Search

Weakly supervised semantic segmentation (WSSS) generally utilizes the Class Activation Map (CAM) to synthesize pseudo-labels. However, the current methods of obtaining CAM focus on salient features of a specific layer, resulting in highlighting the most discriminative regions and further leading to rough segmentation results for WSSS. In this paper, we rethink the potential of the ordinary classifier and find that if features of all the layers are applied, the classifier will obtain CAM with complete discriminative regions. Inspired by this, we propose Fully-CAM for WSSS, which can fully exploit the potential of the ordinary classifier and yield more accurate segmentation results. Precisely, Fully-CAM firstly weights feature with their corresponding gradients to yield CAMs of each layer, then fusing these layers’ CAMs could generate an ultimate CAM with complete discriminative regions. Furthermore, Fully-CAM is encapsulated into a plug-in, which can be mounted on any trained ordinary classifier with convolution layer, and it exceeds its previous performance without extra training.

High control bandwidth is usually restricted in a photoelectric tracking system (PTS) based on a Charge-Couple Device(CCD) with time delay, which hinders a good tracking performance. Generally, a model-based delay-compensation controller called Smith predictor (SP) can help increase the controller gain to promote the bandwidth by separating delay from the control loop. However, the performance promotion is insufficient because the delay still stays in the forward channel which causes errors between output and input. And the increase of the controller gain is still limited due to the effect of model mismatch on stability. In this paper, to solve the problems, a delay-compound-compensation control (DCCC) based on improved SP by trajectory prediction and velocity feedforward is proposed. The additional trajectory prediction is used to further eliminate the effect of delay existing in the forward channel. The additional velocity feedforward is used to further reform the transfer characteristics limited by the controller gain. A Kalman filter-based design method of trajectory prediction is presented and the optimal design principle of feedback and feedforward controllers is given in the face of model mismatch. Experiments demonstrate that the DCCC is valid and could greatly promote the tracking performance in the low frequency.

In this paper, we present a neural network-based framework to compensate for the nonlinear distortion of high-speed continuous variable quantum key distribution (CV-QKD) systems. In practice, the nonlinear impairment of imperfect devices could bring interference to the measured quadrature values and thus compromise the parameter estimation procedure and the secure key rate. Specifically, for a high-speed CV-QKD system, the balanced homodyne detector (BHD) and the following analog-to-digital converter are the primary sources of nonlinear impairment. The nonlinear distortion effect will be more pronounced as the pulse repetition rate increases. We propose an autoencoder-based network to learn the inner patterns of distorted data and then compensate the quadrature measurements in real-time at the receiver’s side. The numerical simulation demonstrates that the excess noise induced by nonlinear impairment is reduced to the $10^{-3}$ level in the shot noise unit. Consequently, such a software approach has the ability to significantly boost the bandwidth efficiency of the BHD and the secure key bit rate.

An improved moments estimation algorithm used in a newly developed K-band wind profiler to obtain accurate 3-D profiles of wind velocity along the altitude is described. The signal-to-noise ratio (SNR) of turbulent echoes detected by millimeter-wave radar is usually low, making it becomes a challenging task to determine accurate Doppler profiles. And if wind profilers can provide quick estimates of wind, they will have greater applicability in automated, real-time environments. The improved method combines the concepts of adaptive Doppler windows and profile chain construction, reduces the search range of prospective spectral peaks by using adaptive Doppler windows, and designs new multiparameter cost functions to determine velocity profiles by studying the wind shear in the surface boundary layer. The algorithm can effectively identify the atmospheric echo components of each range bin even under the condition of low SNR, and the computational time is decreased by at least 18% compared with the previous algorithms. The proposed method is tested on the real wind radar datasets to verify its robustness and reliability. The wind information obtained by four different algorithms is compared with the data obtained from the meteorological mast of the Gaoyou wind farm in China. The comparison results indicate that the proposed method shows a nicer match with the data of the mast, and it derives the winds more accurately in the atmosphere of the surface boundary layer.

P2.5 plug-in hybrid electric vehicles (P2.5-PHEVs) exhibit high transmission efficiencies and no power interruption in the shifting and mode switching process; thus, they have broad application prospects. The power transmission in a PHEV under pure electric startup does not entail the clutch, and the initial torque of the motor is often set under the condition that the maximum allowable slope should not lead to backward sliding, which leads to the problems of excessive jerk in small-slope startup and catapulting when startup occurs downhill. The optimal method is to set different initial torques according to different gradients; however, the vehicle would still be in the pre-startup stage, making it impossible to estimate the slope using a dynamic method. In view of the foregoing problems, according to an analysis of pure electric startup dynamics, a slope-memory-based strategy for estimating and storing the slope during the vehicle movement before parking is proposed. During startup, the initial torque is set according to the memorized slope. In the processes of startup and acceleration, the driver’s startup intention is identified, and different jerk control targets are set. The torque of the acceleration process is controlled according to the set initial torque and jerk target. Simulation results indicate that the maximum jerk of the proposed strategy is reduced by 23.6% for startup on a 5% ramp and by 57.5% for startup at 15% downhill; thus, the strategy mitigates the problems of the excessive jerk and catapult for startup on a small slope.

