Federated learning (FL) and reconfigurable intelligent surfaces (RIS) are pivotal technologies for future Internet of Things (IoT) networks, enhancing user privacy and system efficiency. However, realizing their full potential necessitates a cohesive and synergistic integration, challenging the traditional view of them as disparate components. This paper tackles the complex problem of maximizing energy efficiency (EE)—a critical yet under-explored metric insuch tightly coupled FL-RIS systems. We address this gap by formulating ajoint optimization problem that intrinsically links the FL process with physical layer resource allocation. Our framework maximizes the system’s global EE by concurrently designing the base station’s active beamforming and the RIS’s passive phase shifts,with an FL aggregation mechanism that is explicitly channel-aware and adaptive to the RIS-optimized wireless environment. This co-design ensures RIS actively facilitates FL by establishing robust communication, while FL intelligently leverages these improved channels for efficient and accelerated learning, all under practical FL performance constraints. Simulation results demonstrate that our proposed framework significantly enhances system energy efficiency compared to several benchmark schemes and exhibits robust convergence properties.

This study focuses on the latest research advancements in the field of semantic communication. Traditional communication systems prioritize the transmission of raw data, whilst semantic communication emphasizes conveying the meaning represented by the data. However, the extracted semantic information is often ambiguous and subject to subjective evaluation. To address this problem, this study proposes a model that combines a convolutional neural network (CNN) with a Transformer, called DeepSC‐CT. The model utilizes a CNN to extract semantic information from the data, followed by a Transformer model to capture spatial relationships and contextual information within the semantic content. We utilize federated learning to train the model and propose an adaptive aggregation algorithm to accelerate the convergence process. Moreover, we expand the single‐modality semantic communication model to encompass multiple modalities, such as texts, audio, and images. Furthermore, this study introduces a learnable position‐encoding method for the Transformer. The experimental results and visual effects of audio and image restoration demonstrate that the proposed method exhibits impressive performance and that the proposed model shows robust data restoration capabilities under various signal‐to‐noise ratio conditions.

The Internet of Things (IoT) allows physical devices to be connected over the wireless networks. Although device-to-device (D2D) communication has emerged as a promising technology for IoT, the conventional solutions for D2D resource allocation are usually computationally complex and time consuming. The high complexity poses a significant challenge to the practical implementation of wireless IoT networks. A graph neural network (GNN)-based framework is proposed to address this challenge in a supervised manner. Specifically, the wireless network is modeled as a directed graph, where the desirable communication links are modeled as nodes and the harmful interference links are modeled as edges. The effectiveness of the proposed framework is verified via two case studies, namely the link scheduling in D2D networks and the joint channel and power allocation in D2D underlaid cellular networks. Simulation results demonstrate that the proposed framework outperforms the benchmark schemes in terms of the average sum rate and the sample efficiency. In addition, the proposed GNN approach shows potential generalizability to different system settings and robustness to the corrupted input features. It also accelerates the D2D resource optimization by reducing the execution time to only a few milliseconds.

This paper studies the problem of trajectory design for unmanned aerial vehicle (UAV)-assisted emergency communications, where the ground base station (BS) may be no longer functioning and the UAV acts as an aerial BS to provide emergency communication services to the ground users. In the event of emergency situations, the user distribution and the geographical features of the target area may have changed dramatically while urgent demand for communications are raised by the surviving ground users. In this paper, we model UAV trajectory design problem as a deep reinforcement learning (DRL) process and propose to adopt transfer learning to leverage previously learned knowledge so as to boost up the learning procedure. Simulation results validate that with limited interactions with the environment, the UAV can rapidly and effectively adapt its trajectory to the new environment and achieve much faster convergence speed than DRL based design.

This paper studies the problem of joint edge cache placement and content delivery in cache-enabled small cell networks in the presence of spatio-temporal content dynamics unknown a priori. The small base stations (SBSs) satisfy users' content requests either directly from their local caches, or by retrieving from other SBSs' caches or from the content server. In contrast to previous approaches that assume a static content library at the server, this paper considers a more realistic non-stationary content library, where new contents may emerge over time at different locations. To keep track of spatio-temporal content dynamics, we propose that the new contents cached at users can be exploited by the SBSs to timely update their flexible cache memories in addition to their routine off-peak main cache updates from the content server. To take into account the variations in traffic demands as well as the limited caching space at the SBSs, a user-assisted caching strategy is proposed based on reinforcement learning principles to progressively optimize the caching policy with the target of maximizing the weighted network utility in the long run. Simulation results verify the superior performance of the proposed caching strategy against various benchmark designs.

This paper focuses on joint edge cache placement and content delivery problem at a base station (BS) in the presence of spatio-temporal unknown content dynamics, where the BS can satisfy its users' content demands either directly from its local cache or by fetching from the content server. Unlike the previous works that assume a static content library, we consider a more realistic non-stationary scenario, where new contents are emerging over time at the content library and might be cached at users. We propose that the new contents cached at local users can be utilized by the BS to timely update its flexible portion of cache memory in addition to its routine off-peak main cache update from the content server. We model the caching problem as a non- stationary bandit problem and introduce a user-aided caching algorithm that accounts for the traffic demand variations and the limited caching space at the BS. The proposed algorithm progressively improves the caching policy, with the target of maximizing the weighted content delivery rate to the users in the long run. Simulation results validate that the proposed strategy outperforms various benchmark designs.

