Spatial Channel Information Research Articles

A Spatial-Channel-Temporal-Fused Attention for Spiking Neural Networks.

Spiking neural networks (SNNs) mimic brain computational strategies, and exhibit substantial capabilities in spatiotemporal information processing. As an essential factor for human perception, visual attention refers to the dynamic process for selecting salient regions in biological vision systems. Although visual attention mechanisms have achieved great success in computer vision applications, they are rarely introduced into SNNs. Inspired by experimental observations on predictive attentional remapping, we propose a new spatial-channel-temporal-fused attention (SCTFA) module that can guide SNNs to efficiently capture underlying target regions by utilizing accumulated historical spatial-channel information in the present study. Through a systematic evaluation on three event stream datasets (DVS Gesture, SL-Animals-DVS, and MNIST-DVS), we demonstrate that the SNN with the SCTFA module (SCTFA-SNN) not only significantly outperforms the baseline SNN (BL-SNN) and two other SNN models with degenerated attention modules, but also achieves competitive accuracy with the existing state-of-the-art (SOTA) methods. Additionally, our detailed analysis shows that the proposed SCTFA-SNN model has strong robustness to noise and outstanding stability when faced with incomplete data, while maintaining acceptable complexity and efficiency. Overall, these findings indicate that incorporating appropriate cognitive mechanisms of the brain may provide a promising approach to elevate the capabilities of SNNs.

Deep blind super-resolution for hyperspectral images

Current deep learning methods for single hyperspectral image super-resolution are non-blind ones, which adopt the simplistic bicubic degradation model. These models have poor generalization performance and cannot handle unknown degradation. Additionally, blind super-resolution methods for RGB images neglect rich spectral information in hyperspectral images, which causes spectral distortion in results. To address these issues, we consider the degeneration estimation and propose a single hyperspectral image blind super-resolution algorithm. Specifically, we first use a blur kernel estimation network and a deblurring network to obtain the image without blur. We change the receptive field by exchanging spatial channel information when estimating the kernel, thereby obtaining blurry information at different scales. In addition, to reduce the errors introduced by kernel mismatch and the relevant information among bands ignored by the group convolution, we fuse the spatial information of the blurred image to compensate for the errors and then utilize the spatial and spectral information of other bands to guide the feature extraction of the current band. Finally, we integrate the feature information from different branches in a global fusion network to improve spatial and spectral information fidelity further. Extensive experiments on synthetic and real-world hyperspectral datasets unequivocally demonstrate that our approach outperforms other state-of-the-art approaches in terms of both quantitative and qualitative evaluations. Our code is publicly available at https://github.com/YoungP2001/DBSR.

Towards a Novel Generative Adversarial Network-Based Framework for Remote Sensing Image Demosaicking

During satellite remote sensing imaging, the use of Bayer mode sensors holds significant importance in saving airborne computing resources and reducing the burden of satellite transmission systems. The demosaicing techniques play a key role in this process. The integration of Generative Adversarial Networks (GANs) has garnered significant interest in the realm of image demosaicking, owing to their ability to generate intricate details. However, when demosaicing mosaic images in remote sensing techniques, GANs, although capable of generating rich details, often introduce unpleasant artifacts while generating content. To address this challenge and differentiate between undesirable artifacts and realistic details, we have devised a novel framework based on a Progressive Discrimination Strategy within a Generative Adversarial Network architecture for image demosaicking. Our approach incorporates an artifact-weighted Location Map refinement technique to guide the optimization process towards generating authentic details in a stable and precise manner. Furthermore, our framework integrates a global attention mechanism to boost the interaction of spatial-channel information across different dimensions, thereby enhancing the performance of the generator network. Moreover, we conduct a comparative analysis of various prevalent attention mechanisms in the context of remote sensing image demosaicking. The experimental findings unequivocally demonstrate that our proposed methodology not only achieves superior reconstruction accuracy on the dataset but also enhances the perceptual quality of the generated images. By effectively mitigating artifacts and emphasizing the generation of true details, our approach represents a significant advancement in the field of remote sensing image demosaicking, promising enhanced visual fidelity and realism in reconstructed images.

Out-of-Band Information Aided mmWave/THz Beam Search: A Spatial Channel Similarity Perspective

The transition to higher frequency bands, e.g., millimeter-wave (mmWave) and terahertz (THz), will be capitalized on the long term for future wireless communications. One of challenges relates to rapid establishment of mmWave/THz links with low beam training overhead due to highly directional transmission. A promising solution is to take advantage of the coexistence of sub-6 GHz, mmWave, and THz wireless networks and to use out-of-band spatial information for enabling fast beam search. The success depends on the spatial similarity of radio channels across different frequency bands. In this article we promote a feasibility study of low-frequency spatial channel information assisted high-frequency beam search from a radio channel point of view. We develop multi-band channel similarity measure of desired beam directions extracted from radio channels, which are obtained via filtering propagation paths by different beampatterns at different frequencies. Measurementand ray-tracing-based evaluations across multiple frequencies and environments are performed, which prove the usability of out-of-band information aided beam search strategy in line-ofsight (LOS) dominated scenario and even in non-LOS scenario. Finally, we discuss the challenges associated with exploiting spatial channel similarity.

