Coarse WDM/CDM/TDM concept for optical packet ... - Semantic Scholar

(C) 2000 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE 1928

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 18, NO. 12, DECEMBER 2000

Coarse WDM/CDM/TDM Concept for Optical Packet Transmission in Metropolitan and Access Networks Supporting 400 Channels at 2.5 Gb/s Peak Rate Thomas Pfeiffer, Jens Kissing, Jörg-Peter Elbers, Bernhard Deppisch, Martin Witte, Harald Schmuck, and Edgar Voges

Abstract—To improve the networking flexibility in the metropolitan and access area, the granularity in the optical domain has to be increased above that in the core network requiring more channels at lower bit rates. Pure dense wavelength division multiplexing (DWDM) as it is applied in the core does not seem to meet this requirement at affordable cost. We propose and analyze a network based on hybrid optical multiplexing techniques including wavelength, code, and time division multiplexing. Applied to optical packet transmisssion this approach enables several 100 all-optical channels between end users and headend with average bit rates up to 100 Mb/s per channel while keeping the installation and maintenance cost at a minimum. Index Terms—Code division multiplexing, multiaccess communication, network reliability, optical crosstalk, optical noise, packet switching, time division multiplexing, wavelength division multiplexing.

I. INTRODUCTION

I

N metropolitan area networks (MAN) the required number of independent optical channels is likely to increase to 100 with channel bit rates up to the Gb/s range within the near future. This fine granularity enables flexible utilization of fiber bandwidth and simple network reconfiguration. Network operators will be able to lease single optical channels or groups of optical channels to service providers on demand. Even in access networks with multiple optical channels, the increase of total throughput to the Gb/s range is anticipated [1]. It is, however, questionable whether the straight forward scaling of today’s dense wavelength division multiplexing (DWDM) technology, as it is presently used in the core network, to the specific requirements of MAN and access networks is an affordable option for the realization of 100 optical channels. Specially for packet-based services like Internet with average data rates per user that are up to two orders of magnitude lower than the peak rate [2] the implementation of DWDM in network parts close to the end user will not be a realistic option. The spectral efficiency of such networks would be extremely poor making this approach very expensive.

Manuscript received April 19, 2000. This work was supported in part by the German Ministry for Research within the Photonik II and KomNet projects. T. Pfeiffer, B. Deppisch, M. Witte, and H. Schmuck are with Alcatel Corporate Research Center, D-70499 Stuttgart, Germany J. Kissing J.-P. Elbers, and E. Voges are with Lehrstuhl für Hochfrequenztechnik, University Dortmund, Germany. Publisher Item Identifier S 0733-8724(00)11609-3.

With increasing channel number and taking into account the optical bandwidth of fiber networks that is mostly limited by the gain bandwidth of optical amplifiers the channel spacing will rapidly drop below the current 100 GHz and 50 GHz spacings. The technological and economical effort that has to be spent to realize even smaller optical channel spacings increases dramatically and is hardly justified by the required channel capacity. The main issues to be encountered at the optical transmitters and receivers, when the channel spacing drops down to the 10 GHz range, are selection, tuning and control of the laser center frequency, the requirement for external modulation even for low channel bit rates and the fabrication and control of narrowband optical filters. Even without considering additional impairments from fiber transmission (e.g., channel crosstalk due to nonlinear effects) and optical amplification (gain flatness, transient effects) these systems are expected to become extremely demanding with respect to component specifications and network management. For packet-based transmission systems the additional application of time and/or code division multiplexing (TDM, CDM) techniques in the network is a favorable way to increase the channel number. TDMA (time division multiple access] is a viable option for increasing the channel number in simple fiber network structures with limited capacity to several 100 [3]. CDM techniques can also be applied to more complex network architectures, since some optical CDM approaches can be realized to support asynchronous operation of channels. Moreover, with asynchronous CDM a statistical multiplexing gain can be achieved in packet transmission networks enabling a large number of channels to be allocated. The bursty nature of the traffic helps to reduce the performance degradation due to multiple access interferences. Alternatively, optical TDMA can be applied to a limited part of the network, typically in the access part, whereas in the MAN part WDM and CDM techniques are implemented. There have been different proposals how to implement optical CDM applying time domain [4], [5] or wavelength domain [6], [7] coding. Also, the combination of WDM and CDM techniques has been proposed in some papers [8]–[10]. We favor spectrally encoded CDM due to the fact that, in contrast to most time domain encoded CDM approaches, the different channels do not have to be synchronized to each other to achieve full orthogonality between codes. This guarantees independence of the optical multiplexing scheme from the network architecture. Moreover, in time domain CDM the optical pulses would have

