Passive optical network (PON) is viewed by many as an attractive solution to the first mile problem. With the rapidly increasing number of users and bandwidth intensive applications, upgrading current PON architecture with the wavelength division multiplexing (WDM) technology has become a natural choice. Multiple WDM PON architectures have been proposed, yet to the best of our knowledge none could support deterministic quality of service (QoS) so far. In this paper we propose a two-level WDM EPON solution which provides two main functions: efficient network scaling with bandwidth sharing at the wavelength level, and deterministic QoS provisioning at the EPON level. To guarantee deterministic QoS for multiple classes, an efficient admission control scheme based on network calculus has been developed in conjunction with the scheduling discipline. We evaluate the performance of our scheme by means of extensive simulations, and show that it could not only provide deterministic QoS to bursty traffic but also achieve high utilization of resources.

In this paper, we propose a novel 2-D layered multiple description coding (2DL-MDC) for error-resilient video transmission over unreliable networks. The proposed 2DL-MDC scheme allocates multiple description sub-bitstreams of a 2-D scalable bitstream to two network paths with unequal loss rates. We formulate the 2-D scalable rate-distortion problem and derive the expected distortion for the proposed scheme. To minimize the end-to-end distortion given the total rate budget and packet loss probabilities, we need to optimally allocate source and channel rates for each hierarchical sublayer of the scalable bitstream. The conventional Lagrangian multiplier method can be utilized to solve this problem but with overwhelming computational complexity. Therefore, we consider the use of the genetic algorithm to solve the rate-distortion optimization problem. The simulation results verify that the proposed method is able to achieve significant performance gain as opposed to the conventional equal rate allocation method.

The new IEEE 802.lie standard is designed to provide QoS guarantees to different types of multimedia applications in WLAN. IEEE 802.lie defines two channel access mechanisms: HCCA and EDCA. Basically, HCCA is more suitable for providing strict QoS guarantees than EDCA because of its centralized TXOP allocation manner and enhanced capability of initiating a transmission at any time during the beacon interval. Nevertheless, the HCCA scheduling scheme is not yet consummate in the current standard and it only performs well under CBR real-time traffics. In this paper, we propose a scheduling algorithm and admission control policy based on Service Curve theory that can deal with bursty traffic while strict delay bounds can be provided for traffic streams. We conduct extensive simulations to evaluate our scheme and the experiment results show good improvements upon the reference scheduling algorithm of HCCA.

This paper presents a novel video scheduling algorithm for streaming MPEG video over multi-user shared wireless link, in which video sessions are scheduled based on both location-dependent channel condition and instantaneous frame delay. This algorithm also employs a pro-active frame discard mechanism to discard potential problematic frames. The numerical results show that the proposed algorithm is able to significantly improve link throughput and guarantee long-term fairness.

According to conventional neural network theories, single-hidden-layer feedforward networks (SLFNs) with additive or radial basis function (RBF) hidden nodes are universal approximators when all the parameters of the networks are allowed adjustable. However, as observed in most neural network implementations, tuning all the parameters of the networks may cause learning complicated and inefficient, and it may be difficult to train networks with nondifferential activation functions such as threshold networks. Unlike conventional neural network theories, this paper proves in an incremental constructive method that in order to let SLFNs work as universal approximators, one may simply randomly choose hidden nodes and then only need to adjust the output weights linking the hidden layer and the output layer. In such SLFNs implementations, the activation functions for additive nodes can be any bounded nonconstant piecewise continuous functions g : R --> R and the activation functions for RBF nodes can be any integrable piecewise continuous functions g : R --> R and integral of R g(x)dx not equal to 0. The proposed incremental method is efficient not only for SFLNs with continuous (including nondifferentiable) activation functions but also for SLFNs with piecewise continuous (such as threshold) activation functions. Compared to other popular methods such a new network is fully automatic and users need not intervene the learning process by manually tuning control parameters.

In some practical applications of neural networks, fast response to external events within an extremely short time is highly demanded and expected. However, the extensively used gradient-descent-based learning algorithms obviously cannot satisfy the real-time learning needs in many applications, especially for large-scale applications and/or when higher generalization performance is required. Based on Huang's constructive network model, this paper proposes a simple learning algorithm capable of real-time learning which can automatically select appropriate values of neural quantizers and analytically determine the parameters (weights and bias) of the network at one time only. The performance of the proposed algorithm has been systematically investigated on a large batch of benchmark real-world regression and classification problems. The experimental results demonstrate that our algorithm can not only produce good generalization performance but also have real-time learning and prediction capability. Thus, it may provide an alternative approach for the practical applications of neural networks where real-time learning and prediction implementation is required.

