In the process of identifying non-line-of-sight (NLOS), acoustics-based indoor positioning needs to collect audio recordings of sound fields in multiple rooms and upload them to the central server for training. Once the transmission process and server-side suffer malicious attacks, private data will also be leaked. To solve the training difficulty and privacy issues at the same time, we propose a novel Personalized Federated Learning (PFL) model combined with user frequency and room data capacity, taking into account the significant differences in positioning data with room layout. The proposed model can accurately identify the differences between different room data when aggregating on the server-side. By collecting data in the actual indoor environment and comparing the existing algorithms, the accuracy of the proposed method in the data verification of unfamiliar rooms is 90%.
In this study, an eye blinking re-identification system was proposed. A fast local binary pattern was used for feature extraction because its grayscale invariance and rotational invariance allow for the effective acquisition of feature information even in the presence of noise. Finally, a recurrent neural network and long short-term memory were used for model training. The results indicated that, compared with the model trained using static data, the models based on dynamic features were less affected by environmental noise in terms of accuracy. In addition, the model trained using the recurrent neural network was highly effective in identifying unenrolled users and achieved high overall accuracy.
Hand pose estimation based on a single RGB image has low accuracy due to the complexity of the pose, local self-similarity of finger features, and occlusion. A multiscale feature fusion network (MS-FF) for monocular vision gesture pose estimation is proposed to address this problem. The network can take full advantage of different channel information to enhance important gesture information, and it can simultaneously extract features from feature maps of different resolutions to obtain as much detailed feature information and deep semantic information as possible. The feature maps are merged to obtain the hand pose results. The InterHand2.6M dataset and Rendered Handpose Dataset (RHD) are used to train the MS-FF. Compared with the other methods (which can estimate interacting hand poses from a single RGB image), the MS-FF obtains the smallest average error of hand joints on RHD, verifying its effectiveness.
There can be numerous electronic components on a given PCB, making the task of visual inspection to detect defects very time-consuming and prone to error, especially at scale. There has thus been significant interest in automatic PCB component detection, particularly leveraging deep learning. However, deep neural networks typically require high computational resources, possibly limiting their feasibility in real-world use cases in manufacturing, which often involve high-volume and high-throughput detection with constrained edge computing resource availability. As a result of an exploration of efficient deep neural network architectures for this use case, we introduce PCBDet, an attention condenser network design that provides state-of-the-art inference throughput while achieving superior PCB component detection performance compared to other state-of-the-art efficient architecture designs. Experimental results show that PCBDet can achieve up to 2× inference speed-up on an ARM Cortex A72 processor when compared to an EfficientNet-based design while achieving ∼2-4% higher mAP on the FICS-PCB benchmark dataset.
This study proposed a matched field source localization method based on tensor decomposition. By considering the advantages of tensors in multidimensional data processing, a three-dimensional tensor signal model of space-time-frequency is constructed, and the signal subspace is estimated using high-order singular value decomposition (HOSVD). The source position is estimated by matching the measured data tensor signal subspace with the replica field tensor signal subspace. The S5 event data of SWellEx-96 is processed by the proposed tensor-based matched-field processing (TMFP). The comparison with the results of conventional matched field processing (MFP) shows that TMFP has a better suppression effect on ambient noise under low SNR and better source localization performance.
Computing in-memory technology is a promising way of solving the “memory wall” problem when processing large-scale data,accelerating data processing, and improving energy efficiency. However, calculations in the analog domain are limited in terms of accuracy and sensitivity to process, voltage, and temperature (PVT) changes. In this study, we proposed a computing in memory multiply-and-accumulate (CIM MAC) circuit which could generate pulses automatically and which endows the time-domain approach with a digital character. The MAC result was reflected by the pulse edge and converted into the final digital output using a dual-edge counter quantization circuit, improving the accuracy of the MAC operation and reducing the difficulty of quantization. The performance of the proposed CIM circuit was evaluated using a 28-nm process. It could achieve 4-bit multiplication without errors with an energy efficiency of 40.34 to 883.37 TOPS/W.
