In erbium doped optical fiber amplifiers (EDFAs) used in modern high-capacity wavelength division multiplexing (WDM) systems, the gain flatness of EDFA is very important in wide-band long-haul systems. In the EDFAs using the passive gain equalizers, the gain flatness deteriorates due to gain-tilt when the operating condition of the EDFA changes, while the EDFAs should maintain the gain flatness even if the operating condition has changed. To solve this problem, we have developed an active gain-slope compensation technique of an EDFA using a thulium-doped optical fiber (TDF) as a saturable absorber. The actively gain-slope compensated EDFA with the TDF compensator keeps the gain profile constant for the wide gain dynamic range more than 8 dB with the low noise figure less than 6 dB in the wavelength range of 1539-1564 nm.

In erbium doped optical fiber amplifiers (EDFAs) used in modern high-capacity wavelength division multiplexing (WDM) systems, the gain flatness of EDFA is very important in wide-band long-haul systems. In the EDFAs using the passive gain equalizers, the gain flatness deteriorates due to gain-tilt when the operating condition of the EDFA changes, while the EDFAs should maintain the gain flatness even if the operating condition has changed. To solve this problem, we have developed an active gain-slope compensation technique of an EDFA using a thulium-doped optical fiber (TDF) as a saturable absorber. The actively gain-slope compensated EDFA with the TDF compensator keeps the gain profile constant for the wide gain dynamic range more than 8 dB with the low noise figure less than 6 dB in the wavelength range of 1539-1564 nm.

Highly reliable and high power weakly index guided buried-stripe type 980 nm pump laser diodes developed for undersea applications are reviewed. The 10,000-hour large scale reliability tests of the first generation LD chips shows that 16.7 FIT for the random failure was confirmed at 10C with 60% confidence level at 120 mW output power. We also fabricated a FBG locked co-axial type module using a can-sealed LD with a two-lens system, which showed a stable FBG locked mode oscillation at 980 nm under the temperature range from 5C to 45C. The 5,000-hour heat cycle test of the modules reveals that the cumulative failure rate after 27 years at 10C is expected to be 0.023%. These first generation LD modules were employed in the transoceanic commercial systems such as Pacific Crossing-1 and the Japan-US cable system projects. We have also succeeded in developing the 980 nm LD for higher output operation with optimizing waveguide design. The 1000 µm long LD showed CW kink-free operation up to 545 mW optical output and a maximum output power of over 650 mW, which was limited by thermal rollover. In addition, a preliminary aging test at 350 mW optical output power at 50C has shown stable operation up to 2,300 h. We also confirmed 300 mW kink-free fiber output power with a co-axial type module with the improved coupling efficiency of approximately 78%. These figures are the highest reported operation levels for 980-nm co-axial type modules.

Highly reliable and high power weakly index guided buried-stripe type 980 nm pump laser diodes developed for undersea applications are reviewed. The 10,000-hour large scale reliability tests of the first generation LD chips shows that 16.7 FIT for the random failure was confirmed at 10C with 60% confidence level at 120 mW output power. We also fabricated a FBG locked co-axial type module using a can-sealed LD with a two-lens system, which showed a stable FBG locked mode oscillation at 980 nm under the temperature range from 5C to 45C. The 5,000-hour heat cycle test of the modules reveals that the cumulative failure rate after 27 years at 10C is expected to be 0.023%. These first generation LD modules were employed in the transoceanic commercial systems such as Pacific Crossing-1 and the Japan-US cable system projects. We have also succeeded in developing the 980 nm LD for higher output operation with optimizing waveguide design. The 1000 µm long LD showed CW kink-free operation up to 545 mW optical output and a maximum output power of over 650 mW, which was limited by thermal rollover. In addition, a preliminary aging test at 350 mW optical output power at 50C has shown stable operation up to 2,300 h. We also confirmed 300 mW kink-free fiber output power with a co-axial type module with the improved coupling efficiency of approximately 78%. These figures are the highest reported operation levels for 980-nm co-axial type modules.

To enhance fault tolerance ability of the feedforward neural networks (NNs for short) implemented in hardware, we discuss the learning algorithm that converges without adding extra neurons and a large amount of extra learning time and cycles. Our algorithm modified from the standard backpropagation algorithm (SBPA for short) limits synaptic weights of neurons in range during learning phase. The upper and lower bounds of the weights are calculated according to the average and standard deviation of them. Then our algorithm reupdates any weight beyond the calculated range to the upper or lower bound. Since the above enables us to decrease the standard deviation of the weights, it is useful in enhancing fault tolerance. We apply NNs trained with other algorithms and our one to a character recognition problem. It is shown that our one is superior to other ones in reliability, extra learning time and/or extra learning cycles. Besides we clarify that our algorithm never degrades the generalization ability of NNs although it coerces the weights within the calculated range.