Window-based multicast congestion control protocols such as pgmcc and AER/NCA try to emulate TCP congestion control behaviors for achieving TCP-compatibility. However, they employ NAK-based loss recovery scheme instead of the ACK-based approach, which is used in TCP. The improvement introduced by NAK-based approach on loss recovery capabilities can bring higher throughput for the multicast session under the same network conditions. In this paper, we provide a quantitative evaluation on the influences of NAK-based loss recovery mechanism on the behaviors and throughput properties of window-based multicast congestion control. First, we identify the differences on congestion control behaviors caused by the NAK-based retransmission mechanism. Based on the observations, we mathematically model the throughput of window-based multicast congestion control protocols. Accordingly, the multicast throughput can be expressed as a function of the round-trip time and loss rate experienced by the "worst" receiver. Moreover, selecting the "worst" receiver as the nominee to send congestion control feedbacks are critical to ensure fairness and congestion avoidance in single-rate multicast congestion control schemes. As the TCP throughput equations are used in existing schemes for nominee selection, the obtained multicast throughput equations can be used to enhance the mechanisms since they predict the multicast throughput more accurately than TCP throughput equations.

Support vector machines (SVMs) have been extensively used. However, it is known that SVMs face difficulty in solving large complex problems due to the intensive computation involved in their training algorithms, which are at least quadratic with respect to the number of training examples. This paper proposes a new, simple, and efficient network architecture which consists of several SVMs each trained on a small subregion of the whole data sampling space and the same number of simple neural quantizer modules which inhibit the outputs of all the remote SVMs and only allow a single local SVM to fire (produce actual output) at any time. In principle, this region-computing based modular network method can significantly reduce the learning time of SVM algorithms without sacrificing much generalization performance. The experiments on a few real large complex benchmark problems demonstrate that our method can be significantly faster than single SVMs without losing much generalization performance.

In reliable multicast, buffer management is an important issue that affects the loss recovery performance. Existing reliable multicast protocols do not explicitly address this issue and there is also little known work on it. We focus on the buffer management for active reliable multicast (ARM) when multiple simultaneous reliable multicast sessions are supported by some common active routers. Motivated by the fact that active routers in ARM can perform customized computation and provide soft-stage storage, we propose a new buffer management scheme, called adaptive cache pool (ACP), for cache resource partitioning and buffer replacement to achieve good loss recovery performance. The proposed ACP scheme is validated using ns-2 simulations to compare its performance with that of the original ARM scheme.

In multi-service mobile networks, asymmetric bandwidth allocation has been proposed to satisfy the requirements of the asymmetric traffic load introduced by some data applications. However, it is difficult to promptly adjust bandwidth allocation on the uplink and downlink according to the traffic load dynamics. Superfluous real-time (RT) calls or non-real-time (NRT) calls might be accepted. This may lead to a low bandwidth utilization. In this paper, we propose and evaluate a new call admission control (CAC) scheme to address the problems caused by the mismatch of bandwidth allocation and traffic changes. By determining the admissible bandwidth region for the NRT calls, the proposed scheme prevents the calls from overusing the bandwidth resources. Thus the bandwidth utilization is improved and the blocking probability of the high priority calls can be guaranteed at a low level. The simulation results demonstrate that the proposed CAC scheme can achieve better performance compared to the existing schemes, even those performing well in multi-service mobile networks with asymmetric bandwidth allocation.

A new learning algorithm called extreme learning machine (ELM) has recently been proposed for single-hidden layer feedforward neural networks (SLFNs) to easily achieve good generalization performance at extremely fast learning speed. ELM randomly chooses the input weights and analytically determines the output weights of SLFNs. This paper shows that ELM can be extended to radial basis function (RBF) network case, which allows the centers and impact widths of RBF kernels to be randomly generated and the output weights to be simply analytically calculated instead of iteratively tuned. Interestingly, the experimental results show that the ELM algorithm for RBF networks can complete learning at extremely fast speed and produce generalization performance very close to that of SVM in many artificial and real benchmarking function approximation and classification problems. Since ELM does not require validation and human-intervened parameters for given network architectures, ELM can be easily used.

