Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

In-Wave Computation Aided Stacked Intelligent Metasurfaces in Next-Generation Networks: Challenges and Opportunities

Mengbing Liu, Chau Yuen, Dusit Niyato, Bruno Clerckx, Lajos Hanzo

TL;DR

The state-of-the-art of SIM research is reviewed, including applications, functions, and characteristics, and their potential is demonstrated through case studies on neural-like analog inference and communication enhancement.

Abstract

Stacked intelligent metasurfaces (SIMs) facilitate computation by cascaded programmable layers so that part of the signal processing can be performed in the wave domain during signal propagation, rather than solely after reception. This approach expands the controllable degrees of freedom and supports the joint design of communication, sensing, and computation with the potential for reduced energy usage, shorter end-to-end latency, and improved task execution. Despite these advances, research on the SIM concept is still at an early stage, with challenges in scalability, controllability, nonlinearity, and robustness. This article reviews the state-of-the-art of SIM research, including applications, functions, and characteristics. We also demonstrate their potential through case studies on neural-like analog inference and communication enhancement. Finally, the paper outlines open challenges and future directions toward establishing SIMs as a new signal processing paradigm for in-wave computation in next-generation (NG) networks.

In-Wave Computation Aided Stacked Intelligent Metasurfaces in Next-Generation Networks: Challenges and Opportunities

TL;DR

The state-of-the-art of SIM research is reviewed, including applications, functions, and characteristics, and their potential is demonstrated through case studies on neural-like analog inference and communication enhancement.

Abstract

Stacked intelligent metasurfaces (SIMs) facilitate computation by cascaded programmable layers so that part of the signal processing can be performed in the wave domain during signal propagation, rather than solely after reception. This approach expands the controllable degrees of freedom and supports the joint design of communication, sensing, and computation with the potential for reduced energy usage, shorter end-to-end latency, and improved task execution. Despite these advances, research on the SIM concept is still at an early stage, with challenges in scalability, controllability, nonlinearity, and robustness. This article reviews the state-of-the-art of SIM research, including applications, functions, and characteristics. We also demonstrate their potential through case studies on neural-like analog inference and communication enhancement. Finally, the paper outlines open challenges and future directions toward establishing SIMs as a new signal processing paradigm for in-wave computation in next-generation (NG) networks.
Paper Structure (15 sections, 3 figures, 1 table)

This paper contains 15 sections, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Scenarios, Characteristics, Computational Functions, and Objectives
  3. Scenarios (S)
  4. Characteristics (C)
  5. Computational Functions (F)
  6. Objectives (O)
  7. Benefits, Challenges and Opportunities
  8. Benefits over MIMO and RIS
  9. Challenges and Opportunities
  10. In-wave neural-like computation for analog inference enabled by SIM
  11. Proposed Scheme: SIM-LIA versus Conventional DNN
  12. Task Description and Setup
  13. Confusion Matrix on SIM-LIA and DNN
  14. Convergence Analysis under Different Numbers of Classes
  15. Conclusions and Future Work

Figures (3)

  • Figure 1: Overview of SIM computational functions and objectives. Functions are organized as wave-domain operators serving communication, sensing, and computation goals, with annotated links indicating supporting mechanisms. Two representative prototypes are shown: a five-layer programmable SIM acting as an in-channel decoder for space–time telecommunication liu2022programmable, and a three-layer transmissive SIM enhancing communication and sensing performance wang2024multi.
  • Figure 2: SIM-aided vs. digital inference over a satellite-to-ground link. (a) SIM-LIA performs spatial encoding and in-propagation inference, yielding a compact $K$-dimensional decision without Level-1 focusing. (b) The digital baseline transports high-dimensional I/Q for post-downlink processing and neural inference, increasing energy, latency, and bandwidth.
  • Figure 4: Epochs versus accuracy under different tasks. Wave-domain phase rotation consistently improves accuracy by an average of 45%.