In this paper we discuss several intriguing issues pertaining to wavelength-division-multiplexed optical systems. An indepth treatment is made of topics which include wavelength routing, packet contention resolution, and incorporating Erbium-doped fiber-optic amplifiers.

We propose enhancements to STARNET, a previously proposed wavelength division multiplexed broadband optical local area network. STARNET offers the users both a medium- speed packet-switched ring network, and a high-speed circuit interconnection. The enhanced architecture, STARNET-E, improves performance in both the packet and the circuit interconnect sections over the previous version. Performance analysis of the STARNET-E packet network shows that this data transport facility can be upgraded to Gb/s total throughput with little added complexity with respect to the original STARNET. Moreover, we show that STARNET nodes can be enhanced to allow reconfigurable multihop topologies with little added optical hardware. The throughput of multihop distributed switching over a passive star physical topology in STARNET-E is compared to the throughput of an active centralized switch. STARNET-E offers to all nodes a Gb/s total throughput packet transport facility and a high speed circuit interconnect, simultaneously and independently. In addition, a node can trade the high speed circuit interconnect service for a multihop broadband packet network without the need for any added optical hardware.

In this paper we describe the design of a packet-switched wavelength-division-multiplexed (WDM) metropolitan area fiber-optic network with multiple rf subcarrier channels multiplexed on each wavelength. We also describe the experimental verification of the key features of this network. Each station (node) in the network is assigned a unique wavelength and rf subcarrier for reception and a unique wavelength for transmission. Packet-switching can be accomplished by a combination of rapid tuning between rf subcarriers on the same wavelength at the transmitter and multihop; wavelength tunability is not required. When fully implemented, our network is capable of supporting 32 wavelengths spaced 1 nm apart and five amplitude-shift- keyed (ASK) subcarrier channels, each operating at 200 Mb/s and spaced 400 MHz apart, on each wavelength, and can achieve an aggregate throughput of 7 Gb/s.

The high bandwidth of optical fibers and photonic switching devices offers the potential to increase the usable bandwidth in packet switched networks and interconnects. In this paper we describe several methods to code packetized information using both the time and optical frequency domains. The presence or absence of optical power at discrete optical wavelengths is treated as units of information, resulting in two-dimensional coded packets which may be processed using novel, perhaps more efficient routing architectures. We describe three basic codes and their corresponding routing processor architectures. Several new logic primitives are introduced including optical set logic and lambda-gates. The resulting information capacities for these codes and the associated routing processor complexity, and efficiency in terms of number of mappings per gate are discussed.

This paper describes a family of systems developed to make better use of the vast usable bandwidth provided by optic fibers. All systems described are frequency division multiplexed (FDM). Special attention is paid to the possibility of reducing channel spacing by signal processing using optic filters and of improving receiver sensitivity by using Erbium-doped fiber amplifiers (EDFA). A suitable filter-amplifier configuration is proposed for each possible application, like LANs, WANs, and Long Haul Systems.

What is multigigabit optical networking? This is a term we are using loosely to refer to the networks with multiple gigabit data channels in order to distinguish them from networks whose aggregate bit rate is on the order of 1 Gbit/s. Current computer networks, such as Ethernet, FDDI, and DQDB, suffer from lack of concurrency: at a given time, only a small number (typically one) of computers can transmit new information into the network. Therefore, each computer has to operate at the network aggregate speed, although effectively it has access only to a fraction of that bandwidth. To achieve gigabits/sec throughput, the next generation of computer networks will have to provide multiple high-speed concurrent channels to the nodes. One way to achieve concurrency is to use electronic switching, thereby placing the burden on the switch rather than each computer. Another direction is to place the burden on the optical hardware. Optical transport facilities have been recognized as an excellent choice for gigabits/sec networks due to the high bandwidths and long distances that can be reached. Moreover, WDM optical techniques allow multiple concurrent channels to be created in the same fiber, and with tunable transceivers one can potentially create networks whose topology changes dynamically in response to changing traffic patterns. In this paper, we will review a few current implementations of very high-speed networks as a background context, and then describe several prototype optical networks.

