This paper proposes a probabilistic framework to infer link availability in Low Earth Orbit (LEO) satellite networks from weak but complementary features such as SNR, capacity, and duration, by first mapping these features with adaptive sigmoid functions, and then integrating them using a Product of Experts (PoE) model. A negative log transformation enables lightweight shortest-path routing with additive weights. Theoretical analysis shows temporal stability, while simulations demonstrate consistent improvements: flow drop rates are reduced by up to $80 \%$ over the naive baseline and $30-40 \%$ over strong heuristics, with enhanced throughput under heavy traffic loads.

Federated learning, as a privacy-preserving distributed learning paradigm, faces three major challenges in practical applications: optimization difficulties caused by non-independent and identically distributed (Non-IID) data, distribution shifts introduced by new clients joining in dynamic environments, and privacy threats such as model inversion attacks. Existing methods often address individual challenges at the expense of performance in other aspects, with the conflict between privacy protection and model performance being particularly prominent. This paper proposes a privacy-preserving federated learning framework for dynamic Non-IID data based on differential privacy and adaptive knowledge distillation (DP-ADKD-FL), which addresses multiple challenges through three innovative mechanisms: (1) a sensitivity-based parameter-layered differential privacy mechanism that provides differentiated protection for parameters of varying importance, minimizing the impact of noise on performance; (2) an adaptive knowledge distillation framework that dynamically adjusts knowledge transfer strategies based on client distribution characteristics, effectively responding to distribution changes; and (3) a dynamic privacy budget regulation strategy that optimizes privacy resource allocation according to training stages and the degree of distribution changes. Experiments on real-world datasets demonstrate that DP-ADKD-FL not only significantly outperforms existing methods in privacy protection metrics such as defense against inversion attacks, but also surpasses baseline methods without privacy protection in model performance, especially showing excellent adaptability in dynamic environments.

This article proposes a resource allocation framework for LEO satellite-HAP communication scenario. To achieve maximum resource satisfaction with minimal resource cost and ensure resource allocation fairness, a multi-objective resource allocation model is established that combines minimizing the total resource allocation with maximizing user node resource satisfaction, subject to resource constraints. To address this problem, a GA-based strategy is introduced to manage available resource. This strategy incorporates a demand response priority mechanism for access network users and designs resource allocation ratios to ensure fair distribution of computational, transmission, and time resource. To address the local optimum trap in the GA, the DATLO is introduced to improve convergence speed and solution quality. Simulation results verify the convergence of the proposed algorithm, and a proof is provided based on the fixed point theorem. Furthermore, a fairness index is used to verify the effectiveness of the proposed priority-driven GA-DATLO for different user scales, different user demand distributions, and different resource allocation types. This work provides a systematic approach to balancing efficiency and fairness in dynamic SIN environments.

In the context of 6G networks, efficient information transmission at high speeds with large data volumes to the cloud poses technical challenges, particularly related to data type heterogeneity and information security. This is especially critical when dealing with extensive data from diverse users, leading to issues in data utilization efficiency and security. To tackle these challenges, an innovative approach has been developed, employing the following data processing methods:(1)Users utilize the Mobile Edge Computing server to train the initial model, sign and upload it to the Blockchain. Due to constraints in data block size, off-chain storage is implemented using InterPlanetary File System. Additionally, differential privacy technology safeguards user privacy during this process;(2)Addressing the vast multi-source heterogeneous data in the 6G network, the Federated Transfer Learning mode is employed. This involves combining cross-user characteristics in different business formats after the data processing center acquires user data. Blockchain replaces the centralized aggregator in the federated learning system, enhancing the security of data transmission. By utilizing actual datasets from sources such as sandstorm, typhoon, and flood data, simulation results demonstrate that the proposed model significantly enhances data utilization efficiency, improves the security of data transmission, predicts disasters, and minimizes disaster losses.

