Visible light communication (VLC) gets increasing attention with the continuous advancement of technology. In contrast to conventional wireless networks, VLC possesses numerous innovative advantages. It has an abundant optical spectrum with immunity to radio-frequency interference. More importantly, its compatibility with existing lighting infrastructure demonstrates its broad applicability. However, high-speed VLC systems still need to address numerous challenges during communication such as limited LED bandwidth, inter-symbol interference, device nonlinearity, and receiver noise. Due to the above-mentioned factors, the channel equalization becomes an important physical-layer function. This paper aims to clarify the strengths and limitations of different deep learning-based channel equalization methods with cross-method comparison. Meanwhile, finds suitable application scenarios for different VLC systems. This review explores neural-network architectures including artificial neural networks, convolutional neural networks, long short-term memory networks, and hybrid/model-driven structures. Finally, it also illustrates the importance of lightweight designs. The analysis shows that no single architecture is optimal for all VLC conditions. ANN is effective for nonlinear mapping, CNN is suitable for local feature extraction, while LSTM performs better for channel memory and inter-symbol interference.

Show more

The development of sixth generation (6G) communications is aimed at achieving global connectivity, covering oceans and remote areas with limited ground infrastructure. With the integration of sensing, communication and computing (ISCC) technology, the space-air-ground integrated network (SAGIN) is evolving from a simple data transmission channel to an intelligent task processing platform. This paper uses a narrative review method to analyze the evolution path of resource orchestration technology in heterogeneous environments, and focuses on mathematical programming and data-driven intelligent methods to optimize resources. The existing research is classified according to the method system. This paper also examines the architecture paradigm, multi-dimensional resource modeling and algorithm system. Studies have shown that static methods such as Lagrangian dual decomposition and block coordinate descent provide the necessary performance benchmarks, and deep reinforcement learning can achieve fast response in high mobility links. In addition, this paper discusses the potential of large language model (LLM) as a semantic interpreter, which can accelerate the process of parsing task requirements into technical constraints in specific applications. This study provides a methodological reference for achieving task-centered connectivity, and believes that a hybrid framework that combines the accuracy of deterministic methods and the adaptability of learning methods is a feasible development direction for future space infrastructure.

Show more

A fundamental problem in reinforcement learning is the trade-off between exploration and exploitation. The quality of the final policy is not the only factor that varies with different exploration strategies, so do learning speed and result stability. Based on Q-learning, this study compares three exploration strategies in a custom 5×5 GridWorld environment: fixed ε=0.10, fixed ε=0.30, and linearly decaying ε (1.00→0.05). A baseline experiment is first performed with slip_prob=0.10. The environmental noise is then extended to three levels, 0.00, 0.10, and 0.20, and the number of episodes required for the rolling 50-episode success rate to first reach 50% is used to measure learning speed. The results show that the fixed lower exploration rate learns faster overall under all three noise levels and demonstrates better stability across random seeds. Decaying exploration achieves a higher final success rate under low-noise conditions, but converges much more slowly in the early stage. The fixed higher exploration rate performs relatively poorly overall and has particular difficulty forming a stable policy in high-noise environments. These experiments show that exploration strategies should not be evaluated only by final success rate; environmental randomness, learning speed, and stability across random seeds should also be considered.

Show more

As data in scenarios such as medical, transportation and the Internet of Things continue to be dispersed to institutions and edge devices, how to balance the quality of model training and communication costs without concentrating raw data has become an important issue in the deployment of federated learning. Federated learning faces a coupling bottleneck between client drift and communication overhead under non-independent and non-identically distributed (Non-IID) data. This paper reviews aggregation, compression and hybrid strategies with representative studies, and verifies Federated Averaging (FedAvg), Federated Proximal (FedProx) and their error-feedback Top-k combinations (FedAvg+EF-Top-k and FedProx+EF-Top-k) under the unified setting of Fashion-MNIST. The results show that compression can significantly reduce the cumulative uplink communication required to reach the target accuracy and enable more training rounds under a fixed total communication budget. However, strong heterogeneity is associated with greater run-to-run variability, and communication rounds, transmitted bits, and end-to-end time cannot be substituted for one another. Based on literature comparison and verification results, this paper proposes a selection strategy according to data heterogeneity, bandwidth and system constraints, and recommends the unified reporting of accuracy, communication volume, communication rounds, time, and energy consumption.

Show more