Recent work in optical code division multiple access (CDMA) is reviewed, progressing from incoherent to coherent techniques. It is shown that under appropriate conditions, coherent CDMA can in principle rival wavelength-division multiplexing (WDM) in terms of aggregate network throughput. Furthermore, it is shown that at high data rates, some of the components for WDM and coherent CDMA networks are nearly identical, indicating a similarity between the two approaches. CDMA retains a coding aspect which may prove attractive in security applications.

Signaling via space-time coding is proposed for asynchronous multiple access free-space and fiber-based optical communications and interference mitigation. Preliminary simulation results are carried out to demonstrate operating principles.

In optical code-division multiple access (CDMA), the architecture of optical encoders/decoders is another important factor that needs to be considered, besides the correlation properties of those already extensively studied optical codes. The architecture of optical encoders/decoders affects, for example, the amount of power loss and length of optical delays that are associated with code sequence generation and correlation, which, in turn, affect the power budget, size, and cost of an optical CDMA system. Various CDMA coding architectures are studied in the paper. In contrast to the encoders/decoders used in prime networks (i.e., prime encodes/decoders), which generate, select, and correlate code sequences by a parallel combination of fiber-optic delay-lines, and in 2ⁿ networks (i.e., 2ⁿ encoders/decoders), which generate and correlate code sequences by a serial combination of 2 X 2 passive couplers and fiber delays with sequence selection performed in a parallel fashion, the modified 2ⁿ encoders/decoders generate, select, and correlate code sequences by a serial combination of directional couplers and delays. The power and delay- length requirements of the modified 2ⁿ encoders/decoders are compared to that of the prime and 2ⁿ encoders/decoders. A 100 Mbit/s optical CDMA experiment in free space demonstrating the feasibility of the all-serial coding architecture using a serial combination of 50/50 beam splitters and retroreflectors at 10 Tchip/s (i.e., 100,000 chip/bit) with 100 fs laser pulses is reported.

The steady-state and dynamic properties of a DFB amplifier have been analyzed from theoretical and experimental viewpoints. The linewidth enhancement factor has been estimated from the experimental hysteresis cycle using an inversion technique of the device transmissivity. The dynamic behavior of the DFB has been tested at 622 Mbit/s and the experimental results are supported by numerical simulations based on an assessed model.

This article compares the performance of several optical receiver configurations optimized for 10 GHz bandwidth: (1) PIN/lossy matched pre-amplifier; (2) feedback amplifier; and (3) distributed pre-amplifier. For this purpose a new technique was developed in order to perform circuit optimization taking into account all the important circuit characteristics such as gain, group delay, and noise spectral density. This technique can be implemented using a standard circuit simulator and is applicable both to direct detection (including sub-carrier multiplexing) and coherent receivers. Given the current trend toward the utilization of high capacity transmission systems, the results and the techniques presented can be of major utility for the design of ultra wide bandwidth optical receivers, a key element in such systems.

In this paper we introduce a new laser diode simulator for general multisection and multielectrode structures. Lasing mode features like linewidth, intensity and frequency modulation responses, etc., are calculated considering all the major physical effects. Furthermore, this model allows us to evaluate thermal effects in order to obtain more realistic results also at high injection rates.