On the Feasibility of Out-of-Band Spatial Channel Information for Millimeter-Wave Beam Search

Prolonged beam alignment is the main source of overhead in mobile wireless communications at millimeter-wave (mm-wave) frequencies due to narrow beams following the requirement of high antenna gains. Out-of-band spatial information may be used in initial beam search when lower frequency band radios are operating in conjunction with mm-wave radios. The feasibility of using low-band channel information for coarse estimation of high-band beam directions strongly depends on the spatial congruence between the two frequency bands. In this work, we try to answer two related questions. First, how similar is the power angular spectrum (PAS) of propagation channels between two widely separated frequency bands? Then, what is the impact of practical antenna configurations on spatial channel similarity? We propose a beam directions based metric to assess the power loss and number of false directions if out-of-band spatial information is used instead of in-band information. This metric is more practical and useful than comparing the PASs directly. Point cloud ray tracing and propagation measurement results across multiple frequency bands and environments are used to show that the degree of spatial similarity of beamformed channels is related to antenna beam widths, frequency gap, and radio link conditions.

Multi-attention guided and feature enhancement network for vehicle re-identification

Vehicle Re-Identification (Re-ID) aims to discover and match target vehicles in different cameras of road surveillance. The high similarity between vehicle appearances and the dramatic variations in viewpoints and illumination cause great challenges for vehicle Re-ID. Meanwhile, in safety supervision and intelligent traffic systems, one needs a quick efficient method of identifying target vehicles. In this paper, we propose a Multi-Attention Guided Feature Enhancement Network (MAFEN) to extract robust vehicle appearance features. Specifically, the Fusing Spatial-Channel information multi-receptive fields Feature Enhancement module (FSCFE) is first proposed to aggregate richer and more representative multi-receptive fields features at different receptive fields sizes. It also learned the spatial structure information and channel dependencies of the multi-receptive fields features and embedded them to enhance the feature. Then, we construct the Spatial Attention-Guided Adaptive Feature Erasure (SAAFE) module, which uses spatial attention to erase the most distinguishing features. The network’s attention is shifted to potentially salient features to strengthen the ability of the network to extract salient features. In addition, a multi-loss knowledge distillation (MLKD) method using MAFEN as a teacher network is designed to improve computational efficiency. It uses multiple loss functions to jointly supervise the student network. Experimental results on three public datasets demonstrate the merits of the proposed method over the state-of-the-art methods.

Multiscale Attention Feature Aggregation Network for Cloud and Cloud Shadow Segmentation

Cloud and cloud shadow segmentation is one of the most important problems in remote sensing image processing. Due to the vulnerability to ground object interference, noise interference and other factors, and the lack of generalization ability, traditional deep learning network would inevitably lose details and spatial information, resulting in imprecise segmentation of cloud and cloud shadow boundaries, missing detection and false detection. In order to solve these problems, a multi-scale attention feature aggregation network is proposed, which extracts semantic information at different levels based on residual network. A multi-scale strip pooling attention module is designed to extract multi-scale context information and deep spatial channel information. In order to improve the feature extraction of the model, the deep multi-head feedforward transfer attention module is introduced to enable the adjacent two layers of the backbone network to guide each other for feature mining. Then, a bilateral feature fusion module is used to guide the fusion of low-level semantic information and high-level detail information. Finally, in order to enhance the repair of boundary details of cloud and cloud shadow, a boundary refinement boosting model is designed. The experimental results show that our model can handle complex cloud cover scenes, and has excellent performance on cloud, cloud shadow dataset and CSWV dataset. Its segmentation accuracy is superior to the existing methods, which is of great significance to the related work of cloud detection.

Recurrent Network Knowledge Distillation for Image Rain Removal

Single-image rain removal (SIRR) based on deep learning has long been a problem of great interest in low-level vision systems. However, traditional convolutional neural network (CNN)-based approaches fail to capture long-range location dependencies effectively and may cause the image background blurred. In this article, we propose a knowledge distilling deraining network (KDRN) to address the SIRR problem. In the proposed network, the teacher regards rain streaks as a linear combination of many residual networks. It is used for image reconstruction at different resolutions. With the aid of a teacher network, the proposed deraining network performs better. A spatial channel aggregation residual attention block (SCARAB) is designed to remove the rain streaks. The block not only concentrates on the rain streak features but also captures the spatial-channel information of the image. For the network structure, we used an end-to-end approach to design the teacher and student networks separately. The proposed KDRN obtains the predicted residual image by a combination of the stage-wise results and the original input image. Extensive experiments show that the proposed KDRN obtains better subjective quality than most of the compared methods, on both heavy and light rain data sets.