A novel 3-bit frequency-reconfigurable antenna (FRA) with miniaturized dimensions is realized with a meander line. The frequency reconfiguration of the antenna is achieved by introducing N RF p-i-n diodes into the meander line. The related parts of the meander line with different lengths are bypassed or included into the antenna by switching on or off the diodes, resulting in 2N switchable size lengths of the antenna and equally spaced operating frequencies. A 3-bit meander-line reconfigurable antenna (N=3) is designed, and the simulated and measured results agree well. The antenna provides 23=8 independent switchable states, with the operating frequencies covering a wide switchable frequency range from 1.04 GHz to 1.51 GHz and the working bandwidths varying from 80 MHz to 150 MHz. The number of working states is optimally large, considering the number of switches used. Besides, this work has an acceptable peak gain of 1.59 dBi regarding the miniaturized total dimension of 0.17 λ × 0.07 λ (λ is the wavelength of the lowest working frequency), which is more compact than many published FRAs.
This letter proposes a polarization microwave correlation imaging method based on the orthogonal complement space. It utilizes the orthogonal complement space of the HV antenna radiation field to cross-multiply the echo information, enabling simultaneous correlation imaging and instantaneous polarization measurement of the target. Currently, there is a significant research gap in polarization-driven microwave correlation imaging methods, and the existing relevant studies focus on enhancing the randomness of the radiation field using polarized antenna elements, without incorporating polarization information into the imaging process. Through simulation analysis, this method further improves the quality of microwave correlation imaging and its ability to resist interference. Moreover, under low time-frequency products, the peak sidelobe level (PSL) and isolation (I) of this method are approximately 3.5 dB and 12.5 dB higher, respectively, than those of traditional instantaneous polarization measurement (TIPM)methods.
The knife-edge diffraction model (KED) and the uniform geometrical theory of diffraction (UTD) have been widely used to predict the shadowing effect at millimetre-wave (mmWave) bands. This letter proposes a mathematical derivation to rigorously prove that, for an absorbing screen, UTD applying the narrow-angle Fresnel approximation is equivalent to KED. The simulation scenarios are designed to validate the proposal by comparing KED with UTD in the narrow-angle (less than 20○) and wide-angle (over 20○) regions at mmWave bands (20 GHz - 100 GHz). Simulated results agree with the proposal that KED is identical to UTD with a low error of less than 0.1 dB in the narrow-angle region, while they have a difference with an error of over 1 dB in the wide-angle region. In addition, the average computational time is measured and results in both UTD and KED taking approximately 8.0 ms for one test. From the proposal, it can be theoretically explained the differences and similarities between KED and UTD for an absorbing screen.
A novel stripline diplexer design using frequency dependent couplings to achieve multiple transmission zeros is developed in this paper. The transmission zeros generated by the frequency dependent couplings are flexible and controllable, on the basis of the existing cross-coupled, more transmission zeros are introduced to improve the frequency selection characteristics. Based on this characteristic, we designed a 2.6G Hz diplexer, its transmitting channel filter is 5 order with 4 transmission zeros, and the receiving channel filter is 4 order with 5 transmission zeros. We fabricated and measured it, the synthesis results, simulation results, and the tested results are well matched with each other, which will provide more flexibility in the design of diplexers for wireless communication system.
Considering light absorbing and scattering problems in connection with wavelength can decrease the visibility, contrast and color distortion of images, we propose a new type of convolutional neural network with two training phases. Firstly, the coordinate attention module is integrated into the residual block of the residual group in the backbone network, which is used to strengthen the feature extraction capability of the network. Secondly, since the unrealistic image colors may degrade the image details, an unsupervised method that combines the physical prior knowledge and the real underwater images is proposed to finetune the backbone network. Furthermore, a model protection mechanism is designed to guarantee the successful execution of the training. The experimental results indicate the proposed model can effectively optimize the contrast, color and image quality of the underwater image. Compared with relevant algorithms, our UCIQE and NIQE are respectively 0.525 and 4.149, which further verifies the superiority of the proposed model.
This work is mainly to describe the phenomenon at low frequency in my previous publication. In this paper, the impact of the magnetostriction mechanism is taken into consideration for the main idea. The axisymmetric FEM model of the spiral-coil EMAT is established to implement the simulation. With the help of the simulation, it is demonstrated that the directivity of ultrasonic wave can be manipulated by frequency. And it is found that the direction of Lorentz force that dominates in the rail varies with time, but the magnetostrictive force compels the ultrasonic wave mainly generated by the Lorentz force to the axis. This describes well that the power of two combined mechanisms is greater than that of only the Lorentz-force mechanism at low frequency.
