The rapid development of communication systems poses unprecedented challenges, e.g., handling exploding wireless signals in a real-time and fine-grained manner. Recent advances in data-driven machine learning algorithms, especially deep learning (DL), show great potential to address the challenges. However, waveforms in the physical layer may not be suitable for the prevalent classical DL models, such as convolution neural network (CNN) and recurrent neural network (RNN), which mainly accept formats of images, time series, and text data in the application layer. Therefore, it is of considerable interest to bridge the gap between signal waveforms to DL amenable data formats. In this paper, we develop a framework to transform complex-valued signal waveforms into images with statistical significance, termed contour stellar image (CSI), which can convey deep level statistical information from the raw wireless signal waveforms while being represented in an image data format. In this paper, we explore several potential application scenarios and present effective CSI-based solutions to address the signal recognition challenges. Our investigation validates that CSI is a promising method to bridge the gap between signal recognition and DL.

Deep learning (DL) has been recognized as an effective solution for automatic modulation cassification (AMC). However, most recent DL based AMC works are based on realvalued operations and representations. In this correspondence, we aim to demonstrate the high potential of complex-valuenetworks for AMC. We present the design of several key building blocks for complex-valued networks, such as complex convolution, complex batch-normalization, complex weight initialization, and complex dense strategies. We then provide a comparison study of three different neural network models and their complex-valued counterparts using the RadioML 2016.10A dataset. Our results validate the superior performance in AMC achieved by the complex-valued networks.

Automatic modulation classification (AMC) plays an important role in both civilian and military applications. Today, increasingly more researchers apply a deep learning (DL) framework in AMC. However, few papers consider that a typical deep model is difficult to deploy on resource constrained devices. In this paper, we propose a new filter-level pruning technique based on activation maximization (AM) that discards the less important convolution filter. Compared to other network pruning techniques, the convolution neural network (CNN) pruned via the AM method obtains equal or higher classification accuracy in the RadioML2016.10a dataset.

Deep learning (DL) models are vulnerable to adversarial attacks, by adding a subtle perturbation which is imperceptible to the human eye, a convolutional neural network (CNN) can lead to erroneous results, which greatly reduces the reliability and security of the DL tasks. Considering the wide application of modulation recognition in the communication field and the rapid development of DL, by adding a well-designed adversarial perturbation to the input signal, this article explores the performance of attack methods on modulation recognition, measures the effectiveness of adversarial attacks on signals, and provides the empirical evaluation of the reliabilities of CNNs. The results indicate that the accuracy of the target model reduce significantly by adversarial attacks, when the perturbation factor is 0.001, the accuracy of the model could drop by about 50% on average. Among them, iterative methods show greater attack performances than that of one-step method. In addition, the consistency of the waveform before and after the perturbation is examined, to consider whether the added adversarial examples are small enough (i.e., hard to distinguish by human eyes). This article also aims at inspiring researchers to further promote the CNNs reliabilities against adversarial attacks.

Better neural networks, more powerful computer hardware and signal Big Data make deep learning increasingly important in Specific Emitter Identification (SEI). However, its implementation uses large amounts of resources and releases CO2. With the requirement of green and low-carbon resource sustainability, deep learning-based SEI faces the great challenge of compressing the model while ensuring its performance. In this article, we propose a novel green and low resource specific emitter identification method using complex networks and Fisher pruning (GLR-SEI). Specifically, we divide the learning process into two stages: knowledge distillation and fisher pruning, which involves transferring knowledge from complex number networks to real number networks. The proposed approach is evaluated using aircraft Automatic Dependent Surveillance Broadcast (ADS-B) data collected in the real world and many pruning experiments are performed. In addition, we add descriptive experiments on Raspberry Pi to provide a more effective data presentation in order to better demonstrate our research results. The experimental results show that the proposed lightweight network in this article can reduce the recognition rate by only 0.7% and the inference time by 10.1% compared to the complex neural network.

It has been recognized that fixed spectrum and channel allocation will lead to waste of spectrum resources when multiple agents communicate at the same time. Dynamic allocation of channels is proposed to maximize the utilization of spectrum resources. In the environment of multiple unmanned aerial vehicles (UAVs), it is necessary to ensure that each UAV can communicate successfully without interfering with other UAVs. Dynamic allocation of channels plays an important role in such systems. In this paper, we propose a dynamic channel allocation scheme based on deep reinforcement learning for multi-UAV systems. A slotted time system is used by all the UAVs. Di2642erent from the traditional method, the occupancy of each channel is scanned first in each time slot. Then a channel will be selected for data transmission, with feedback from the environment when the transmission is over. The proposed channel allocation scheme incorporates a long short-term memory (LSTM) into the deep reinforcement learning framework, to better learn from the past experience and better adapt to the the highly dynamic environment in a multi-UAV system. The experimental results show that compared with the traditional reinforcement learning method (Q- learning and Deep Q Network (DQN)), the proposed method achieves faster convergence and better performance with respect to average collision rate, average reward, and average successful communication rate.