High-capacity Networks Research Articles

4G to 6G: disruptions and drivers for optical access [Invited

The exponential growth of diverse smart mobile applications such as smart monitoring, holograms, and autonomous vehicles are driving mobile technologies toward a more intelligent and software-defined communication system. This growing demand has led to the next giant leap in next-generation wireless communication technology, i.e., sixth-generation (6G) mobile technology, which anticipates providing 1 Tbps data rates and ultralow latency over ubiquitous 3D coverage areas. However, the transport network, which connects hundreds of thousands of cell sites and the network core to enable intelligence, virtualization, and other key features of 6G, has not advanced enough to cater the dense cell deployment expected in future 6G networks. Because many aspects of 6G remain undefined, this provides us with the opportunity to take optical transport networks into design consideration to realize the benefits that 6G has to offer. To this end, in this paper, we present a comprehensive view on the role of optical access networks in supporting fourth-generation (4G), fifth-generation (5G), and beyond wireless technologies. In particular, we discuss design and deployment strategies and their challenges when using optical access as a transport solution during the evolution of wireless access technologies. For this purpose, we first identify how an optical access network is used to support high-capacity transport networks in 4G, followed by the challenges that 5G brought into optical transport networks and its diverse solutions. We conclude the paper by providing insights into how an optical transport network can be designed to support 6G.

Efficient integrated tri-modal coupler for few-mode fibers.

This paper demonstrates a high-efficiency vertical grating coupler for the LP01x, LP11ax, and LP11bx modes of a graded-index few-mode fiber. The coupler is composed of a non-uniform straight bidirectional grating that was inverse-designed to address the desired fiber modes, combined with two mode-selective directional couplers and two tapers. The device was fabricated by e-beam lithography with a minimum feature size of 100 nm and presented coupling efficiencies of -3.0 dB, -3.6 dB, and -3.4 dB for the LP01x, LP11ax, and LP11bx modes, respectively. The high efficiency of the proposed CMOS-compatible coupler demonstrates its potential as a key device for high-capacity networks exploiting space division multiplexing on few-mode fibers.

What we can do to ensuie that the GIGA School concept does not end up being a pipe dream.

One terminal per student and a high-speed, high-capacity communication network were established in 2018 under the GIGA school concept set forth by the Ministry of Education, Culture, Sports, Science and Technology. Moreover, the situation in schools is now required to change at a dizzying pace, as the new coronavirus outbreak has pushed forward the target year from 2023. However, some schools and municipalities are having a tough time making the GIGA school concept a reality. In this issue, we would like to clarify the actual situation, consider the key points for success from the viewpoints of children, teachers, and administrators, and consider measures to promote the spread of the program.

Self-Augmentation Based on Noise-Robust Probabilistic Model for Noisy Labels

Learning deep neural networks from noisy labels is challenging, because high-capacity networks attempt to describe data even with noisy class labels. In this study, we propose a self-augmentation method without additional parameters, which handles noisy labeled data based on small-loss criteria. To this end, we use small-loss samples by introducing a noise-robust probabilistic model based on a Gaussian mixture model (GMM), in which small-loss samples follow class-conditional Gaussian distributions. With this sample augmentation using the GMM-based probabilistic model, we can effectively solve over-parameterization problems induced by label inconsistency in small-loss samples. We further enhance the quality of the small-loss samples using our data-adaptive selection strategy. Consequently, our method prevents networks from over-parameterization and enhances their generalization performance. Experimental results demonstrate that our method outperforms state-of-the-art methods for learning with noisy labels on several benchmark datasets. The proposed method produced a remarkable performance gap of up to 12% compared with the previous state-of-the-art methods on CIFAR dataset.