Adding intelligent decision module with artificial intelligence algorithm as the core in satellite communication process is a powerful measure to improve the overall anti-interference capability of communication system. This paper focuses on the intelligent decision making of communication waveform under different interference environments. By introducing the perceived environmental information and waveform parameters into the designed intelligent decision module, the decision information is obtained based on module feedback and the waveform parameters are switched, and finally the information is transmitted reliably, thus improving the anti-interference performance of the communication system. At the same time, the feasibility of the algorithm and module is proved by a series of simulations.

Movable antenna (MA) arrays beamforming can achieve full array gain while suppressing interference with specific structure, whereas fixed-position antenna (FPA) arrays beamforming has to consider a trade-off between these objectives. However, non-half wavelength antenna spacing can result in a high peak sidelobe level (PSLL) and delayed response due to longer movement times. In this letter, we aim to minimize PSLL through joint optimization of antenna position vector (APV) and antenna weight vector (AWV), subject to full array gain and null steering constraints. We reveal that it is essentially equal to minimize the array size, allowing us to transform the formulated non-convex problem into several equivalent problems. Specifically, we propose an approximation method for PSLL and then formulate an integer linear programming problem. Numerical results demonstrate that the proposed algorithm significantly minimizes the array size and achieves full array gain with decent PSLL compared to other benchmark schemes.

Multi-network integration represents the future trend in measurement, control and communication. To provide optimal service, it is significant to study the network topology strategy of heterogeneous nodes. However, the mobility and node damage and failures caused by interference pose a huge challenge in reconfiguring the topology to ensure network performance. To address this issue, this paper proposes a topology reconfiguration algorithm based on minimum connectivity for a cross-domain multi-system measurement and control network, which integrates air, ground, and sea. This algorithm designs an optimal topology reconfiguration strategy considering connectivity, energy consumption, and position accuracy, and constructs a reliable measurement and control system. Meanwhile, by implementing the improved simulated annealing optimization, a robust measurement and control network topology can be obtained. Theoretical analysis and simulation results demonstrate that this algorithm can effectively resist external interference and ensure positioning accuracy, proved to be feasible and effective compared to the Global K-connected Energy-aware Topology-control Algorithm (GKETA) and greedy algorithm.

Satellite Internet technology is currently booming. Due to the high cost of satellites and the limited number of orbits, a satellite must cover a large area and serve more users than before. Given that the random access process is the initial step in establishing communication between satellites and terrestrial users, congestion control has emerged as a focal point of research. In this paper, we address the random access congestion control problem in satellite Internet scenarios and introduce the soft actor-critic (SAC) method to dynamically set the access class barring (ACB) factors. The throughput of random access is optimized to allow a single satellite to support a greater number of users without experiencing congestion. The SAC-ACB method is verified by simulation to converge faster and perform better than the proximal policy optimization (PPO) method.

In this paper, we propose a novel network architecture named FracNet, which aims to address the problem of OTFS fractional Doppler channel estimation based on a superimposed pilot structure. Inspired by the idea of image super-resolution, we treat the channel matrix that only accounts for the integer-grid Doppler as a low-resolution image, and the channel matrix that incorporates fractional Doppler as a high-resolution image. By learning the mapping function between the two matrices, we seek to recover the effects caused by fractional Doppler shifts. The received pilots in the OTFS domain help estimate the channel parameters, which are then used in the MMSE algorithm for signal demodulation. Simulation results show that, under the same low-power spectral conditions, the proposed method outperforms conventional methods in terms of estimation error, and enhances the spectral efficiency of channel estimation.

In this letter, we propose a novel method to improve the localization accuracy for an array based satellite system, where we aim at minimizing the Cramér–Rao lower bound (CRLB) of the position by adjusting the attitude of the satellite. The Lie group method is adopted to provide an unconstrained representation of the rotation operation, which avoids using the conventional orthogonal matrix representation. Then the problem can be solved by gradient based methods. The proposed method has lower complexity than the traditional manifold method. Numerical simulations verify the effectiveness of our proposed approach.

