In this paper, we propose a novel discrete-time recurrent neural network aiming to resolve a new class of multi-constrained K-winner-take-all (K-WTA) problems. By facilitating specially designed asymmetric neuron weights, the proposed model is capable of operating in a fully parallel manner, thereby allowing true digital implementation. This paper also provides theorems that delineate the theoretical upper bound of the convergence latency, which is merely O(K). Importantly, via simulations, the average convergence time is close to O(1) in most general cases. Moreover, as the multi-constrained K-WTA problem degenerates to a traditional single-constrained problem, the upper bound becomes exactly two parallel iterations, which significantly outperforms the existing K-WTA models. By associating the neurons and neuron weights with routing paths and path priorities, respectively, we then apply the model to a prioritized flow scheduler for the data center networks. Through extensive simulations, we demonstrate that the proposed scheduler converges to the equilibrium state within near-constant time for different scales of networks while achieving maximal throughput, quality-of-service priority differentiation, and minimum energy consumption, subject to the flow contention-free constraints.

In this paper, we propose an energy and cost efficient WDM/OFDMA PON system architecture, referred to as NEWOPS. Based on a two-tiered semi-tree topology, NEWOPS allows a large number of ONUs to be connected to OLT using only a handful of colorless transceivers. NEWOPS' unique architecture not only facilitates high spectral efficiency and flexible sub-channel bandwidth allocation, but also enables the system to achieve high energy and cost efficiency. We conduct a theoretical study to identify and resolve a new data-rate performance issue due to the colorless and wavelength re-use design of the system. Experimental results show that when only 5-GHz bandwidth is allotted for each wavelength, NEWOPS successfully achieves symmetric 20-Gb/s downstream and upstream OFDM-16QAM transmissions via the same fiber trunk. As a demonstration of our system, we show how NEWOPS achieves energy efficiency that is two to twelve times higher than that of a typical tree-based WDM/OFDM PON when supporting 40 ONUs.

Prioritized flow scheduling is essential for providing QoS differentiation and energy saving in data center network. While dynamically manipulating the routing paths of incoming flows becomes practical via enabling technology of software-defined network (SDN), scheduling a large amount of flows in real time remains challenging. In this paper, we propose an Iterative Parallel Grouping Algorithm (IPGA) to the prioritized flow-scheduling problem. Thanks to the inherent nature of parallelism, the IPGA-based flow scheduler not only performs fast scheduling but also can be efficiently realized by a CUDA system within the SDN controller. Via simulations, we demonstrate that the IPGA scheduler achieves QoS differentiation and promising energy saving for a Fat-tree-based network.

Packet scheduling for WDM optical switching systems requires exceedingly low latency processing, making it impractical to be realized by non-parallel based algorithms. In this paper, we propose a new recurrent discrete-time synchronous ranked neural-network (DSRN) for parallel prioritized scheduling. The DSRN is structured with ranked neurons and is capable of operating in a fully parallel (i.e., synchronous) discrete-time manner, and thus can be implemented in digital systems. We then design a DSRN scheduler for a previously proposed experimental WDM optical switching system (WOPIS). For newly arriving packets, the DSRN scheduler determines in real time an optimal set of input/output paths within WOPIS, achieving maximal throughput and priority differentiation subject to the switch- and buffer-contention-free constraints. We delineate via a theorem that DSRN will converge to the optimal solution. The theorem also provides a theoretical upper bound of the convergence latency, O(H), where H is the switch port count. Finally, we demonstrate that, via CUDA-based simulations, the DSRN scheduler achieves near-optimal throughput and prioritized scheduling, with nearly O(logH) convergence latency.

OFDM PON has been considered to be one of the most prominent next-generation broadband wired access solutions. However, based on the current tree-architecture, existing OFDM PON systems face severe challenges when increasing the scalability and data-rate performance. In this paper, we propose a novel virtual-tree architecture of an OFDMA PON system (VTOPS). Coupled with the use of inexpensive direct modulation, VTOPS simultaneously achieves high reliability, scalability, and spectrum efficiency (i.e., high data rate) in a cost-effective manner. Theoretical and experimental results demonstrate that, by applying the pre-emphasis algorithm and subchannel adaptive modulation, VTOPS achieves 40-Gb/s downlink and uplink transmissions using low-cost 10-GHz directly modulated lasers.