The supply of coal is related to the stability of social development in China. Coal mining faces are generally located under farmland and wasteland to avoid destroying the livelihood and property of residents, which seriously affects the selection of permanent scatterer (PS) with time-series interferometric synthetic aperture radar (TS-InSAR) method. The advanced TS-InSAR (ATS-InSAR) method combining PS and distributed scatterer (DS) monitoring modules is a useful tool to increase measurement points. However, the number of DS and the accuracy of measurement are different with different thresholds of temporal coherence. On the basis of ATS-InSAR, we propose a modified method applying the multi-level processing strategy to obtain more reliable deformation information in this paper. 16 sentinel-1A images are used to monitor mining subsidence in Peixian, China. The results of three different methods are cross-verified. Meanwhile, reliable DS pixels are identified by using the thresholding of both temporal coherence and Pearson correlation coefficient through the hierarchical processing. The results show that the deformation of the modified method reveals a large subsidence with the maximum rate of −563 mm/yr. The number of measurement points selected by the modified method is about 6.6 times that of the TS-InSAR method, and 1.3 times that of the ATS-InSAR method. The modified strategy can extract a great number of reliable pixels and reduce error propagation to ensure measurement accuracy. This research offers information to relevant departments for risk management purpose.

A wideband dual-polarized antenna is proposed for 4G and 5G communication applications. The antenna element consists of a pair of opened-loop dipoles for dual-polarization and its three inherent resonant modes are shifted closer to form a wide bandwidth. With additional U-shaped slots etched around the feed point, the input impedance of the antenna element is matched to 50 Ohm. The fabricated antenna element operates from 1.8 to 4.0 GHz, having 75.9% (VSWR ≤ 2) impedance bandwidth and high port-to-port isolation (>25 dB). It also achieves a 67±1° beamwidth in H-plane and 68.7 ± 3.3° in V-plane with 8.5 ± 1 dBi gain across the supported bands. By using six antenna elements and designing RF phase shifting module (RFPSM) based on vector modulators, a six-element dual-polarized array with electrically downtilt is also fabricated and measured, realizing a peak gain of 16.8 dBi, a similar beamwidth in H-plane as single antenna element and an electrically downtilt in V-plane from 0° to 12° which is preferred in communication applications. This antenna array could be used for future 5G communication and other applications.

In this paper, we have introduced a compact eight-element antenna array for triple-band MIMO (multi-input multi-output) operation in 5G (fifth generation) mobile terminals, which consists of two symmetric four-element sub-arrays disposed, respectively, along two long side-edges of the system ground plane of the mobile terminal. This antenna element can operate at two dual-wideband of 3250-3820 MHz and 4790-6200 MHz with three resonance frequencies which can fully cover 5G NR (New Radio) BandN78 (3300-3800 MHz), China 5G-Band of 4800-5000 MHz and LTE Band-46 (5150-5925 MHz). The main parameters that characterize the performance of the proposed MIMO antenna system, such as bandwidth, reflection coefficient, isolation (more than 10.5 dB), total radiation efficiency (more than 43%), ECC (envelope correlation coefficient, less than 0.12) and CC (channel capacity, 34.9-37.6 bps/Hz), are analyzed and presented in this paper. Besides, the effects of single-hand and dual-hand on performance of the MIMO antenna system are also discussed. The structure of antenna element is only 15 mm × 3 mm (0.175 λ0 × 0.035 λ0, λ0 is the free-space wavelength at the frequency of 3.5 GHz) with a very small ground clearance of 13 mm × 2 mm on the main board of the proposed mobile terminal. The antenna system has the possibility to be used in 5G mobile terminals with triple-band operation, narrow frame and large display. The corresponding antennas prototype is fabricated and measured; and a quite good agreement between simulation and measurement is obtained.