In this paper, we investigate the problem of tracking the angles of arrival (AoAs) of multiple sources in millimeter wave (mmWave) systems with a limited number of radio frequency (RF) chains. Considering the time-varying nature of the channel, we propose a deep neural network (DNN)-based active learning scheme for adaptive analog beamforming and multi-source AoA tracking. The proposed scheme consists of a DNN-based beamformer and a subspace tracking-based multiple signal classification (MUSIC) estimator. Specifically, the DNN generates the beamformer using soft AoA estimates from the previous time block, and the MUSIC estimator exploits the measured signal by the beamformer to estimate the AoAs in the current time block. The proposed scheme is first applied to the uniform linear array (ULA) scenario, and then extended to the uniform rectangular array (URA) scenario. Particularly, in the URA scenario, to reduce the computational complexity, we modify the reduced-dimension MUSIC (RD-MUSIC) algorithm to a beam-space form. Furthermore, we adopt a partially connected analog beamforming scheme for the large-scale URA scenario to further reduce the hardware costs. We conduct numerical experiments to evaluate the tracking performance of the proposed scheme in the ULA and URA scenarios, and show that the proposed scheme significantly outperforms the existing codebook-based beamformer methods.

Automatic modulation recognition is a critical step between signal detection and signal demodulation, and it is a critical technology for ensuring proper communication. Since the introduction of 5G technology, wireless communication systems have had massive data throughput, so combining deep learning technology with modulation recognition technology is now one of the most mainstream development directions in the field of communication. In order to effectively improve the robust recognition accuracy of modulation signals in low SNR condition, this paper proposes an end-to-end AMR model based on deep learning, called residual convolution long short memory Improved Transformer-Encoder deep neural network model (RLITNN). First, convolutional network and LSTM network extract the initial features from the communication signals of different modes. Second, the proposed Improved Transformer-Encoder module is used to capture the global and local focus of the extracted features. Then, these features are fused. Finally, the high quality features are further captured from the fused feature vector to enhance the feature representation ability of the model. Experiments on RML2016.10A and RML2016.10B datasets show that the proposed RLITNN has better feature learning ability and better recognition accuracy than other advanced (SOTA) techniques.

Contemporary frequency hopping systems, generating frequency hopping sequences based on chaotic sequences is widely used. In this paper, we study the way of tracking jamming based on GRU networks in non-cooperative chaotic map frequency hopping systems. The selection of the frequency hopping sequence is crucial for frequency hopping systems. While most frequency hopping sequences generated using chaotic map exhibit strong correlation performance and randomness, after conducting the verification, it can be discovered that the accuracy of small sliding window prediction using GRU networks is astonishingly high. Thus, it is imperative that chaotic frequency hopping systems adopt a more sophisticated generation approach to effectively combat tracking jamming based on neural networks. To meet this need, a new chaotic map is proposed in this paper, which combines IST map and Cat map. This paper evaluates the performance of the frequency hopping sequences generated by this approach using a range of metrics, including Hamming correlation, Lyapunov exponent, sequence complexity, uniformity, and anti-prediction test based on the GRU network. The simulation results demonstrate that the frequency hopping sequence generated using this method exhibits superior uniformity, wider interval property, and better anti-prediction performance than traditional chaotic frequency hopping sequences.

In recent times, as a result of the emergence of high-throughput satellites, there is an increasing need for high-speed and high-quality communication. The resources available on these satellites are valuable, and the rational allocation of these resources has become the focal point of research. This paper establishes a resource allocation model for OFDM multi-beam high-throughput satellites that takes into account inter-beam interference, inter-sub carrier interference, and satellite downlink loss. The objective of this model is to ensure fairness among satellite users, and this is achieved through the use of quantum genetic algorithms for resource allocation. Compared with the fixed allocation method, our algorithm has better performance.

Knowledge graphs have a wide range of applications in areas such as intelligent question and answer, personalized recommendation and intelligent decision making. Knowledge graph representation learning aims to address the sparsity and incompleteness of entities and relationships in knowledge graphs. Traditional representation learning methods based on translation operations such as TransE and TransH usually only consider the triadic information of the knowledge graph to learn the representation in isolation, ignoring the rich textual information, type information, etc., resulting in not mining the semantics carried by the entities in the knowledge graph itself. To address the shortcomings of the existing methods, this paper proposes a representation learning method that fuses entity description and type, introducing entity description and type information on the basis of the SRotatE model, aiming to learn triadic knowledge while fusing triadic factual information with entity description and type information, which can better express the semantics between entities and relationships, thus improving the performance of the knowledge graph representation learning model performance of the knowledge graph representation learning model. Experimental results on real datasets show that the proposed method outperforms mainstream models such as TransE, ComplEx and RotatE.

