Towards addressing emerging energy challenges posed by unfair heterogeneous Semantic Communication (SC) codec updates within future wireless networks, this paper presents a novel Anti-discrimination Federated learning (AdFed) approach. Inspired by the economics of discrimination, unique fairness-associated energy concerns in SC systems are formulated as model discrimination challenges, with the SC-deployed wireless network conceptualized as an anti-discrimination labor market. A novel “affirmative action” strategy, based on training epochs, is proposed and adopted according to historical training unfairness results. To address the reverse discrimination issues in “affirmative action” caused by quota fairness impacting training energy cost, we formulate this problem as a coupled integer non-linear programming problem. Moreover, a new quota trade-off mechanism based on the Rubinstein bargaining game is also designed. Simulation results verify that AdFed outperforms SC training baselines, effectively addressing the unique model discrimination challenges of SC codec model heterogeneity updating. The efficacy of the game theoretical trade-off mechanism is demonstrated in achieving optimal outcomes.

To navigate the carbon emission and functional challenges associated with edge caching within heterogeneous Industrial Internet of Things (IIoT) spanning energy use, cache hit rate, and bandwidth usage, this paper proposes a novel Game Theory Empowered Carbon-Intelligent Federated Multi-Edge Caching framework (GT-FMC). The proposed framework enables distributed collaborative caching by intelligently coordinating edge nodes to optimize content decisions while efficiently integrating content providers (CPs), edge nodes, and users with energy-aware strategies. In GT-FMC, a lightweight federated content popularity prediction method based on Temporal Convolutional Networks (TCN) is introduced to collaboratively learn global content popularity while reducing prediction energy cost. The energy-aware utilities of the three involved parties are jointly formulated as a coupled non-linear optimization problem. To address this challenge, a two-stage game-theoretic algorithm is designed. Experimental results on a real-world testbed show that GT-FMC achieves up to 77.9% of Oracle in cache hit rate and 10.6%–32.4% reduction in transmission energy consumption compared to baseline methods. Complementary evaluations also validate the game-theoretic design’s effectiveness.

In navigating the challenges of real-time semantic communication (SC) codec updates in the 6G-era non-terrestrial network (NTN)-assisted vehicular networks (NTN-VNs), a crucial component of intelligent transportation systems (ITS), this article introduces a novel approach inspired by human society. Facing complexities like 3-dimensional updating, network dynamism, and updating costs, NTN-VNs are treated as social networks. The proposed NTN-VN federated learning (NTN-VN-FL) framework asynchronously addresses challenges such as uplink and downlink SC codec updates, device decentralization, and asynchronous updating. By viewing device behaviors during updating as social behaviors with economic costs, an NTN-VN social management system ensures the proper functioning of the social network in the context of NTN-VN-FL. An economical social behavior selection mechanism, based on the reverse auction game for NTN-VN-FL, minimizes training delay and device energy costs, considering social relationships. The article also presents a two-stage Stackelberg game with the Vickrey auction rule to maximize social welfare in the auction. Simulation results highlight the superiority of NTN-VN-FL over existing potential application algorithms, effectively addressing the unique challenges of SC codec updating in NTN-VN. The efficacy of the social management system and social behavior selection mechanism is demonstrated in achieving optimal outcomes.

Machine learning-based semantic communication is a promising enabler for future-generation wireless network systems such as 6G networks. In practice, effective semantic communication requires online training for unknown content. In highly mobile vehicular networks, however, reliable, and efficient model training becomes significantly challenging. The existing distributed learning approaches are also unable to effectively operate in highly dynamic vehicular semantic communication networks. To address these challenges, we propose a novel mobility-aware split-federated with transfer learning (MSFTL) framework based on vehicle task offloading scenarios in this paper. To enable adaptation to the complex vehicle semantic communication, the proposed framework divides the training of the model into four parts and uses the proposed new splitfederated learning. Furthermore, to improve training efficiency, model accuracy, and the ability to adapt in highly mobile environments, we also present a new transfer learning approach integrated into the proposed framework. Particularly, we propose a high-mobility training resource optimisation mechanism based on a Stackelberg game for MSFTL to further reduce training costs and adapt vehicle mobility scenarios. We also investigate the performance of the proposed schemes through extensive simulations. The results validate the proposed approach and indicate its superiority compared to the conventional learning frameworks for semantic communication in vehicular networks.

The low earth orbit (LEO) satellite-borne edge cloud (SEC) and machine learning (ML) based semantic communication (SemCom) are both enabling technologies for 6G systems facilitating computation offloading. Nevertheless, integrating SemCom into the SEC networks for user computation offloading introduces semantic coder updating requirements as well as additional semantic extraction costs. Offloading user computation in SEC networks via SemCom also results in new functional challenges considering, e.g., latency, energy, and privacy. In this paper, we present a novel SemCom-assisted SEC (SemCom-SEC) framework for computation offloading of resource-limited users. We then propose an adaptive pruning-split federated learning (PSFed) method for updating the semantic coder in SemCom-SEC. We further show that the proposed method guarantees training convergence speed and accuracy. This method also improves the privacy of the semantic coder while reducing training delay and energy consumption. In the case of trained semantic coders in service, for the users processing computational tasks, the main objective is to minimise the users’ delay and energy consumption, subject to sustaining users’ privacy and fairness amongst them. This problem is then formulated as an incomplete information mixed integer nonlinear programming (MINLP) problem. A new computational task processing scheduling (CTPS) mechanism is also proposed based on the Rubinstein bargaining game. Simulation results demonstrate the proposed PSFed and game theoretical CTPS mechanism outperforms the baseline solutions reducing delay and energy consumption while enhancing users’ privacy.

Collaborative Intrusion Detection System (CIDS) is an essential technology that enables vehicular ad hoc networks (VANET) to protect against malicious intrusions. CIDS, however, is unable to prevent accidents if an anomalous vehicle is detected. Detecting anomalies and notifying vehicles in the VANET rapidly is thus essential, considering technical challenges such as communication efficiency, vehicle velocity and privacy. In this paper, we propose a novel two-layer privacy-aware trust evaluation CIDS framework, termed 2PT-CIDS, tailored to VANET. In 2PT-CIDS, vehicles and roadside units (RSUs) cooperate efficiently to enhance anomalous vehicle detection and notification. Considering its potential privacy leakage, we then present two types of game-theoretic information incentive mechanisms. In the case of traffic congestion, the privacy-aware incentive mechanism is presented based on the Stackelberg game. A Barycentric Lagrange interpolation (BLI) based algorithm is then proposed to speedy achieve the Nash equilibrium (NE). In the case of traffic smooth, the varying high velocities of vehicles are involved and a noncooperative game-based mechanism is proposed. The optimal NE decision selection is reconstructed as a Markov decision process (MDP) and the NE point is obtained via the designed novel reward-shaping double duelling deep Q network (D3QN) learning algorithm. Simulation results highlight the superiority of 2PT-CIDS over existing CIDS and potential application algorithms for VANET, effectively enhancing anomaly detection and notification considering communication cost and vehicle privacy.