Search

It is a promising technology to replace traditional transceiver design with neural network (NN)-based autoencoder to realize end-to-end communication. The joint optimization of multiple blocks helps the communication system to achieve better performance. However, this approach needs a differentiable channel modeling to train both NNs at the transceiver. For a non-differentiable channel in a real scenario, gradient backpropagation cannot be performed. In this paper, we propose a Kalman-based autoencoder framework to achieve gradient estimation and backpropagation by adding a control layer at the last layer of the encoder NN, which combines the rich feature representations learned by the NN with the Bayesian filter method based on Extended Kalman Filter (EKF) and Cubature Kalman Filter (CKF). According to simulation results, the proposed framework has a fast convergence rate and strong robustness in both additive white Gaussian noise (AWGN) and Rayleigh block fading (RBF) of perturbation scenarios.

Product co-innovation community is increasingly used by firms to involve customers and leverage user knowledge for products and services innovation. The purpose of this paper is to investigate: what users do, what their roles are, and what they contribute to the co-innovation. This paper investigates user behaviors with netnography, a qualitative methodology which adapts ethnographic research techniques to study online community. The research setting is a typical product co-innovation project in an international community. The results reveal that: User participation, both creation and interaction behaviors, is characterized by core-periphery model. Only a small number of active individuals are valuable contributors; Users act as six different roles; Correspondingly, users make various contributions not only in forms but also in quality. This study provides community managers a comprehensive and workable way to grasp user behaviors. Besides, it suggests managerial guidance for community practitioners on improving design efficiency and participation convenience and effectiveness. It contributes to a theoretical understanding of user performance in the product co-innovation community and expands the research in online community.

According to well-stablished results in the literature, the Long Short Term Memory (LSTM) model is one of learning models most widely used in stock price prediction given its characteristic feature. In this paper, we employ a novel neural network, Gated Recurrent Unit (GRU), in performing individual stock price prediction task in Chinese A-share market. As shown by the experiment results, GRU has comparable performance with LSTM and both them outperform the conventional Recurrent Neural Network (RNN) model. Further, regression analysis indicates that there may exist quadratic relationship between prediction accuracy and training data size. Thereby attempts have been made on adding nonlinear time-weight functions to substantially improve the prediction accuracy with the LSTM model.

In secure communication, finding bits that used for steganography is essential for effective and high-quality speech information hiding. This paper presents an adaptive speech information hiding method based on K-means, which can quickly find hidden bits. This method uses K-means algorithm to detect effective hidden bits through clustering and embeds secret information with different matrices to achieve good efficiency utilization of hidden bits. This proposed method embeds secret information according to the clustering result and the matrix coding at the transmitting end and extracts the secret information according to the corresponding embedding matrix ate the receiving end. Thereby implementing secure communication over VoIP. The experimental results indicate that the proposed method has better speech quality (PESQ is reduced by about 1.5%) while achieving a higher embedding capacity (up to 976 bit/s), and it shows strong concealment.

Based on Planer near-field translation theory and Nyquisty sampling theory, this paper studies the influence of sampling spacing on the result in planar near-field measurement by means of theoretical analysis and simulation, and acquires some laws and useful results, and proves the applicability and correctness of Nyquisty undersampling method in narrowband antenna measurement. The test efficiency can be effectively improved by increasing sampling spacing.

This paper describes an SNO (Simultaneous Nadir Overpass) method based on observation elements matching for radiometric cross-calibration of China HJ-1B and MODIS thermal infrared channels. Firstly, a DTC (Diurnal Temperature Cycle) model with four parameters and ECMWF data are introduced for time matching. Then a BRDF model updated to the TIR domain is built for angle matching. Combining matching coefficients with the spectral matching factor, the TOA radiance of MODIS B31 can be converted into TOA radiance of HJ-1B. Finally, the radiometric calibration results using the images of Qinghai Lake show that the proposed method is effective. Compared with the strict SNO method, the accuracy is improved by 0.73K.

U-type ironless permanent magnet linear motors (IL-PMLMs) are widely used in high-performance applications due to the advantages of high speed, acceleration, precision, and low wear. The magnetic field in the back iron of the traditional U-shaped IL-PMLMs appears a strong and weak alternating phenomenon. And the peak magnetic density is too high, which raises the risk of saturation. By employing a magnetic permeable block at the end of IL-PMLMs, the homogeneity of the back iron magnetic field can be optimized. Also, the design method of the best size of the magnetic permeable block is proposed. Finally, the optimization results of the magnetic field are verified by the finite element model.