In adaptive array beamforming, when the interference position changes rapidly, the antenna receiving platform vibrates and moves, or the update speed of adaptive weights is slow, the interference may move out of the null position and cannot be effectively cancelled, and then the conventional method may be completely invalid. Therefore, in this paper, we propose an interference null widening algorithm based on the U-V domain expansion of the steering vector of interference source. The algorithm solves the problem of coupling the pitch angle and the azimuth angle of the interference signal by adding the adjacent interference, introducing the covariance matrix taper method, and mapping the expansion of angle domain to the expansion of independent U-V domain. The simulation results show that the algorithm has a deep depth of null in the range of null widening, and can effectively improve the anti-interference ability of the system in dynamic environment.

Compression-sensing-based DOA estimation method can overcome the disadvantages of the traditional spatial spectral estimation algorithm, but the calculation task is heavy and time-consuming, which is limited by the power and processing capability on satellite. At the same time, with the change of the space environment and the aging of the devices, the satellite array error problem will appear. Therefore, we proposed a joint subspace decomposition and compressed sensing approximation algorithm to solve the problem of DOA estimation. We first use the Fast Root-MUSIC algorithm to correct the satellite array error and perform a preliminary DOA estimation to reduce the search range of the compressed sensing algorithm, then performed an accurate estimation of the DOA by the L1-svd algorithm. Simulation results show that the proposed algorithm has better performance and less computation complexity under low signal to noise ratio, small fast beat number, as well as in the presence of array error.

Downlink and uplink decoupling access (DUDA) has recently been shown to significantly improve uplink transmission performance in wireless networks, such as in terrestrial and unmanned aerial vehicle (UAV) networks. In this paper, we propose a novel DUDA scheme for non-geostationary orbit (NGEO) heterogeneous satellite networks (HSN) with the avoidance of in-line interference for spectral coexistence of geostationary Earth orbit (GEO) and NGEO satellites. Concretely, the in-line interference caused by the high main lobe gain of the directional antenna in NGEO satellites explicitly described by utilizing geocentric angle-based spherical surface model. For both uplink and downlink transmission of the proposed DUDA scheme, each NGEO user equipment (UE) is allowed to access the NGEO satellite with the largest average received signal power under the different NGEO satellites exclusion angles constraint. The performance is evaluated in terms of transmission rate. Simulation results show notable enhancement of DUDA compared with counterpart access schemes.

An important characteristic of a low earth orbit (LEO) satellite communication system is a large number of satellites. Therefore, it is common that the number of receiving antennas is less than that of transmitting antennas. In this scenario, to save resources and improve spectral efficiency, an underdetermined blind separation algorithm can be considered. In this paper, under the underdetermined condition, the number of source signals is greater than that of the mathematical constraint equation, and the prior knowledge or characteristics of the source signal are usually used to add constraints. In order to obtain better signal processing performance, and combined with the information transmission characteristics of LEO satellite communication system, a greedy optimization algorithm can be executed. In order to improve the algorithm performance, during algorithm iteration, the sampling points with the highest cost performance are selected and added to the sample set at each time. The simulation results show that the algorithm has good performance.

This paper investigates the physical layer security (PLS) problems of satellite-terrestrial links for LEO satellite communication systems, ultimately aiming to improve LEO satellite communications’ security. A scheme derived from the handoff process of LEO satellite communication system has been proposed to adapt special characteristics of LEO satellite channel, especially the time-varying and block fading in Land Mobile Satellite (LMS) channel. Through a series of simulations, the superiority of the proposed scheme compared with the traditional model could be illustrated.