Semiconductor laser amplifiers have been the object of increasing interest in recent years. Such an interest is due to the possibility of low noise direct amplification of optical signals in future high speed and coherent optical transmission systems. The travelling wave amplifier (TWA) is a Fabry-Perot amplifier (FPA) with anti-reflection coated end facets. TWAs present the disadvantage of a higher spontaneous emission noise level, which in FPAs is reduced by the wavelength selective cavity effect. On the other hand, the spectral characteristics of FPAs also mean a reduced signal bandwidth and the possibility of undesirable amplified signal reflections. With respect to homogeneously pumped laser amplifiers, multi-section structures for TWAs (MS-TWAs) seem to be attractive from the point of view of both the adjustable parameters (electrical pumping of each section) and the device performances in terms of gain saturation characteristics and four-wave mixing. In this paper we consider the possibility of optimizing 1.55 micrometers MS-TWAs with respect to the length and the electrical injection level of each homogeneously pumped region and to the input signal wavelength. The optimization is accomplished by taking into account the continuous-wave (cw) gain-output power characteristics and the thermal effects within the device. The analysis of the transient behavior shows that optical input pulses can undergo considerable amplitude distortion, if the amplifier is not properly designed.

We demonstrated experimentally relative frequency stabilization of a set of 1551 nm semiconductor lasers using optical phase locking. A semiconductor laser is locked to 6 consequent sidebands of an rf phase-modulated semiconductor master laser with a frequency stability better than 700 kHz.

In this paper, we present an `almost-all' optical packet switch architecture that does not rely on recirculating loops for storage implementation. Our architecture is based on two rearrangeably non-blocking stages interconnected by optical delay lines with different amount of delay. We've investigated the probability of loss as a function of link utilization and the size of the switch. In general, with proper setting of the number of delay lines, the switch can achieve arbitrarily low probability of loss. The latency characteristics of the switch were also investigated. Possible design of the electronic control circuit was shown. We estimated that transparent-bit-rate 32 X 32 switches may easily be implemented with ECL logic.

Recently, there has been an increased awareness of the impact queuing plays in the performance and design of high-performance packet switches. For instance, the importance of memory sharing and trunking is generally accepted by researchers and developers today, especially when one considers `real' traffic that is `bursty.' Burstiness can result from the sources themselves, and can be exacerbated by store-and-forward networks. In this paper, we address some new features and characteristics of queuing in optical packet switches. All of the fundamental results (e.g., the performance advantages of output queuing and the memory reduction of buffer sharing) still apply to optical packet switches. However, the limited state of optical technology (e.g., the lack of an optical random access memory) puts constraints on the architecture of optical packet switches. The lack of an optical random access memory led many in the past to propose hybrid switch architectures that exploit the advantages of both optics and electronics, using electronics for the queuing of packets. Here, we instead focus our attention on the fundamental performance limitations associated with `all-optical' packet switches, in which the packet buffering is implemented via fiber delay lines.

This paper describes some novel architectures for optical TDM switching, and an experimental system working at 720 Mb/s; this includes a complete BER characterization. The architectures are composed of 2 X 2 optical switches and delay lines. The trade-offs between the various factors affecting their design are explored, and a wide range of architectures are presented suitable for different applications.

Wavelength division multiple access (WDMA) as a technique for achieving very high capacity local area networks (LANs) is introduced. New efficient WDMA demand assignment protocols suitable for operation in such a high speed environment are presented and their performances are investigated via discrete event simulation. Symmetric traffic loading is initially assumed before investigation protocol performance for the more realistic case of a network with asymmetrically distributed traffic loads. For each traffic loading type performance characteristics are compared with those of a single channel equivalent operating with the same overall network transmission rate. Under symmetric traffic loading, multichannel protocol performance is shown to exhibit an improvement which is particularly significant at higher speeds. It is then demonstrated that the multichannel protocol performance is much less sensitive to the degradation caused by asymmetric traffic loads resulting in a substantial improvement in the maximum throughput achievable as compared to the single channel equivalent. Finally, the potential capacity of a wavelength division multiplexed optical fiber LAN using such a protocol is explored.

For OFDM systems based on multistage interconnection networks with space-frequency interstage patterns the number of crossed channels and the interconnection length in the frequency domain is aimed to be a minimum. A solution which only requires nearest-neighbor interconnections in the frequency domain is proposed and discussed.

