Semantic communication has emerged as a promising paradigm for next-generation wireless networks, offering enhanced efficiency by reducing transmission data. However, implementing semantic communication faces significant challenges, particularly in resource-constrained devices that cannot support the complex artificial intelligence (AI) models required for semantic extraction. This paper addresses this challenge by proposing a novel unmanned aerial vehicle (UAV)-enabled semantic-bit coexisting relay system, where the UAV serves as intermediate nodes to assist transmissions from resource-limited users to the base station. By deploying semantic extraction models at the UAV, the proposed system solves the computational resource limitations for user devices while minimizing transmission latency via data size reduction. In such a system, we formulate a system latency minimization problem that jointly considers semantic compression model selection and bandwidth allocation. To address this complex problem, we develop an effective solution method by decomposing the original problem into a semantic compression model selection based on performance-latency trade-offs and a bandwidth-allocation optimization via convex optimization techniques. Extensive numerical evaluations demonstrate that the proposed framework consistently outperforms conventional schemes across diverse network settings and compression parameters, significantly reducing end-to-end latency while maintaining high-quality semantic communication.

The domain adaptation issues in semantic communications become critical when transmitter and receiver operate across different multiple domains or when input data during inference have different distributional characteristics than the data used to train semantic encoders and decoders. In this paper, we introduce the Multidomain Adaptive Deep Semantic Communication (MA-DeepSC) framework, designed to enhance semantic communications across multiple domains. Our framework consists of two core components: the Multidomain Adaptive Semantic Coding Network (MASCN), inherently designed to adapt semantic encoding and decoding across multiple domains, and the multidomain data adaptation network (MDAN), which transforms actual observable data into the data on which the system was initially trained, thus obviating the need for retraining the existing pre-trained semantic coding network. We validate our approach through experiments on digit datasets and CelebA, observing significant outperformance over existing techniques. In addition, we analyze the strategic benefits and drawbacks of both MASC and MDAN, assessing their applicability under various scenarios. The source code for MA-DeepSC is available at https://github.com/wongdongwook/JSAC_MA-DeepSC.

Federated learning (FL) is a framework for realizing distributed machine learning in an environment where training samples are distributed to each device. Recently, FL has employed over-the-air computation enabling all devices to transmit learning model updates simultaneously. This work proposes a communication-efficient sparse one-bit analog aggregation (SOBAA) method, incorporating new power control, layer-wise scaled one-bit quantization, layer-wise sparsification, and an error-feedback mechanism. We derive a tight upper bound of the expected convergence rate of the proposed SOBAA as a closed-form expression. From this expression, we explicitly identify the relationship between the convergence rate and compression and aggregation errors. Based on the theoretical convergence analysis, we formulate a joint optimization problem of the compression ratio and power control to minimize compression and aggregation errors, leading to the fastest convergence. In each communication round, the optimization problem is decomposed, and solved in a computationally efficient and feasible way. From this solution, we characterize the trade-off between learning performance and communication cost. Through extensive experiments on well-known MNIST and CIFAR-10 datasets, we confirm that the proposed method provides an enhanced trade-off performance between test accuracy and communication costs and a faster convergence rate than the other state-of-the-art methods. In addition, it is proven that the proposed method is more effective for more complex datasets and learning models.