The gradient-based learning algorithms are usually used to train feedforward neural networks. In these algorithms, the parameters of the network are adjusted iteratively according to the partial gradients of the user-defined performance functions. Such algorithms usually require tens to hundreds of learning epochs to reach the required accuracy. If it sticks in the local minimum in the learning process, the situation tends to be even worse. In Huang et al., a novel fast learning algorithm called extreme learning machine (ELM) for single-hidden-layer neural networks (SLFNs) has been proposed where a constructive method is used instead of a gradient-based learning algorithm. In this paper, we further verify the performance of ELM on two benchmark artificial problems.

Nowdays, neural networks have been widely used. In some areas, such as bioinfomatics and data mining, the problems are usually large-scaled and time-concerned. Thus, the requirement of fast learning algorithm and network architecture for large scale problems is stressed. In this paper, a novel network architecture consisting of several network modules is proposed. In this architecture, each network module is trained to learn a subset of the whole training data, and a number of neural quantizers are used to activate the corresponding network module to output. In principle, this modular network architecture (MNA) can effectively reduce the learning time and achieve the similar gerneralization performance. The results of experiments on several real-world benchmarking problems shown here to make the algorithm convincible.

It is clear that the learning speed of feedforward neural networks is in general far slower than required and it has been a major bottleneck in their applications for past decades. Two key reasons behind may be: 1) the slow gradient-based learning algorithms are extensively used to train neural networks, and 2) all the parameters of the networks are tuned iteratively by using such learning algorithms. Unlike these traditional implementations, this paper proposes a new learning algorithm called extreme learning machine (ELM) for single-hidden layer feedforward neural networks (SLFNs) which randomly chooses the input weights and analytically determines the output weights of SLFNs. In theory, this algorithm tends to provide the best generalization performance at extremely fast learning speed. The experimental results based on real-world benchmarking function approximation and classification problems including large complex applications show that the new algorithm can produce best generalization performance in some cases and can learn much faster than traditional popular learning algorithms for feedforward neural networks.

Server-based loss recovery for reliable multicast can provide significant performance improvement in terms of loss recovery latency and bandwidth consumption. Appropriate congestion control mechanisms can provide fairness and maintains a high network throughput and link utilization. Data delivery, including loss recovery, and congestion control in reliable multicast are not independent issues and should be addressed simultaneously. In this paper, we propose to jointly perform local delivery and congestion control (LDCC). Mechanisms including local loss recovery, acknowledgement processing, buffer management, and congestion control are carefully designed. We demonstrate through simulations that the proposed scheme can achieve significantly lower loss recovery latency without sacrificing the network throughput, compared to existing local recovery protocols such as AER/NCA. It is also shown that our scheme conforms to TCP-compatible requirement.

A growing application area in mobile communications is mobile teleconference in which a group of mobile users collaborate in an interactive procedure, such as a board meeting, a task force, a scientific discussion, or even a virtual classroom. Wireless communications transmit conversations via radio, making them more susceptible to eavesdropping and unauthorized access than are conversations carried via wires. Therefore, it is crucial to ensure confidentiality and authenticity in a mobile teleconference. The authors design a new secure conference scheme for mobile communications. Based on a modular square root technique, this scheme is secure against eavesdropping, impersonating, and tracking attacks and allows a participant to join or quit a mobile teleconference dynamically.

In this paper, we propose a scheme for implementing multiparty videoconferencing service on SONET/ATM rings. We focus on the architectural design and bandwidth demand analysis. Different multicasting methods on SONET/ATM rings are discussed and compared. A new multicast virtual path (VP) called "Multidrop VP" which is particularly suitable for SONET/ATM rings is proposed. An add-drop multiplexer (ADM) structure for rings capable of multidropping is also presented. Several VP assignment schemes are proposed and their bandwidth utilizations are compared.

The problem of finding a minimum-cast multicast tree (Steiner tree) is known as NP-complete. Heuristic-based algorithms for this problem to achieve good performance are usually time-consuming. In this paper, we propose a new strategy called tree-caching for efficient setup of multicast connections in connection-oriented networks. In this scheme, the tree topologies that have been computed are cached in a database of the source nodes and can be used for setup of subsequent connection requests which have some common multicast members. This can reduce the connection establishment time by an efficient reuse of cached trees without having to rerun a multicast routing algorithm for the whole group. This gain is obtained by eliminating, whenever possible, the expensive tree computation algorithm that has to be performed in setting up a multicast connection. We first formulate the problem of tree-caching. We then propose a tree-caching algorithm to reduce the complexity of the tree computations when a new connection is to be established. Through simulations, we find that the proposed tree-caching strategy perform well and can significantly reduce the computation complexity for setting up multicast connections.