A parallel processing architecture based on multiple channel optical communication is described and compared with existing interconnection strategies for parallel computers. The proposed multiple channel architecture (MCA) uses MQW-DBR lasers to provide a large number of independent, selectable channels (or virtual buses) for data transport. Arbitrary interconnection patterns as well as machine partitions can be emulated via appropriate channel assignments. Hierarchies of parallel architectures and simultaneous execution of parallel tasks are also possible. Described are a basic overview of the proposed architecture, various channel allocation strategies that can be utilized by the MCA, and a summary of advantages of the MCA compared with traditional interconnection techniques. Also describes is a comprehensive multiple processor simulator that has been developed to execute parallel algorithms using the MCA as a data transport mechanism between processors and memory units. Simulation results -- including average channel load, effective channel utilization, and average network latency for different algorithms and different transmission speeds -- are also presented.

This paper is a preliminary report on an ongoing experimental effort aimed at the implementation of a prototype of a lightwave digital coherent transmission system based on polarization modulation, or polarization shift keying (POLSK). Several devices specifically designed and built for the experiment are presented. The state of the experimental project is assessed.

Applying a subcarrier multiplexing technique (SCM) in an optical frequency division multiplexing system (OFDM) will improve utilization of the optical channel, at the same time increasing the flexibility in designing an optical network. In this paper, system examples based on the combination of these two techniques ar discussed. Particular attention is given to the application of SCM in OFDM systems which employs FM direct detection techniques. Component characteristics which affect system performance are discussed and results presented.

The performance of Manchester-coded optical wavelength division multiplexing (WDM) systems is evaluated taking into account the shot noise and the four wave mixing (FWM) caused by fiber nonlinearities. The result is compared to conventional non-return-to-zero (NRZ) systems for ASK and DPSK modulation formats. Further, the dynamic range, defined as the ratio of the maximum transmitter power (limited by the FWM) to the minimum transmitter power (limited by receiver sensitivity) is evaluated. For 1.55 micrometers 16 channel WDM systems, the dynamic range of Manchester coded systems show as 2 dB improvement with respect to the NRZ; DPSK systems outperform ASK systems by 5.5 dB. This result holds true for both dispersion-shifted fiber and conventional fiber; it has been obtained for 10 GHz channel spacing, 1 Gbps/channel bit rate and 100 Km transmission length.

Direct frequency modulation of a DFB laser followed by FM to IM conversion in an interferometer is a promising technique to produce low chirp transmitters to be used in high speed direct detection systems. In this paper, the performance of FSK-IM systems operating at 10 Gb/s is analyzed, and the sensitivity and the linewidth requirements are estimated. A comparison between FSK-IM systems and FSK heterodyne systems is also performed, showing that for 1 dB sensitivity penalty, the linewidth requirements for heterodyne systems are higher.

A principal factor in realizing a high-performance bandwidth-efficient fiber communication system at an acceptable cost is the choice of modulation format on the optical carrier. In this context pulse time modulation (PTM) techniques represent an attractive alternative to purely digital or analogue methods. This paper reviews the PTM family and explores their potential for use in high-speed fiber systems intended for transmission of analogue data.

In n-ary PPM, a single high energy pulse is used to convey M bits of information. In this paper we propose that advantage be taken of the fiber non-linearity such that the n-ary PPM pulse becomes a temporal soliton. We examine the maximum achievable bit rate of such a system and compare it to that of PCM. We show that under ideal conditions (no fluctuations of the input soliton pulse width) n-ary PPM can be used to achieve a higher bit-rate than PCM. However, in a practical situation fluctuations of the input soliton pulse width are translated to variations in the soliton pulse position. We show that this has a severe effect on the error probability of n-ary PPM and that in this situation PCM offers greater immunity. Finally, we consider the implications of the Gordon-Haus effect on the maximum transmission distance of n-ary PPM and show that it is less than that for PCM due to the narrower pulse arrival time window.