In this paper, we propose a novel ranked Hopfield neural-network (RHNN) parallel scheduler for a WDM optical interconnection system (WOPIS), containing a set of Clos-like optical switches and a handful of output FDL-based optical buffers. The RHNN scheduler determines an optimal set of neurons (I/O paths) to be enabled, achieving maximal system throughput and priority differentiation subject to the switch- and buffer-contention-free constraints. The RHNN is specially structured with ranked neurons. With each neuron being associated with an input/output path within WOPIS, the RHNN allows higher-rank neurons (higher-priority and/or lower-delay paths) to disable lower-rank neurons that were enabled during previous iterations. Ranking the neurons unfortunately gives rise to a convergence problem. We present two theorems that supply the sufficient conditions for the RHNN scheduler to converge to the optimal solution. We demonstrate via simulation results that, with the computation time of less than one system time slot, the RHNN scheduler achieves near 100% throughput and multi-level prioritized scheduling.

In this paper, we propose a novel architecture for next-generation orthogonal frequency division multiple access (OFDMA)-based passive optical networks (PON's), referred to as ROFPON. Besides carrying local broadband OFDMA data, ROFPON seamlessly supports radio-over-fiber (RoF) transports between the central office and multiple remote antennas at end users without using costly WDM lasers. We analytically and experimentally study the receiver sensitivity to OFDMA signals and the radio frequency (RF) signal's performance. By corroborating simulation results with experimental results, we discuss the determination of crucial system parameters, such as the optimal broadband-to-radio power ratio, and the exploitation of a notch filter for removing RF interference. Experimental results show that the integrated 10 Gb/s OFDMA and three 20 MHz RF signals are successfully transported both downstream and upstream over a 20 km single-mode-fiber PON. Finally, experimental results demonstrate that QPSK-encoded WiMAX format RF signals are transmitted/relayed upstream with E-O-E conversion at each optical network unit (ONU), and received error-free at the optical line terminal after cascading 32 ONU's.

This paper proposes a nonlinear-predictive QoS-promoted dynamic bandwidth allocation (NQ-DBA) scheme for upstream transmission in Ethernet passive optical network (EPON) systems. The proposed NQ-DBA scheme divides incoming packets of voice, video, data service into six priorities, whereas packets having less room before QoS requirements violation will be dynamically promoted to high priority. It predicts packets arriving at a prediction interval for ONUs using a nonlinear pipelined structure of recurrent neural network (RNN) with extended recursive least squares (ERLS) update so that the bandwidth allocation can be more up-to-date and then accurate. Simulation results show that the proposed NQ-DBA scheme achieves higher system utilization than the DBAM scheme by 4%, and 21%, 90%, 43%, respectively.

In this paper, we present the access and control design of a high-performance optical packet-switched WDM metro ring network (HOPSMAN). HOPSMAN has been designed for networks and nodes to be unconstrained by the number of wavelengths. It includes a handful of nodes that are equipped with fast optical slot erasers making bandwidth reusable and achieving greater bandwidth efficiency. In essence, HOPSMAN incorporates a versatile medium access control (MAC) scheme, which embodies efficient and fair bandwidth allocation in accordance with a quota being exerted probabilistically. The quota is analytically derived with the number of slot-eraser-nodes taken into account. The scheme also employs a new notion of credit to regulate flexible access of remaining bandwidth that is suitable for the metro environment with bursty traffic. With the MAC scheme, HOPSMAN is shown to achieve exceptional throughput, delay, and fairness performance under a wide range of traffic settings via simulation results.

Optical Packet Switching (OPS) has been envisioned as a prominent future optical networking technology for datacentric IP over Wavelength Division Multiplexing (WDM) networks, or optical Internet. Such OPS technology however raises significant transport and Quality of Service (QoS) challenges due to technological limitations. To circumvent OPS limitations, we have proposed a new Optical Coarse Packet Switching (OCPS) paradigm, which uses in-band-controlled per-burst switching and advocates traffic control enforcement to achieve high packet-loss performance and differentiated services. Based on OCPS, we have constructed an experimental IP-over-WDM network, referred to as OPSINET. OPSINET consists of two major types of nodes-edge routers, and Optical Label Switched Routers (OLSRs), and is facilitated with an out-of-band Generalized Multi-protocol Label Switching (GMPLS) control network. In this paper, we first introduce the OCPS paradigm. We then present the architecture of OPSINET, describe the in-band header/payload modulation technique, and detail the operations of the edge routers, OLSRs, and GMPLS control.

