The future mobile communication networks (beyond 5th generation (5G)) are evolving toward the service-based architecture where network functions are fine-grained, thereby meeting the dynamic requirements of diverse and differentiated vertical applications. Consequently, the complexity of network management becomes higher, and artificial intelligence (AI) technologies can improve AI-native network automation with their ability to solve complex problems. Specifically, deep reinforcement learning (DRL) technologies are considered the key to intelligent network automation with a feedback mechanism similar to that of online closed-loop architecture. However, the 0-delay assumptions of the standard Markov decision process (MDP) of traditional DRL algorithms cannot directly be adopted into real-world networks because there exist random delays between the agent and the environment that will affect the performance significantly. To address this problem, this paper proposes a random-delay-corrected framework. We first abstract the scenario and model it as a partial history-dependent MDP (PH-MDP), and prove that it can be transformed to be the standard MDP solved by the traditional DRL algorithms. Then, we propose a random-delay-corrected DRL framework with a forward model and a delay-corrected trajectory sampling to obtain samples by continuous interactions to train the agent. Finally, we propose a delayed-deep-Q-network (delayed-DQN) algorithm based on the framework. For the evaluation, we develop a real-world cloud-native 5G core network prototype whose management architecture follows an online closed-loop mechanism. A use case on the top of the prototype namely delayed-DQN-enabled access and mobility management function (AMF) scaling is implemented for specific evaluations. Several experiments are designed and the results show that our proposed methodologies perform better in the random-delayed networks than other methods (e.g., the standard DQN algorithm).
23 December 2024 in IEEE Transactions on Cognitive Communications and Networking
Deep learning (DL)-enabled semantic communications offer promising prospects for higher data transmission efficiency. However, most existing studies focus on point-to-point transmissions with perfect channel state information (CSI) and fixed channel bandwidth ratio (CBR). This paper investigates multi-point transmission in cellular networks with spectrum resource sharing and proposes a signal-to-interference plus noise ratio (SINR) adaptive and CBR-controllable semantic cellular communication system (SACC). In this system, we design a semantic encoder by a Transformer-convolutional neural network (CNN) mixture block to capture non-local and local latent image features simultaneously and a SINR-adaptive module based on the channel-spatial soft attention mechanism to scale image features according to SINR conditions. A model adaption strategy is proposed to handle varying inter-cell co-channel interference proportions under the same SINR. Additionally, a semantic-channel encoder-decoder forwarding paradigm, termed semantic translation, enables dynamic networking among intra-cell users. Extensive simulation results show that the proposed SACC achieves graceful degradation and better image reconstruction performance versus the current engineered “JPEG+LDPC+QAM” and DeepJSCC benchmarks across different wireless environments and CBRs. Furthermore, experimental results demonstrate substantial improvements in FLOPs, model sizes, and computing latency over the state-of-the-art ADJSCC and WITT with comparable performance.
1 February 2025 in Digital Communications and Networks
Deep learning (DL)-enabled semantic communication has been widely regarded as a promising 6G technology due to its efficient wireless transmission capabilities, providing a novel technological foundation for bandwidth-constrained Internet of Things (IoT) scenarios. Combining semantic communication with classical multicell network architectures, which are commonly employed in long-distance IoT communications, is essential for enabling its large-scale deployment in future 6G systems. However, the performance optimization of semantic transmission in this scene still faces significant challenges due to the complex coupling among compression, transmission, and computation under resource-limited conditions, which is difficult to solve by traditional methods due to the closed box nature of the artificial intelligence models used. To address this problem, in this article, we formulate an optimization problem to minimize the weighted sum of end-to-end latency and information reconstruction performance by adjusting the compression control strategy, radio block assignment, and computing resource allocation decision. The case of image transmission is investigated. To address this nonconvex mixed-integer nonlinear programming problem, we adopt a lower bound approximation method to decompose the formulated problem and design an optimization method combining the Lagrange multiplier method and multiagent deep reinforcement learning algorithm. We also introduce an image reconstruction performance prediction model to empower the agent with semantic-aware optimization ability. The results demonstrate that the proposed method can achieve superior performance compared to other approaches.
Most existing joint source-channel coding-based semantic communication systems utilize static statistical channel models, leading to poor performance in time-frequency selective fading channels. To fill this gap, this paper proposes an orthogonal frequency division multiplexing (OFDM)-guided mobility-robust JSCC (MR-JSCC) scheme for wireless image transmission in multipath high-mobility channels. Leveraging an attention mechanism-based signal restorer, the scheme mitigates time-frequency selectivity effects and refines feature information using pilots. The restorer captures channel state information to separate and map aliased image features caused by inter-carrier interference. Experiments demonstrate that MR-JSCC outperforms existing JSCC schemes in multipath high-mobility scenarios, even beyond the model’s training range.