An original model is established and used to predict the spectral properties of optical fiber pulse position modulation (PPM). A suitable random variable is built into the model to represent the stochastic data sequence encoded. The spectral prediction is then used to examine the problem of slot synchronization under two conditions, first when the channel dispersion is negligible, then when Gaussian type pulses are received. The model is shown to be capable of catering for the different PPM parameters and is used to optimize slot synchronization using these parameters under each of the two conditions. Practical results measured on a PPM system are given and shown to agree with the theoretical predictions, concluding with those conditions that optimize slot synchronization.

Data carrying pulse train compression up to an ultra-high repetition rate based on four-wave mixing is proposed. The method gives an opportunity to encode information at a low repetition rate, then to compress a pulse train up to 10¹¹ - 10¹² bit/s rate and to stretch it to common bit rate at the receiver. A theoretical description of the pulse train compression is derived. The effect of the accompanying processes as stimulated Raman scattering, self-phase modulation, and dispersion broadening are also discussed.

STARNET is a coherent optical wavelength division multiplexed (WDM) local area network. In STARNET, the use of low level amplitude modulation of phase modulated light has been proposed to implement a packet-switched ring network in addition to a high speed circuit- switched network. To verify the feasibility and characterize the effects of amplitude shift- keyed (ASK) modulation on a phase shift-keyed (PSK) receiver, we have measured the sensitivity of a 1.244 Gb/s PSK synchronous heterodyne receiver in the presence of 125 Mb/s ASK modulation. The experiments show that the unmodified (i.e., designed for no ASK modulation) PSK receiver successfully operates with a bit error ratio less than 10^-9 and maintains phase-lock in the presence of ASK modulation for ASK modulation indexes less than 0.5.

A group delay in an optical fiber is one of the major factors limiting the transmission distance of a long-distance, high bit-rate optical fiber communication system using in-line erbium doped fiber amplifiers (EDFAs). The group delay can be compensated by a delay equalizer (a stripline section, for instance) inserted in the intermediate frequency stage of a coherent receiver. The purpose of this paper is to propose a method for compensating the fiber group- delay precisely by adjusting the width distribution of a nonuniform, open-type stripline compensator and thus optimizing its group-delay characteristics. It is shown that, by using the optimized nonuniform stripline, the overall group-delay dispersion is reduced to approximately 1/20 of the value obtained with an ordinary, uniform stripline compensator. The effectiveness of the optimized stripline was confirmed by a computer simulation of the entire system.

In this paper, we describe a demonstration of the optical equalization of polarization dispersion in direct-detection lightwave systems. We use a polarization controller to adjust the polarization into a fiber to one of the principal states of polarization of the fiber, which eliminates first-order polarization dispersion. Results for a 2.5 Gbps, externally modulated system with a fiber with 120 ps rms of polarization dispersion, show that by using the equalizer we maintain a 10^-9 BER, while without the technique the BER varies with time from 10^-5 to 10^-9. At 10 Gbps, the equalizer allows reliable bit detection even though the eye is closed without equalization, demonstrating an order of magnitude increase in the dispersion-limited B²L product, in agreement with our analytical and computer simulation results. We also describe a demonstration of the technique at the receiver, and show how to implement an adaptive polarization controller to continuously track the principal states using a gradient search algorithm. This technique provides bit-rate-independent equalization of first-order polarization dispersion.

The performance and operation of an optically processed coherent optical CDMA network based on electronically reconfigurable optical phase-addressing is described. It is compared to both coherent and incoherent all-optical approaches based on time-addressing and shown to be an attractive system when flexible and extensive multiple access is required.