In heterogeneous ultra-dense networks (HUDNs), load balancing is always considered by offloading users from the macro base station (MBS) tier to the small base station (SBS) tier, to alleviate the data traffic of the MBS tier, while it causes a serious problem that the offloaded users will suffer strong interference from the MBS tier, and hence the signal-to-interference ratio (SIR) is in low region for the offloaded users. In order to improve the performance of these offloaded users, an interference mitigation scheme based on resource partitioning is employed in the literature. Specifically, a fraction of resource blocks (RBs) is reserved for the offloaded users, and then the SIRs of these users will be enhanced because there is only interference from the SBS tier. In summary, the combination of offloading and resource partitioning is called joint offloading and resource partitioning scheme, which has been investigated in the HUDNs built on the 2D space, while it has rarely been deliberated in 3D HUDNs that are used to model the scenario of high-rise buildings. Moreover, a two-tier 3D HUDN is proposed in this paper, and the MBS and SBS tiers are modeled as two independent Poisson point processes. Then, the SIR and rate meta distributions, which are used to characterize the fraction of users that achieve a desired link reliability given SIR or rate thresholds, are formulated for the proposed 3D HUDN. Lastly, numerical results show that with the usage of the joint offloading and resource partitioning scheme, the performance of the proposed 3D HUND can be significantly improved.

Being a promising candidate technology for the fifth-generation cellular systems, the ultra-dense network (UDN) is capable of improving the system throughput by deploying more small cells (SCs) in the existing macro-cells (MCs). In most cases, base stations (BSs) from both the MCs tier and SCs tier are modeled as two independent homogeneous Poisson point processes (PPPs) or other point processes derived from a PPP. In the previous works, the area spectral efficiency (ASE) and energy efficiency (EE) are evaluated from a typical user's perspective, and this method works well in single-user scenarios. However, there are multiple users sharing spectral resources in practical cellular systems, and the above-mentioned method cannot work in these scenarios. Therefore, we investigate the ASE and EE from the perspective of a typical BS. Analytical results show that the ASE increases with the increasing SCs density, which means that deploying more SCs in the MCs coverage area is a feasible way to enhance the system throughput. However, the energy consumption of the entire network is increasing as the SCs density increases. Therefore, we employ the firefly algorithm to jointly optimize the ASE and EE for the two-tier heterogeneous UDN. Finally, the optimal system parameters can be found under two fixed weights for the ASE and EE, respectively.

Millimeter-Wave (mmWave) antennas for 5G mobile terminals require a wide bandwidth, large-angle beam scanning, low-profile design, and multi-substrate compatibility for module-level integration. In this paper, we propose one candidate design employing a patch structure with the shorting pin to particularly generate extra zero modes. By taking advantages of the 2nd zero-mode with the TM01 mode, we can obtain a wide bandwidth covering 23.5~28 GHz, a large-angle beam scanning with ±60°, as well as keep the substrate as low profile as 0.508 mm. Thanks to the zero-mode induced patch-type design, it is compatible to multi-layer configuration possessing the extensibility and flexibility for further module design. We experimentally valid the design of a 4 × 2 array with multi-layer configuration in a cell phone environment. Good RF performances with ±60° scanning in the wide bandwidth indicate this proposed design can be an appropriate candidate for 5G mobile terminals.

A dual-functional antenna array operating in the long term evolution (LTE) band 42 (3.4-3.6 GHz) and 5.8-GHz wireless wide area network (WLAN) band (5.725-5.875 GHz) for multiple-input multiple-output (MIMO) applications in 5G smartphones is proposed. The proposed 8-antenna array is composed of two different antenna elements, namely, the folded L-shaped antenna and couple-fed U-shaped antenna. It can be applied in the 8 × 8 (by using 8 antenna elements) and 4 × 4 (by using 4 antenna elements) MIMO system across the LTE band 42 and 5.8-GHz WLAN band, respectively. The quasi-orthogonal polarization, balanced mode excitation and pattern diversity are properly used to achieve high isolation within the operation bands. A prototype of the proposed antenna array was fabricated and measured. The measured isolations in the LTE band 42 (6-dB impedance matching) and 5.8-GHz WLAN band (10-dB impedance matching) are better than 17.1 dB and 34.6 dB, respectively. The total efficiencies are 45%-62% and 52%-59% in the two operation bands, respectively. Besides, the measured envelope correlation coefficients (ECCs) are less than 0.045 and 0.0001, the calculated peak ergodic channel capacities are 37.9 bps/Hz and 19.3 bps/Hz in the 8 × 8 or 4 × 4 MIMO system across the operation bands. Furthermore, user's hand and head effects are investigated and desirable results are obtained. The above results show that this proposed antenna array is a good candidate for MIMO applications in smartphones.