In this paper, we propose an efficient approximation approach, called Lagrangean relaxation with heuristics (LRH), aimed to resolve routing and wavelength assignment (RWA) for multi-granularity WDM networks facilitating fiber, waveband, and lambda switching capabilities. The task is first formulated as a combinatorial optimization problem in which the bottleneck link utilization is to be minimized. The LRH approach performs constraint relaxation and derives a lower-bound solution index according to a set of Lagrangean multipliers generated through subgradient-based iterations. In parallel, using the generated Lagrangean multipliers, the LRH approach employs a new heuristic algorithm to arrive at a near-optimal upper-bound solution. With lower and upper bounds, we delineate the performance of LRH with respect to accuracy and convergence speed under different parameter settings. We further draw comparisons between LRH and a typical linear programming (LP) approach via experiments over the widely-used NSFNET and three randomly generated networks. Numerical results demonstrate that LRH outperforms the LP approach in both accuracy and computational time complexity particularly for larger sized networks.

Optical wavelength-division multiplexed (WDM) networks often include optical cross-connects with multigranularity switching capability, such as switching on a single lambda, a waveband, or an entire fiber basis. In addition, it has been shown that routing and wavelength assignment (RWA) in an arbitrary mesh WDM network is an NP-complete problem. In this paper, we propose an efficient approximation approach, called Lagrangean relaxation with heuristics (LRH), aimed to resolve RWA in multigranularity WDM networks particularly with lambda and fiber switches. The task is first formulated as a combinatorial optimization problem in which the bottleneck link utilization is to be minimized. The LRH approach performs constraint relaxation and derives a lower-bound solution index according to a set of Lagrangean multipliers generated through subgradient-based iterations. In parallel, using the generated Lagrangean multipliers, the LRH approach employs a new heuristic algorithm to arrive at a near-optimal upper-bound solution. With lower and upper bounds, we conduct a performance study on LRH with respect to accuracy and convergence speed under different parameter settings. We further draw comparisons between LRH and an existing practical approach via experiments over randomly generated and several well-known large sized networks. Numerical results demonstrate that LRH outperforms the existing approach in both accuracy and computational time complexity, particularly for larger sized networks.

For IP-over-WDM networks, optical coarse packet switching (OCPS) has been proposed to circumvent optical packet switching limitations by using in-band-controlled per-burst switching and advocating traffic control enforcement to achieve high bandwidth utilization and quality-of-service (QoS). In this paper, we first introduce the OCPS paradigm. Significantly, we present a QoS-enhanced traffic control scheme exerted during packet aggregation at ingress nodes, aiming at providing delay and loss class differentiations for OCPS networks. Serving a dual purpose, the scheme is called (/spl psi/,/spl tau/)-Scheduler/Shaper, where /spl psi/ and /spl tau/ are the maximum burst size and burst assembly time, respectively. To provide delay class differentiation, for IP packet flows designated with delay-associated weights, (/spl psi/,/spl tau/)-Scheduler performs packet scheduling and assembly into bursts based on their weights and a virtual window of size /spl psi/. The guaranteed delay bound for each delay class is quantified via the formal specification of a stepwise service curve. To provide loss class differentiation, (/spl psi/,/spl tau/)-Shaper facilitates traffic shaping with larger burst sizes assigned to higher loss priority classes. To examine the shaping effect on loss performance, we analytically derive the departure process of (/spl psi/,/spl tau/)-Shaper. The aggregate packet arrivals are modeled as a two-state Markov modulated Bernoulli process (MMBP) with batch arrivals. Analytical results delineate that (/spl psi/,/spl tau/)-Shaper yields substantial reduction, proportional to the burst size, in the coefficient of variation of the burst interdeparture time. Furthermore, we conduct extensive simulations on a 24-node ARPANET network to draw packet loss comparisons between OCPS and just-enough-time (JET)-based OBS. Simulation results demonstrate that, through burst size adjustment, (/spl psi/,/spl tau/)-Shaper effectively achieves differentiation of loss classes. Essentially, compared to JET-based OBS using out-of-band control and offset-time-based QoS strategy, OCPS is shown to achieve invariably superior packet loss probability for a high-priority class, facilitating better differentiation of loss traffic classes.

