arXiv subjects

Sheng Chen

Publications and source records attributed to Sheng Chen.

At least 163 records · Page 9Linked to original sources

arXiv2019

Attention-based Walking Gait and Direction Recognition in Wi-Fi Networks

The study of human gait recognition has been becoming an active research field. In this paper, we propose to adopt the attention-based Recurrent Neural Network (RNN) encoder-decoder framework to implement a cycle-independent human gait and walking direction recognition system in Wi-Fi networks. For capturing more human walking dynamics, two receivers together with one transmitter are deployed in different spatial layouts. In the proposed system, the Channel State Information (CSI) measurements from different receivers are first gathered together and refined to form an integrated walking profile. Then, the RNN encoder reads and encodes the walking profile into primary feature vectors. Given a specific recognition task, the decoder computes a corresponding attention vector which is a weighted sum of the primary features assigned with different attentions, and is finally used to predict the target. The attention scheme motivates our system to learn to adaptively align with different critical clips of CSI data sequence for human walking gait and direction recognitions. We implement our system on commodity Wi-Fi devices in indoor environment, and the experimental results demonstrate that our system can achieve average F1 scores of 89.69% for gait recognition from a group of 8 subjects and 95.06% for direction recognition from 8 walking directions, in addition, the average accuracies of these two recognition tasks both exceed 97%.

cs.HC↗

arXiv2018

Boosting Fronthaul Capacity: Global Optimization of Power Sharing for Centralized Radio Access Network

The limited fronthaul capacity imposes a challenge on the uplink of centralized radio access network (C-RAN). We propose to boost the fronthaul capacity of massive multiple-input multiple-output (MIMO) aided C-RAN by globally optimizing the power sharing between channel estimation and data transmission both for the user devices (UDs) and the remote radio units (RRUs). Intuitively, allocating more power to the channel estimation will result in more accurate channel estimates, which increases the achievable throughput. However, increasing the power allocated to the pilot training will reduce the power assigned to data transmission, which reduces the achievable throughput. In order to optimize the powers allocated to the pilot training and to the data transmission of both the UDs and the RRUs, we assign an individual power sharing factor to each of them and derive an asymptotic closed-form expression of the signal-to-interference-plus-noise for the massive MIMO aided C-RAN consisting of both the UD-to-RRU links and the RRU-to-baseband unit (BBU) links. We then exploit the C-RAN architecture's central computing and control capability for jointly optimizing the UDs' power sharing factors and the RRUs' power sharing factors aiming for maximizing the fronthaul capacity. Our simulation results show that the fronthaul capacity is significantly boosted by the proposed global optimization of the power allocation between channel estimation and data transmission both for the UDs and for their host RRUs. As a specific example of 32 receive antennas (RAs) deployed by RRU and 128 RAs deployed by BBU, the sum-rate of 10 UDs achieved with the optimal power sharing factors improves 33\% compared with the one attained without optimizing power sharing factors.

cs.IT↗

arXiv2018

Exponential Scaling in Early-stage Agglomeration of Adhesive Particles in Turbulence

We carry out direct numerical simulation together with an adhesive discrete element method calculation (DNS-DEM) to investigate agglomeration of particles in homogeneous isotropic turbulence (HIT). We report an exponential-form scaling for the size distribution of early-stage agglomerates, which is valid across a wide range of particle inertia and inter-particle adhesion values. Such scaling allows one to quantify the state of agglomeration using a single scale parameter. An agglomeration kernel is then constructed containing the information of agglomerate structures and the sticking probability. An explicit relationship between the sticking probability and microscale particle properties is also proposed based on the scaling analysis of the equation for head-on collisions. Our results extend Smoluchowski's theory to the condition of non-coalescing solid adhesive particles and can reproduce DNS-DEM results with a simple one-dimensional simulation.