Network Pruning for Remote Sensing Images Classification Based on Interpretable CNNs

Convolutional neural network (CNN)-based research has been successfully applied in remote sensing image classification due to its powerful feature representation ability. However, these high-capacity networks bring heavy inference costs and are easily overparameterized, especially for the deep CNNs pretrained on natural image datasets. Network pruning is regarded as a prevalent approach for compressing networks, but most existing research ignores model interpretability while formulating pruning criterion. To address these issues, a network pruning method for remote sensing image classification based on interpretable CNNs is proposed. More specifically, an original interpretable CNN with a predefined pruning ratio is trained at first. The filters, namely channels in the high convolutional layer, are able to learn specific semantic meanings in proportion to the predefined pruning ratio. The filters without interpretability are supposed to be removed. As for other convolutional layers, a sensitivity function is designed to assess the risk of pruning channels for each layer, and furthermore, the pruning ratio for each layer is corrected adaptively. The pruning method based on the proposed sensitivity function is effective and requires little computational costs to search abandoned channels without damaging classification performance. To demonstrate the effectiveness, the proposed method is implemented on different scales of modern CNN models, including VGG-VD and AlexNet. The experimental results, obtained on the UC Merced dataset and NWPU-RESISC45 dataset, prove that our method significantly reduces the inference costs and improves the interpretability of networks.

Protection Scheme for a Wavelength-Division-Multiplexed Passive Optical Network Based on Reconfigurable Optical Amplifiers

This paper demonstrates a wavelength-division-multiplexed passive optical network (WDM-PON) scheme based on novel reconfigurable optical amplifiers (ROAs). The measured switching characteristics of the ROA3 constructed with a 2 × 2 crossbar optical switch and a four-port reversible optical circulator (OC) and a conventional EDFA can meet the requirements of most network management and surveillance. The self-made four-port reversible OC’s response time is less than 2 ms, and its insertion losses are about 1 dB or less for all the transmission paths and switching states. An optimal design of ROAs is proposed and evaluated for bidirectional optical amplifier protection, in which ROA3 has an EDF length of 7.5 m long with a 1480 nm pump laser and possesses a backward or forward pumping configuration with the corresponding pump power of 200 mW or 50 mW. We verified the scheme’s feasibility through a simulation of WDM-PON systems with 40 downstream and upstream channels. This scheme enables the intelligent protection switching in practical operation scenarios for high-capacity multi-wavelength networks.

Intelligent Reflecting Surface-Aided Vehicular Networks Toward 6G: Vision, Proposal, and Future Directions

As one of the typical applications in the 6G mobile network, autonomous driving will significantly benefit from the extremely high network throughput and capacity, and an increasing number of computation-aggressive tasks will be generated by vehicles and offloaded to edge and cloud servers. Since many tasks generated by vehicles are delay sensitive, it is of great importance to ensure the transmission rate and stability. However, signals transmitted with high-frequency bands, even the terahertz, in 6G are seriously affected by physical obstacles, while the vehicle mobility can cause the dynamic channel conditions.

Generation of higher-order orbital angular momentum squeezed light

The continuous-variable orbital angular momentum squeezed states is a crucial resource in the implementation of quantum precise measurement, super-resolution quantum imaging, and quantum storage. Currently, only the lower-order orbital angular momentum squeezed states can be generated by an optical parametric oscillator. However, the higher-order orbital angular momentum squeezed states are more useful for reliable quantum information processing. Here, we use a vortex plate to generate a series of higher-order orbital angular momentum squeezed states. The measured 20th-order orbital angular momentum squeezing is − 3.5 dB. The method may find important applications in high-dimensional quantum information processing and high-capacity quantum communication networks.