Spoofing jamming with satisfactory concealment and strong aggression can effectively mislead the operation of the non-cooperative system. This is a kind of effective jamming signal. However, it has costly construction and limited application since a lot of prior information is required. In this paper, a new high-performance generation method for spoofing jamming signals is proposed. We used a one-dimensional generative adversarial network (GAN) to learn the latent distribution of the DS/SS signals and construct a generated signal set. Then, we choose the signals in this set with highly aggressive for the DSSS system as spoofing jamming signals. Simulation results show that this method can more easily generate spoofing signals with strong aggressiveness to the DS/SS system, and effectively solve the problem that it is difficult to construct spoofing jamming under the condition of limited prior information. Our proposed method provides a new idea for the construction of spoofing jamming signals, which is helpful for its application in the field of communication countermeasures.

As the potential candidate of next generation communication networks, rate splitting multiple access (RSMA) has drawn great attentions. Usually to maximize the sum rate of RSMA, weighted minimum mean square error (WMMSE) algorithm is often used. But its computational complexity hinders its practical application. In this paper, we apply a deep learning network called Deep Unfolding (DU) to RSMA in Satellite-Terrestrial Integrated Network (STIN) with the intention of boosting downlink capacity. The momentum accelerated projection gradient descent (PGD) algorithm is adopted to substitute the complex operations and speed up the computations. By selecting the momentum and step size as the trainable parameters, the simulation results indicate that the deep learning approach outperforms the original WMMSE algorithm in sum rate and speed.

In this paper, we develop a novel visual-inertial navigation system with motion constraint (VINS-Motion), which extends the visual-inertial navigation system (VINS) to incorporate vehicle motion constraints for improving the autonomous vehicles localization accuracy. Besides the prior information, IMU measurement residual, and visual measurement residual utilized in VINS, vehicle orientation/velocity constraint is first exploited to constitute motion residual. We minimize the sum of priors and Mahalanobis norms of three kinds of residuals to obtain a maximum posteriori estimation, thus increasing system consistency and accuracy. Stop detection is also added to help eliminate the abnormal jitter of the estimated poses during stopping, thus ensuring reasonability of the trajectory. The pro-posed approach is validated on public datasets and compared against state-of-the-art algorithms, which demonstrates that VINS-Motion achieves significantly higher positioning accuracy.

The air-space-ground integrated network architecture is one of the core directions of 6G. Its network architecture is based on the ground cellular mobile network, combined with the characteristics of wide coverage, flexible deployment, and efficient broadcasting of broadband satellite communications, and through the deep integration of a variety of heterogeneous networks to achieve the full coverage of sea, land and air, which will bring new opportunities for the marine, airborne, transnational, space and ground applications. However, the satellite network is a highly complex heterogeneous multi-layer system, how to realize the capacity management among multi-layer satellites is still a challenging. In this paper, we first investigate the problem of capacity management in the three-layer heterogeneous satellite network, using Quality of Experience (QoE) as the optimization target for resource allocation. Then, we built a complex network model architecture of mobile users, base stations, low-orbit satellites, and high-orbit satellites. Based on the genetic algorithm to optimize capacity allocation, we evaluate and compare the proposed algorithm with traditional optimization algorithms in terms of performance and algorithm complexity.

Dirty data is ubiquitous in real-world, and data cleaning is a long-standing problem. The importance of data cleaning is growing in the era of big data. In this paper we propose a novel data repairing approach by leveraging statistical relational learning (SRL). We learn Bayesian networks of attributes from the dirty data, then transform the dependency relationships among attributes into first-order logic formulas. We calculate the weight of each formula based on the mutual information of the attributes involved in the formula and obtain Markov logic network (often abbreviated as MLN) by assigning weight to each first-order logic formula. Then we transform Markov logic networks into inference rules and conduct these inference rules on DeepDive. The inference results are utilized to repair dirty data at last. Experiments on real-world datasets demonstrate that our approach has higher accuracy in terms of different situations and is universal for different kinds of datasets.