physics.flu-dyn↗

arXiv2018

On non-feasible edge sets in matching-covered graphs

Let $G=(V,E)$ be a matching-covered graph and $X$ be an edge set of $G$. $X$ is said to be feasible if there exist two perfect matchings $M_1$ and $M_2$ in $G$ such that $|M_1\cap X|\not \equiv|M_2\cap X|\ (\mbox{mod } 2)$. For any $V_0\subseteq V$, $X$ is said to be switching-equivalent to $X\oplus \nabla_G(V_0)$, where $\nabla_G(V_0)$ is the set of edges in $G$ each of which has exactly one end in $V_0$ and $A \oplus B$ is the symmetric difference of two sets $A$ and $B$. Lukot'ka and Rollová showed that when $G$ is regular and bipartite, $X$ is non-feasible if and only if $X$ is switching-equivalent to $\emptyset$. This article extends Lukot'ka and Rollová's result by showing that this conclusion holds as long as $G$ is matching-covered and bipartite. This article also studies matching-covered graphs $G$ whose non-feasible edge sets are switching-equivalent to $\emptyset$ or $E$ and partially characterizes these matching-covered graphs in terms of their ear decompositions. Another aim of this article is to construct infinite many $r$-connected and $r$-regular graphs of class 1 containing non-feasible edge sets not switching-equivalent to either $\emptyset$ or $E$ for an arbitrary integer $r$ with $r\ge 3$, which provides negative answers to problems asked by Lukot'ka and Rollová and He, et al respectively.

math.CO↗

arXiv2018

Time-Sequence Channel Inference for Beam Alignment in Vehicular Networks

In this paper, we propose a learning-based low-overhead beam alignment method for vehicle-to-infrastructure communication in vehicular networks. The main idea is to remotely infer the optimal beam directions at a target base station in future time slots, based on the CSI of a source base station in previous time slots. The proposed scheme can reduce channel acquisition and beam training overhead by replacing pilot-aided beam training with online inference from a sequence-to-sequence neural network. Simulation results based on ray-tracing channel data show that our proposed scheme achieves a $8.86\%$ improvement over location-based beamforming schemes with a positioning error of $1$m, and is within a $4.93\%$ performance loss compared with the genie-aided optimal beamformer.

cs.IT↗

arXiv2018

Inferring Remote Channel State Information: Cramér-Rao Lower Bound and Deep Learning Implementation

Channel state information (CSI) is of vital importance in wireless communication systems. Existing CSI acquisition methods usually rely on pilot transmissions, and geographically separated base stations (BSs) with non-correlated CSI need to be assigned with orthogonal pilots which occupy excessive system resources. Our previous work adopts a data-driven deep learning based approach which leverages the CSI at a local BS to infer the CSI remotely, however the relevance of CSI between separated BSs is not specified explicitly. In this paper, we exploit a model-based methodology to derive the Cramér-Rao lower bound (CRLB) of remote CSI inference given the local CSI. Although the model is simplified, the derived CRLB explicitly illustrates the relationship between the inference performance and several key system parameters, e.g., terminal distance and antenna array size. In particular, it shows that by leveraging multiple local BSs, the inference error exhibits a larger power-law decay rate (w.r.t. number of antennas), compared with a single local BS; this explains and validates our findings in evaluating the deep-neural-network-based (DNN-based) CSI inference. We further improve on the DNN-based method by employing dropout and deeper networks, and show an inference performance of approximately $90\%$ accuracy in a realistic scenario with CSI generated by a ray-tracing simulator.

cs.IT↗

arXiv2018

Unified Framework of KKT Conditions Based Matrix Optimizations for MIMO Communications

For multi-input multi-output (MIMO) communication systems, many transceiver design problems involve the optimization of the covariance matrices of the transmitted signals. The derivation of the optimal solutions based on Karush-Kuhn-Tucker (KKT) conditions is a most popular method, and many results have been reported for different scenarios of MIMO systems. In this overview paper, we propose a unified framework in formulating the KKT conditions for general MIMO systems. Based on this framework, the optimal water-filling structures of the transmission covariance matrices are derived rigorously, which are applicable to a wide range of MIMO systems. Our results show that for seemingly different MIMO systems with various power constraints and objective functions, the derivations and water-filling structures for the optimal covariance matrix solutions are fundamentally the same. Thus, our unified framework and solution reveal the underlying relationships among the different water-filling structures of the covariance matrices. Furthermore, our results provide new solutions to the covariance matrix optimization of many complicated MIMO systems with multiple users and imperfect channel state information which were unknown before.

cs.IT↗

arXiv2018

Exploiting Wireless Channel State Information Structures Beyond Linear Correlations: A Deep Learning Approach