Efficient Folded Attention for Medical Image Reconstruction and Segmentation

Recently, 3D medical image reconstruction (MIR) and segmentation (MIS) based on deep neural networks have been developed with promising results, and attention mechanism has been further designed for performance enhancement. However, the large size of 3D volume images poses a great computational challenge to traditional attention methods. In this paper, we propose a folded attention (FA) approach to improve the computational efficiency of traditional attention methods on 3D medical images. The main idea is that we apply tensor folding and unfolding operations to construct four small sub-affinity matrices to approximate the original affinity matrix. Through four consecutive sub-attention modules of FA, each element in the feature tensor can aggregate spatial-channel information from all other elements. Compared to traditional attention methods, with the moderate improvement of accuracy, FA can substantially reduce the computational complexity and GPU memory consumption. We demonstrate the superiority of our method on two challenging tasks for 3D MIR and MIS, which are quantitative susceptibility mapping and multiple sclerosis lesion segmentation.

CANet: A Channel Attention Network to Determine Informative Multi-channel for Image Classification from Brain Signals.

Changes in brain state that depend on various visual image stimulations have been investigated recently; however, it is difficult to decode visual image information from brain signal information. Recently, deep learning techniques have been applied to classify brain signals in various experiments, such as motor imagery and steady state visual evoked potential. However, although the deep learning model seems powerful, it is understood poorly, and thus, can be considered a black box. Accordingly, when multi-channel brain signals are trained, which channels include important information is not understood clearly. In this paper, we proposed a channel attention network (CANet) and investigated the way the deep learning network may determine which channels contain more important information that represents brainwaves' characteristics and the way it may visualize that information. Using such spatial channel information, we found that our proposed deep learning architecture outperforms basic approaches (spatial channel information is not considered) to classifying categorized images from visual evoked magnetoencephalographic (MEG) brain signals.

Compressive Initial Access and Beamforming Training for Millimeter-Wave Cellular Systems

Initial access (IA) is a fundamental physical layer procedure in cellular systems where user equipment (UE) detects nearby base station (BS) as well as acquire synchronization. Due to the necessity of using antenna array in millimeter-wave (mmW) IA, the channel spatial information can also be inferred. The state-of-the-art directional IA (DIA) uses sector sounding beams with limited angular resolution, and thus requires additional dedicated radio resources, access latency and overhead for refined beam training. To remedy the problem of access latency and overhead in DIA, this work proposes to use a quasi-omni pseudorandom sounding beam for IA, and develops a novel algorithm for joint initial access and fine resolution initial beam training without requiring extra radio resources. We provide the analysis of the proposed algorithm miss detection rate under synchronization error, and further derive Cram\'er-Rao lower bound of angular estimation under frequency offset. Using QuaDRiGa simulator with mmMAGIC model at 28 GHz, the numerical results show that the proposed approach is advantageous to DIA with hierarchical beam training. The proposed algorithm offers up to two order of magnitude access latency saving compared to DIA, when the same discovery, post training SNR, and overhead performance are targeted. This conclusion holds true in various propagation environments and 3D locations of a mmW pico-cell with up to 140m radius.

Three-dimensional optical observations of an upward lightning triggered by positive cloud-to-ground lightning

Based on dual-station optical records, electric field and radar data, a detailed analysis of an instance of upward lightning initiated from Canton Tower, in the city of Guangzhou, Guangdong Province, China, is presented. The upward lightning was triggered by a positive cloud-to-ground (PCG) flash nearby. The upward lightning channel and part of the horizontal PCG lightning channel obtained by dual-station optical systems were reconstructed in three dimensions, providing us with spatial channel information on how the PCG lightning process triggered the upward lightning from Canton Tower. The upward lightning occurred in the trailing stratiform cloud area behind a squall line system. The radar echo intensity in this area was relatively weak (30–40 dBZ), and the height of the strongest radar echo center was about 4 km. From the optical records, the two-dimensional length of the horizontal PCG lightning channel was longer than 17.2 km. The length of the three-dimensional reconstructed part of the PCG lightning channel was about 3660 m. Its height increased from 3410 m to 4170 m along the development direction, with an average height of 3640 m. Based on the above result, as well as previous studies on the charge structure of the stratiform region, we infer that the height of the positive charge layer was roughly consistent with the height of the strongest echo center.

