Unmanned aerial vehicles as base stations (UAV-BSs) are recognized as effective means for tackling eruptive communication service requirements especially when terrestrial infrastructures are unavailable. Quality of service (QoS) received by ground terminals (GTs) highly depends on the spatial movement of UAV-BSs. In this paper, we investigate the cooperative trajectory design problem of multiple UAV-BSs towards fair throughput maximization of GTs. Considering the restriction of coverage and sensing, we first propose a heterogeneous-graph-based formulation of relations between GTs and UAV-BSs. Subsequently, we design a framework named graph vision and communication (GVis&Comm) to 1) let each UAV-BS efficiently manage time-varying local observations; 2) facilitate cooperation between UAV-BSs through explicit information exchange. To further reduce the overhead of over-the-air cooperation, we realize discretization of the message passing process among UAV-BSs while still enabling end-to-end training. By leveraging multi-agent reinforcement learning (MARL), UAV-BSs as agents learn a distributed trajectory design policy. Extensive numerical simulation shows that our framework on the one hand achieves remarkable efficiency in processing local observations of each UAV-BS, and on the other improves the overall network performance via close cooperation among UAV-BSs.

Detecting ever increasing anomalous signals is critical to effective spectrum management. In this letter, we present a radio anomaly detection algorithm based on modified generative adversarial network (GAN). Firstly, short time fourier transform (STFT) is applied to obtain the spectrogram image from the received signal. Then, a novel encoder-GAN (E-GAN) structure is proposed by incorporating an encoder network into the original GAN to reconstruct the spectrogram. As a result, the existence of anomalies can be detected based on the reconstruction error and discriminator loss. In addition, the reconstruction error can also be exploited to locate the anomalies in time-frequency domain. Simulation results show that the proposed algorithm brings a performance improvement of up to 10 dB compared with the spectrum anomaly detector with interpretable features (SAIFE).

In wireless Ad Hoc networks, optimal routing is closely coupled with radio resource allocation. This paper adopts a cross-layer approach and improves the performance of multi-hop transmission by joint routing and power control. A hop-by-hop optimization framework based on deep reinforcement learning (DRL) is proposed for efficient and decentralized decisions on relay nodes and transmit power. For decision at each hop, graphs are used by the agent to flexibly characterize irregular topological and interference relations among neighboring nodes. Then, a novel graph neural network (GNN)-based model architecture is designed as the agent’s policy to exploit these graph relations and handle input/output of fluctuating dimensions. Particularly, information exchange among nodes during optimization strictly adheres to practical limitations on their sensing ability, fitting the distributed configuration of Ad Hoc networks. The learnt policy can quickly establish high-quality paths between sources and destinations, exhibiting strong generalization ability to random network topologies. Numerical results show that our proposed method outweighs both heuristic benchmarks and state-of-the-art DRL counterparts in terms of both throughput and path connection probability.

In this paper, the probability distribution of the peak to average power ratio (PAPR) is analyzed for the mixed numerologies transmission based on orthogonal frequency division multiplexing (OFDM). State of the art theoretical analysis implicitly assumes continuous and symmetric frequency spectrum of OFDM signals. Thus, it is difficult to be applied to the mixed-numerology system due to its complication. By comprehensively considering system parameters, including numerology, bandwidth and power level of each subband, we propose a generic analytical distribution function of PAPR for continuous-time signals based on level-crossing theory. The proposed approach can be applied to both conventional single numerology and mixed-numerology systems. In addition, it also ensures the validity for the noncontinuous-OFDM (NC-OFDM). Given the derived distribution expression, we further investigate the effect of power allocation between different numerologies on PAPR. Simulations are presented and show the good match of the proposed theoretical results.

Iterative clipping and filtering (ICF) is a straightforward method for reducing the peak-to-average power ratio (PAPR) of signals in orthogonal frequency-division multiplexing (OFDM) system. Recently, convex optimization has been used to find the optimal filter coefficients that minimize the error vector magnitude (EVM) and meet the PAPR constraint. However, high computation complexity may be incurred when solving convex optimization problem. Therefore, we develop an efficient PAPR reduction method that uses the time-domain kernel matrix to generate the PAPR-reduction signal. Besides, we relax the assumption that the clipping noise is a series of uncorrelated parabolic pulses and apply the proposed method to more general cases. Based on the instantaneous observation of clipping noise, the proposed method constructs a simple time-domain kernel matrix and employs the curve fitting approach to optimize the corresponding scaling factors. Simulation results show that the proposed method can achieve very close performance to that using convex optimization in terms of both the PAPR reduction and EVM while the computational cost is greatly reduced. In addition, due to the decrease of iteration numbers and computational complexity, the proposed method is more efficient than some existing clipping and filtering methods.