Mitigation of Distortions in Radio-Over-Fiber Systems Using Machine Learning

Introduction: The ever-growing number of users connected to internet via mobile devices has driven to increase the research in the paradigm of hybrid optical networks called Radio-over-Fiber. These networks take advantages of the bandwidth given by the optical fiber and the mobility given by wireless transmissions, avoiding the bottleneck of optical-to-electrical conversion interfaces. However, the chromatic dispersion of the optical fiber generates distortions in the radiofrequency signals optically modulated, limiting the reach of transmission. Objective: To improve the performance of a Radio-over-Fiber system in terms of bit-error-rate, using nonsymmetrical demodulation by means of the machine learning algorithm Support Vector Machine. Methodology: A Radio-over-Fiber System is simulated in the specialized software VPIDesignSuite. The radiofrequency signals are modulated at 16 and 64-QAM formats with different laser linewidths and transmitted over optical fiber. The Support Vector Machine algorithm is applied to carry out nonsymmetrical demodulation. Results: The implementation of the machine learning algorithm for signal demodulation significantly improves the network performance, reaching transmissions up to 30 km. It implies a reduction of the bit-error-rate up to two Introduction: The ever-growing number of users connected to internet via mobile devices has driven to increase the research in the paradigm of hybrid optical networks called Radio-over-Fiber. These networks take advantages of the bandwidth given by the optical fiber and the mobility given by wireless transmissions, avoiding the bottleneck of optical-to-electrical conversion interfaces. However, the chromatic dispersion of the optical fiber generates distortions in the radiofrequency signals optically modulated, limiting the reach of transmission. Objective: To improve the performance of a Radio-over-Fiber system in terms of bit-error-rate, using nonsymmetrical demodulation by means of the machine learning algorithm Support Vector Machine. Methodology: A Radio-over-Fiber System is simulated in the specialized software VPIDesignSuite. The radiofrequency signals are modulated at 16 and 64-QAM formats with different laser linewidths and transmitted over optical fiber. The Support Vector Machine algorithm is applied to carry out nonsymmetrical demodulation. Results: The implementation of the machine learning algorithm for signal demodulation significantly improves the network performance, reaching transmissions up to 30 km. It implies a reduction of the bit-error-rate up to two orders of magnitude in comparison with conventional demodulation. Conclusions: Mitigation of distortions in terms of bit-error-rate is demonstrated in a Radio-over-Fiber system using nonsymmetrical demodulation by using the Support Vector Machine algorithm. Thus, the proposed technique can be suitable for future high-capacity access networks.

Chaotic local search-based levy flight distribution algorithm for optimizing ONU placement in fiber-wireless access network

Fiber-wireless (FiWi) access networks smartly integrate the mobility, robustness, and ubiquity of wireless networks and the high capacity of fiber networks. A demanding task in FiWi networks is to find the optimal placement of optical network units (ONUs), which leads to resource, throughput, and cost optimization. Several studies have been carried out to solve the ONU placement problem. Most approaches achieve encouraging results; however, they suffer from trapping in local optima, and their performance is far from the ideal state. This paper introduces a chaotic local search-based levy flight distribution (CLS-LFD) algorithm to optimize the position of ONUs in the FiWi network. The proposed algorithm models the ONU placement as a constraint multi-modal optimization problem. The objective is to minimize the overall average distance between ONUs and users. With a test on four ONU placement benchmarks, the proposed CLS-LFD algorithm has shown significant improvement over the canonical LFD algorithm and several counterpart algorithms in terms of convergence performance and cost value.

Training strategies to minimize interchannel interference effects using supervised learning in gridless Nyquist-WDM systems.

The dynamism envisioned in future high-capacity gridless optical networks requires facing several challenges in distortion mitigation, such as the mitigation of interchannel interference (ICI) effects in any optical channel without information of their adjacent channels. Machine learning (ML)-based techniques have been proposed in recent works to estimate and mitigate different optical impairments with promising results. We propose and evaluate two training strategies for supervised learning algorithms with the aim to minimize ICI effects in a gridless 3×16-Gbaud 16-quadrature amplitude modulation (QAM) Nyquist-wavelength-division multiplexing (WDM) system. One strategy, called updating strategy, is based on symbol training sequence, and the other one, called characterization strategy, is based on an offline training using a previous system characterization. Artificial neural networks (ANN), support vector machine (SVM), K-nearest neighbors (KNN), and extreme learning machine (ELM) algorithms are explored for both training strategies. Experimental results showed a bit error rate (BER) improvement at low training lengths for both training strategies, for instance, gains up to ∼4dB in terms of optical signal-to-noise ratio were achieved in a back-to-back scenario. Besides, the KNN and ELM algorithms showed significant BER reduction in transmission over 250 km optical fiber. Additionally, we carried out a brief computational complexity analysis where ELM presented only 1.9% of ANN processing time. Hence, the use of ML-based techniques could enhance the optical gridless networks performance and consequently fulfill future traffic demands.