Due to satellite physical constraints in terms of available power and area, data processing capacity is low, storage and security are limited. It is a challenge to protect satellite network from illegal information access and use storage space effectively. In this paper, a blockchain technology-based on authentication and privacy protection scheme is proposed for a satellite communication network. to this aim, an architecture consisting of both conventional and restricted devices connected to the blockchain via a wireless heterogeneous network is deployed. The communication is carried out through registration, authentication and revocation. In this scheme, the satellite will forward the collected information to the ground base station, which will record all key parameters on the distributed blockchain and all malicious node certificates will be cleared from the blockchain by the ground base station. The simulation results show that the scheme has been greatly improved in terms of communication security and communication overhead.

This paper proposes an adaptive direction of arrival (DOA) estimation method, which satisfies the dynamic signal to noise ratio (SNR) and low complexity requirement of space communications. Integrating with subspace class algorithms, this paper uses information of covariance matrix corresponding to the array received signal as feature input, adding instance selection while minimizing distribution distance of source domain and target domain, whose samples go through different SNRs and snapshots. Error of the angle estimation is employed as cost function to construct the regression model. The simulation results show that after using the training set of a specific SNR and snapshot number to get the estimation model, good angle estimation performance can be observed under a certain range of SNR and the snapshot number, which demonstrates the adaptability of the proposed algorithm.

The openness of satellite communication channel makes satellite signal easy to be intercepted and decoded by non-partners, so it’s necessary to adopt secure communication technologies in satellite communication systems. Based on the existing hybrid spread-frequency technology, this paper introduces chaotic sequence to control system frequency hopping and code-hopping, puts forward a new type of safe communication method combined with the characteristics of chaotic sequence. This method improves the anti-interception performance of satellite communication system and has a good application prospect.

This paper discusses the issue of multi-beam forming for ground station (GS) in LEO satellite communication system. A store-load phased array structure is proposed, which doesn’t require complicated real-time computation. Meanwhile, for generating the pre-stored array pattern, a synthesis approach of array pattern for multi-beam is proposed. Based on a reference pattern designed by using Chebyshev method, the second-order cone programming (SOCP) constraint is applied to obtain optimized array patterns. Numerical simulations have been conducted in our unified platform containing Matlab and SeDuMi tool. The results show that the proposed approach can effectively synthesize array pattern with two beams and keep constant beamwidth at different directions and frequencies.

In satellite network, when the data is transmitted to the nearest base station on the ground with satellite cross-links, the attack appears to come from one satellite, but it's actually from multiple satellites. Especially in non-cooperative communication, the separation of received signals is a difficult task. In this article, solve the problem with blind source separation method is feasible. The contributions in this paper is as follows, firstly, signal sparsity is completed by computing the Short Time Fourier Transform (STFT), secondly, we separate the mixed source signals by improved PSO clustering. At last, we verify the proposed method by several simulations. The experimental results demonstrate the effectiveness of the proposed method.

In this paper, we propose a novel secure communication method for terrestrial-satellite system called hybrid-power communication. Hybrid-power communication, which belongs to physical layer security technology, not only eliminates the need for spread spectrum technology, but also is compatible with spread spectrum technology. Compared with artificial noise, hybrid-power communication method ensures the security of confidential user more efficiently, utilizing non-confidential users whose communication has no need to be encrypted. By allocating different power to the signals in accordance with hybrid-power factors, secure communication is achieved. Hybrid-power communication is able to improve the security of confidential user at a very low cost without affecting the communication performance of non-confidential user. In the last place, the simulation verifies our conclusion that compared with other available technologies, with the communication performance guaranteed, hybrid-power technology can effectively improve the security of the physical layer with the assistance of non-confidential user, while only costs tolerable resource loss and hardware requirements.

De Bruijn sequences have many applications in communication system, coding theory, and cryptography due to their attractive characteristics, such as long period, being balanced, large linear complexity, and so forth. For any given nonsingular feedback shift registers, its cycle structure is key point for the construction of de Bruijn sequences by using the cycle joining algorithm, which is the most popular method to construct de Bruijn sequences. In this paper, a new approach for analyzing the cycle structure of a class of linear feedback shift registers is proposed, and we come to a conclusion how many cycles each of these linear feedback shift registers can produce. In addition, one example is presented to demonstrate the algorithm.