Recently, significant progress was achieved in the creation of high bit-rate optic fiber communication systems based on single-mode technology. The reduction of semiconductor lasers emission band, the increase in the stability of their parameters, as well as the development of lightguides preserving polarization of the propagating optic emission have provided the possibility to apply efficient methods of signal transmission and processing allowing a great increase of the capacity and the range of communication. Heterodyning (homodyning), angular modulation of carriers, and signal division with a relatively small frequency spacing are only a few of them. Single-mode technology has also allowed implementation of discrete phase modulation with all the advantages inherent to it. However, together with the simplicity of technical implementation of the phase modulators based on the transversal electro-optic effect certain problems arise in the creation of demodulator functional units. They are caused due to a rather small wavelength (lambda) of the optic emission compared to the dimensions of optic elements and result in inaccurate signal phase relationships at various points on the optic circuit unless special measures are taken. Besides, it is extremely difficult to perform synchronization of the heterodyne oscillator required for coherent reception. Under these conditions the application of DPSK (or PDM) modulation, in particular, of higher orders becomes more preferable. In this case signal demodulation with PDM can be performed based on autocorrelation, correlation, or on coherent methods possessing their own peculiarities and restrictions in the optic frequency range.

A review is presented of multilevel coherent and DD optical transmission systems with the aim of a comparative analysis between conventional modulation formats, such as N-PSK, N-QAM and new modulation/demodulation techniques, such as N-4QSK, N-SPSK and PM-DD. Besides the conceptual relevance of new multilevel modulation formats proposed to exploit the transmission characteristics of single-mode optical fibers, important applications can be foreseen like, for example, multiple parallel paths between network nodes or transmission between different processing units of supercomputers.

The objective of this paper is to examine the combined effects of weak phase noise and polarization mode dispersion (PMD) on a coherent receiver employing phase and polarization diversity reception. The receiver is assumed to be subjected to the following: transmitter and receiver polarization misalignment relative to the principal states of the optical fiber, phase noise, polarization mode dispersion, and shot noise. The receiver outputs are investigated for ASK demodulation using square-law and envelope detection. The results show that for the assumed receiver configuration, square law detection provides an output which is independent of PMD, phase noise, and polarization misalignment. Envelope detection results in a receiver output which is dependent on all of these parameters. Furthermore, when phase noise and PMD are simultaneously present, the resulting probability of bit error is no greater than the probability of bit error under worst-case operating conditions when polarization mode dispersion and phase noise are absent.

A detailed theoretical analysis is given for the impact of finite frequency deviation on the sensitivity of dual-filter heterodyne frequency shift keying (FSK) lightwave systems. Our analysis provides closed-form signal-to-noise ratio (SNR) results for estimating the bit-error- ratio (BER) performance of the system. These closed-form results provide an insight into the impact of finite frequency deviation 2(Delta) f_d, laser linewidth (Delta) (nu) , bit rate R_b, and IF filter bandwidths on the system performance. Simulation results indicate that the accuracy of the approximate theory presented in this paper is within 1 dB for linewidths up to 22% when BER equals 10^-9. It is shown that there is a well-defined relationship between the choice of frequency deviation and the tolerable amount of laser phase noise. When there is no phase noise, a frequency deviation of 2(Delta) f_d equals 0.72 R_b is sufficient for 1 dB sensitivity penalty with respect to infinite frequency deviation case; whereas for a linewidth of (Delta) (nu) equals 0.50 R_b the required frequency deviation increases to 2(Delta) f_d equals 3.42 R_b for the same sensitivity penalty. The sensitivity degradation can be very severe for a fixed linewidth as the frequency deviation gets smaller: for a linewidth of 20% the sensitivity penalty is only 0.54 dB when the frequency deviation is infinite whereas it is 3.48 dB when the frequency deviation is 2(Delta) f_d equals R_b.

The exact expression of the bit error rate (BER) for optical soliton communication systems with lumped amplifiers is proposed in consideration of the combined effect of timing jitter and energy fluctuation. Some numerical examples are given. The dependence of the BER and the optimal fiber dispersion on the amplification period and transmission distance are discussed.

Phase diversity homodyne receivers are analyzed when there is a difference in pathlengths between the two quadrature components. Bit error rate (BER) floors due to laser phase noise are calculated. The result shows more sensitivity than one would estimate from heuristic calculations.