User association for load balancing in coordinated multipoint green HetNets: A Quasi-Newton-based approach

The demand for high capacity network services with stringent quality of service requirements is at a rapidly accelerating rate due to the exponential rise in the numbers of mobile-connected devices. This demand has motivated the use of the heterogeneous network (HetNet) architectures. However, even though small-cell base-stations have relatively low power consumption, the overall aggregate power consumption of a dense HetNet is significant. Due to high inter-cell-interference and imbalanced loads in dense HetNets with conventional user association techniques, cell-edge users perceive dramatically less quality of service than their cell-center counterparts. The use of a Coordinated Multipoint (CoMP) association can augment the service perceived by cell-edge users by allowing a single user to be jointly served by two base-stations. In this work, we propose a load balancing scheme for CoMP-enabled HetNets with hybrid energy supplies that jointly optimizes user latency and green energy utilization. The proposed scheme employs a fractional solution to the user association problem to decide CoMP transmission for cell-edge users, ultimately improving their data rates. Performance evaluations of the proposed scheme show a reduction in latency of 79% and on-grid power consumption by 99% compared to conventional user association schemes that associate users based on the maximum received signal strength. Furthermore, an improvement in the network sum-rate for cell-edge users by 24% has been achieved compared to the traditional association scheme and as much as 40% over other existing schemes.

Design and analysis of 5G network architecture with orthogonal frequency division multiple access based passive optical network

The demand for high-speed internet with a high-capacity network is increasing at an exponential rate. It is quite challenging to accommodate a large number of users with multiple communication devices without degrading the quality of service at an acceptable cost. Therefore, strong communication technology is required that can provide high data rates along with the provision of multiple connecting devices for different users. The 5G network can adjust large users with multiple device connectivity by providing high bandwidth with a minimum delay which depends on the underlying network architecture for connecting different components of the systems. Optical Fiber is used as a transmission media to connect the densely deployed Remote Radio Units (RRUs) with the Baseband Units (BBUs) at the Centralized Radio Access Network (C-RAN) that makes the architecture quite complex and expensive in terms of deployment and maintenance cost. Here in this paper, we propose an architecture by replacing an optical fiber with Free Space Optics (FSO) at the front-end of the 5G network. In addition, an Orthogonal Frequency Division Multiple Access (OFDMA) is implemented to provide high spectrum efficiency and high data rate in an adverse channel environment. The performance analysis of the proposed system is done in Optisystem by using the constellation points which shows that the proposed architecture can support data rate up to 40 Gbps using 512 subcarriers and 10 Gbps per BBU with QPSK over 4 km of FSO medium. In the end, a comparative cost analysis of the proposed architecture with the conventional architecture is done, which shows a cost reduction of 21 % by the proposed system.

Orbital Angular Momentum Multiplexed Quantum Dense Coding.

To beat the channel capacity limit of conventional quantum dense coding (QDC) with fixed quantum resources, we experimentally implement the orbital angular momentum (OAM) multiplexed QDC (MQDC) in a continuous variable system based on a four-wave mixing process. First, we experimentally demonstrate that the Einstein-Podolsky-Rosen entanglement source coded on OAM modes can be used in a single channel to realize the QDC scheme. Then, we implement the OAM MQDC scheme by using the Einstein-Podolsky-Rosen entanglement source coded on OAM superposition modes. In the end, we make an explicit comparison of channel capacities for four different schemes and find that the channel capacity of the OAM MQDC scheme is substantially enhanced compared to the conventional QDC scheme without multiplexing. The channel capacity of our OAM MQDC scheme can be further improved by increasing the squeezing parameter and the number of multiplexed OAM modes in the channel. Our results open an avenue to construct high-capacity quantum communication networks.

Experimental Single-Copy Entanglement Distillation.