In long-haul wavelength-division-multiplexed (WDM) transmission systems, signals with shorter and longer wavelengths have self-phase modulation group-velocity-dispersion (SPM-GVD) penalty caused by to the dispersion slope even after the dispersion-compensation at the receiver has been optimized. As a countermeasure, we have already proposed both pre-compensation and post-compensation of chromatic dispersion at the transmitter and receiver for each channel. This method can decrease the channel variation of path-averaged chromatic dispersion along the transmission line, and it can improve the eye opening of the waveform after transmission. We investigated the optimized parameter of chromatic dispersion and chirping at the transmitter. The optimized pre-dispersion compensation parameter R was about 50%. The optimized chirping parameter α was about 3 when the signal wavelength was less than the mean zero-dispersion wavelength. In a single-channel, 5.3-Gbit/s NRZ signal transmission experiment over a 4,760-km straight line, this method decreased SPM-GVD penalty. In a 32-channel, 5.3-Gbit/s WDM transmission experiment over 9,879 km using a circulating loop, this method improved Q-factors for the 1st and 32nd channels by more than 1.5 dB.

In an optical submarine cable transmission system, small size, low consumption power, and high reliability are required for inline repeaters. The structure of the inline repeater should be a simple single stage. The design of erbium doped fiber (EDF) itself is very important for the inline repeater to achieve broad bandwidth, high output power, and low noise figure. We designed and developed high alumina co-doped erbium doped fiber amplifiers (EDFAs) for long-haul, high-capacity WDM transmission systems. We investigated the trade-off relationship between the gain flatness and the output power to optimize the EDF length. We obtained high performance, including a slightly sloped gain flatness of +0.04 dB/nm at 1550 nm, a superior noise figure of 4.7 dB, and a relatively large output power of +11.5 dBm for an EDF length of 5 m using a 1480-nm pumping laser diode. We applied gain-equalizers (GEQs) using Mach-Zehnder type filters with different FSRs to accurately compensate for the EDFAs ' gain-wavelength characteristics. The main GEQs have free-spectral-ranges (FSRs) of 48-nm, which are about 2 times as long as the wavelength difference between a 1558-nm EDFA gain peak and a 1536-nm EDFA gain valley. Using a circulating loop with the above EDFAs and GEQs, we performed the broad wavelength bandwidth. The achieved signal wavelength bandwidth after 5,958-km transmission was 20 nm. We successfully transmitted 700-Gbit/s (66 10.66-Gbit/s) WDM signals over 2,212 km. The combination of high alumina co-doped silica EDFA and large FSR GEQ is attractive for long-haul, high-capacity WDM transmission systems.

In pattern recognition using neural networks, it is very difficult for researchers or users to design optimal neural network architecture for a specific task. It is possible for any kinds of neural network architectures to obtain a certain measure of recognition ratio. It is, however, difficult to get an optimal neural network architecture for a specific task analytically in the recognition ratio and effectiveness of training. In this paper, an evolutional method of training and designing feedforward neural networks is proposed. In the proposed method, a neural network is defined as one individual and neural networks whose architectures are same as one species. These networks are evaluated by normalized M. S. E. (Mean Square Error) which presents a performance of a network for training patterns. Then, their architectures evolve according to an evolution rule proposed here. Architectures of neural networks, in other words, species, are evaluated by another measurement of criteria compared with the criteria of individuals. The criteria assess the most superior individual in the species and the speed of evolution of the species. The species are increased or decreased in population size according to the criteria. The evolution rule generates a little bit different architectures of neural network from superior species. The proposed method, therefore, can generate variety of architectures of neural networks. The designing and training neural networks which performs simple 3 3 and 4 4 pixels which include vertical, horizontal and oblique lines classifications and Handwritten KATAKANA recognitions are presented. The efficiency of proposed method is also discussed.

Coaxial-core erbium-doped fiber amplifiers (EDFA's) having a property of self-regulated gain spectrum are developed. The operation of a coaxial-core EDFA is based on the partial separation of the light paths for different wavelength channels in the directionally-coupled waveguides of a coaxial-core geometry. The degree of channel equalization depends on the geometrical and optical parameters of the coaxial-core EDFA and on relative channel power levels. A numerical analysis based on the coupled-mode theory and on the rate equation shows that, under fully optimized conditions, a coaxial-core EDFA provides equalization rates in excess of -0.4 dB per dB of input-power imbalance in the case with two WDM channels. A cascade experiment demonstrates the effect of coaxial-core EDFA's toward channel-power equalization in fiber links with a small number of WDM channels.