0733–8724/00$10.00 © 2000 IEEE

(C) 2000 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE PFEIFFER et al.: OPTICAL PACKET TRANSMISSION SYSTEM

to be much shorter than the bit period, so that the optical components must satisfy Gb/s specifications even for Mb/s channel bit rates. In this paper, we present a detailed system proposal for the implementation of CDM enhanced WDM based on previous experimental and theoretical studies that we have performed on a special form of spectrally encoded optical CDM. Our CDM approach applies periodic spectral encoding of broadband optical sources like light emitting diodes (LEDs) [11]. This approach has some advantageous features like stability and insensitivity of system performance with respect to component drift and specifications making it very attractive for MAN and access applications. On the networking side the benefits are asynchronous operation of different optical channels, support of different signal formats and ease of redistributing electrical bandwidth among optical channels. After a short review of the operating principle these aspects will be discussed in the next section with reference to experiments that have been performed in multichannel system demonstrators transmitting continuous data streams. Then the basic system limitations, namely channel crosstalk and optical intensity noise, will be discussed yielding some simple rules of thumb to estimate the achievable channel number and bit rate. A full numerical simulation is used for system evaluation when using different transmitter and receiver configurations. From this analysis two network scenarios for optical multichannel transmission are derived that also include time domain features like statistical multiplexing and synchronous TDMA within limited network areas. At the end impacts from transmission of broadband optical signals are discussed, specially from chromatic dispersion induced signal distortions.

II. CONCEPT

AND

FEATURES OF PERIODIC SPECTRAL ENCODING

The concept of periodic spectral encoding as proposed by Möller [11] is a generalization of coherence multiplexing [6] in the sense that all kind of optical filters having periodic power transmission characteristics are considered for encoding and decoding. The optical output power of a thermal light source, like an LED, is intensity modulated by the electrical data. After emission the broadband spectrum is optically shaped using passive filters with periodic transmission functions like Mach–Zehnder (MZ) or Fabry–Perot (FP) filters (Fig. 1). The periodicity of the transmission function for these type of filters is given in terms of the free spectral range (FSR) which is defined by the filter round trip . The round trip time is determined time via by the filter geometry and given as for for FP filters, where MZ filters and and are the differential delay and the cavity length of the is the group index of the cavity mafilters, respectively, terial and is the vacuum speed of light. In the network different FSR are allocated to different optical transmitters and define the codes in the system. At the receiver an optical filter with periodic transmission function is tuned by matching its FSR to the desired channel. The filter types

1929

at transmitter and receiver do not have to be identical, only and , respectively, the FSR or the round trip times must be matched (Fig. 2). To achieve nearly perfect orthogonality between channel codes a differential receiver set-up is used that removes the offset of the received signal power in . In the theoretical analysis of this Fig. 2 at paper, (cf. Section III) it will be assumed that the common mode rejection ratio (CMRR) of the differential receiver is infinite, i.e., the signal level and group delay for the undesired channels are assumed to be identical on the signal arm and on the reference arm of the receiver. In our 155 Mb/s differential receiver the CMRR was better than 40 dB for frequencies up to 200 MHz [12]. The optical source spectra are assumed to be much broader than the FSR of the transmitter filters, or in other words, the source coherence time defined by the source optical power density spectrum as [13] be much smaller than . Then the relative positioning the filter round trip time of the source spectrum center frequency and the filter transmission function is irrelevant which leads to the first benefit of this approach: the selection of the source spectra and the and filter transmission is not critical as long as for every the difference of the round trip times and is in the order of the source pair of transmitters coherence time or larger. [In practice the gain spectra of the optical amplifiers define the coherence time, since usually they are narrower ( 30 nm) than the source spectra ( 50 nm).] This has been experimentally verified in an 8 155 Mb/s system where the transmitter LEDs were taken from two different suppliers (MRV and Anritsu) with different spectral 53–68 nm] width [FWHM (full width at half maximum) and center wavelengths (1539–1551 nm) [12]. The filter FSR in this system (MZ at the transmitters, FP with finesse 4 at the receiver) were chosen in the range 10–20 GHz, the power extinction of the transmitter filters was 10–13 dB and slightly polarization dependent, the FP receiver filter extinction ratio was only 9 dB. Nevertheless the system operated with bit . The second benefit of peerror rate (BER) riodic encoding is that only the difference in filter roundtrip for any two transmitters needs to be larger times is not than some picoseconds. The absolute value of the important as long as they are much larger than . From that it follows that even temperature changes of more than 100 C do not affect the system performance as proven by experiment [14]. The reason for this unsensitivity to temperature changes is the low temperature coefficient of the refractive C for fused silica [15]). Our index of glass (about 10 MZ encoding filters were made from standard 3 dB fiber for the above tempercouplers, so that ature range. The fine tuning of the receiver filter to drifting transmitter filter FSR was accomplished by applying a local low-frequency dithering technique in the receiver [16], but no control of the transmitter filter FSR was required to keep [14]. The robustness of the system with respect to component specifications and drift as demonstrated above is unique to the approach applying periodic spectral encoding with broad source spectra and small filter FSR. Other code multiplex systems like


Fig. 1.


Principle set-up of CDM system applying periodic spectral encoding.

Fig. 2. Principle of periodic spectral encoding and decoding (left). Received optical power with single-ended detection around point of optimum tuning (right). Encoding and decoding filter type have been exchanged with respect to Fig. 1 for more instructive visualization. Gaussian source spectrum is assumed with 2 ln(2)= .