The phenomenon of entanglement marks one of the furthest departures from classical physics and is indispensable for quantum information processing. Despite its fundamental importance, the distribution of entanglement over long distances through photons is unfortunately hindered by unavoidable decoherence effects. Entanglement distillation is a means of restoring the quality of such diluted entanglement by concentrating it into a pair of qubits. Conventionally, this would be done by distributing multiple photon pairs and distilling the entanglement into a single pair. Here, we turn around this paradigm by utilizing pairs of single photons entangled in multiple degrees of freedom. Specifically, we make use of the polarization and the energy-time domain of photons, both of which are extensively field tested. We experimentally chart the domain of distillable states and achieve relative fidelity gains up to 13.8%. Compared to the two-copy scheme, the distillation rate of our single-copy scheme is several orders of magnitude higher, paving the way towards high-capacity and noise-resilient quantum networks.

A study and Description of the Femtocells Wireless Communication Networks

In this paper the femtocell base station called as Femtocell Access Point (FAP) was descriped and studied. The femtocell (FAP) is fully user deployed and hence reduces the infrastructure, maintenance and operational cost of the operator while at the same time providing good Quality of Service (QoS) to the end user and high network capacity gains. The surest way to increase the system capacity and data rates can be achieved efficiently in a wireless system by getting the transmitter and receiver closer to each other. In a network with nomadic users, this inevitably involves deploying more infrastructure, typically in the form of microcells, hotspots, distributed antennas, or relays. Femtocells deployed in the macrocell significantly improve the indoor coverage and provide better user experience.
 In this paper the four indoor propagation path loss models with simulated data for different scenarios at 2 GHz are presented In additions the Simulation were carried out in two different scenarios LOS and NLOS for max 50m femtocell and max 100m macrocell network. The simulation results was implemented using Matlab code software.

Semi-supervised training of deep convolutional neural networks with heterogeneous data and few local annotations: An experiment on prostate histopathology image classification

Convolutional neural networks (CNNs) are state-of-the-art computer vision techniques for various tasks, particularly for image classification. However, there are domains where the training of classification models that generalize on several datasets is still an open challenge because of the highly heterogeneous data and the lack of large datasets with local annotations of the regions of interest, such as histopathology image analysis. Histopathology concerns the microscopic analysis of tissue specimens processed in glass slides to identify diseases such as cancer. Digital pathology concerns the acquisition, management and automatic analysis of digitized histopathology images that are large, having in the order of 100′0002 pixels per image. Digital histopathology images are highly heterogeneous due to the variability of the image acquisition procedures. Creating locally labeled regions (required for the training) is time-consuming and often expensive in the medical field, as physicians usually have to annotate the data. Despite the advances in deep learning, leveraging strongly and weakly annotated datasets to train classification models is still an unsolved problem, mainly when data are very heterogeneous. Large amounts of data are needed to create models that generalize well. This paper presents a novel approach to train CNNs that generalize to heterogeneous datasets originating from various sources and without local annotations. The data analysis pipeline targets Gleason grading on prostate images and includes two models in sequence, following a teacher/student training paradigm. The teacher model (a high-capacity neural network) automatically annotates a set of pseudo-labeled patches used to train the student model (a smaller network). The two models are trained with two different teacher/student approaches: semi-supervised learning and semi-weekly supervised learning. For each of the two approaches, three student training variants are presented. The baseline is provided by training the student model only with the strongly annotated data. Classification performance is evaluated on the student model at the patch level (using the local annotations of the Tissue Micro-Arrays Zurich dataset) and at the global level (using the TCGA-PRAD, The Cancer Genome Atlas-PRostate ADenocarcinoma, whole slide image Gleason score). The teacher/student paradigm allows the models to better generalize on both datasets, despite the inter-dataset heterogeneity and the small number of local annotations used. The classification performance is improved both at the patch-level (up to κ=0.6127±0.0133 from κ=0.5667±0.0285), at the TMA core-level (Gleason score) (up to κ=0.7645±0.0231 from κ=0.7186±0.0306) and at the WSI-level (Gleason score) (up to κ=0.4529±0.0512 from κ=0.2293±0.1350). The results show that with the teacher/student paradigm, it is possible to train models that generalize on datasets from entirely different sources, despite the inter-dataset heterogeneity and the lack of large datasets with local annotations.

