Abstract:The performance of modern wireless communications systems depends critically on the quality of the available channel state information (CSI) at the transmitter and receiver. Several previous works have proposed concepts and algorithms that help maintain high quality CSI even in the presence of high mobility and channel aging, such as temporal prediction schemes that employ neural networks. However, it is still unclear which neural network-based scheme provides the best performance in terms of prediction quality, training complexity and practical feasibility. To investigate such a question, this paper first provides an overview of state-of-the-art neural networks applicable to channel prediction and compares their performance in terms of prediction quality. Next, a new comparative analysis is proposed for four promising neural networks with different prediction horizons. The well-known tapped delay channel model recommended by the Third Generation Partnership Program is used for a standardized comparison among the neural networks. Based on this comparative evaluation, the advantages and disadvantages of each neural network are discussed and guidelines for selecting the best-suited neural network in channel prediction applications are given.
What problem does this paper attempt to address?
The paper attempts to address the problem of how to predict wireless channel state information (CSI) using neural networks in wireless communication systems, in order to improve prediction quality, reduce training complexity, and ensure practical feasibility. Specifically, the paper focuses on how to select the most suitable neural network model for high-quality channel prediction in the presence of high mobility and channel aging.
### Background and Motivation
The performance of modern wireless communication systems highly depends on the quality of the channel state information (CSI) available at the transmitter and receiver. Although many studies have proposed various concepts and algorithms to maintain high-quality CSI, especially in the presence of high mobility and channel aging, it is still unclear which neural network-based scheme performs best in terms of prediction quality, training complexity, and practical feasibility.
### Main Contributions
1. **Review of Existing Work**: The paper first reviews the latest neural network methods suitable for channel prediction and compares their prediction performance.
2. **Quantitative Comparison**: The paper provides an original quantitative comparison of four promising neural networks with different prediction time ranges and compares them with the traditional Kalman filter.
3. **Performance Discussion**: Based on numerical evaluation results, the paper discusses the advantages and disadvantages of each neural network and provides guidelines for selecting the most suitable neural network for channel prediction applications.
### Research Methods
- **Dataset**: A standardized comparison was conducted using a simulated dataset generated by the 3GPP recommended standard tapped delay line (TDL-A) model.
- **Experimental Setup**: Prediction time ranges from 1-40 milliseconds were considered, including both noise-free and noisy datasets.
- **Model Architectures**: The comparison included five neural network models: multilayer perceptron (MLP), convolutional neural network (CNN), recurrent neural network (RNN), long short-term memory network (LSTM), gated recurrent unit (GRU), and transformer.
### Experimental Results
- **Noise-Free Data**: In the short prediction time range, LSTM and GRU outperformed the Kalman filter; as the prediction time range increased, the mean squared error (MSE) of all neural networks gradually increased, reaching a relatively stable level at around 15 milliseconds.
- **Noisy Data**: In noisy environments, the performance of all neural networks was worse than in noise-free environments, but in the short prediction time range, the performance of RNN, MLP, CNN, and transformer remained quite close.
- **Kalman Filter**: In noise-free environments, the Kalman filter performed excellently in the short prediction time range; in noisy environments, its performance was similar to that of neural networks, but in the long prediction time range, RNN (especially GRU) outperformed the Kalman filter.
### Conclusion
Through detailed experiments and analysis, the paper provides valuable references for selecting the most suitable neural network model for wireless channel prediction. Particularly in the long prediction time range, GRU performed excellently and is a recommended choice. Additionally, the paper points out future research directions, including further optimization of neural network models and exploring performance over longer prediction time ranges.