In this paper, we consider a simple and effective blind source separation (BSS) method based on geometric segmentation in satellite communications. In the geometric segmentation, we analyze the correlation between any two vectors with canonical correlation analysis. There are two contributions in this paper. Firstly, a three-dimensional geometrical hybrid model is proposed based on the source signal's sparseness. Secondly, a novel geometrical hybrid model of feature extraction is proposed according to the idea of canonical correlation analysis. At last, we discuss the performance of the proposed method and verify the novel method based on several simulations. The experimental results demonstrate the effectiveness of the proposed method.

In satellite communication systems, it is necessary to implement fast and effective interference suppression for unknown interference signals to ensure the safety and reliability of satellite communication, and successful recognition of the types of interference is a prerequisite for ensuring high-efficiency anti-interference. In this paper, the incremental learning method is introduced into the process of satellite interference signals recognition to lower the hardware conditions required by the machine learning algorithm, reduce the memory consumption and shorten the computing time. Firstly, a variety of frequency domain features are extracted from five kinds of interference signals that often appear in satellite communication systems as classification feature parameters. Secondly, the incremental support vector machine learning model proposed by Gert Cauwenberghs et al. is extended to a multiclass model by one-versus-one, decision tree and directed acyclic graph respectively. Then the three improved models are used to train the characteristic parameters of the interference signals to obtain the recognition results. Simulation experiments show that compared with the conventional multi-category support vector machine model, the proposed models have little difference in recognition accuracy, but the computing time and the memory occupation of the training models are greatly reduced. These results are consistent with the actual requirements of satellite communication systems.

Recently, hybrid Satellite-Terrestrial (S-T) communication system, especially satellite communications integration with 5G/6G, is regarded as future development trend. Considering physical layer access waveform design, this paper proposed a novel Generalized Frequency Division Multiplexing- Code Division Multiple Access (GFDM-CDMA) scheme for hybrid S-T communication system. GFDM has many advantages such as good adaptability and low out-of-band (OOB) radiation. However, because of its interference between adjacent subcarriers, the multiple access performance is degraded. In this paper, we introduced (CDMA) technology in GFDM. The odd and even subcarriers of GFDM are spread spectrum respectively. Moreover, this scheme integrates the slot-ALOHA protocol and GFDM-CDMA, which extends access freedom in frequency, time and code domain. The simulation and analysis results show that the proposed GFDM-CDMA scheme is effective and suitable for hybrid S-T communication system.

As the demand of communications continuously increasing, researches of satellite communication systems have received more and more attention. Among the studies, one of the important problems is access and handover, due to the high-speed movement of satellites and the mobility of the terminals. Meanwhile, the channel allocation strategy is a crucial part of the technology. Therefore, this paper studies the channel reservation strategy, thinking of how to reserve channels reasonably. Finally, a dynamic channel reservation strategy is proposed based on grey prediction to get better access and handover performance. This strategy dynamically adjusts the channel reservation number with the grey model prediction by predicting whether the calls need to handover or not, and then count all the calls that are possible to handoff.

As neural network became one of the indispensable technologies in many fields, a large number of data processing methods emerged one after another. Convolutional neural network (CNN) soon became a useful tool to classify images. However, the drawback of CNN is quite obviously, which is the training time would be tremendous long if the network running on CPU. To tackle this issue, this paper improves the BP algorithm based on the weight freezing method and entropy weight data processing method, and uses the experiment to verify the effectiveness of this approach. With MNIST handwritten database, CIFAR-10 database and PUBFIG83 and LFW database as background, the original LeNet-5 model has been optimized and the back propagation (BP) algorithm used in the original network training is improved. The results show that the improved algorithm can effectively reduce the complexity of the algorithm, and the training time reduces to 8%.