We present the dynamic cognitive network (DCN) which is an extension of the fuzzy cognitive map (FCM). Each concept in the DCNs can have its own value set, depending on how precisely it needs to be described in the network. This enables the DCN to describe the strength of causes and the degree of effects that are crucial to conducting meaningful inferences. The arcs in the DCN define dynamic, causal relationships between concepts. Structurally, DNCs are scalable and more flexible as compared to FCMs. A DCN can be as simple as a cognitive map and FCM, or as complex as a nonlinear dynamic system. To demonstrate the potential applications of DCNs, we present some simulation results. This paper represents our first attempt to develop a dynamic fuzzy inference system using causal relationships. There are many interesting and challenging theoretical and practical issues in DCNs open to further research.

In this paper, we propose a two-layer hierarchical cache architecture for enhancing TCP performance over heterogeneous networks with both wired and wireless links. A new network-layer protocol, called New Snoop, is designed. The main idea is to cache the unacknowledged packets at both the mobile switch center (MSG) and base station (BS), thus forming a two-layer cache hierarchy. If a packet is lost due to transmission errors in the wireless link, the BS takes the responsibility to recover the loss. When a handoff occurs during a TCP connection session, the packets cached in MSC can help to minimize the latency of retransmissions due to temporal disconnection. Simulation results show that using New Snoop is significantly more robust in dealing with unreliable wireless inks and handoffs as compared with the Snoop scheme (Balakrishnan et al. 1995) as well as other existing TCP enhancements.

Active reliable multicast (ARM) is a newly proposed loss recovery scheme for reliable multicast over the Internet. For a given amount of total cache available at each active router, the performance of ARM depends on how the amount of cache is partitioned to each multicast session. We call it the cache-partitioning problem. Following the approach of minimizing the total loss recovery traffic in the backbone network, an optimal cache-partitioning scheme is proposed and an analytical model is constructed. The performance of using the proposed optimal cache partitioning is compared with that using uniform cache partitioning and proportional partitioning. A significant performance improvement is found.

Local loss recovery for reliable multicast can provide significant performance improvement in terms of loss recovery latency, bandwidth consumption and network throughput. An analytical model for studying the optimal cache allocation for active reliable multicast (ARM) is constructed using standard optimization techniques, the optimal cache allocation pattern can be found. Our numerical results show that using optimal cache allocation yields significantly smaller loss recovery latency than using uniform cache allocation. To further enhance the loss recovery performance when the amount of cache at an active router is limited, we propose a probabilistic caching policy. We derive the optimal caching probabilities for each active router in a given multicast tree with a given cache allocation pattern. We show that with the use of probabilistic caching policy, a further reduction in loss recovery latency can be obtained.

We propose a confidential information gathering system which employs a software agent traversing a list of information resource servers on the Internet to gather confidential information. Both software agent technology and cryptographic technology are applied to automate and secure the confidential information gathering process.

Changeability is one of the major concerns in software development. OO (object-oriented) technology itself is not enough to solve this problem. In this paper, we first discuss how the changeability issue is handled in current software development methods and what is the most important prerequisite for solving the software changeability problem. Based on these analyses, we argue that most of current OO analysis methods have missed an important activity in their analysis phase, i.e. change analysis. Thus, OO software systems developed according to these methods often cannot exhibit the expected maintainability, extensibility and reusability. In order to improve current OO analysis methods, we have proposed a conceptual framework for conducting a systematic change analysis in the software analysis phase. A detailed case study of the application of this framework in a real software project has also been described in this paper.

We present a dynamic cognitive network (DCN) which is a systematic extension of the fuzzy cognitive map (FCM). Each concept in a dynamic cognitive network can have its own value set, depending on how precisely it needs to be described in the network. This enables the dynamic cognitive network to describe the strength of the cause and the degree of the effect. This ability also enables DCN to carry out inferences via numerical calculation instead of symbolic deduction. The arcs of DCN define dynamic relationships between concepts and describe the causal procedures. DCN overcomes the problems of FCM without losing the essential advantages.