magazine, vol. 29, no. 10, 1991. Special Issue “Dense wavelength division multiplexing techniques for high capacity and multiple access com- munications systems,” IEEE J. Selected Areas in Communication, vol. 8, no. 6, 1990. Further Information J. G. Nellist, Understanding Telecommunications and Lightware Systems, 2nd ed. IEEE Press, 1996. J. Gibson, The Mobile Communication Handbook, Boca Raton, Fla.: CRC Press, 1996. S. Betti, Coherent Optical Communications Systems, New York: Wiley, 1995. B. Saleh, Fundamentals of Photonics, New York: Wiley, 1992. I. P. Kaminow and T. L. Koch, Optical Fiber Telecommunications, III, New York: Academic Press, 1997. ? 2000 by CRC Press LLC