=

-sequence encoded systems [7] are not expected to show similar robustness due to the requirement of exact matching of the aperiodic spectra at transmitter and receiver. With systems as shown above, where the individual CDM channels were modulated with independent continuous data streams, different transmission experiments have been per155 Mb/s over 111 km field formed. Transmission of 8 installed standard single-mode fiber (SMF) was demonstrated [17] as well as over 204 km Alcatel TeraLight™ fiber ps/nm km @ 1550 nm) with for ( each channel (Fig. 3). Transmission and splitting losses were compensated by three cascaded erbium-doped fiber amplifiers (EDFAs) and the chromatic dispersion of the fiber link was compensated by using matched lengths of dispersion compensating fiber (DCF). In more complex networks including parallel groups of transmitters in a tree-like configuration the

capability of the chosen CDM approach to support different channel bit rates was experimentally demonstrated [12]. In these experiments, it was also found that the length tolerance of the dispersion compensation using DCF was at least 4 km SMF at this channel bit rate (cf. Section V). Even transmission of quasi-analog electrical signals (16-QAM subcarriers) over optical CDM parallel to baseband signals within the same network was experimentally demonstrated [14]. These last features cannot easily be realized with CDM systems based on time domain coding.

III. BASIC LIMITS ON SYSTEM PERFORMANCE After having reviewed the basic features of the optical CDM approach applying periodic spectral encoding we will now dis-


Fig. 3. Transmission of 8 3 155 Mb/s optical CDM signals over 204 km Alactel TeraLight™ fiber.

cuss the main factors that pose an upper limit to the achievable system capacity. The BER is evaluated using the -factor approximation , where erfc denotes the complementary error function and is defined as with the current and for the “1” and “0” signal and the and . The noise on the “1” associated standard deviations and on the “0” is assumed to be identical and is set to 0, so with the differential receiver current and its that standard deviation . The main contributions to originate from (thermal) receiver noise, shot noise, optical intensity noise, and crosstalk. The variance of the latter two disturbances increases linearly with electrical signal power and are responsible for the potential occurrence of a BER floor, hence they decide on the feasibility of a system configuration. They will now shortly be discussed, under simplifying assumptions, how they are influenced by the width of the optical source spectra and by the optical filters. We first focus the discussion on Gaussian source spectra and on combinations of MZ and FP filters. The results are characteristic of the discussed system and are only slightly modified if other spectra or filters are used. In the next section of this paper realizations are presented that are more useful in systems with many channels. When in the following we talk about source spectra, it is assumed that they are not modified during transmission due to e.g., optical amplifiers. Otherwise they have to be replaced in the formulae by the respective spectra at the receiver. Some basic considerations about crosstalk have been discussed in [18] (note: the symbols used here slightly differ from those in [18]). Here, we derive the number of optical channels that could be allocated in a system, if crosstalk was the only limiting factor taking into account the data statistics. For broad source spectra (as the compared to the filter FSR) with total power photocurrent after the receiver filter is given by the transmitter and and receiver filter transmission functions as where denotes the average over the optical frequency . The photodetector response is omitted here

1931

for simplicity. With a differential receiver the signal current is given by , where is the difference between the receiver transmission function in the filtered arm and the reference arm with and with the periodicity . is detuned from the transmitter , the signal When varies in an oscillatory manner (Fig. 2) the details of which depend on the specific choice of filters and source spectra. Specially with a FP–FP combination for transmitter and receiver filter the envelope of the tuning curve yields maxima also around is a rational number. This is in contrast points, where to systems, where, for example, the receiver filter is an MZ filter is an inyielding maxima only around points, where teger. So in an FP–FP system the allocation of transmitter FSR as the ratio of large is preferably done by choosing prime numbers, whereas in an –MZ system ( being either MZ or FP) the can be allocated to the transmitters equidistantly within one octave with or larger [18]. To simplify the discussion the latter case is assumed with identical Gaussian source spectra for all transmit. ters described by Then the envelope of the tuning curve for an MZ receiver filter is described as function of with as [18] (1) . The with FWHM given by parameter in (1) is the second coefficient of the transmitter filter cos-series

with the

value for MZ or for FP with finesse . In the following, it is assumed that all transmitters continuously send NRZ signals (non return to zero) with equal for “0” and “1.” This results in a probability binomial probability distribution of “1” signals at the receiver which for many interfering transmitters is approximated by a Gaussian distribution of the crosstalk current with the mean value , where is calculated from (1) for all transmitters with . The standard deviation is given , where for channels by in the transmitting synchronously to each other and asynchronous case [19]. With these results the crosstalk-limited -factor for the channel with ( denotes the integer part of its argument) can now be expressed as

(2) with source FWHM and . The number of addressable codes is very large (cf. Fig. 5) if transmitter or receiver filter or both are MZ like filters. In case of an FP/FP combination the crosstalk around points



where is a rational number strongly reduces the useful number of codes. However, even with MZ filters not all codes can be simultaneously used, since the intensity noise at the receiver imposes a more restrictive limit to the achievable BER. For single ended detection, i.e., no differential receiver, the intensity noise of a broadband thermal light source with ( for Gaussian and coherence time for rectangular spectra) is given by [20]. Here is the photocurrent and is the noise equivalent bandwidth defined via the normalized electrical as . It receiver transfer function was assumed that the detected light is linearly polarized. For partly polarized light the electrical noise power is reduced where denotes the degree of polarization by and for polarized and unpolarized light, with respectively [13]. If the source spectrum is passively filtered by and a periodic optical filter with transmission function the relative intensity noise is enhanced by a factor [ for MZ and for FP] which can be regarded as the noise figure of the optical filter for thermal light. The effective coherence time independent passively filtered thermal light sources is of with the superposed optical . For power density spectrum identical source spectra with coherence time this reduces to