Coaxial-core erbium-doped fiber amplifiers (EDFA's) having a property of self-regulated gain spectrum are developed. The operation of a coaxial-core EDFA is based on the partial separation of the light paths for different wavelength channels in the directionally-coupled waveguides of a coaxial-core geometry. The degree of channel equalization depends on the geometrical and optical parameters of the coaxial-core EDFA and on relative channel power levels. A numerical analysis based on the coupled-mode theory and on the rate equation shows that, under fully optimized conditions, a coaxial-core EDFA provides equalization rates in excess of -0.4 dB per dB of input-power imbalance in the case with two WDM channels. A cascade experiment demonstrates the effect of coaxial-core EDFA's toward channel-power equalization in fiber links with a small number of WDM channels.

In learning of feed-forward neural networks, so-called 'training error' is often minimized. This is, however, not related to the generalization capability which is one of the major goals in the learning. It can be interpreted as a substitute for another learning which considers the generalization capability. Admissibility is a concept to discuss whether a learning can be a substitute for another learning. In this paper, we discuss the case where the learning which minimizes a training error is used as a substitute for the projection learning, which considers the generalization capability, in the presence of noise. Moreover, we give a method for choosing a training set which satisfies the admissibility.

Wavelength-division multiplexing (WDM) transmission systems have been intensely researched in order to increase the transmission capacity. One of the most important key devices for this use is erbium-doped fiber amplifiers (EDFAs) which feature a flattened gain, a high pumping efficiency and a low noise figure (NF), simultaneously. To fulfill these requirements, hybrid silica-based EDFAs (EDSFAs) composed of Al codoped and P/Al codoped EDSFs have been proposed so far. They are also attractive from the viewpoint of productivity, reliability, and cost-effectiveness. On the other hand, the optical bandwidth has been around 15 nm at most. In this paper, we have proposed newly designed hybrid EDSFAs for more than 25 nm optical bandwidth. The gain peak around 1. 53 µm can be suppressed through the saturation degree control in both EDSFs. The remaining obstacle is the diparound 1. 54 µm, which results in the relative gain non-uniformity of 10. 7% over the wavelength range from 1535 to 1560 nm. Owing to the glass composition optimization, the relative gain non-uniformity has been reduced to 5.8% without gain equalizers(GEQs), which is comparable to that of EDFFAs. As another solution, the hybrid EDSFA including two-stage Fabry Perot etalons as the GEQ has been proposed. In this configuration, the hybrid EDSFA has been designed to exhibit the gain profile similar to the summation of two sinusoidal curves, and the relative gain non-uniformity has been reduced to 3. 7%, which is almost equal to that of the hybrid EDFAs composed of EDSF and EDFF. Moreover, it has been demonstrated that newly developed hybrid EDSFAs exhibit a higher pumping efficiency and a lower NF than EDFFAs and hybrid EDSF/EDFFAs.

In this paper, a dynamic constructive algorithm for fault tolerant feedforward neural network, called DCFTA, is proposed. The algorithm starts with a network with single hidden neuron, and a new hidden unit is added dynamically to the network whenever it fails to converge. Before inserting the new hidden neuron into the network, only the weights connecting the new hidden neuron to the other neurons are trained (i. e. , updated) until there is no significant reduction of the output error. To generate a fault tolerant network, the relevance of each synaptic weight is estimated in each cycle, and only the weights which have their relevance less than a specified threshold are updated in that cycle. The loss of a connections between neurons (which are equivalent to stuck-at-0 faults) are assumed. The simulation results indicate that the network constructed by DCFTA has a significant fault tolerance ability.

Considering the pattern classification/recognition tasks, the influence of the activation function on fault tolerance property of feedforward neural networks is empirically investigated. The simulation results show that the activation function largely influences the fault tolerance and the generalization property of neural networks. It is found that, neural networks with symmetric sigmoid activation function are largely fault tolerant than the networks with asymmetric sigmoid function. However the close relation between the fault tolerance and the generalization property was not observed and the networks with asymmetric activation function slightly generalize better than the networks with the symmetric activation function. First, the influence of the activation function on fault tolerance property of neural networks is investigated on the XOR problem, then the results are generalized by evaluating the fault tolerance property of different NNs implementing different benchmark problems.

A new learning algorithm is proposed to enhance fault tolerance ability of the feedforward neural networks. The algorithm focuses on the links (weights) that may cause errors at the output when they are open faults. The relevances of the synaptic weights to the output error (i.e. the sensitivity of the output error to the weight fault) are estimated in each training cycle of the standard backpropagation using the Taylor expansion of the output around fault-free weights. Then the weight giving the maximum relevance is decreased. The approach taken by the algorithm described in this paper is to prevent the weights from having large relevances. The simulation results indicate that the network trained with the proposed algorithm do have significantly better fault tolerance than the network trained with the standard backpropagation algorithm. The simulation results show that the fault tolerance and the generalization abilities are improved.