With the increasing innovation of network communication technology, short burst communication has been developed to a certain extent in the time compression technology, which brings great challenges to the radio signal blind reception technology .In the blind reception of non-cooperative signals in short-time burst communication, signal time domain detection is the premise and key of signal modulation parameter estimation, signal demodulation, decoding, interpretation, information acquisition and interference guidance. In this paper, the blind demodulation method of short-time burst FM-MFSK signal is studied under the condition of low signal-to-noise ratio. A short-time burst FM-MFSK blind demodulation method based on STFT is proposed. When SNR is higher than 2dB, the correct demodulation probability for FM-MFSK signal can reach 90%.
In this work, we report the first demonstration of an ultraviolet light-emitting diode (LED) with boron-containing multiple quantum wells. Electroluminescence emission from the BAlGaN LED was observed at 350 nm, with higher intensity compared to the AlGaN reference LED. A higher operating voltage compared to the reference LED was also observed which may be attributable to a nanomasking behaviour of boron in (Al)GaN alloys.
Using Jensen's inequality and integration by parts, we derive some tight upper bounds on the Gaussian Q-function. The tightness of the bounds obtained by Jensen's inequality can be improved by increasing the number of exponential terms, and one of them is invertible. We obtain a piece-wise upper bound and show its application in the analysis of the symbol error probability of various modulation schemes in different channel models.
Consider general minimum variance distortionless response (MVDR) robust adaptive beamforming problems based on the optimal estimation for both the desired signal steering vector and the interference-plus-noise covariance (INC) matrix. The optimal robust adaptive beamformer design problem is an array output power maximization problem, subject to three constraints on the steering vector, namely, a (convex or nonconvex) quadratic constraint ensuring that the direction-of-arrival (DOA) of the desired signal is separated from the DOA region of all linear combinations of the interference steering vectors, a double-sided norm constraint, and a similarity constraint; as well as a ball constraint on the INC matrix, which is centered at a given data sample covariance matrix. To tackle the nonconvex problem, a new tightened semidefinite relaxation (SDR) approach is proposed to output a globally optimal solution; otherwise, a sequential convex approximation (SCA) method is established to return a locally optimal solution. The simulation results show that the MVDR robust adaptive beamformers based on the optimal estimation for the steering vector and the INC matrix have better performance (in terms of, e.g., the array output signal-to-interference-plus-noise ratio) than the existing MVDR robust adaptive beamformers by the steering vector estimation only.
Automatic Modulation Recognition (AMR) is a fundamental research topic in the field of signal processing and wireless communication, which has widespread applications in cognitive radio, non-collaborative communication, etc. However, current AMR methods are mostly based on unimodal inputs, which suffer from incomplete information and local optimization. In this paper, we focus on the modality utilization in AMR. The proxy experiments show that different modalities achieve a similar recognition effect in most scenarios, while the personalities of different inputs are complementary to each other for particular modulations. Therefore, we mine the universal and complementary characteristics of the modality data in the domain-agnostic and domain-specific aspects, yielding the Universal and Complementary subspaces accordingly (dubbed as UCNet). To facilitate the subspace construction, we propose universal and complementary losses accordingly, where the former minimizes the heterogeneous feature gap by an adversarial constraint and the latter consists of an orthogonal constraint between universal and complementary features. The extensive experiments on the RadioML2016.10A dataset demonstrate the effectiveness of UCNet, which has achieved the highest recognition accuracy of 93.2% at 10 dB, and the average accuracy is 92.6% at high SNR greater than zero.
To improve the detection rate of pulmonary nodules in early lung cancer screening, a low-dose CT pulmonary nodule detection algorithm based on 3D CNN-CapsNet (3D convolution neural network and capsule network) was presented. However, the convolution kernel size of the traditional CNN is relatively simple at each layer, and it is difficult to obtain more abundant features, which is not effective for medical images with a hierarchical structure and does not fully consider the spatial information of medical sequence data. CapsNet is a new network architecture that can be used to classify, using a group of neurons as a capsule to replace the traditional neural networks, it may be made to the attribute information and spatial feature extraction. The network structure we designed includes FCN and CapsNet. First, the convolution kernels of different sizes are used to extract features at different scales. Then enter the initial feature map to obtain the first part into the designed CapsNet to get the final classification result. Through the experimental verification of the ELCAP database, the nodule detection rate is 95.19%, the sensitivity is 92.31%, the specificity is 98.08% and the F1-score is 0.95 which are much better than other baseline methods.