(3)

denotes the average over the optical where again frequency . This spectrum is filtered at the receiver and differentially detected. With the effective receiver filter function the electrical noise power at the output of the differential receiver is (4) denotes the photocurrent that would be measured if the total was directly detected with spectrum a single photodiode: . With the signal current , given by , the optical for transmitter in case of identical noise limited -factor for transmitter Gaussian source spectra and with MZ receiver filter is

(5)

It was assumed that the noise originating from all transmitters is slightly larger is identical although the photocurrent from than the others. In practice, however, the above approximation is valid as supported from experiments. Furthermore, the optical has only entered the formula by 50% to account for power the data statistics [19]. The -factor in (5) can be optimized

Fig. 4. Optimum transmitter filter finesse for FP–MZ filer combination.

by the choice of the transmitter filter type and specifications. , For FP transmitter filters with finesse [ for ] the term in the denominator is minimized with the optimum finesse (6)

This relation is depicted in Fig. 4 showing optimum finesse values that can be easily realized from a technological point of view. The reason for the existence of an optimum finesse lies in the fact that with increasing the differential signal current degrades and the noise figure of the transmitter filters increases, but on the other side the interchannel beat noise is reduced. With the optimum finesse the optical noise limited -factor in the floor is related to the number of active channels by

(7) . EquaThe last approximation is well suited up to tion (7) indicates that systems, where is valid, show the same error floor under the above assumptions, specially that the filter finesse is optimized to the number of active transmitters. This is, of course, not feasible in practice. However, the dependence of the -factor on filter finesse is weak at large so that 10–20 is a good choice in many cases. Equation (7) indicates that the system employing FP/MZ filters performs slightly better as compared to MZ/MZ systems like coherence multiplexed systems [6]. The electrical signal-to-noise ratio (SNR) with respect to MZ/MZ systems is improved by about 3 dB for six channels and depends more weakly on channel number as compared to MZ/MZ systems. This improvement originates from the use of FP transmitter filters which reduce the optical interchannel beat noise. that are allowed to be active The number of channels at the same time is reduced considerably as compared to the . The noise limited -factor number of addressable codes for given is determined by the ratio [see (7)].


Fig. 5. Number of addressable codes (for parameters see text).

N

and of active channels

1933

N

For a given number of channels to be addressed, on the other hand, the crosstalk limited -factor is determined by the [see (2)]. The number of addressable codes ratio and the number of active channels are compared as function of the source spectral width in Fig. 5 for FP transmitter filters with FSR 10–20 GHz and optimized finesse. The requirement to in the floor ( ) be met was at 155, 622, and 2500 Mb/s (noise equivalent electrical bandwidth 122, 488, 1961 MHz with fourth-order Bessel filters). It was assumed that the channels were synchronized and that the detected light was linearly polarized, both assumptions yielding lower numbers as compared to the opposite cases. There are many simplifications made in the calculations shown in Fig. 5, as indicated in the text. However, comparison with results from a full numerical model show that the results presented in the figure can serve as a guideline to the performance to be expected with such kind of systems. The calculations presented in the next section are based on a more complex mathematical description. IV. SYSTEM DESIGNS FOR OPTICAL PACKET MODE TRANSMISSION In this section, we discuss how an optical CDM network must be dimensioned to allow large channel numbers. Using a similar but more complete numerical model [19]–[21] as described in the last section we evaluated the number of channels that can be , if the source spectra and the opsupported with tical coding filters are optimized. Throughout this section, the source spectra are assumed to be unpolarized and the different CDM channels are assumed to be unsynchronized to each other. For reference (Fig. 6) the optical channel number is evaluated that may transmit simultaneously with Gaussian source spectra THz, i.e., nm) and for transmitter/receiver ( filter combinations MZ/MZ, FP/MZ, and FP/FP. The FSR were allocated within 10–20 GHz, the inverse FSR being equidistantly distributed within one octave for /MZ combination and with large prime number ratios in the case FP/FP. The FP filter in all cases. The FP/ combination finesse was set to

Fig. 6. Allowed number of simultaneously active channels with a 45-nm-wide Gaussian spectrum (FP finesse 10).