For WDM networks with multi-granularity switching, optical tunnel allocation (OTA) deals with the real-time establishment of optical tunnels between optical nodes through various optical multi-granularity switching devices. OTA is in principle a dynamic routing and wavelength assignment (RWA) problem with multi-granularity switching devices taken into account. In this paper, we propose a novel approximation approach, called Lagrangean relaxation with heuristics (LRH), aimed to resolve RWA considering both fiber and lambda switches. Such RWA is first formulated as a combinatorial optimization problem in which the bottleneck link utilization is to be minimized. To tackle the problem, the LRH approach performs constraint relaxation and derives a lower-bound solution index according to a set of Lagrangean multipliers generated through subgradient-based iterations. In parallel, using the generated Lagrangean multipliers, the LRH approach employs a new heuristic algorithm to arrive at a near-optimal upper-bound solution. Through numerical results and comparisons, we delineate that the LRH approach achieves a near-optimal solution, which is profoundly tight to its lower bound, at the expense of low computational time complexity.

Prevailing weight-based scheduling algorithms advocate static or coarse-grained simplification of timestamp computation, resulting in performance bottlenecks and unnecessary downgrades under heavy and normal computation loads, respectively. Our goal is to design a scheduler that enables quantitative trade-off balance between performance and complexity. In this paper, we propose a stepwise QoS scheduler (SQS), enabling packet to burst scheduling under normal to heavy computation loads, based on a new notion of window. Window sizes of unity and greater than one correspond to packet and burst scheduling, respectively. SQS exerts simple FIFO service within the window and guarantees stepwise weight-proportional service at the window boundary. As a result, quantitative trade-off balance can be facilitated via the adjustment of the window size. We further formally specify the SQS service guarantee for a flow by a stepwise service curve, which is a function of W and /spl theta//sub min/-the least-upper-bound delay incurred by the first packet of the busy period of the flow. Finally, we demonstrate through experimental results that, by applying a small window (W/spl les/10), SQS outperforms weighted fair queueing (WFQ) and performs as superior as worst-case fair weighted fair queueing (WF/sup 2/Q) with respect to throughput fairness, mean delay, and worst-case delay fairness.

We propose a new contention access protocol (CAP) augmented with a dynamic bandwidth allocator (DBA) for wireless ATM networks supporting ABR and signaling control (SCR) traffic. CAP incorporates a dynamic tree-splitting collision resolution algorithm parameterized by an optimal splitting depth (SD). DBA performs estimation and on-line prediction of ABR self-similar traffic characteristics. It in turn determines the optimal SD per frame, satisfying ABR throughput and SCR blocking probability requirements while retaining maximal aggregate throughput. Simulation results postulate the optimal SDs under various ABR and SCR traffic conditions. These results are then off-line trained and constructed by a backpropagation neural network (BPNN), which is used on-line for optimal bandwidth allocation.

Two major challenges pertaining to wireless asynchronous transfer mode (ATM) networks are the design of multiple access control (MAC), and dynamic bandwidth allocation. While the former draws more attention, the latter has been considered nontrivial and remains mostly unresolved. We propose a new intelligent multiple access control system (IMACS) which includes a versatile MAC scheme augmented with dynamic bandwidth allocation, for wireless ATM networks. IMACS supports four types of traffic-CBR, VBR, ABR, and signaling control (SCR). It aims to efficiently satisfy their diverse quality-of-service (QoS) requirements while retaining maximal network throughput. IMACS is composed of three components: multiple access controller (MACER), traffic estimator/predictor (TEP), and intelligent bandwidth allocator (IBA). MACER employs a hybrid-mode TDMA scheme, in which its contention access is based on a new dynamic-tree-splitting (DTS) collision resolution algorithm parameterized by an optimal splitting depth (SD). TEP performs periodic estimation and on-line prediction of ABR self-similar traffic characteristics based on wavelet analysis and a neural-fuzzy technique. IBA is responsible for static bandwidth allocation for CBR/VBR traffic following a closed-form formula. In cooperation with TEP, IBA governs dynamic bandwidth allocation for ABR/SCR traffic through determining the optimal SD. The optimal SDs under various traffic conditions are postulated via experimental results, and then off-line constructed using a back propagation neural network (BPNN), being used on-line by IBA. Consequently, with dynamic bandwidth allocation, IMACS offers various QoS guarantees and maximizes network throughput irrelevant to traffic variation.

