Abstract
Deep learning methods have seen increasing importance and rapid advancements in timeseries forecasting. These methods, which leverage the power of neural networks, have proven to be highly effective in capturing complex patterns and dependencies in data, offering significant improvements over traditional forecasting techniques. With the ability to model nonlinear relationships and learn from big data, deep learning has revolutionised time-series forecasting, leading to more accurate and robust predictions across a diversity of domains.In wireless communications, numerous time-series forecasting problems arise, e.g., predicting channel states and user mobility. Addressing these challenges is crucial for optimising network performance, enhancing energy efficiency, and ensuring robust communications. Deep learning provides powerful tools to tackle these problems by learning from historical data and making precise predictions, enabling adaptive network management.
This thesis presents two significant applications of deep learning for solving time-series forecasting problems in wireless communications:
First, we propose a novel deep learning-based algorithm for channel prediction and energy efficiency (EE) optimisation in an intelligent reflecting surface (IRS) aided Terahertz communication system. Specifically, a multi-antenna base station with an IRS with massive reflecting elements is designed to serve multiple moving users. A deep learning-based prediction-optimisation scheme is presented where we first propose a transformer encoder with channel index embedding (TE-CIE) deep learning model for time-varying channel prediction. With the help of channel prediction, an EE optimisation algorithm is then designed to maximise the EE in advance based on the predicted channel state information (CSI). Finally, we combine both methods to construct a deep learning-based predictionoptimisation scheme. Specifically, the future CSI is predicted by TE-CIE and the IRS phaseshift and precoding matrices are optimised in advance. Simulation results demonstrate that our proposed channel prediction method achieves close-to-optimal performance in terms of low mean absolute error and a much faster speed than the two baseline models. We demonstrate that the proposed EE optimisation algorithm outperforms the baseline algorithms in terms of much better EE under diverse parameter settings. Finally, the proposed prediction-optimisation scheme achieves at least twice the EE improvement compared to the baseline methods in the literature.
Second, we focus on designing a robust deep-learning model to predict user mobility under malicious Global Navigation Satellite System (GNSS) spoofing attacks for unmanned aerial vehicle (UAV) swarm position optimisation. UAV swarms have become a promising solution to enhance modern wireless communication in complicated environments. However, due to the existence of real-world malicious attacks, the performance of prediction and optimisation methods used for UAV swarms are easily degraded. In this paper, we propose a novel deep learning-based user mobility prediction, user assignment and drone position optimisation scheme for a UAV swarm-enabled wireless communication system with the existence of malicious GNSS spoofing attackers. Specifically, a robust deep learning-based user mobility prediction model, namely denoising autoencoder recurrent transformer (DART), is designed and various efficient user assignment and drone position optimisation methods are proposed. Simulation results show that the proposed deep learning-based prediction-optimisation scheme can provide up to 30% higher overall sum rate compared with the adversarial trained long short-term memory (LSTM) baseline and almost doubled the overall sum rate compared with the vanilla LSTM baseline.
To reduce the computational complexity of the DART model without compromising its performance, we employ a technique called knowledge distillation for sustainable purposes. By distilling the knowledge learned by the complex DART model into a simpler and more computationally efficient architecture, such as a smaller Gated Recurrent Unit (GRU) model, we aim to retain the essential information for user mobility prediction and drone position optimisation. This distilled model can offer much faster inference times and reduced resource requirements while preserving much of the performance achieved by the original DART model, making it more practical for real-time deployment in UAV swarm-enabled wireless communication systems under the threat of malicious GNSS spoofing attacks. Simulation results demonstrate that the optimised sum rate using the distilled GRU student model’s predicted user locations can achieve almost 99% compared to the Transformer teacher model. Meanwhile, the inference time of the student model is only 4% compared to the teacher model.
In conclusion, our research emphasises the potential of deep learning for time-series forecasting in next-generation wireless communication scenarios. By addressing key forecasting problems, e.g., predicting channel states and user mobility, our deep learningbased algorithms demonstrate significant improvements in energy efficiency and network performance. The proposed solutions for Terahertz communications, IRS systems, and UAV swarm networks show the robustness and accuracy of deep learning models in complex and dynamic environments. Future research will continue to explore innovative deep-learning techniques to solve additional time-series forecasting challenges and further optimise wireless communication systems.
| Date of Award | 1 Sept 2024 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Mohammed Shikh-Bahaei (Supervisor) & Mohammad Nakhai (Supervisor) |