Executive Summary : | Deep learning (DL) has gained popularity in various fields, including computer vision, speech, natural language processing, healthcare diagnosis, and product recommendations. Recently, there has been an increasing interest in developing solutions for wireless communications through artificial intelligence. DL-based solutions, realized by deep neural networks, have the potential to learn end-to-end strategies from radio frequency (RF) information due to the significant increase in computational resources. Traditionally, wireless communications have used model-based techniques, but DL has been introduced to handle situations where no good models are available or DL has significant computational advantages. DL techniques in wireless communications have shown remarkable success in various applications, such as autoencoder communication systems, channel encoding/decoding, and RF modulation classification. However, these systems are exposed to new security issues, such as inference, adversarial, and poisoning attacks, model extraction, and leakage of information. This project aims to develop reliable and robust DL models for wireless communications, particularly in the context of DL applications. The project aims to envision practical attacks on DL models for future wireless applications from an adversary point of view, devise countermeasure techniques for DL solutions, and develop an RF dataset for indoor human activity sensing using wireless signals. This work will establish a framework for the deployment of secured and reliable DL models for various applications beyond 5G/6G wireless communication systems, including Indian defense security surveillance, elderly/children care monitoring, and healthcare. |