Quality of service (QOS) for diverse multimedia applications can be satisfied by the provision of synchronization platforms furnishing intramedia/intermedia synchronization and efficient process multitasking at end systems. As a result, the design of viable platforms should confront two concerns: network delay and jitter, and system multitasking. In this paper, we propose an intelligent synchronization platform, called SYNC*, satisfying given QOSs in terms of skew and throughput. SYNC* is composed of three components: intelligent video smoother (IVS), synchronizer, and multi-thread priority manager (MPM). While IVS and synchronizer deal with the network delay/jitter concern, MPM manages the challenge of system multitasking. We demonstrate via prototyping results that SYNC* achieves superior playout with respect to skew and throughput. We also observe that, the absence of either IVS/synchronizer or MPM results in unsatisfactory playout.

When transporting voice data with silence suppression over the Internet, the problem of jitter introduced from the network often renders the speech unintelligible. It is thus indispensable to offer intramedia synchronization to remove jitter while retaining minimal playout delay (PD). We propose a neural network (NN)-based intravoice synchronization mechanism, called the intelligent voice smoother (IVoS). The IVoS is composed of three components: (1) the smoother buffer; (2) the NN traffic predictor; and (3) the constant bit rate (CBR) enforcer. Newly arriving frames, assumed to follow a generic Markov modulated Bernoulli process (MMBP), are queued in the smoother buffer. The NN traffic predictor employs an online-trained back propagation NN (BPNN) to predict three traffic characteristics of every newly encountered talkspurt period. Based on the predicted characteristics, the CBR enforcer derives an adaptive buffering delay (ABD) by means of a near-optimal simple closed-form formula. It then imposes the delay on the playout of the first frame in the talkspurt period. The CBR enforcer in turn regulates CBR-based departures for the remaining frames of the talkspurt, aiming at assuring minimal mean and variance of distortion of talkspurts (DOT) and mean PD. Simulation results reveal that, compared to three other playout approaches, the IVoS achieves superior playout, yielding negligible DOT and PD, irrespective of traffic variation.

For transporting voice data with silence suppression over the Internet, the problem of jitter introduced from the network often renders the speech unintelligible. It is thus indispensable to offer intramedia synchronization to remove jitter while retaining minimal playout delay. We propose a neural-network-based intra-voice synchronization mechanism, called the intelligent voice smoother (IVoS). The IVoS is composed of three components: smoother buffer, neural network (NN) traffic predictor, and constant bit rate (CBR) enforcer. Newly arriving frames, being assumed to follow a generic Markov modulated Bernoulli process (MMBP), are queued in the smoother buffer. The NN traffic predictor employs an on-line-trained backpropagation neural network (BPNN) to predict three traffic characteristics of every newly encountered talkspurt period. Based on the predicted characteristics, the CBR enforcer derives an adaptive buffering delay. It then imposes such delay on the playout of the first frame in the talkspurt period. The CBR enforcer in turn regulates CBR-based departures for the remaining frames of the talkspurt, aiming at assuring minimal mean and variance of distortion of talkspurts (DOT) and mean playout delay (PD). Simulation results reveal that, compared to three other playout approaches, IVoS achieves superior playout yielding negligible DOT and PD irrespective of traffic variation.

