Stacked Intelligent Metasurface for End-to-End OFDM System

Yida Zhang, Qiuyan Liu, Hongtao Luo, Yuqi Xia, Qiang Wang, Fuchang Li, Xiaofeng Tao, Yuanwei Liu

Abstract

Stacked intelligent metasurface (SIM) and dual-polarized SIM (DPSIM) enabled wave-domain signal processing have emerged as promising research directions for offloading baseband digital processing tasks and efficiently simplifying transceiver design. However, existing architectures are limited to employing SIM (DPSIM) for a single communication function, such as precoding or combining. To further enhance the overall performance of SIM (DPSIM)-assisted systems and achieve end-to-end (E2E) joint optimization from the transmitted bitstream to the received bitstream, we propose an SIM (DPSIM)-assisted E2E orthogonal frequency division multiplexing (OFDM) system, in which traditional communication tasks such as channel coding, modulation, precoding, combining, demodulation, and channel decoding are performed synchronously within the electromagnetic (EM) forward propagation. Furthermore, inspired by the idea of abstracting real metasurfaces as hidden layers of a neural network, we propose the EM neural network (EMNN) to enable the control of the E2E OFDM communication system. In addition, transfer learning is introduced into the model training, and a training and deployment framework for the EMNN is designed. Simulation results demonstrate that both SIM-assisted E2E OFDM systems and DPSIM-assisted E2E OFDM systems can achieve robust bitstream transmission under complex channel conditions. Our study highlights the application potential of EMNN and SIM (DPSIM)-assisted E2E OFDM systems in the design of next-generation transceivers.

Introduction

As the telecommunications industry pivots toward 6G, the demand for ultra-high-speed and reliable data transmission is reaching unprecedented levels. Traditional Multiple-Input Multiple-Output (MIMO) systems have historically met these demands through large antenna arrays. However, the requirement for a massive number of radio frequency (RF) chains in these systems leads to prohibitive hardware costs and extreme power consumption, particularly for mobile terminals.

While Reconfigurable Intelligent Surfaces (RIS) have been proposed to manipulate electromagnetic (EM) waves, their single-layer structure often limits processing flexibility. Most current research utilizes these surfaces for isolated tasks like beamforming. This paper introduces a holistic approach: an end-to-end (E2E) optimization framework that connects the transmitted bitstream to the received bitstream through the physical propagation environment itself using Stacked Intelligent Metasurfaces (SIM).

System Architecture

The proposed architecture offloads traditional digital baseband tasks—such as modulation, precoding, and decoding—directly into the "wave domain." By utilizing SIM or Dual-Polarized SIM (DPSIM), the system performs "analog computing" in the air.

SIM-assisted E2E OFDM system architecture

In this setup, tasks like channel coding and modulation are performed synchronously as the EM wave passes through the metasurface layers. This wave-domain processing significantly reduces the reliance on power-hungry digital components. The DPSIM variant further exploits two orthogonal polarization directions, providing additional degrees of freedom for signal processing and increasing robustness against interference.

Electromagnetic Neural Network (EMNN)

The core innovation of this work is the abstraction of physical metasurface layers as hidden layers within a neural network, termed the Electromagnetic Neural Network (EMNN). The forward propagation of an EM wave through the SIM is mathematically equivalent to the forward pass of a deep learning model.

EMNN architecture and real-space mapping

As illustrated in the EMNN architecture, the system establishes a bijective (one-to-one) mapping between neural network weights and the physical phase shifts of the metasurface units. This "real-space mapping" allows the entire communication link to be treated as an Autoencoder (AE). The transmitter (encoder) and receiver (decoder) are jointly trained to minimize the Bit Error Rate (BER). To handle the time-varying nature of wireless channels, transfer learning is employed: a base model is pre-trained using statistical Channel State Information (CSI) and then fine-tuned in real-time with instantaneous CSI, drastically accelerating deployment.

Experimental Results and Insights

The performance of the SIM-assisted E2E OFDM system was evaluated using a geometric multi-path channel model at 28 GHz with a 100 MHz bandwidth. Key findings include:

  • Hardware Efficiency: A 16T9R SIM-assisted system achieved bit-error performance comparable to a massive 256T49R MIMO system, demonstrating significant hardware savings and reduced RF chain requirements.
  • Scalability: Performance consistently improved as the number of metasurface layers increased (from 1 to 5), proving the effectiveness of the stacked architecture in providing more degrees of freedom.
  • Robustness: The system maintained reliable transmission even under complex, frequency-selective fading conditions, with the DPSIM architecture showing particular resilience.
  • Computational Speed: By adjusting the phase shifts of thousands of EM units, the SIM performs complex matrix multiplications at the speed of light with near-zero energy consumption.

Conclusion and Future Outlook

The SIM-assisted E2E OFDM system represents a paradigm shift where the physical environment acts as a computational engine. While challenges remain—such as hardware calibration for manufacturing imperfections and the initial computational overhead for training at the Base Station—the potential for energy-efficient, high-performance transceivers is clear. Future research may explore integrating advanced architectures like Transformers into the digital DNN components to further enhance scalability and performance as the number of users or subcarriers scales.

Summary updated

Abstract

Stacked intelligent metasurface (SIM) and dual-polarized SIM (DPSIM) enabled wave-domain signal processing have emerged as promising research directions for offloading baseband digital processing tasks and efficiently simplifying transceiver design. However, existing architectures are limited to employing SIM (DPSIM) for a single communication function, such as precoding or combining. To further enhance the overall performance of SIM (DPSIM)-assisted systems and achieve end-to-end (E2E) joint optimization from the transmitted bitstream to the received bitstream, we propose an SIM (DPSIM)-assisted E2E orthogonal frequency division multiplexing (OFDM) system, in which traditional communication tasks such as channel coding, modulation, precoding, combining, demodulation, and channel decoding are performed synchronously within the electromagnetic (EM) forward propagation. Furthermore, inspired by the idea of abstracting real metasurfaces as hidden layers of a neural network, we propose the EM neural network (EMNN) to enable the control of the E2E OFDM communication system. In addition, transfer learning is introduced into the model training, and a training and deployment framework for the EMNN is designed. Simulation results demonstrate that both SIM-assisted E2E OFDM systems and DPSIM-assisted E2E OFDM systems can achieve robust bitstream transmission under complex channel conditions. Our study highlights the application potential of EMNN and SIM (DPSIM)-assisted E2E OFDM systems in the design of next-generation transceivers.
BibTeX
@article{zhang2026stacked,
  author = {Yida Zhang and Qiuyan Liu and Hongtao Luo and Yuqi Xia and Qiang Wang and Fuchang Li and Xiaofeng Tao and Yuanwei Liu},
  title = {{Stacked Intelligent Metasurface for End-to-End OFDM System}},
  journal = {IEEE Trans. Wireless Communications},
  year = {2026},
  doi = {10.1109/twc.2026.3705756},
}
Figure