Smart SIM project evaluates an FFT-based direction-of-arrival DoA estimation algorithm with reinforcement learning on the waveform domain.
SIM (Stacked Intelligent Metasurface) constitutes a MATLAB-based framework that executes the 2D Discrete Fourier Transforms (2D-DFT) explicitly within the waveform domain. The repository replicates and advances the findings of reference [1] by incorporating a reinforcement learning (RL) component, also via a SIM. This approach is designed to efficiently localize multiple users within an indoor environment, facilitated by the RL agent.
The system archicture is represented in Fig. 1, where SIM 1 develops the 2D-DFT, and its output is passed to the SIM 2 that estimates the electric angles of arrival. The architecture operates as follows:
- Mobile users transmit baseband single-carrier pulses modeled as
$\boldsymbol{a}(\psi_x,\psi_y)\times s$ over the symbol time-interval$T_s$ , where$s$ is constant, and$\boldsymbol{u}(\psi_x,\psi_y)$ is a vector that represents the Kronecker product of the spatial sequences$e^{j\psi_x(n_x-1)}$ and$e^{j\psi_y(n_y-1)}$ , see [1, Eq. (3)-(6)]. The complex exponential indicates the phase introduced by the users' spatial positions, determined by the electric angles$\psi_x$ and$\psi_y$ , and indices$n_x$ and$n_y$ referring to the first layer in SIM 1. - The received signal at the first layer of the SIM 1 is modeled with a clustered-delay-line (CDL) channel. Specifically, to model indoor scenarios in industrial environments.
- The SIM 1 evaluates the 2D-DFT of the emitted signals by the mobile users.
Its toput evaluates magnitude the of the 2D-DFT and its peaks signals the coordinates of the electrica angles in the
$x$ and$y$ axis. See an example in Fig. 2, as the output produced by SIM 1. The SIM 1 operates as indicated in [1, Sec. III], where three main parameters are defined$N$ , which is the number of elements in the first layer, and$T$ , which is the total number of time slots where the 2D-DFT is evaluated. - The SIM 2 is interconnected to the output of the SIM 1 and develops a fullly-connected layer neural network (NN).
The ouput of SIM 2 provides the estimated angles
$\psi_x$ and$\psi_y$ of the peak in the 2D-DFT plain. That is the values for$\hat\psi_x$ and$\hat\psi_y$ .
Fig. 1: Representation of the system model with the mobile user (MU) and the two SIMs.
Fig. 2: 2D-DFT output as derived at the output of the SIM 1.
This code is tested in MATLAB 2025a, and the required toolboxes are listed in the table below.
| Matlab Toolbox | Version |
|---|---|
| Signal Processing Toolbox | 25.1 |
| Communications Toolbox | 25.1 |
| Phased Array System Toolbox | 25.2 |
| WLAN Toolbox | 25.1 |
| 5G Toolbox | 25.2 |
This project directly runs from the Matlab accesible on each folder.
SIM/
├── DoA/ # Includes the code to estimate the electric angles
├── Channel Model/ #This folder includes a clustered-delay-line (CDL) model for the received signal, which mimics indoor industrial scenarios.
- Modeling the Wave-Domain Computing: The stacked metasurface performs computation as EM waves propagate, without digital hardware.
- DoA Estimation: This code develops a direct mapping between direction-of-arrival and SIM output intensities.
Interested contributors can contact the project owners. Please refer to the Contact Information below. We identify further developments for more complex scenarios like estimating the distance to multiple cancer cells.
This project was supported in part by the Federal Ministry of Education and Research (BMBF, Germany) within the 6G Research and Innovation Cluster 6G-RIC under Grant 16KISK020K..
[1]: J. An et al., "Two-Dimensional Direction-of-Arrival Estimation Using Stacked Intelligent Metasurfaces," in IEEE Journal on Selected Areas in Communications, vol. 42, no. 10, pp. 2786-2802, Oct. 2024. Link
-
Name: Jorge Torres Gómez
-
Name: Karel Toledo de la Garza
