Search

Fast array diagnosis method is of great importance for ensuring reliable array performances in the fifth generation (5G) communication systems. In this paper, a fast array diagnosis method is presented for antenna arrays composed of several subarrays. The objective is to detect the failures based on themeasured complex array signals in short measurement distance. A single probe is required to record the array response in the near-field of the array. All the array elements are excited simultaneously, and only phase shift states of 0◦ and 180◦ are required in the measurements. For a base station (BS) antennaarray with N subarrays, N + 1 measurements are required, resulting in a fast diagnosis process. Finally, the proposed method was validated in an antenna array composed of 4 subarrays with 3 antenna elements in each subarray and successful diagnosis results (both for the whole subarray and single antenna element failure cases) can be observed.

The decoupling ground technique is applied to a 1×4 dipole array. The isolation between each array element is improved by optimizing the decoupling ground to make the free-space coupling wave and the surface current (introduced by the ground plane) out of phase. The mutual coupling can be reduced below -25 dB with improved radiation patterns and larger scanning gains. It is also shown that grating lobe reduction can be achieved by reducing the inter-element spacing of the array.

Pedestrian detection is widely used in daily life. many fields require high accuracy and fast speed of pedestrian detection is an urgent problem to be solved. the depth, convolution kernel size, and feature layer selection of neural networks have a great impact on the performance of target detection. in this paper, based on Single Shot MultiBox Detector (SSD), a pedestrian detection method based on Inception-SSD of sparse connections and multi-scale fusion is proposed. this algorithm achieves good performance in both detection speed and detection accuracy. Through comparing the experimental data on the PASCAL VOC and CUHK Occlusion image data sets, it shows that some of the optimized designs adopted in this paper have higher accuracy than the original algorithm, and the detection speed reaches 31 fps to meet the real-time requirements.

A 2×4 dual-polarized multiple-input-multiple-output (MIMO) array with baffle structures is proposed for 5G base station application. The baffle is composed of two metal split ring resonators (SRRs) printed on a 1-mm thick FR4 substrate. It works as band-stop filter which can block the coupling waves from different directions within the array. Some modifications are also made on the ground to further improve the array performance. Simulated results show that the mutual coupling for both the cross-polarization and co-polarization can be reduced to almost −25 dB or lower in 3.4-3.8 GHz.

In this paper, a defected wall structure (DWS) is proposed to reduce the mutual coupling within a 2×4 dual-polarized multiple-input multiple-output (MIMO) bowtie antenna array. The proposed DWS has frequency band-gap which can block the surface-waves propagation in the target band. The periodic DWS is inserted between the adjacent antennas both along the x-axis and y-axis. The 2×4 antenna array with DWS can achieve 25dB isolation for all co-polarization ports in 3.4-3.6 GHz. The DGS method will not enlarge the dimensions of the array or deteriorate the radiation characteristics.

Online Sequential Extreme Learning Machine (OS-ELM) can help to achieve high diagnostic accuracy, strong generalization performance, and high efficiency in processing big data, so this algorithm is widely used in the field of classification. However, parameters in the two hidden-layers, i.e., the input weight coefficients and thresholds are randomly set. Once the parameters are not feasible, the performance of the classification model will severely degenerate. In order to solve this problem, this paper combines OS-ELM and GA. Based on the GA, the optimal parameters of the hidden-layer are searched for better performance of the online sequential limit learning machine. By comparing the classification accuracy of the algorithm and the traditional online sequential limit learning machine algorithm in the fault diagnosis of the liquad filter of the analog circuit, it is found that the online sequential limit learning machine algorithm based on genetic algorithm can significantly improve the accuracy of the model.

For traditional machine learning and image processing, it is difficult to extract features, the quality of feature extraction will directly affect the classification accuracy, another problem is that the large difference between the original datasets and the target datasets in the transfer learning, lead to unsatisfactory results. In addition, the original convolution model network is too shallow, recognition rate is not high; Therefore, this paper presents an improved convolutional neural network method for pneumonia detection based on deep learning model. First, the image size in the original datasets are fixed and the appropriate batch size are used as the input of the network, Then the convolution layers and pooling layers are added on the basis of lenet-5 model, Secondly, a feature integration layer is added to construct an optimal model for accurate classification. The proposed method not only avoids the complex feature extraction process but also has fewer parameters than other classical convolutional neural networks. By using two public datasets, the second which provided by the RSNA® joint Kaggle medical image pneumonia recognition competition, after a series of experiments of single dataset and fusion of two datasets, the final accuracy rate reached 98.83% and 98.44% respectively, the test accuracy also reached 97.26% and 91.41% respectively. Compared with existing transfer learning, GoogleNet Inception V3+Data Augmentation(GIV+DA), GIV3+RF models, this model has many improvements.