In this paper, a K-band antenna with ±60° beam scanning is proposed for a mmWave (millimeter-Wave) radar sensor. Especially, we take advantages of the magnetic dipole principle, and design the antenna by employing via holes in three sides, so as to form a cavity which can be equivalently considered as magnetic dipole, resulting in the 3 dB beam width as wide as 140° in E-plane. We cascade the mmWave antenna to 1×4 and 4×4 arrays for radar transmitting and receiving, respectively. The antenna array is fabricated and implemented to a K-band radar and its wide-angle beam scanning performance is evaluated by the radar's distance and angular detections.

A 12-port antenna array operating in the long term evolution (LTE) band 42 (3400-3600 MHz), LTE band 43 (3600-3800 MHz), and LTE band 46 (5150-5925 MHz) for 5G massive multiple-input multiple-output (MIMO) applications in mobile handsets is presented. The proposed MIMO antenna is composed of three different antenna element types, namely, inverted π-shaped antenna, longer inverted L-shaped open slot antenna, and shorter inverted L-shaped open slot antenna. In total, eight antenna elements are used for the 8×8 MIMO in LTE bands 42/43, and six antenna elements are designed for the 6×6 MIMO in LTE band 46. The proposed antenna was simulated, and a prototype was fabricated and tested. The measured results show that the LTE bands 42/43/46 are satisfied with reflection coefficient better than -6 dB, isolation lower than -12 dB, and total efficiencies of higher than 40%. In addition to that, the proposed antenna array has also shown good MIMO performances with an envelope correlation coefficient lower than 0.15, and ergodic channel capacities higher than 34 and 26.5 b/s/Hz in the LTE bands 42/43 and LTE band 46, respectively. The hand phantom effects are also investigated, and the results show that the proposed antenna array can still exhibit good radiation and MIMO performances when operating under data mode and read mode conditions.

A multi-band 10-antenna array working at the sub-6-GHz spectrum (LTE bands 42/43 and LTE band 46) for massive multiple-input multiple-output (MIMO) applications in future 5G smartphones is proposed. To realize 10 × 10 MIMO applications in three LTE bands, 10 T-shaped coupled-fed slot antenna elements that can excite dual resonant modes are integrated into a system circuit board. Spatial and polarization diversity techniques are implemented on these elements so that the improved isolation and mitigated coupling effects can be achieved. The proposed antenna array was manufactured and experimentally measured. Desirable antenna efficiencies of higher than 42% and 62% were measured in the low band and high band, respectively. Vital results, such as the envelope correlation coefficient, channel capacity, and mean effective gain ratio, have also been computed and analyzed. The calculated ergodic channel capacities of the 10 × 10 MIMO system working in the LTE bands 42/43 and LTE band 46 reached up to 48 and 51.4 b/s/Hz, respectively.

The title of data scientist has been described as one of the sexiest jobs of the 21st century. Numerous efforts have been made to define the job of a data scientist in a qualitative manner by, for example, listing the job functions and required skill sets of data scientists. However, to the best of our knowledge, no attempt has been made to define the term data scientist in a scientific manner. In this paper, we address this issue by using a data-driven approach to answer three questions: (1) What is a proper definition of the term data scientist from a market-demand perspective? (2) Do self-described data scientists meet the market demand? and (3) Finally, how can companies efficiently recruit data scientists that match their openings? To answer these questions, we crawl two data sets for the supply and demand sides. For the former, we collect a set of data scientist user profiles from LinkedIn; for the latter, we collect a set of data scientist job descriptions from Monster. We first parse the set of data scientist job descriptions via natural language processing techniques and derive a scientific definition of the job of a data scientist via a clustering algorithm. Second, we use the same approach to determine that, under the aforementioned definition, self-claimed data scientists on the market would meet the market demand with a high probability. Finally, we introduce a distance-metric learning approach that can be used by companies to find data scientist candidates that match their openings. We achieve an average precision of 12.31%; i.e., one in ten candidates with matching qualifications would accept a given offer. The application of this quantitative approach could significantly reduce the human-resource costs incurred by companies in recruiting matching data scientists.

A practical method using single-shot ptychographic imaging (SPI) in the transform domain is proposed for multiple-image hiding. The excellent compressibility of SPI can bring high multiplexing capacity and imperceptibility to the proposed multiple-image hiding system. The security is mainly supported by the imperceptibility and the large quantity of scrambling modes, and security will be improved greatly with the increase of hidden images. Furthermore, the successful combination with the discrete cosine transform-domain algorithm can not only further enhance the imperceptibility and security but indicates that SPI-based multiple-image hiding is compatible with transform domain algorithms as well. Optical experiments demonstrate the high feasibility, multiplexing capacity, imperceptibility, and security of the proposed technique. The proposed method may open up a novel research perspective for multiple-image hiding combined with various efficient transform domain algorithms.