This paper studies the problem of max-min fairness power allocation in distributed small cell networks operated under the same frequency bandwidth. We introduce a calibrated learning enhanced time division multiple access scheme to optimize the transmit power decisions at the small base stations (SBSs) and achieve max-min user fairness in the long run. Provided that the SBSs are autonomous decision makers, the aim of the proposed algorithm is to allow SBSs to gradually improve their forecast of the possible transmit power levels of the other SBSs and react with the best response based on the predicted results at individual time slots. Simulation results validate that in terms of achieving max-min signal-to-interference-plus-noise ratio, the proposed distributed design outperforms two benchmark schemes and achieves a similar performance as compared to the optimal centralized design.

This paper introduces a combined calibrated learning and bandit approach to online distributed power control in small cell networks operated under the same frequency bandwidth. Each small base station (SBS) is modelled as an intelligent agent who autonomously decides on its instantaneous transmit power level by predicting the transmitting policies of the other SBSs, namely the opponent SBSs, in the network, in real-time. The decision making process is based jointly on the past observations and the calibrated forecasts of the upcoming power allocation decisions of the opponent SBSs who inflict the dominant interferences on the agent. Furthermore, we integrate the proposed calibrated forecast process with a bandit policy to account for the wireless channel conditions unknowna priori, and develop an autonomous power allocation algorithm that is executable at individual SBSs to enhance the accuracy of the autonomous decision making. We evaluate the performance of the proposed algorithm in cases of maximizing the long-term sum-rate, the overall energy efficiency and the average minimum achievable data rate. Numerical simulation results demonstrate that the proposed design outperforms the benchmark scheme with limited amount of information exchange and rapidly approaches towards the optimal centralized solution for all case studies.

Time-varying wireless channel as well as the variability of renewable energy supply and energy prices are practically unknown in advance. To address such dynamic statistics of wireless networks, this letter develops an adaptive strategy inspired by combinatorial multiarmed bandit model for energy storage management and cost-aware coordinated load control at the base stations. The proposed strategy makes online foresighted decisions on the amount of energy to be stored in storage to minimize the average energy cost over long-time horizon. Simulation results validate the superiority of the proposed strategy over a recently proposed storage-free learning-based design.

This paper studies a robust beamforming optimization problem of minimizing total transmit power in a distributed manner in the presence of imperfect channel state information (CSI) in multicell interference networks. Due to the fact that worst- case is a rare occurrence in practical network, this problem is constrained to satisfying a set of signal-to-interference-plus-noise-ratio (SINR) requirements at user terminals with certain SINR outage probabilities. This problem is numerically intractable due to the cross-link coupling effect among base stations (BSs) operating under the same frequency bandwidth and the robust constraints that involve instantaneous CSI uncertainties. The intractable problem is first converted to a semidefinite programming form with linear matrix inequality constraints via Schur complement, S-procedure and semidefinite relaxation technique, and then decomposed into a set of independent subproblems at individual BSs and solved via subgradient iterations with a light inter-BS communication overhead. Simulation results demonstrate the advantage of the proposed strategy in terms of providing larger SINR operational range as compared with recent proposed designs.

This paper addresses the problem of maximizing the weighted signal-to-interference-plus-noise-ratio (SINR) targets at user terminals in a distributed manner in multicell interference networks. The optimization is constrained to strict individual base station (BS) transmit power limitations in the presence of imperfect channel state information (CSI). This problem is numerically intractable due to the coupling effect among a cluster of BSs operating under the same frequency bandwidth and robust constraints that involve the imperfect CSI. We first convert the original problem into a dual total transmit power minimization problem subject to a set of robust SINR constraints in the centralized worst-case scenario. Then the resulting global optimization problem is decomposed into a set of independent subproblems at individual BSs. Finally, a multi-arm bandit based algorithm is proposed to optimally scale the SINR targets in a distributed manner based on individual BS power budgets, and coordinate intercell interference among the BSs with a light inter-BS communication overhead. Simulation results demonstrate the advantage of the proposed scheme in terms of providing larger SINR operation range and robustness to the CSI uncertainties.