In this paper, a new method to compare the effectiveness of a stirrer by calculating the number of characteristic modes of the stirrer is proposed. We find that, for Z-shaped stirrers, the more characteristic modes a stirrer has, the better it works in the reverberation chamber. The number of characteristic modes refer to the number of modes whose characteristic angels fall between 135 and 225 degree. The effectiveness of a stirrer is measured in terms of the field uniformity in the reverberation chamber. The field uniformity is defined as the standard deviation (STD)of the average power of the total electric fields at the eight vertices of the working volume. Three types of stirrers with the same rotational volume are used for verifications.

Online Sequential Extreme Learning Machine (OS-ELM) has high accuracy and fast update speed in the areas of classification, such as fault diagnosis and anomaly detection. However, OS-ELM selects hidden layer parameters randomly leads to unstable output, which reduces the reliability of OS-ELM seriously. In this paper, a ensemble method based on OS-ELM and Naive Bayes(Bayes-OS-ELM) has been developed. The ensemble model establishes parallel sub-classifiers with OS-ELM and a secondary classifier with Naive Bayes to fuse the results of the former sub-classifiers. Because of the parallel structure, the ensemble model can greatly reduce the disturbance caused by the random set of hidden layer parameters of OS-ELM and make the classification result more stable. Besides, as an accurate and stable algorithm, Naive Bayes effectively promote the accuracy and stability of the classification model. Several UCI data sets have been involved to verify the proposed classification model. Experimental results show that this method has high accuracy, stable result and great generalization performance compared with the existing approach.

A 26GHz lens antenna for 5G communications is presented in this paper. It consists of a specially designed dielectric lens and a feed, which is made of five linear subarrays with two microstrip patches per subarray (10 elements in total) etched on a 0.5-mm-thick substrate. A prototype has been designed and manufactured. The proposed antenna is capable of beam steering and achieving high gain with a light polyamide lens. The simulation and measured results on return loss and radiation characteristics are shown in this paper.

The memristor has the abilities of storage and computation, and is compatiable with the standard CMOS technology. Consequently, the memristor has been thought to potentially break the bottlenecks of memory and power in the traditional computer architecture. In this work, we present optimized full adders using memristor-based material implication (IMPLY) logic operation. Compared with the latest works, the execution steps of N-bit full adder with a serial structure is reduced from 23N to 16N while using one less memristor, and the number of execution steps and memristors used in full adders with a parallel structure are reduced from 5N + 18 to 3N +9 and from 9N to 8N, respectively. Additionally, we present a look-ahead carry adder based on the IMPLY logic on the memristor crossbar structure, which consumes (3N + 8) × (N + 3) memristors and needs only 3N +15 execution steps for an N-bit addition.

Locating multiple optima of a problem is an important and challenging task for many real-world applications. In this paper, a random walk mutation strategy is proposed for differential evolution (DE) to handle multimodal optimization problems. The mutation strategy is able to find a balance between exploitation and exploration. First, the neighborhood and fitness information of individuals is incorporated into mutation to guide the formation of donor vectors. This facilitates the evolution of individuals toward their nearby optima. Second, the exploration ability of the mutation strategy is preserved by simulating a random walk process. Moreover, an archive technique is designed to detect converged subpopulations. The converged individuals are then reinitialized to search for other optima. This enhance the algorithm's exploration ability. Meanwhile, found optima can be maintained throughout the optimization process by using the archive technique. The random walk mutation strategy and the archive technique are integrated with DE to make a competitive multimodal algorithm. The resulting algorithm is tested on a recently proposed benchmark function set. Experimental results show that the proposed algorithm is able to provide better performance than a number of state-of-the-art multimodal algorithms.

As the significance of the third industry grows, the tourism has gained more momentum, which renders the analysis of tourism imagery quite urgent. To meet the need, this thesis, firstly, established ECSPT model (Acronym of Environment, Culture, Service, Promotion and Tourist), the factor model based on former domestic and international tourism imagery studies. Secondly, by using the self-made software ROST NAT, data about 40 sight spots across the whole networks were captured. Combined with self-made software ROST EA and Emotion Formula, the 40 sight spots were structured in the ECSPT model, and the weight of each factor is measured.

Oil drilling materials are an essential material used in the process of oil exploration and development, it is important to predict the consumption quantity of oil drilling materials scientific. Firstly this paper proposes a guiding significance to the demand Forecasting, reviews the research on methods of demand forecasting. Secondly the basic principle of a forecasting model of the virtual variables is introduced. Thirdly the process of selecting the model variables is illustrated. Finally, the process of construction of virtual variable model is analyzed in detail. Its accuracy and a significant are tested. The results show that the predictability of the model is good. It is conducive to the oil drilling material management and control of oil companies.