This work proposes deploying stacked intelligent metasurface (SIM) on individual Internet of Things (IoT) devices to enhance the uplink transmission capability under a finite blocklength (FBL) regime. Aiming to maximize the achievable sum rate, a joint transmit power allocation, SIM phase shifts, and receiving beamforming design optimization problem is formulated. By decomposing the original problem into three sub-problems, reducing the intractable quadratic fraction of signal-to-interference-plus-noise ratio (SINR), the nonconvex channel dispersion function, and the constant modulus constraints to linear forms, we propose an iterative algorithm to obtain the solutions. Numerical results demonstrate that in a multi-user uplink FBL network, the incorporation of SIM yields approximately a 40% enhancement in the sum rate. The optimization of phase shifts leads to an improvement of nearly 70% in the sum rate compared to a random phase setting scheme, highlighting the crucial role of proper phase shift configuration in realizing the significant performance gains offered by SIM. The performance of the proposed algorithm is validated to be close to the slack upper bound. Furthermore, with the same total number of metasurface elements, the multi-layer SIM performs better than the traditional single-layer RIS, which reveals the advantages of multi-layer structure.

The multispectral and hyperspectral image fusion tasks aim to obtain the hyperspectral image(HSI) with a high spatial resolution. However, existing fusion methods usually utilize degradation simulation for training due to the inaccessible ground truth, which inevitably leads to spatial distortion and parameter bias during application. Besides, neglect of the prior information inherent in inputs and the deep-learning network’s low-frequency preference eventually leads to lower interpretability and limited performance. To this end, we proposed a model-embedded two-stage diffusion method(MTDiff) for unsupervised reconstruction of high spatial resolution HSI. In the first stage, the degradation model is estimated by the prior information inherent input pairs. Meanwhile, in the second stage, embedded with the degradation model, the dual-resolution diffusion model reconstructs high-resolution HSI. Specifically, treating the degradation process as a fixed diffusion step, an unsupervised paradigm is established through a mapping from upsampled low-resolution HSI to high-resolution HSI. Besides, with the estimated degradation model, the well-designed dual-resolution diffusion model step-by-step perturbs and then denoises the image in both native and degraded resolutions for a fidelity reconstruction with great interpretability. Furthermore, to establish a high-frequency shortcut for network learning, a discrete cosine injection module is designed to flatten the frequency information to a clear 2D domain with a huge high-frequency area, achieving sharp textures and clear structures in fusion results. Extensive systematic experiments across three datasets indicate the superior performance of MTDiff in multispectral and hyperspectral fusion tasks.