Very-low-voltage 1.2-V mixed-signal CMOS technology is a device/circuit solution aimed at ultra-low-power portable systems such as digital cellular terminals and PDAs. We have developed an experimental 1.2-V mixed analog and digital LSI circuit/device technology. This technology is based on a new transistor structure that has a 0.3-µm gate length and a low Vth of 0.4 V, and that suppresses the short-channel effect. In this paper, we will mainly discuss low-voltage analog circuit design that uses this technology. We show that low Vth is essential not only to digital circuits, but also to 1.2-V analog amplifier, A/D converter and analog switch designs. To achieve high-conversion rate A/D converters, a pipeline architecture is used for low-voltage operation. To increase the attainable gain-bandwidth of the operational amplifier of the converter, a feedforward phase-compensated three-stage amplifier is proposed. The addition of a feedforward capacitor allows a high frequency signal to pass directly to the second stage, which optimizes use of the second stage bandwidth. Pole-zero canceling is used to achieve a fast settling of the amplifier. Although gain precision is degraded by the positive feedback through the feedforward capacitor, this can be offset by increasing the equivalent second-stage gain with an inner feedforward compensated amplifier. The gain-bandwidth of the proposed double feedforward amplifier is two to three times wider than with the conventional Miller compensation. With these techniques, we used 1.2-V mixed-signal CMOS technology to create a basic logic gate with a 400-ps delay and 0.4-µW/MHz power, and a 9-bit 2-Msample/s pipeline A/D converter with power dissipation of only 4 mW.

We report here a pulsed lightwave frequency synthesizer system that is composed of a pulsed lightwave sweep frequency generator and a tracking generator. The key advance in the sweep generator is the use of a dynamically gain controlled EDFA. The combination of feedback and feed forward dynamic gain control effectively compensates EDFA gain fluctuation and equalizes fiber loop loss so that the initial pulse wave form and amplitude is retained in the loop at large circuit numbers. Over 1000 pulsed lightwave frequencies are synthesized in 250MHz steps by the sweep generator. Almost flat response (0.55dB variation) is realized up to 240GHz. The power spectrum decreases by 67% (1.7dB down) at 250GHz. The peak level of the pulses output from the loop is about -4dBm. Tracking generator and total synthesizer system performance are evaluated by (a) beat frequency between the tracking generator and the master lightwave source, (b) beat frequency between two tracking generators, and (c) a frequency chain between the master lightwave source and another HCN stabilized lightwave source via the synthesizer system. A continuous lightwave frequency locked to a frequency selected from the pulsed sweep frequency signal is demonstrated at over 200GHz to have an instability of 5MHz. Absolute accuracy of the lightwave frequency emitted from the synthesizer system is about 10MHz. Therefore, the relative accuracy of the lightwave frequency is as high as 510-8.

Feasibility of 20 Gbit/s single channel transoceanic soliton transmission systems with a simple EDFA repeaters configuration has been studied. Both a simple and versatile soliton pulse generator and a polarization insensitive optical demultiplexer, which can provide a almost square shape optical gate with duration of full bit time period, have been proposed and demonstrated by using sinusoidally modulated electroabsorption modulators. The optical time-division multiplexing/demultiplexing scheme using the optical demultiplexer results in drastic improvement of bit error rate characteristics. We have experimentally confirmed that the use of alternating-amplitude solitons is an efficient way to mitigate not only soliton-soliton interaction but also Gordon-Haus timing jitter constraints in multi-ten Gbit/s soliton transmission. Timing jitter reduction using relatively wide band optical filter bas been investigated in 20 Gbit/s loop experiments and single-carrier, single-polarization 20 Gbit/s soliton data transmission over 11500 km with bit error rate of below 10-9 has been experimentally demonstrated, using the modulator-based soliton source, the optical demultiplexer, the alternation-amplitude solitons, and wide-band optical filters. Obtained 230 Tbit/skm transmission capacity shows the feasibility of 20 Gbit/s single channel soliton transoceanic systems using fully practical technologies.

Soliton transmission control has already proved to be an outstanding technique and enable a soliton to be transmit over one million kilometers. This technique is not only applicable to vast distances but also to shorter distances where the amplifier spacing is greater than that of conventional systems. A combination of time and frequency domain control eliminates the noise accumulation and timing jitter caused by soliton interaction and the Gordon-Haus effect, that are the main impediments to extending the transmission distance. In this paper we describe soliton control techniques applied over an astronomical transmission distance of 180,000,000 km, and to a terrestrial system with a large amplifier spacing of up to 100km. We also report the possibility of realizing a sub-tera bit/s soliton transmission system operating over more than 5,000 km in which the soliton self-frequency shift is controlled with the soliton control technique.

We describe 2.5Gb/s 4 channel WDM transmission over 1060km using 18 EDFAs. Gain bandwidth narrowing in concatenated EDFAs has been successfully suppressed using unsaturated EDFAs and a 1.53µm ASE rejection filter.