Comprehensive Review on ML-based RIS-enhanced IoT Systems: Basics,
Research Progress and Future Challenges
Abstract
Sixth generation (6G) internet of things (IoT) networks will modernize
the applications and satisfy user demands through implementing smart and
automated systems. Intelligence-based infrastructure, also called
reconfigurable intelligent surfaces (RISs), have been introduced as a
potential technology striving to improve system performance in terms of
data rate, latency, reliability, availability, and connectivity. A huge
amount of cost-effective passive components are included in RISs to
interact with the impinging electromagnetic waves in a smart way.
However, there are still some challenges in RIS system, such as finding
the optimal configurations for a large number of RIS components. In this
paper, we first provide a complete outline of the advancement of RISs
along with machine learning (ML) algorithms and overview the working
regulations as well as spectrum allocation in intelligent IoT systems.
Also, we discuss the integration of different ML techniques in the
context of RIS, including deep reinforcement learning (DRL), federated
learning (FL), and FL-deep deterministic policy gradient (FL-DDPG)
techniques which are utilized to design the radio propagation atmosphere
without using pilot signals or channel state information (CSI).
Additionally, in dynamic intelligent IoT networks, the application of
existing integrated ML solutions to technical issues like user movement
and random variations of wireless channels are surveyed. Finally, we
present the main challenges and future directions in integrating RISs
and other prominent methods to be applied in upcoming IoT networks.