Performance of space-division multiple-access system using preprocessing based on feedback of vector-quantized channel spatial information

In this paper, we report the performance of multi-user transmitter preprocessing assisted multiple-input multiple-output downlink (DL) communication, when the channel impulse responses (CIRs) required to formulate the preprocessing matrix are fed back to the base station (BS) via feedback channels that endure noise and fading. Specifically, in our work the CIRs are estimated at the mobile stations (MSs) and are then decomposed by singular value decomposition. The magnitudes and phases associated with the right-hand side unitary matrix of each of the MSs are then vector-quantized (VQ) and conveyed to the BS as channel spatial information. This spatial information is then used to conceive the preprocessing matrix. Our study shows that the symbol-error-rate and capacity achieved with the ideal feedback scenario is close to that attained with perfect spatial information assumption. Hence, we advocate that VQ-spatial information based preprocessing comprises an efficient technique to deal with the DL multi-user interference.

Stochastic Modeling of THz Indoor Radio Channels

Providing the basis for fast system simulations and the adequate design of upcoming THz communication systems, a novel stochastic 300 GHz indoor channel model is introduced. It combines both the modeling in time as well as in frequency domain in order to account for the significant frequency dispersion of ultra broadband THz channels. Not only amplitude, phase and temporal, but also spatial channel information is considered. That way, MIMO systems as well as novel antenna concepts can be simulated. Verified and calibrated frequency domain ray tracing simulations in an office scenario provide the data basis for the derivation of model parameters. Model channel realizations are tested against ray tracing predictions and channel measurements. A complete scenario-specific parameter set is given for the considered environment, so that the model can be implemented for further use and future THz communication links can be designed under consideration of realistic propagation conditions.

Space–Time Precoding for Downlink Transmission in Multiple Antenna CDMA Systems

This paper introduces a linear space-time precoding scheme for downlink transmission in a multiple-transmit single-receive antenna code-division multiple-access (CDMA) system. The proposed scheme relies on using space-time spreading sequences that are constructed from the time-domain user signatures combined with spatial channel information that corresponds to multiple transmit antennas. Numerical results show that the proposed scheme significantly reduces the multiple-access interference in both flat- and multipath-fading channels. At the same time, it achieves the maximum possible diversity order that is offered by the combination of multiple transmit antennas and multipath fading. To further improve the performance in overloaded CDMA systems, the proposed linear precoding method is combined with a nonlinear operation known as vector perturbation (based on applying modulo operation to minimize the transmit energy). Throughout the paper, we assume that 1) the channel state information is known at the transmitter, 2) the modulation is binary phase-shift keying, and 3) the receivers are simply based on linear filters that are matched to the time-domain user signatures. We use a recent work of Reynolds et al. as a baseline for comparison. Numerical results show that our proposed method outperforms that of Reynolds et al. in both flat and multipath channels.

Efficient feedback methods for MIMO channels based on parameterization

In this paper, we propose two efficient low-complexity quantization methods for multiple-input multiple-output (MIMO) systems with finite-rate feedback based on proper parameterization of the information to be fed back followed by quantization in the new parameter domain. For a MIMO channel which has multiple orthonormal vectors as channel spatial information, we exploit the geometrical structure of orthonormality while quantizing the spatial information matrix. The parameterization is of two types: one is in terms of a set of unit-norm vectors with different lengths, and the other is in terms of a minimal number of scalar parameters. These parameters are shown to be independent for the i.i.d. flat-fading Rayleigh channel, facilitating efficient quantization. In the first scheme, each of the unit-norm vectors is independently quantized with a finite number of bits using an optimal vector quantization (VQ) technique. Bit allocation is needed between the vectors, and the optimum bit allocation depends on the operating SNR of the system. In the second scheme, the scalar parameters are quantized. In slowly time-varying channels, the scalar parameters are also found to be smoothly changing over time, leading to the development of a simple quantization and feedback method using adaptive delta modulation. The results show that the proposed feedback scheme has a channel tracking feature and achieves a capacity very close to perfect feedback with a reasonable feedback rate

Channel Feedback Quantization for High Data Rate MIMO Systems

In this work, we study a multiple-input multiple-output (MIMO) wireless system where the channel state information is partially available at the transmitter through a feedback link. Based on singular value decomposition, the MIMO channel is split into independent sub-channels. Effective feedback of the required spatial channel information entails efficient quantization/encoding of a unitary matrix. We propose two schemes for quantizing unitary matrices via Givens rotations and examine the performance for a scenario where the rates allocated to the sub-channels are selected according to their corresponding gains. Numerical results show that the proposed schemes offer a significant performance improvement as compared to that of MIMO systems without feedback, with a negligible increase in the complexity