Knowledge of information about the propagation channel in which a wireless system operates enables better, more efficient approaches for signal transmissions. Therefore, channel state information (CSI) plays a pivotal role in the system performance. The importance of CSI is in fact growing in the upcoming 5G and beyond systems, e.g., for the implementation of massive multiple-input multiple-output (MIMO). However, the acquisition of timely and accurate CSI has long been considered as a major issue, and becomes increasingly challenging due to the need for obtaining CSI of many antenna elements in massive MIMO systems. To cope with this challenge, existing works mainly focus on exploiting linear structures of CSI, such as CSI correlations in the spatial domain, to achieve dimensionality reduction. In this article, we first systematically review the state-of-the-art on CSI structure exploitation; then extend to seek for deeper structures that enable remote CSI inference wherein a data-driven deep neural network (DNN) approach is necessary due to model inadequacy. We develop specific DNN designs suitable for CSI data. Case studies are provided to demonstrate great potential in this direction for future performance enhancement.

cs.IT↗

arXiv2018

Device-to-Device Communications Enabled Multicast Scheduling with the Multi-Level Codebook in mmWave Small Cells

With the exponential growth of mobile data, there are increasing interests to deploy small cells in millimeter wave (mmWave) bands to underlay the conventional homogeneous macrocell network as well as in exploiting device-to-device (D2D) communications to improve the efficiency of the multicast service that supports content-based mobile applications. To compensate for high propagation loss in the mmWave band, high-gain directional antennas have to be employed, while it is critical to optimize multicast service in order to improve the network performance. In this paper, an efficient multicast scheduling scheme is proposed for small cells in the mmWave band, called MD2D, where both D2D communications in close proximity and multi-level antenna codebook are utilized. Specifically, a user partition and multicast path planning algorithm is proposed to partition the users in the multicast group into subsets and to determine the transmission node for each subset, so as to achieve optimal utilization of D2D communications and multi-level antenna codebook. Then a multicast scheduling algorithm schedules the transmission for each subset. Furthermore, in order to optimize the network performance, the optimal choice of user partition thresholds is analyzed. Performance evaluation demonstrate that the MD2D achieves the best performance, in terms of network throughput and energy efficiency, compared with other existing schemes. MD2D improves the network throughput compared with the second-best scheme by about 27\%.

cs.NI↗

arXiv2018

PP-MCSA: Privacy Preserving Multi-Channel Double Spectrum Auction

Auction is widely regarded as an effective way in dynamic spectrum redistribution. Recently, considerable research efforts have been devoted to designing privacy-preserving spectrum auctions in a variety of auction settings. However, none of existing work has addressed the privacy issue in the most generic scenario, double spectrum auctions where each seller sells multiple channels and each buyer buys multiple channels. To fill this gap, in this paper we propose PP-MCSA, a Privacy Preserving mechanism for Multi-Channel double Spectrum Auctions. Technically, by leveraging garbled circuits, we manage to protect the privacy of both sellers' requests and buyers' bids in multi-channel double spectrum auctions. As far as we know, PP-MCSA is the first privacy-preserving solution for multi-channel double spectrum auctions. We further theoretically demonstrate the privacy guarantee of PP-MCSA, and extensively evaluate its performance via experiments. Experimental results show that PP-MCSA incurs only moderate communication and computation overhead.

cs.CR↗

arXiv2018

Unifying size-topology relations in random packings of dry adhesive polydisperse spheres

We study the size-topology relations in random packings of dry adhesive polydisperse microspheres with Gaussian and lognormal size distributions through a geometric tessellation. We find that the dependence of the neighbour number on the centric particle size is always quasilinear, independent of the size distribution, the size span or interparticle adhesion. The average local packing fraction as a function of normalized particle size for different size variances is well regressed on the same profile, which grows to larger values as the relative strength of adhesion decreases. As for the local coordination number-particle size profiles, they converge onto a single curve for all the adhesive particles, but will gradually transfer to another branch for non-adhesive particles. Such adhesion induced size-topology relations are interpreted theoretically by a modified geometrical "granocentric" model, where the model parameters are dependent on a dimensionless adhesion number. Our findings, together with the modified theory, provide a more unified perspective on the substantial geometry of amorphous polydisperse systems, especially those with fairly loose structures.