For distinctively transporting voice data with silence suppression over asynchronous transfer mode (ATM) networks via the variable bit rate (VBR) service, the problem of jitter introduced from the network often renders the speech unintelligible. It is thus indispensable to offer intramedia synchronization to remove jitter while retaining minimal playout delay. We propose a neural-network-based intra-voice synchronization mechanism, called the intelligent voice smoother (IVoS). The IVoS is composed of three components: smoother buffer, neural network (NN) traffic predictor, and constant bit rate (CBR) enforcer. Newly arriving frames, being assumed to follow a generic Markov modulated Bernoulli process (MMBP), are queued in the smoother buffer. The NN traffic predictor employs an on-line-trained backpropagation neural network (BPNN) to predict three traffic characteristics of every newly encountered talkspurt period. Based on the predicted characteristics, the CBR enforcer derives an adaptive buffering delay by means of a near-optimal, simple, closed-form formula. It then imposes such delay on the playout of the first frame in the talkspurt period. The CBR enforcer in turn regulates CBR-based departures for the remaining frames of the talkspurt, aimed at assuring minimal mean and variance of distortion of talkspurts (DOT) and mean playout delay (PD). Simulation results reveal that, compared to three other playout approaches, the IVoS achieves superior playout yielding negligible DOT and PD irrespective of traffic variation.

Multimedia communications often require intramedia synchronization for video data to prevent potential playout discontinuity resulting from network delay variation (jitter) while still achieving satisfactory playout throughput. In this paper, we propose a neural network (NN) based intravideo synchronization mechanism, called the intelligent video smoother (IVS), operating at the application layer of the receiving end system. The IVS is composed of an NN traffic predictor, an NN window determinator, and a window-based playout smoothing algorithm. The NN traffic predictor employs an on-line-trained back-propagation neural network (BPNN) to periodically predict the characteristics of traffic modeled by a generic interrupted Bernoulli process (IBP) over a future fixed time period. With the predicted traffic characteristics, the NN window determinator determines the corresponding optimal window by means of an off-line-trained BPNN in an effort to achieve a maximum of the playout quality (Q) value. The window-based playout smoothing algorithm then dynamically adopts various playout rates according to the window and the number of packets in the buffer. Finally, we show that via simulation results and live video scenes, compared to two other playout approaches, IVS achieves high-throughput and low-discontinuity playout under a mixture of IBP arrivals.

Call admission control (CAC) has been accepted as a potential solution for supporting diverse, heterogeneous traffic sources demanding different quality of services in asynchronous transfer mode (ATM) networks. Besides, CAC is required to consume a minimum of time and space to make call acceptance decisions. We present an efficient neural-network-based CAC (NNCAC) mechanism for ATM networks with heterogeneous arrivals. All heterogeneous traffic calls are initially categorized into various classes. Based on the number of calls in each class, the NNCAC efficiently and accurately estimates the cell delay and cell loss ratio of each class in real time by means of a pre-trained neural network. According to our decent study which exhibits the superiority of the employment of analysis-based training data over simulation-based data, we particularly construct the training data from heterogeneous-arrival dual-class queueing model M/sup [N1]/+I/sup [N2]//D/1/K, where M and I represent the Bernoulli process and interrupted Bernoulli process, and N/sub 1/ and N/sub 2/ represent the corresponding numbers of calls, respectively. Analytic results of the queueing model are confirmed by simulation results. Finally, we demonstrate the profound agreement of our neural-network-based estimated results with analytic results, justifying the viability of our NNCAC mechanism.

Multimedia communications often require intra-media synchronization for video data to prevent potential playout discontinuity resulting from network delay variation (jitter) while still achieving satisfactory playout throughput. We propose a neural-network-based intra-media synchronization mechanism, called neural network smoother (NNS). The NNS is composed of a neural network (NN) traffic predictor, an NN window determinator, and a window-based playout smoothing algorithm. The NN traffic predictor employs an on-line-trained backpropagation neural network (BPNN) to periodically predict future traffic characteristics. With the predicted traffic characteristics, the NN window determinator determines the corresponding optimal window by means of an off-line-trained BPNN in an effort to achieve a maximum of the playout quality (Q) value. The window-based playout smoothing algorithm then dynamically adopts various playout rates according to the window and the number of packets in the buffer. Compared to two other playout approaches, simulation results show that NNS achieves high-throughput and low-discontinuity playout under a variety of traffic arrivals.