Universal optical network architecture for future wireless LANs [Invited

Ultra-high-capacity indoor wireless connectivity is considered a key technology enabler to support end-to-end delivery of new services in the beyond-5G and 6G era. With physical layer data rates well beyond 10 Gbps, optical wireless communication (OWC) technology is positioning itself as a prospective candidate to support these services. While OWC is capable of handling multi-gigabit data rates in the wireless domain, the underlying distribution network architecture is not designed to transport the projected high volumes of data as well as the expected added complexity of managing and coordinating multiple narrow beams, beamsteering, and femto cells. To future proof wireless local area networks (LANs) including those using the radio spectrum, we propose a universal network architecture that features centralized software/hardware resource pools as well as control and data plane separation in conjunction with virtualization support and a high-capacity fiber distribution network. Such an architecture has the potential to realize interoperability between existing and future high-capacity wireless LAN technologies such as Wi-Fi 6 and Wi-Fi in the 60 GHz range, as well as dynamic and flexible allocation of resources in response to changing user bandwidth demand. Our results demonstrate the benefits of the universal network architecture in supporting high throughput performance, low delay, and increased resiliency.

Fractional Matching Preclusion for Data Center Networks

An edge subset [Formula: see text] of [Formula: see text] is a fractional matching preclusion set (FMP set for short) if [Formula: see text] has no fractional perfect matchings. The fractional matching preclusion number (FMP number for short) of [Formula: see text], denoted by [Formula: see text], is the minimum size of FMP sets of [Formula: see text]. A set [Formula: see text] of edges and vertices of [Formula: see text] is a fractional strong matching preclusion set (FSMP set for short) if [Formula: see text] has no fractional perfect matchings. The fractional strong matching preclusion number (FSMP number for short) of [Formula: see text], denoted by [Formula: see text], is the minimum size of FSMP sets of [Formula: see text]. Data center networks have been proposed for data centers as a server-centric interconnection network structure, which can support millions of servers with high network capacity by only using commodity switches. In this paper, we obtain the FMP number and the FSMP number for data center networks [Formula: see text], and show that [Formula: see text] for [Formula: see text], [Formula: see text] and [Formula: see text] for [Formula: see text], [Formula: see text]. In addition, all the optimal fractional strong matching preclusion sets of these graphs are categorized.

Missing value imputation on multidimensional time series

We present DeepMVI, a deep learning method for missing value imputation in multidimensional time-series datasets. Missing values are commonplace in decision support platforms that aggregate data over long time stretches from disparate sources, whereas reliable data analytics calls for careful handling of missing data. One strategy is imputing the missing values, and a wide variety of algorithms exist spanning simple interpolation, matrix factorization methods like SVD, statistical models like Kalman filters, and recent deep learning methods. We show that often these provide worse results on aggregate analytics compared to just excluding the missing data. DeepMVI expresses the distribution of each missing value conditioned on coarse and fine-grained signals along a time series, and signals from correlated series at the same time. Instead of resorting to linearity assumptions of conventional matrix factorization methods, DeepMVI harnesses a flexible deep network to extract and combine these signals in an end-to-end manner. To prevent over-fitting with high-capacity neural networks, we design a robust parameter training with labeled data created using synthetic missing blocks around available indices. Our neural network uses a modular design with a novel temporal transformer with convolutional features, and kernel regression with learned embeddings. Experiments across ten real datasets, five different missing scenarios, comparing seven conventional and three deep learning methods show that DeepMVI is significantly more accurate, reducing error by more than 50% in more than half the cases, compared to the best existing method. Although slower than simpler matrix factorization methods, we justify the increased time overheads by showing that DeepMVI provides significantly more accurate imputation that finally impacts quality of downstream analytics.