cond-mat.soft↗

arXiv2018

A fast adhesive discrete element method for random packings of fine particles

Introducing a reduced particle stiffness in discrete element method (DEM) allows for bigger time steps and therefore fewer total iterations in a simulation. Although this approach works well for dry non-adhesive particles, it has been shown that for fine particles with adhesion, system behaviors are drastically sensitive to the particle stiffness. Besides, a simple and applicable principle to set the parameters in adhesive DEM is also lacking. To solve these two problems, we first propose a fast DEM based on scaling laws to reduce particle Young's modulus, surface energy and to modify rolling and sliding resistances simultaneously in the framework of Johnson-Kendall-Roberts (JKR)-based contact theory. A novel inversion method is then presented to help users to quickly determine the damping coefficient, particle stiffness and surface energy to reproduce a prescribed experimental result. After validating this inversion method, we apply the fast adhesive DEM to packing problems of microparticles. Measures of packing fraction, averaged coordination number and distributions of local packing fraction and contact number of each particle are in good agreement with results simulated using original value of particle properties. The new method should be helpful to accelerate DEM simulations for systems associated with aggregates or agglomerates.

cond-mat.soft↗

arXiv2018

Global Norm-Aware Pooling for Pose-Robust Face Recognition at Low False Positive Rate

In this paper, we propose a novel Global Norm-Aware Pooling (GNAP) block, which reweights local features in a convolutional neural network (CNN) adaptively according to their L2 norms and outputs a global feature vector with a global average pooling layer. Our GNAP block is designed to give dynamic weights to local features in different spatial positions without losing spatial symmetry. We use a GNAP block in a face feature embedding CNN to produce discriminative face feature vectors for pose-robust face recognition. The GNAP block is of very cheap computational cost, but it is very powerful for frontal-profile face recognition. Under the CFP frontal-profile protocol, the GNAP block can not only reduce EER dramatically but also boost TPR@FPR=0.1% (TPR i.e. True Positive Rate, FPR i.e. False Positive Rate) substantially. Our experiments show that the GNAP block greatly promotes pose-robust face recognition over the base model especially at low false positive rate.

cs.CV↗

arXiv2018

Regularized Zero-Forcing Precoding Aided Adaptive Coding and Modulation for Large-Scale Antenna Array Based Air-to-Air Communications

We propose a regularized zero-forcing transmit precoding (RZF-TPC) aided and distance-based adaptive coding and modulation (ACM) scheme to support aeronautical communication applications, by exploiting the high spectral efficiency of large-scale antenna arrays and link adaption. Our RZF-TPC aided and distance-based ACM scheme switches its mode according to the distance between the communicating aircraft. We derive the closed-form asymptotic signal-to-interference-plus-noise ratio (SINR) expression of the RZF-TPC for the aeronautical channel, which is Rician, relying on a non-centered channel matrix that is dominated by the deterministic line-of-sight component. The effects of both realistic channel estimation errors and of the co-channel interference are considered in the derivation of this approximate closed-form SINR formula. Furthermore, we derive the analytical expression of the optimal regularization parameter that minimizes the mean square detection error. The achievable throughput expression based on our asymptotic approximate SINR formula is then utilized as the design metric for the proposed RZF-TPC aided and distance-based ACM scheme. Monte-Carlo simulation results are presented for validating our theoretical analysis as well as for investigating the impact of the key system parameters. The simulation results closely match the theoretical results. In the specific example that two communicating aircraft fly at a typical cruising speed of 920 km/h, heading in opposite direction over the distance up to 740 km taking a period of about 24 minutes, the RZF-TPC aided and distance-based ACM is capable of transmitting a total of 77 Gigabyte of data with the aid of 64 transmit antennas and 4 receive antennas, which is significantly higher than that of our previous eigen-beamforming transmit precoding aided and distance-based ACM benchmark.

cs.IT↗

arXiv2018

MobileFaceNets: Efficient CNNs for Accurate Real-Time Face Verification on Mobile Devices

We present a class of extremely efficient CNN models, MobileFaceNets, which use less than 1 million parameters and are specifically tailored for high-accuracy real-time face verification on mobile and embedded devices. We first make a simple analysis on the weakness of common mobile networks for face verification. The weakness has been well overcome by our specifically designed MobileFaceNets. Under the same experimental conditions, our MobileFaceNets achieve significantly superior accuracy as well as more than 2 times actual speedup over MobileNetV2. After trained by ArcFace loss on the refined MS-Celeb-1M, our single MobileFaceNet of 4.0MB size achieves 99.55% accuracy on LFW and 92.59% TAR@FAR1e-6 on MegaFace, which is even comparable to state-of-the-art big CNN models of hundreds MB size. The fastest one of MobileFaceNets has an actual inference time of 18 milliseconds on a mobile phone. For face verification, MobileFaceNets achieve significantly improved efficiency over previous state-of-the-art mobile CNNs.