Taking into account physical layer waveform design for integrated Satellite-Terrestrial (S-T) communication systems, this paper proposes a new interference-free generalized frequency division multiplexing (GFDM) transmission scheme. Many advantages such as low peak- average-power-ratio (PAPR), low out-of-band (OOB) emission and good adaptability indicate GFDM is suitable for S-T communications. However, due to intrinsic inter-carrier interference (ICI), conventional GFDM receivers have poor performance and high complexity. In this paper, ICI is firstly evaluated in frequency domain, and corresponding anti- interference matrix is derived and added on the sending side. By introducing active inverse interference, low- complexity match filtering (MF) receiver can also get rid of intrinsic ICI completely. The simulation and analysis prove that the proposed transmission scheme is effective and easy to practice. In the additional white Gaussian noise (AWGN) channel and typical satellite multipath channel, the proposed scheme has better symbol error ratio (SER) performance than that of conventional GFDM systems significantly.

The problem of underdetermined blind source separation for satellite communications is proposed in this paper. In underdetermined blind separation, people suppose the source is sparse and the number of source signals is known when they estimate the mixture matrix. In fact, the sparsity is often not satisfied and the number of source signals is unknown. This paper presents a novel Rough Set algorithm (RS algorithm) based on rough set theory, which can get the source signal sparse points and accurately estimate the number of sources and the mixture matrix respectively, by which source signals can be reconstructed. The last simulations show the good performance of the paper's algorithm.

In a massive Low Earth Orbit(LEO) satellite network without inter-satellite links(ISLs), connection between source node and destination node can be established through satellite-ground links. Utilizing satellite network topology in time virtualization algorithm and assuming that the current traffic of satellite-ground links which is used as the value of the topology correlation matrix can be perceived, the network uses the Dijkstra algorithm to obtain the shortest path of the current total network traffic and thus implements load balancing of the constellation system. Due to the large number of satellites, the network topology is complex and time varying, it is difficult to find the optimal path in a short time, so a suboptimal routing algorithm based on network topology is proposed in this paper. The local network topology is obtained by using the positions of the source and destination gateway, and the number of satellites and gateways used to establish the correlation matrix of the Dijkstra algorithm is reduced. Through the simulation analysis, the average computational complexity of this algorithm is much lower than that of the global optimal routing algorithm, and the average propagation delay of routing is almost the same. When the network load is normal, the system performance is almost the same with the global optimal routing algorithm.

Aiming to calculate the system capacity of a given satellite constellation, this paper proposes a time-varying traffic model based on varying traffic with varying time-zone and focuses on a MEO satellite constellation with global coverage. Using a load balanced routing algorithm, this paper analyzes the system capacity with time slicing method. The time-varying traffic model makes analysis more reliable and effective by comparison with equal traffic model.

In order to meet the critical 5G requirements both in terms of global connectivity and large throughput, satellite communications provide a valuable solution to complement and extend terrestrial mobile communications. Considering physical layer transmission waveform integration in hybrid Satellite-Terrestrial (S-T) communication systems, generalized frequency division multiplexing (GFDM) technology is introduced to apply in satellite communications. As potential candidate waveform of 5G, GFDM has many advantages which is appropriate for S-T communications but the traditional GFDM receivers have high complexity due to its intrinsic inter-carriers-interference (ICI). In this paper, a new data-aided interference cancellation (IC) GFDM receiver is proposed for satellite communications. Based on pilot subcarriers and new match filter, the proposed scheme can solve the receiving problem of raised cosine (RC) filtering GFDM. The simulation results and the complexity analysis prove that the proposed GFDM receiver performed significantly better than traditional GFDM receiver at reasonable computational cost, which can promote the application of GFDM in hybrid S-T communication systems.

This paper takes advantage of the mutual statistical independence of subcarriers' signals of OFDM through the application of independent component analysis (ICA) for strengthening the signal recovery capabilities. The ICA, also known as blind source separation (BSS), can increase spectral efficiency and anti-noise performance of OFDM system compared to pilot-assisted method. First we build the ICA model for OFDM system. Then the blind source recovery is implemented via the ICA based natural gradient algorithm. Numerical studies indicate that the proposed ICA-aided receivers enhance the conventional detection capabilities of OFDM equalization method relying on channel identification. This consists with the facts that ICA is able to utilize the independence of the original signals, and that ICA does not explicitly depend on the erroneous channel estimation.