=

allows a larger number of channels as compared to the MZ/MZ system by up to a factor of 2. But obviously with this system design the channel numbers are not large enough for future metro and access networks, specially at high channel bit rates. To increase the number of channels we take advantage of the fact that FP/MZ systems scale with (cf. last section) implying that the total channel number is increased if the available optical bandwidth is divided into several WDM slices and optical CDM is applied within each of those slices. As discussed in the last section with /MZ combinations large numbers of codes can be addressed, while the number of simultaneously active transmitters must be considerably lower. Such systems are attractive for optical packet mode transmission, taking advantage from statistical multiplexing to increase the optical channel number. In Fig. 7, nearly rectangular optical spectra are considered (mathematically described with ) with by Super-Gaussian spectra spectral width of 10 nm and 1.6 nm (200 GHz). In the former case, 84 codes can be addressed (FSR 10–20 GHz), in the latter case, there are 27 codes (FSR 5–10 GHz) allowed before crosstalk increases the BER to . The number of transmitters that may be active at the same time is limited to 34 and 11, respectively, due to optical noise limits. The 10-nm slice can be used to transmit optical packets with, e.g., 622 Mb/s ), the 1.6 nm slice, with, e.g., 155 Mb/s peak rate ( ). The channel spacing required in an equivalent ( DWDM system would be 190 GHz and 40 GHz, respectively, if the number of simultaneously active channels is taken as reference. However, in packet mode operation the required channel spacing for an equivalent DWDM system would be 16 GHz and and . In a packet mode 7.4 GHz for CDM system either the packet statistics or the network manchannels are agement must assure that no more than received at the same time. Violations of these limits might be corrected applying forward error correction in the receivers. If the available optical bandwidth is, for example, 50 nm, it can be divided into five (10 nm) or 31 (1.6 nm) WDM slices. and These system proposals then enable optical packet mode channels within that bandwidth, not



Fig. 7. Allowed number of simultaneously active channels in 10 nm wide (top, 84) and 1.6 nm (200 GHz) wide (bottom, = 27) nearly rectangular WDM slice.

N

=

N

taking into account guard bands. This is more than an order of magnitude increase as compared to the system with broad Gaussian source spectra (see Fig. 6). To receive these many channels, the WDM slices are demultiplexed using conventional WDM filters before applying the afore mentioned differential detection scheme to each CDM channel. The average channel bit rate amounts to about 29 Mb/s for 155 Mb/s peak rate and to 46 Mb/s for 622 Mb/s peak rate and is almost independent of the two slice bandwidths considered in Fig. 7. These throughput values are attractive for access networks. The total system capacity ranges from 12.2 Gb/s to 38.5 Gb/s depending on the specific configuration. From a networking point of view the grouping of channels into wavelength bands might have some added value for the management and reconfiguartion of the network. It also allows the parallel operation of additional Gb/s DWDM channels for high end users over the same network. The 1.6-nm slice option was selected, because of its compatibility with the ITU DWDM grid. However, it must be mentioned that

problems are to be expected with the optical power budget in that case. In the above simulations, it was assumed that the optical source power was 1 mW within the considered bandwidth before the optical encoding filter at the transmitters. In the network scenario developed above there is one differential receiver for each optical channel. Since mechanically tunable filters are used, this requires much floor space and will be an expensive solution as long as the average channel capacity is well below 100 Mb/s. In this section, we propose an approach where the number of optical CDM channel is again reduced, but now each CDM channel carries many optical TDM signals. The proposed network comprising metropolitan and access fiber networks is depicted in Fig. 8 [22]. The optical CDM technique is supposed to be only applied in the upstream direction from the optical network units (ONU) to the headend, so the discussion is restricted to transmission in that direction. For downstream transmission a combination of DWDM and TDM seems better suited [22]. There are three cascaded optical multiplexing techniques in the network. Starting from the headend, there is a coarse WDM on the primary ring and optical CDM on the secondary ring. The link between both rings is realized by the indicated a coarse WDM coupler. The combined WDM/CDM operation is the same as discussed above, with the only exception that now the CDM channels are made up of sequential optical packets from a TDM network (cf. below) and are quasi-continuously active. So the CDM system must be optimized for continuous data transmission in contrast to the situation discussed in the preceding section. The best performance is achieved, if we use FP filters both at the transmitters and at the receiver (FP/FP combination). The finesse is set to . To reduce crosstalk and optical noise originating from the overlap of multiple inteand the gers of the filter FSR in terms like FSR are allocated within the semi-octave 15–20 GHz using ratios of large prime numbers. The number of continuously active CDM channels that can now be supported within broad WDM slices is depicted in Fig. 9. Now transmission even at 2.5 Gb/s channel bit rate is possible with six (10) CDM channels within 10 nm (16 nm) WDM slices. However, the modulation transfer function of the FP filters is reduced to bandwidths in the order of 0.75–1.0 GHz leading to signal distortions at 2.5 Gb/s. Using an electrical decision feedback equalizer (DFE) after detection these distortions can be compensated. DFEs have been successfully applied to compensate for even larger ratios of electrical signal bandwidth to equivalent electrical filter bandwidth in dispersive LED transmission [23]. Each optical CDM channel in Fig. 8 carries optical packets from a TDM PON (passive optical network). The all optical conversion from TDM to CDM applies a technique that has been recently demonstrated for bit rates up to 10 Gb/s [24]. It is based on the optical modulation of the backward ASE (amplified spontaneous emission) output power of an SOA (semiconductor optical amplifier) or an LED by means of an intensity modulated laser diode (Fig. 10). The logic of the modulated ASE signals is inverted with respect to the input. This is, however, easily reverted in the electrical domain after detection. The modulated ASE is optically encoded using FP filters as above and now behaves like the other optical CDM channels discussed in this paper. Since the laser diode emission wavelengths may


Fig. 8.

1935

Optical network incorporating different cascaded optical multiplexing techniques (WDM denotes a wavelength selective coupler).

