The application of deep neural networks to address automatic modulation recognition (AMR) challenges has gained increasing popularity. Despite the outstanding capability of deep learning in automatic feature extraction, predictions based on low-data regimes with imbalanced classes of modulation signals generally result in low accuracy due to an insufficient number of training examples, which hinders the wide adoption of deep learning methods in practical applications of AMR. The identification of the minority class of samples can be crucial, as they tend to be of higher value. However, in AMR tasks, there is a lack of attention and effective solutions to the problem of Imbalanced-class in a low-data regime. In this work, we present a practical automatic data augmentation method for radio signals, called SigAugment, which incorporates eight individual transformations and effectively improves the performance of AMR tasks without additional searches. It surpasses existing data augmentation methods and mainstream methods for solving low-data and imbalanced-class problems on multiple datasets. By simply embedding SigAugment into the training pipeline of an existing model, it can achieve state-of-the-art performance on benchmark datasets and dramatically improve the classification accuracy of minority classes in the low-data imbalanced-class regime. SigAugment can be trained for uniform use on different types of models and datasets and works right out of the box.

In recent years, pedestrian dead reckon (PDR) is considered a popular localization solution for its reliable accuracy. To tackle its intrinsic problem of accumulative errors, various methods are proposed to refine the trajectory by indoor space information to promote localization accuracy. However, these methods usually require the initial motion state, such as starting location and absolute orientation to infer the coordinates. They may be limited because the initial location has to be determined by other highly precise sensors and required people holding the phone ahead in the fixed position to ensure the x-axis is consistent with the heading. In this paper, we release the constraints by utilizing the gyroscope and accelerometer sensor sequence to estimate the orientation change and trajectory length, which are insensitive to the initial motion state, to build a HMM model to infer the initial motion and refine the trajectory by spatial information. Firstly, the indoor environment is divided into grids to construct a directed map. Then, we use the gyroscope and accelerometer to collect the signal sequence to represent the rotation angle and length information of the trajectory. Finally, the Viterbi algorithm is used for the inference of the whole trajectory including the initial location and orientation. The experiments show that the initial location and absolute orientation can be predicted as the convergence of the matching candidates when the length of the trajectory increases. And the localization error of the proposed method is superior to the compared method with the benefit of rotation angle and trajectory length.

Identifying the modulation type of radio signals is challenging in both military and civilian applications such as radio monitoring and spectrum allocation. This has become more difficult as the number of signal types increases and the channel environment becomes more complex. Deep learning-based automatic modulation classification (AMC) methods have recently achieved state-of-the-art performance with massive amounts of data. However, existing models struggle to achieve the required level of accuracy, guarantee real-time performance at edge devices, and achieve higher classification performance on high-performance computing platforms when deployed on various platforms. In this paper, we present a family of AMC models based on communication domain knowledge for various computing platforms. The higher-order statistical properties of signals, customized data augmentation methods, and narrowband convolution kernels are the domain knowledge that is specifically employed to the AMC task and neural network backbone. We used separable convolution and depth-wise convolution with very few residual connections to create our lightweight model, which has only 4.61k parameters while maintaining accuracy. On the four different platforms, the classification accuracy and inference time outperformed those of the existing lightweight models. Meanwhile, we use the squeeze-and-excitation attention mechanism, channel shuffle module, and expert feature parallel branch to improve the classification accuracy. On the three most frequently used benchmark datasets, the high-accuracy models achieved state-of-the-art average accuracies of 64.63%, 67.22%, and 65.03%, respectively. Furthermore, we propose a generic framework for evaluating the complexity of deep learning models and use it to comprehensively assess the complexity strengths of the proposed models.

As online social networks play a more and more important role in public opinion, the large-scale simulation of social networks has been focused on by many scientists from sociology, communication, informatics, and so on. It is a good way to study real information diffusion in a symmetrical simulation world by agent-based modeling and simulation (ABMS), which is considered an effective solution by scholars from computational sociology. However, on the one hand, classical ABMS tools such as NetLogo cannot support the simulation of more than thousands of agents. On the other hand, big data platforms such as Hadoop and Spark used to study big datasets do not provide optimization for the simulation of large-scale social networks. A two-tier partition algorithm for the optimization of large-scale simulation of social networks is proposed in this paper. First, the simulation kernel of ABMS for information diffusion is implemented based on the Spark platform. Both the data structure and the scheduling mechanism are implemented by Resilient Distributed Data (RDD) to simulate the millions of agents. Second, a two-tier partition algorithm is implemented by community detection and graph cut. Community detection is used to find the partition of high interactions in the social network. A graph cut is used to achieve the goal of load balance. Finally, with the support of the dataset recorded from Twitter, a series of experiments are used to testify the performance of the two-tier partition algorithm in both the communication cost and load balance.

Distributed heterogeneous multiagent systems (DHMASs) link geographically dispersed agents through networks, harnessing information technology to foster collaboration. Considering the mainstream status of wireless communication in modern multiagent systems and the differences in the performance of interagent communication devices, we believe that it is appropriate to use directed heterogeneous networks (DHNs) to model distributed heterogeneous multiagent systems. This model not only reflects the directionality of interagent communication but also reflects the complexity of communication due to performance differences, thus providing a more accurate framework for understanding and optimizing system behavior. The study of disintegration in DHNs is vital for enhancing the decision-making agility of DHMAS. We introduce Directed heterogeneous network Attack sequence inference through graph attention MAtrix Generation Embedding and reinforcement learning (DAMAGE), an algorithm that integrates graph neural networks and reinforcement learning within an inductive reasoning framework. DAMAGE is designed to optimize the generation of disintegration strategies, improving the efficiency of network breakdown processes. Our approach includes a directed network embedding technique with a graph attention matrix generation module, which enhances the utilization of imperfect network structure information. Through ablation studies, we demonstrate that DAMAGE not only increases the effectiveness of network disintegration under perfect topological conditions but also maintains robustness in scenario with imperfect topological information.