Search

In recent years, unmanned aerial vehicles (UAVs) have been widely used in ocean target tracking and image acquisition for processing. Due to the limited energy of the UAV and the high computational complexity associated with image processing tasks, a lightweight energy-saving target tracking scheme is designed for the UAV, and the unmanned surface vehicle (USV) based mobile edge computing (MEC) networks are adopted to share the computing load of the UAV. Due to the randomness of the environment, we formulate data processing, computation offloading, resource allocation, and target-tracking as a joint stochastic optimization problem. This paper investigates a two-stage optimization scheme to address the problem. First, we employ a Lyapunov-based approach to convert the stochastic optimization problem into a deterministic per-time slot problem under communication and computing resources constraints. Then, we develop a real-time target tracking scheme for the UAV based on the Elman neural network. Numerical results validate that the designed tracking scheme can effectively minimize propulsion energy consumption while maintaining a high success rate in tracking. Furthermore, the proposed method balances data-related energy consumption, image detection accuracy, and stability of the data storage queue.

In satellite communications, the power amplifier (PA) plays a vital role in enhancing the transmission performance of signals. However, the non-linearity of PAs often distorts the transmitted signals, degrading communication performance. Digital pre-distortion (DPD) is one of the most effective linearization technologies to address the abovementioned issue and improve PA performance. Neural networks (NNs) outperform traditional models in PA behavioral modeling and linearization due to their strong nonlinear fitting capability, but the lack of learning data distributions limits the performance. Hence, this article proposes a novel behavioral and linearization model for PA, utilizing the residual generative adversarial network (Res-GAN). The proposed model can extract deep features of the PA with enhanced accuracy, thereby improving PA linearization performance. Simulation results show that the proposed Res-GAN behavioral model achieves the normalized mean squared error (NMSE) of −55.737 dB. Additionally, the Res-GAN DPD outperforms traditional DPD models by a reduction in the NMSE of about 2 dB and the adjacent channel power ratio (ACPR) of about 2-4 dBc.

The distributed subgradient (DSG) method is a widely used algorithm for coping with large-scale distributed optimization problems in machine-learning applications. Most existing works on DSG focus on ideal communication between cooperative agents, where the shared information between agents is exact and perfect. This assumption, however, can lead to potential privacy concerns and is not feasible when wireless transmission links are of poor quality. To meet this challenge, a common approach is to quantize the data locally before transmission, which avoids exposure of raw data and significantly reduces the size of the data. Compared with perfect data, quantization poses fundamental challenges to maintaining data accuracy, which further impacts the convergence of the algorithms. To overcome this problem, we propose a DSG method with random quantization and flexible weights and provide comprehensive results on the convergence of the algorithm for (strongly/weakly) convex objective functions. We also derive the upper bounds on the convergence rates in terms of the quantization error, the distortion, the step sizes, and the number of network agents. Our analysis extends the existing results, for which special cases of step sizes and convex objective functions are considered, to general conclusions on weakly convex cases. Numerical simulations are conducted in convex and weakly convex settings to support our theoretical results.

Intelligent reflecting surfaces (IRS) have been proposed in millimeter wave (mmWave) and terahertz (THz) systems to achieve both coverage and capacity enhancement, where the design of hybrid precoders, combiners, and the IRS typically relies on channel state information. In this paper, we address the problem of uplink wideband channel estimation for IRS aided multiuser multiple-input single-output (MISO) systems with hybrid architectures. Combining the structure of model driven and data driven deep learning approaches, a hybrid driven learning architecture is devised for joint estimation and learning the properties of the channels. For a passive IRS aided system, we propose a residual learned approximate message passing as a model driven network. A denoising and attention network in the data driven network is used to jointly learn spatial and frequency features. Furthermore, we design a flexible hybrid driven network in a hybrid passive and active IRS aided system. Specifically, the depthwise separable convolution is applied to the data driven network, leading to less network complexity and fewer parameters at the IRS side. Numerical results indicate that in both systems, the proposed hybrid driven channel estimation methods significantly outperform existing deep learning-based schemes and effectively reduce the pilot overhead by about 60% in IRS aided systems.

Edge artificial intelligence (AI) has been a promising solution towards 6G to empower a series of advanced techniques such as digital twins, holographic projection, semantic communications, and auto-driving, for achieving intelligence of everything. The performance of edge AI tasks, including edge learning and edge AI inference, depends on the quality of three highly coupled processes, i.e., sensing for data acquisition, computation for information extraction, and communication for information transmission. However, these three modules need to compete for network resources for enhancing their own quality-of-services. To this end, integrated sensing-communication-computation (ISCC) is of paramount significance for improving resource utilization as well as achieving the customized goals of edge AI tasks. By investigating the interplay among the three modules, this article presents various kinds of ISCC schemes for federated edge learning tasks and edge AI inference tasks in both application and physical layers.