Fig. 9. Allowed number of simultaneously active channels in different slice widths for optimized FP/FP filter combinations.

vary over a 50-nm wide window without significantly degrading the conversion efficiency a tight selection of transmitter lasers is not necessary [24]. We demonstrated the feasibility of the concept by time domain multiplexing 2 4 DFB lasers with 127 bit packets at 622 Mb/s into two CDM channels (8.6 nm spectral width) and achieved factors of 10 [22] (the width of the guard bands between packets in Fig. 10 can be substantially reduced in practice). Now the network in Fig. 8 comprises 20 TDM channels per PON at 2.5 Gb/s peak rate, collected into five optical CDM channels that are multiplexed into four WDM slices (10 nm wide Super-Gaussian with 3 nm guard bands between them). This gives an all optical connection between 400 ONUs and headend with 100 Mb/s average channel bit rate (including 20% TDM guard bands between packets). The number of optical CDM receivers is now reduced to 20, because demultiplexing

of the TDM packets is performed electrically after optical decoding and detection. Although the optical transmission length between ONU and headend may be up to 100 km the stringent timing requirements of the TDM approach apply only to the PON section that is assumed to be spread over 20 km. With the TDM-in-CDM approach the primary and secondary rings can be managed as is usually done including any kind of protection switching schemes. This could not be achieved, if the TDM approach was applied throughout the entire network, e.g., within a common DWDM channel for each secondary ring. In that case the temporal ranging of the ONUs would have to be changed, if the optical path length changes due to protection switching requiring a reinitialization of the network after a failure. The advantage of the proposed system approach lies in the low cost per channel and the inherent stability and robustness of the applied multiplexing techniques. At the ONUs almost unspecified laser diodes can be used and the effort for generation and encoding/decoding of broadband signals is shared among 20 high-speed ONUs additionally taking advantage of the above mentioned temperature stability of the CDM technique. WDM applying 10-nm slices requires only loosely specified optical filters that are expected to operate in a very stable way. V. IMPACTS FROM TRANSMISSION OF BROADBAND OPTICAL SIGNALS We will now highlight two issues arising from the broad spectral width of the optical signals: 1) optical amplification and 2) transmission over dispersive fibers. Since both crosstalk and noise depend critically on the shape and width of the source spectrum as seen by the receiver, any deformation of the spectrum during transmission may degrade the system performance. However, the spectral gain shape of the optical amplifiers is not as critical as it might seem from first sight. With the exception of one fluoride-based EDFA we used conventional silica-based EDFAs with the well-known


Fig. 10.


All optical TDM-to-CDM conversion (top) and experimental demonstration at 622 Mb/s (bottom).

ASE peak around 1530 nm for low input signals. At high input 10 dBm) the corresponding gain peak vanishes power ( leaving a sufficiently flat gain response over almost 30 nm width. If the signal input power was so low that the gain peak around 1530 nm became too large we added a small piece of unpumped erbium-doped fiber after the silica-based EDFAs, thus equalizing the gain shape in a very simple yet effective way. This measure was accompanied by a 2–3 dB loss in total optical signal power. With narrowband optical signal spectra as proposed in the last section the requirement of gain flatness is more related to power equalization among the WDM slices, rather than to CDM specific crosstalk and noise issues. So the impact on system performance is not CDM specific in that case. In the remainder of this section, we will concentrate on chromatic dispersion induced signal distortions. The chromatic dispersion of the fiber link has a potential impact both on the signal and on the noise in systems operating with thermal light sources [25]. However, for broad source spectra the electrical noise power in the receiver is not significantly affected by the chromatic dispersion, even if transmission

takes place far from the wavelength of zero fiber dispersion [19]. The degradation of the SNR is primarily due to distortions of the signal shape. In our experiments we used matched lengths of DCF to restore the optical signal shape before detection. The question with optical dispersion compensation is, how exactly the DCF dispersion value must match the SMF dispersion to enable error free transmission, or in other words, how large is the length tolerance of optical compensation schemes. To answer that question we investigated the optical power penalty (eye opening penalty, EOP) that is induced, when high-speed thermal light pulses around 1550 nm are transmitted over uncompensated SMF links. The EOP is the additional optical power that is required to maintain a given BER in the presence of signal distortions. The value after transmission over fiber length can be estimated from the relative pulse peak amplitude for a single isolated “1” (losses neglected) [26]. It is calculated from (8)


1937

Assuming an electrical Gaussian pulse shape after low pass filtering in the receiver the peak amplitude of the detected op, can be expressed tical pulse, relative to the input value at in terms of the parameter . is a normalized dispersion and yields the asymptotic behavior of the relative broadof NRZ pulses at bit ening of the width rate . Here is the fiber dispersion parameter (usually given in ps/nm km) and is the FWHM of the optical signal spectrum. is taken to be constant over the entire optical spectrum considered which is a valid assumption for transmission far from the wavelength of zero dispersion. The factor is given as for Gaussian source spectrum (9a)

for rectangular source spectrum

(9b)