The linear motor system is a low-damping system, which will cause residual vibration especially in the high-speed start-stop motion process, and then deteriorates the accuracy of the equipment. By employing an eddy current damper, the residual vibration can be controlled and efficiently reduced. In this paper, an improved dual-sided hybrid excitation eddy current damper (ECD) by using steel-copper plate secondary is proposed. An analytical model is established to explore the relation between the thickness of the steel layer and the copper layer. According to the finite element analysis (FEA), with the increase of the steel thickness, the damping force first increases and then decreases. The maximum damping force is increased by 36.15% compared with that of the pure copper plate. Experimental tests are carried out on the prototype to verify the improvement of the optimized damper. The results show that the peak amplitude of the position error and the settling time are reduced by 1.94 times and 1.80 times, respectively, compared with that of the pure copper plate.

Recently, a parallel decoding algorithm of $G_{N} -$coset codes was proposed. The algorithm exploits two equivalent decoding graphs. For each graph, the inner code part, which consists of independent component codes, is decoded in parallel. The extrinsic information of the code bits is obtained and iteratively exchanged between the graphs until convergence. This algorithm enjoys a higher decoding parallelism than the previous successive cancellation algorithms, due to the avoidance of serial outer code processing. In this work, we present a hardware implementation of the parallel decoding algorithm, it can support maximum $N =16384$. We complete the decoder’s physical layout in TSMC $16 nm$ process and the size is $999. 936 \mu m \times 999.936 \mu \mathrm{m}, \approx 1. 00 mm^{2}$. The decoder’s area efficiency and power consumption are evaluated for the cases of $N = 16384,K =13225$ and $N = 16384,K =14161$. Scaled to $7 nm$ process, the decoder’s throughput is higher than $477{Gbps/mm}^{2}$ and $533{Gbps/mm}^{2}$ with five iterations.

Recently, a parallel decoding framework of G N -coset codes was proposed. High throughput is achieved by decoding the independent component polar codes in parallel. Various algorithms can be employed to decode these component codes, enabling a flexible throughput-performance tradeoff. In this work, we adopt successive cancellation (SC) as the component decoders to achieve the highest-throughput end of the tradeoff. The benefits over soft-output component decoders are reduced complexity and simpler (binary) interconnections among component decoders. To reduce performance degradation, we integrate an error detector and a log-likelihood ratio (LLR) generator into each component decoder. The LLR generator, specifically the damping factors therein, is designed by a genetic algorithm. This low-complexity design can achieve an area efficiency of 533$Gbps/mm^{2}$ under 7nm technology.

Deep neural networks have been widely used in crowd counting that aims to give the number of objects in images and videos. The performance of crowd counting models is greatly affected by the size of datasets with high quality annotations. However, collecting and annotating large-scale crowd counting dataset is labor-intensive and time-consuming. In this work, we exploit unlabeled web images to boost the performance of crowd counting models in a semi-supervised manner. Based on the observation that the rotation and splitting operations will not change the number of object in image, we design three auxiliary tasks to improve the feature representation ability of deep models. A semi-supervised multi-task learning framework is proposed by introducing three auxiliary tasks with respect to unlabeled data and our framework can be easily extended to other crowd counting models. An unlabeled dataset (Web-Crowd) with 8679 web images are collected for semi-supervised crowd counting. Experiments shows that our semi-supervised multi-task learning framework can effectively boost the performance of crowd counting models on UCF-QNRF dataset and ShanghaiTech dataset.

In the research on firepower strike optimization of intelligent battlefield multi-weapon platform, in order to improve the overall strike effectiveness of weapons, it is necessary to establish a weapon target allocation model to obtain real-time and accurate strike plans, and obtain strike-related strike parameters through data model comparison. To this end, this article first introduces the concept, classification and basic mathematical model of the WTA problem, and sorts out the progress of the weapon-target assignment (WTA) problem at home and abroad. For the analysis of the WTA problem, you can use the genetic algorithm "survival of the fittest, survival of the fittest" mechanism, the particle swarm algorithm iterative process is simple and fast, and the ant colony algorithm has clear goals and strong scalability, etc. to obtain a precise strike plan to solve the problem of low real-time firepower strikes, The problem of low accuracy, finally pointed out the lack of horizontal comparison between the current WTA problem and the intelligent optimization algorithm research, and clarified the next development direction from the three perspectives of the feasibility, effectiveness and scalability of the WTA problem.

