The 802.11n networks with MIMO technique provide a spatial degree of freedom for dealing with co-channel interference. In this letter, our proposed spatial interference coordination scheme is achieved by distributed precoding for the downlink and distributed multi-user detection for the uplink. Simulation results validate the proposed scheme in terms of the downlink and uplink maximum achievable rates at each AP.

Non-Terrestrial-Network (NTN) can provide seamless and ubiquitous connectivity of massive devices. Thus, the feeder links between satellites and gateways need to provide essentially high data transmission rates. In this paper, we focus on a typical high-capacity scenario, i.e., LEO-IoT, to find an optimal satellite selection schema to maximize the capacity of feeder links. The proposed schema is able to obtain the optimal mapping among all the satellites and gateways. By comparing with maximum service time algorithm, the proposed schema can construct a more balanced and reasonable connection pattern to improve the efficiency of the gateways. Such an advantage will become more significant as the number of satellites increases.

In this letter, we propose a novel and effective haze removal method by using the structure-aware atmospheric veil. More specifically, the initial atmospheric veil is first estimated based on dark channel prior and morphological operator. Furthermore, an energy optimization function considering the structure feature of the input image is constructed to refine the initial atmospheric veil. At last, the haze-free image can be restored by inverting the atmospheric scattering model. Additionally, brightness adjustment is also performed for preventing the dehazing result too dark. Experimental results on hazy images reveal that the proposed method can effectively remove the haze and yield dehazing results with vivid color and high scene visibility.

Infrared and visible image fusion can combine the thermal radiation information and the textures to provide a high-quality fused image. In this letter, we propose a hybrid variational fusion model to achieve this end. Specifically, an ℓ0 term is adopted to preserve the highlighted targets with salient gradient variation in the infrared image, an ℓ1 term is used to suppress the noise in the fused image and an ℓ2 term is employed to keep the textures of the visible image. Experimental results demonstrate the superiority of the proposed variational model and our results have more sharpen textures with less noise.

LSTM network have shown to outperform in facial expression recognition of video sequence. In view of limited representation ability of single-layer LSTM, a hierarchical attention model with enhanced feature branch is proposed. This new network architecture consists of traditional VGG-16-FACE with enhanced feature branch followed by a cross-layer LSTM. The VGG-16-FACE with enhanced branch extracts the spatial features as well as the cross-layer LSTM extracts the temporal relations between different frames in the video. The proposed method is evaluated on the public emotion databases in subject-independent and cross-database tasks and outperforms state-of-the-art methods.

Deep neural networks have achieved great success in visual tracking by learning a generic representation and leveraging large amounts of training data to improve performance. Most generic object trackers are trained from scratch online and do not benefit from a large number of videos available for offline training. We present a real-time generic object tracker capable of incorporating temporal information into its model, learning from many examples offline and quickly updating online. During the training process, the pre-trained weight of convolution layer is updated lagging behind, and the input video sequence length is gradually increased for fast convergence. Furthermore, only the hidden states in recurrent network are updated to guarantee the real-time tracking speed. The experimental results show that the proposed tracking method is capable of tracking objects at 150 fps with higher predicting overlap rate, and achieves more robustness in multiple benchmarks than state-of-the-art performance.

This paper proposes a coarse-fine Time-to-Digital Converter (TDC), based on a Ring-Tapped Delay Line (RTDL). The TDC achieves the picosecond's level timing resolution and microsecond's level dynamic range at low cost. The TDC is composed of two coarse time measurement blocks, a time residue generator, and a fine time measurement block. In the coarse blocks, RTDL is constructed by redesigning the conventional Tapped Delay Line (TDL) in a ring structure. A 12-bit counter is employed in one of the two coarse blocks to count the cycle times of the signal traveling in the RTDL. In this way, the input range is increased up to 20.3µs without use of an external reference clock. Besides, the setup time of soft-edged D-flip-flops (SDFFs) adopted in RTDL is set to zero. The adjustable time residue generator picks up the time residue of the coarse block and propagates the residue to the fine block. In the fine block, we use a Vernier Ring Oscillator (VRO) with MOS capacitors to achieve a scalable timing resolution of 11.8ps (1 LSB). Experimental results show that the measured characteristic curve has high-level linearity; the measured DNL and INL are within ± 0.6 LSB and ± 1.5 LSB, respectively. When stimulated by constant interval input, the standard deviation of the system is below 0.35 LSB. The dead time of the proposed TDC is less than 650ps. When operating at 5 MSPS at 3.3V power supply, the power consumption of the chip is 21.5mW. Owing to the use of RTDL and VRO structures, the chip core area is only 0.35mm × 0.28mm in a 0.35µm CMOS process.

To implement the parallel acceleration of convolution operation of Convolutional Neural Networks (CNNs) on field programmable gate array (FPGA), large quantities of the logic resources will be consumed, expecially DSP cores. Many previous researches fail to make a well balance between DSP and LUT6. For better resource efficiency, a typical convolution structure is implemented with LUT6s in this paper. Besides, a novel convolution structure is proposed to further reduce the LUT6 resource consumption by modifying the typical convolution structure. The equations to evaluate the LUT6 resource consumptions of both structures are presented and validated. The theoretical evaluation and experimental results show that the novel structure can save 3.5-8% of LUT6s compared with the typical structure.

The conventional algorithms in the echo canceling system have drawback when they are faced with double-talk condition in noisy environment. Since the double-talk and noise signal are exist, then the error signal is contaminated to estimate the gradient correctly. In this paper, we define a new class of adaptive algorithm for tap adaptations, based on the correlation function processing. The computer simulation results show that the Correlation LMS (CLMS) and the Extended CLMS (ECLMS) algorithms have better performance than conventional LMS algorithm. In order to implement the ECLMS algorithm, the Frequency domain Extended CLMS (FECLMS) algorithm is proposed to reduce the computational complexity. However the convergence speed is not sufficient. In order to improve the convergence speed, the Wavelet domain Extended CLMS (WECLMS) algorithm is proposed. The computer simulation results support the theoretical findings and verify the robustness of the proposed WECLMS algorithm in the double-talk situation.

We combine the siamese network and the recurrent regression network, proposing a two-stage tracking framework termed as SiamReg. Our method solves the problem that the classic siamese network can not judge the target size precisely and simplifies the procedures of regression in the training and testing process. We perform experiments on three challenging tracking datasets: VOT2016, OTB100, and VOT2018. The results indicate that, after offline trained, SiamReg can obtain a higher expected average overlap measure.