where erf( ) denotes the error function. The corresponding power penalties as calculated from (8) are valid as long as , or equivalently as long as and for Gaussian and rectangular spectra, respectively. After these values are reached a BER floor will be measured. In Fig. 11 the calculated penalties are depicted for Gaussian and rectangular source spectra as function of the normalized dis) the calculated curves are persion . At small values ( verified by experimental values that were extracted from transMb/s NRZ signals in mission measurements of the range around 1550 nm over different SMF lengths using nm wide Gaussian and nm wide rectangular optical spectra. These types of spectra correspond to the situations discussed in the section about the optical CDM system designs. For large values, approaching the limits of uncompensated and , the experiments showed transmission higher penalties than predicted by the simple model. The measured fiber length limits, where a BER floor set in, were somewhat smaller than calculated (around 8 km and 77 km instead of 9.3 km and 93 km). This may be due to not taking into account electrical noise in the calculations or that the Gaussian pulse shape approximation is not good enough. Nevertheless, the model is useful for a quick estimation of the expected system performance in terms of chromatic dispersion induced degradations. If the fiber lengths are larger than the above given limits or if the penalties are too large to be tolerated, an electrical postcompensation applying a DFE can be used to improve the performance. Using a two stage DFE a normalized dispersion of ( km SMF, nm with Gaussian spectrum), i.e., beyond the practically measured limit (cf. Fig. 11), has been successfully compensated with only 4.0 dB residual penalty [23]. Approaches applying linear electrical filters to compensate the low pass characteristic of the fiber-LED system have been found not to be as effective as the DFE approach in that the power penalty easily exceeds 10 dB [27]. In practical system implementations the larger part of the transmission fiber dispersion should be compensated using appropriate lengths of DCF. The residual dispersion with normal-

=

1

Fig. 11. Optical power penalty versus normalized dispersion BDL for dispersive LED transmission with rectangular and with Gaussian spectrum. Measurements were performed at 155 Mb/s with 71 nm wide Gaussian and with 8.6 nm wide rectangular spectrum.

ized values up to around can then either be tolerated or the induced power penalty may be reduced by applying the DFE approach. To give a number that is relevant for the proposed systems of the preceding section the theoretical limit for uncompensated transmission of 2.5 Gb/s signals with 10 nm rectangular spectra lies at 4.9 km. This is the order of magnitude to which the length of the DCF must be matched to the SMF length. The dispersion slope of commercially available DCF is well enough matched to the slope of the SMF to enable compensation over broad spectral ranges for long fiber links. In 8 155 Mb/s system experiments with about nm wide spectra we compensated the dispersion of up to 111 km field installed SMF [17]. With a dispersion slope of 0.09 ps/nm km for the ps/nm km) and 0.2 ps/nm km for the DCF SMF ( ( ps/nm km) the maximum differential group delay due to mismatched dispersion slope alone for the wavelength nm and nm wide spectra components of amounts to 6.4 ps and 0.7 ps per kilometer SMF, respectively. So if the dispersion is perfectly compensated at the center wavelength of the spectra, severe signal degradation due to dispersion slope mismatch would only be measured with signals at 2.5 Gb/s with 30 nm spectra after about 30 km SMF. In the case of narrow rectangular spectra as they have been proposed in the system design section of this paper the residual mismatch in total dispersion is irrelevant even at 2.5 Gb/s over 200 km SMF. VI. CONCLUSION We have proposed and analyzed a hybrid optical multiplexing approach for metropolitan and access networks. Based on a mix of wavelength, optical code and time division multiplexing a very flexible and robust implementation of multiuser optical networks can be realized. The basic technologies applied have been demonstrated experimentally. The network was analyzed using a detailed theoretical approach that takes into account crosstalk and intensity noise in wavelength sliced optical CDM