The purpose of this study is to develop an automatic technique to segment the membrane of the maxillary sinus with morphological changes (e.g. thickened membrane and cysts) for the detection of abnormalities. The first step is to segment the sinus bone cavity in the CBCT image using fuzzy C-mean algorithm. Then, the vertices of inner bone walls of sinus in the mesh model are screened with vertex normal direction and angular based mean-distance filtering. The resulted vertices are then used to generate the bony sinus cavity mesh model by using Poisson surface reconstruction. Finally, the sinus membrane morphological changes are segmented by subtracting the air sinus segmentation from the reconstructed bony sinus cavity. The proposed method has been applied on 5 maxillary sinuses with mucosal thickening and has demonstrated that it can segment thin membrane thickening (< 2 mm) successfully within 4.1% and 3.5% error in volume and surface area respectively. Existing methods have issues of leakages at openings and thin bones, and inaccuracy with irregular contours commonly seen in maxillary sinus. The current method overcomes these shortcomings.

This paper combines the principle of two degrees of freedom with internal model PID control, and designs an internal model PID controller based on two degrees of freedom, which is applied in the aeration system of sewage treatment to control the concentration of dissolved oxygen. Through the comparison and Simulation of MATLAB, it is proved that the controller makes the actual output value of dissolved oxygen concentration track the set value rapidly, it overcomes the adverse effect caused by the model mismatch of dissolved oxygen concentration control, it optimizes the target tracking performance and anti-interference performance in the dissolved oxygen control of the aeration system, it improves the control effect, and it has important use value.

In recent years, extreme weather such as low temperature freezing rain and snow has caused great threat to the safe and stable operation of the power grid. The DC ice melting device is a key technical means to cope with extreme weather and can effectively improve the ability of the power grid to resist freezing disasters. In 2019, the Kunming Power Supply Bureau 500kV Qidian substation installed DC ice melting device. Since the rectifier unit of the DC ice melting device uses traditional half-controlled power electronic components, a large amount of harmonics will be generated on the AC/DC side. The zero-power debugging of the DC ice-melting device and the line melting test show that the harmonic will not only affect the system protection action, but also affect the measurement of the leakage current of the metal oxide arrester, thus affecting the judgment of the insulation characteristics of the system equipment. In this paper, the DC melting device in Qidian substation and its harmonic characteristics are studied, and the influence of harmonics generated by DC ice melting device on the leakage current measurement of metal oxide arrester in the system is studied. The theoretical derivation and analysis of this process are given. In order to further guide the operation of the DC ice-melting device and the elimination of harmonics, the theoretical basis for the influence of the leakage current monitoring of the metal oxide arrester is provided.

It has become a trend for computers to automatically mark examination papers. This paper presents an algorithm for automatically correcting blank filling questions. We first use edge detection algorithm to find the location of the answer, and cut the storage. Then, the connected domain is used to cut out the characters in the answer and store them as MAT files. Finally, we send these characters to the improved lenet-5 convolutional neural network for recognition. Experiments show that the algorithm can correctly review the blank questions.

A trajectory of a destination-directed moving object (e.g. an aircraft from an origin airport to a destination airport) has three main components: an origin, a destination, and motion in between. We call such a trajectory that end up at the destination destination-directed trajectory (DDT). A class of conditionally Markov (CM) sequences (called CM L ) has the following main components: a joint density of two endpoints and a Markov-like evolution law. A CM L dynamic model can describe the evolution of a DDT but not of a guided object chasing a moving guide. The trajectory of a guided object is called a guided trajectory (GT). Inspired by a CM L model, this paper proposes a model for a GT with a moving guide. The proposed model reduces to a CM L model if the guide is not moving. We also study filtering and trajectory prediction based on the proposed model. Simulation results are presented.

The sharing economy emerged just twenty years ago. It was hard to envision its importance for many industries back then, but today, it is closely incorporated into our daily lives. Airbnb is a prime example of this phenomenon; it brought a brand new business model to the hospitality industry. The Airbnb platform has created millions of micro-businesses with a distinctive pricing strategy. Understanding the pricing strategy will provide insights into this new business model.This paper analyzes a sample of 48 896 listings in New York City from Airbnb.com, and builds a price prediction model with natural language processing and machine learning techniques. Different methods ranging from linear regression, to a generalized additive model, to a deep neural network, to random forest, to XGBoost and bagging (merging the previous five methods) were explored for the creation of the final prediction model. Of these, bagging, XGBoost and random forest demonstrated the strongest performance with the test data collected from the Airbnb website on January 28th, 2020.

From a systematic perspective, Industrial Internet involves five kinds of industrial models, including object information model, mechanism model, data-driven model, mixed model and simulation model. Firstly, the contents of five kinds of industrial models and the relationships between them are introduced. Secondly, the existing modeling methods in academia and industry are summarized from different perspectives, such as object information model, mechanism model, data-driven model, mixed model and simulation model. Finally, the development trend of modeling technology in the field of Industrial Internet and making the future research prospects are summarized.