Benefiting from the ability to process and integrate data from various modalities, multi-modal foundation models (FMs) facilitate potential applications across a range of fields, including computer vision (CV), natural language processing (NLP), and diverse multi-modal applications such as imagetext retrieval. Currently, FMs are deployed on computing clusters for training and inference to meet their considerable computational demands. In the foreseeable future, the parameter size of FMs is expected to evolve further, posing challenges to both computation resources and energy supply. Fortunately, leveraging the next-generation wireless networks (6G) to aggregate substantial computation resources and multi-modal data from myriad wireless devices holds promise for handling the aforementioned challenges. In this work, we delve into state-of-the-art artificial intelligence (AI) techniques, specifically focusing on pipeline parallelism, data parallelism, and multi-modal learning, with the aim of supporting the sustainable development of distributed multi-modal FMs in the 6G era. In the context of pipeline parallelism, compressing activations and gradients while intelligently allocating communication resources can overcome communication bottlenecks caused by unstable wireless links. For data parallelism, federated learning (FL) with over-the-air computation (AirComp) seamlessly integrates communication and computation, significantly expediting gradient aggregation. Furthermore, by following the recent success of large language models (LLMs) and incorporating multi-modal learning into FMs, we can seamlessly integrate NLP and CV, along with the broader AI community, establishing the cornerstone for the intrinsic AI within 6G wireless networks.

Metaverse is envisioned to merge the actual world with a virtual world to bring users unprecedented immersive feelings. To ensure user experience, federated learning (FL) has been expected as a critical enabler to provide metaverse users with high-quality sensing, communicating, and rendering. However, considering the limitation of wireless communication resources and the stringent requirements of users, collaborating with massive metaverse users to realize FL still has tremendous challenges. Most pioneer works on improving the performance of FL assume that the system states are static, which is unsuitable in the metaverse. Because the FL in the metaverse is always a complicated long-term iteration process, where the fluctuations of channel status and available computing resources of users are inevitable, a changeless strategy may lead to poor results. Therefore, this paper proposes an efficient FL scheme relying on dynamic user selection, gradient quantization, and resource allocation. Specifically, we derive the convergence error bound to reveal the impact of user selection, wireless transmission error, and gradient quantization error of each iteration on FL’s convergence. Based on the theoretical analysis, we jointly and dynamically optimize the user selection, gradient quantization, and resource allocation to minimize the error bound with time and energy consumption budgets. Furthermore, to make the formulated sequential decision-making problem tractable, we transform it into a Markov decision process and design a soft actor-critic-based solution. Extensive experiments validate that our proposed scheme has superior performance compared to conventional schemes in dynamic-changing network environments.

Theincreasing number of smart vehicles is leading to an increasing scarcity of spectrum resources for the internet of vehicles (IoV), which has given rise to an urgent requirement for automatic modulation classification (AMC) in cognitive radio (CR) systems. Meanwhile, for the flexibility of unmanned aerial vehicles (UAVs), the AMC implemented based on UAVs is considered an effective method to achieve reliable communication between intelligent vehicles. However, for distributed UAV task implementation, real-time radio data needs to be transmitted between UAVs and a cloud server. This process requires maintaining a high-capacity, secure channel environment, which is difficult to accomplish. In this article, we propose a federated transfer learning framework to implement AMC in a distributed scenario, which avoids radio data transmission in each UAV. To reduce data dependence, the pre-trained deep learning (DL)-based model is sent to each UAV node and performs transfer learning, which brings more focused learning of the channel environment in which various UAVs are located. The simulation results show that federated transfer learning-based AMC offers better recognition accuracy than centralized approach. Compared to the centralized training methods, the federated transfer learning algorithm achieves an improvement of 1.04% to 12.05% in classification accuracy for each node with less training data. Besides, the effect of different fine-tuning layers on the accuracy is investigated, showing that fine-tuning three layers could achieve optimal accuracy. Additionally, different numbers of UAVs are employed to verify the impact on the results. The experimental results show that the number of UAVs can improve the results but to a limited extent. Furthermore, we evaluate the proposed method by various measurements, such as accuracy, precision, and F1-score. Accordingly, compared with the baseline methods, the proposed scheme achieves an improvement of 1% to 14% over them.