In this paper, we propose a coordinated base-station energy management (CoBEM) technique in which the BSs in a coordinated multipoint (CoMP) with rate-limited backhaul links collaborate to keep the demand and supply balanced using local renewable energies. We formulate two sparse beamforming techniques as ¿0-norm optimisation problems and apply a method that can replace ℓ0-norm with ℓ1-norm and iteratively updated the weight factors to obtain a sufficient sparsity solution called reweighted ℓ1-norm minimisation. For user-centric clustering technique, each user terminal selects a cluster of BSs, whereas, for BS-centric clustering technique, the BSs with a shortage of power budget are authorised to select an optimal number of user terminals based on their available energy budget. In both techniques, we investigate the optimal tradeoff between the BS-receiver cooperation links, the overall energy consumption by the BSs and the energy purchased by the retailer from the realtime market, whilst accounting for the quality-of-service (QoS) requirements for simultaneous wireless information and power transfer (SWIPT). Extensive simulation results confirm that the proposed sparse beamforming techniques significantly improve the infeasibility of a full cooperation in CoMP-SWIPT networks and reveal that the BS-centric clustering is more profitable than the user-centric clustering in real-time energy balancing.

Without a proper observation of the energy demand of the receiving terminals, the retailer may be obliged to purchase additional energy from the real-time market and may take the risk of losing profit. This paper proposes two combinatorial multi-armed bandit (CMAB) strategies in green cloud radio access network (C-RAN) with simultaneous wireless information and power transfer under the assumption that no initial knowledge of forthcoming energy demand and renewable energy supply are known to the central processor. The aim of the proposed strategies is to find the set of optimal sizes of the energy packages to be purchased from the day-ahead market by observing the instantaneous energy demand and learning from the behaviour of cooperative energy trading, so that the total cost of the retailer can be minimized. Two novel iterative algorithms, namely, ForCMAB energy trading and RevCMAB energy trading are introduced to search for the optimal set of energy packages in ascending and descending order of package sizes, respectively. Simulation results indicate that CMAB approach in our proposed strategies offers the significant advantage in terms of reducing overall energy cost of the retailer, as compared to other schemes without learning-based optimization.

We propose a robust coordinated game theoretic approach that distributively minimizes the aggregate downlink transmit power in a multicell interference network in the presence of imperfect channel state information (CSI). The optimization is constrained to satisfying the signal-to-interference-plus-noise-ratio (SINR) requirements at individual user terminals within certain predefined SINR outage probabilities. This problem is numerically intractable due to the cross-link coupling effect among a cluster of base stations (BSs) as well as the robust constraints that involve the second order statistical CSI estimation error. By employing cumulative distribution function of standard normal distribution, Lemma 2 and semidefinite relaxation technique, we first convert the original problem to a linear matrix inequality form. Then, we introduce an iterative subgradient algorithm that decomposes the multicell-wise general problem into a set of parallel subproblems at individual BSs to find the global optimality. We show that the proposed design co-ordinates intercell interference among the BSs with a light inter-BS communication overhead. Simulation results demonstrate the advantages of the proposed chance-constrained distributed design in terms of power efficiency and achievable robustness trade-off.

With enormously increasing demand for mobile data and high data rates, the aggregated power requirements by user terminals may exceed the amount of power budget at the remote radio heads (RRHs). Hence, the retailer has to purchase additional energy from the real-time market, i.e., the grid, and may take a risk of losing the profit. To address the aforementioned issue, this paper studies the real-time energy management for green cloud radio access network (C-RAN) using the local renewable energy generation at RRHs. We develop three different cooperative real-time energy trading strategies, namely, power-shortage management by partial cooperation, power-shortage management by full cooperation, and overall network energy management by full cooperation, to jointly minimize the energy consumption and the real-time energy trading under the constraints of demand and supply power balancing at RRHs and quality of service satisfaction at user terminals. For the first strategy, we formulate a sparse beamforming problem as an ℓ0-norm optimization problem and solve it using semidefinite relaxation (SDR) and iterative reweighted ℓ1-norm approximation of ℓ0-norm. We formulate the second and the third strategies as numerically tractable optimization problems and solve them using the SDR approach. Our simulation results show that the second and the third strategies perform closely up to medium signal-to-interference-plus-noise ratio range and both outperform the first strategy in terms of reducing the total energy cost of the retailer. Furthermore, in terms of cost reduction, all of our three joint cooperative transmission and energy trading strategies achieve significant performance gain, as compared to a baseline scheme that separately optimize the cooperative transmission and energy trading.

Recently, equipping the base stations (BSs) with renewable energy harvesters for green communications has been considered as a promising technique to benefit both the environment and the retailer. However, the uneven distribution of renewable energy generation and the drastically increasing demand for mobile data has put forward the new challenge for the retailer. To tackle this problem, we propose a real-time base-station energy management (RBEM) strategy using renewable energy sources for green communications in coordinated multipoint networks with simultaneous wireless information and power transfer, where all the BSs collaborate to maintain the real-time supply and demand power balancing at the BSs, in order to reduce the overall energy cost of the retailer. The RBEM strategy is formulated as a non-convex linear combinational optimization problem to jointly minimize the energy consumption and energy trading, whilst accounting for the quality-of-service requirements for information receiving terminals and power transfer requirements for energy receiving terminals (ETs). Moreover, an iterative algorithm is proposed to determine the sources of energy harvested by the ETs. The original non-convex problem can be transformed into a numerically tractable form and solved using the semidefinite relaxation approach. Our simulation results confirm that the proposed RBEM strategy significantly reduces overall energy cost of the retailer.