networks. By taking advantage of the fact that optical CDM supports more codes than can be used simultaneously an optical packet transmission system with more than 400 optical channels was proposed that relies on statistical multiplexing gain. Alternatively, using a novel optical TDM-to-CDM conversion technique the implementation of a 400-channel system with average channel bit rates up to 100 Mb/s was discussed. In the latter approach, the optical packet transmission is realized sequentially within each CDM channel. The applied optical CDM technology as well as the TDM-to-CDM conversion has been shown to be very insensitive to poor component specifications and drift, thus keeping the cost low for components and network supervision in the physical layer. REFERENCES [1] B. E. Lemoff, “Technology alternatives for 10 Gbit/s LAN’s,” in Proc. 25th OFC, Baltimore, MD, 2000, paper WI1. [2] D. D. Clark, “Fiber-based metropolitan access networks for Internet traffic,” in Proc. 25th OFC, Baltimore, MD, 2000, paper WD1. [3] I. Van de Voorde, C. Martin, H. Slabbinck, L. Gouwy, B. Stubbe, X. Z. Qui, J. Vandewege, and P. Solina, “Evaluation of the SuperPON lab demonstrator,” in Proc. 23rd ECOC, Edinburgh, U.K., 1997, pp. 3.331–3.334. [4] A. Grunnet-Jepsen, A. E. Johnson, E. S. Maniloff, T. W. Mossberg, M. J. Munroe, and J. N. Sweetser, “Demonstration of all-fiber sparse lightwave CDMA based on temporal phase encoding,” IEEE Photon. Technol. Lett., vol. 11, pp. 1283–1285, 1999. [5] I. Andonovic, L. Tancevski, M. Shabeer, and L. Bazgaloski, “Incoherent all-optical code recognition with balanced detection,” J. Lightwave Technol., vol. 12, pp. 1073–1080, 1994. [6] G. J. Pendock and D. D. Sampson, “Capacity of coherence-multiplexed CDMA networks,” Opt. Commun., vol. 143, pp. 109–117, 1997. [7] M. Kavehrad and D. Zaccarin, “Optical code-division multiplexed systems based on spectral encoding of noncoherent sources,” J. Lightwave Technol., vol. 13, pp. 534–545, 1995. [8] W. Huang and I. Andonovic, “OCDMA/WDMA networks based on OCDMA systems with interference cancellation,” in Proc. 25th ECOC, vol. Proc.-I, Nice, France, 1999, pp. 192–193. [9] K. Kitayama, “OCDM/WDM networks for gigabit access: 1.24 Gbit/s, 2 OCDM by 2 WDM experiment,” in Proc. 25th ECOC, vol. Proc.-I, Nice, France, 1999, pp. 194–195. [10] M. J. L. Cahill, G. J. Pendock, and D. D. Sampson, “Hybrid coherence multiplexing/coarse wavelength-division multiplexing passive optical network for customer access,” IEEE Photon. Technol. Lett., vol. 9, pp. 1032–1034, 1997. [11] L. Möller, “An optical CDMA method based on periodic spectrum encoding,” in Proc. 13th EFOC-N, Brighton, U.K., 1995, pp. 178–181. [12] T. Pfeiffer, B. Deppisch, M. Witte, and R. Heidemann, “Optical CDMA transmission for robust realization of complex and flexible multiple access networks,” in Proc. OFC 99, San Diego, CA, 1999, Paper WM-51. [13] J. W. Goodman, Statistical Optics. New York: Wiley, 1985. [14] T. Pfeiffer, B. Deppisch, H. Schmuck, and M. Witte, “Temperature insensitive spectrally encoded optical multichannel system for simultaneous transmission of digital baseband and subcarrier signals,” Electron. Lett., vol. 36, pp. 451–453, 2000. [15] I. H. Malitson, “Interspecimen comparison of the refractive index of fused silica,” J. Opt. Soc. Amer., vol. 55, pp. 1205–1209, 1965. [16] T. Pfeiffer, B. Deppisch, M. Kaiser, and R. Heidemann, “High speed optical network for asynchronous multiuser access applying periodic spectral coding of broadband sources,” Electron. Lett., vol. 33, pp. 2141–2142, 1997. [17] T. Pfeiffer, B. Deppisch, and M. Witte, “Transmission of broadband coherence multiplexed signals over long distance fiber feeder,” in Proc. 25th ECOC, vol. Proc.-I, Nice, France, 1999, pp. 186–187.

2

[18] T. Pfeiffer, B. Deppisch, M. Witte, and R. Heidemann, “Operational stability of a spectrally encoded optical CDMA system using inexpensive transmitters without spectral control,” IEEE Photon. Technol. Lett., vol. 11, pp. 916–918, 1999. [19] J.-P. Elbers, “Modellierung und Simulation optischer Zugangsnetze mit Breitbandquellen,” Ph.D. dissertation, Univ. of Dortmund, Germany, 2000. [20] M. Tur, E. Shafir, and K. Bløtekjaer, “Source-induced noise in optical systems driven by low-coherence sources,” J. Lightwave Technol., vol. 8, pp. 183–189, 1990. [21] J.-P. Elbers, C. Glingener, J. Kissing, E. Voges, and T. Pfeiffer, “Performance evaluation of a CDMA system using broadband sources,” in Proc. 24th ECOC, vol. 1, Madrid, Spain, 1998, pp. 341–342. [22] T. Pfeiffer, H. Schmuck, B. Deppisch, M. Witte, and J. Kissing, “TDM/CDM/WDM approach for metro networks with 200 optical channels,” in Proc. 26th ECOC, vol. 3, Munich, Germany, 2000, pp. 77–78. [23] M. Witte, F. Buchali, and T. Pfeiffer, “Reducing the optical power penalty for electronically dispersion compensated LED pulse transmission by using multi-bit-shift decision feedback,” Electron. Lett., vol. 36, pp. 450–451, 2000. [24] T. Pfeiffer, “High-speed optically modulated broadband light source,” Proc. 26th ECOC, vol. 3, pp. 51–52, 2000. [25] G. J. Pendock and D. D. Sampson, “Signal-to-noise ratio of modulated sources of ASE transmitted over dispersive fiber,” Photon. Technol. Lett., vol. 9, pp. 1002–1004, 1997. [26] J. L. Gimlett and N. K. Cheung, “Dispersion penalty analysis for LED/single-mode fiber transmission systems,” J. Lightwave Technol., vol. LT-4, pp. 1381–1392, 1986. [27] T. Pfeiffer, M. Witte, and B. Deppisch, “High-speed transmission of broadband thermal light pulses over dispersive fibers,” IEEE Photon. Technol. Lett., vol. 11, pp. 385–387, 1999.

Thomas Pfeiffer, photograph and biography not available at the time of publication.

Jens Kissing, photograph and biography not available at the time of publication.

2

Jörg-Peter Elbers, photograph and biography not available at the time of publication.

Bernhard Deppisch, photograph and biography not available at the time of publication.

Martin Witte, photograph and biography not available at the time of publication.

Harald Schmuck, photograph and biography not available at the time of publication.

Edgar Voges, photograph and biography not available at the time of publication.