cs.CV↗

arXiv2018

Differential Evolution Algorithm Aided Turbo Channel Estimation and Multi-User Detection for G.Fast Systems in the Presence of FEXT

The ever-increasing demand for broadband Internet access has motivated the further development of the digital subscriber line to the G.fast standard in order to expand its operational band from 106 MHz to 212 MHz. Conventional far-end crosstalk (FEXT) based cancellers falter in the upstream transmission of this emerging G.fast system. In this paper, we propose a novel differential evolution algorithm (DEA) aided turbo channel estimation (CE) and multi-user detection (MUD) scheme for the G.fast upstream including the frequency band up to 212 MHz, which is capable of approaching the optimal Cramer-Rao lower bound of the channel estimate, whilst approaching the optimal maximum likelihood (ML) MUD's performance associated with perfect channel state information, and yet only imposing about 5% of its computational complexity. Explicitly, the turbo concept is exploited by iteratively exchanging information between the continuous value-based DEA assisted channel estimator and the discrete value-based DEA MUD. Our extensive simulations show that 18 dB normalized mean square error gain is attained by the channel estimator and 10 dB signal-to-noise ratio gain can be achieved by the MUD upon exploiting this iteration gain. We also quantify the influence of the CE error, of the copper length and of the impulse noise. Our study demonstrates that the proposed DEA aided turbo CE and MUD scheme is capable of offering near-capacity performance at an affordable complexity for the emerging G.fast systems.

cs.IT↗

arXiv2018

Stress anisotropy in shear-jammed packings of frictionless disks

We perform computational studies of repulsive, frictionless disks to investigate the development of stress anisotropy in mechanically stable (MS) packings. We focus on two protocols for generating MS packings: 1) isotropic compression and 2) applied simple or pure shear strain $γ$ at fixed packing fraction $ϕ$. MS packings of frictionless disks occur as geometric families (i.e. parabolic segments with positive curvature) in the $ϕ$-$γ$ plane. MS packings from protocol 1 populate parabolic segments with both signs of the slope, $dϕ/dγ>0$ and $dϕ/dγ<0$. In contrast, MS packings from protocol 2 populate segments with $dϕ/dγ<0$ only. For both simple and pure shear, we derive a relationship between the stress anisotropy and dilatancy $dϕ/dγ$ obeyed by MS packings along geometrical families. We show that for MS packings prepared using isotropic compression, the stress anisotropy distribution is Gaussian centered at zero with a standard deviation that decreases with increasing system size. For shear jammed MS packings, the stress anisotropy distribution is a convolution of Weibull distributions that depend on strain, which has a nonzero average and standard deviation in the large-system limit. We also develop a framework to calculate the stress anisotropy distribution for packings generated via protocol 2 in terms of the stress anisotropy distribution for packings generated via protocol 1. These results emphasize that for repulsive frictionless disks, different packing-generation protocols give rise to different MS packing probabilities, which lead to differences in macroscopic properties of MS packings.

cond-mat.soft↗

arXiv2017

Adaptive Coding and Modulation for Large-Scale Antenna Array Based Aeronautical Communications in the Presence of Co-channel Interference

In order to meet the demands of `Internet above the clouds', we propose a multiple-antenna aided adaptive coding and modulation (ACM) for aeronautical communications. The proposed ACM scheme switches its coding and modulation mode according to the distance between the communicating aircraft, which is readily available with the aid of the airborne radar or the global positioning system. We derive an asymptotic closed-form expression of the signal-to-interference-plus-noise ratio (SINR) as the number of transmitting antennas tends to infinity, in the presence of realistic co-channel interference and channel estimation errors. The achievable transmission rates and the corresponding mode-switching distance-thresholds are readily obtained based on this closed-form SINR formula. Monte-Carlo simulation results are used to validate our theoretical analysis. For the specific example of 32 transmit antennas and 4 receive antennas communicating at a 5 GHz carrier frequency and using 6 MHz bandwidth, which are reused by multiple other pairs of communicating aircraft, the proposed distance-based ACM is capable of providing as high as 65.928 Mbps data rate when the communication distance is less than 25\,km.

cs.IT↗