Codebook Configuration for RIS-aided Systems via Implicit Neural Representations

TL;DR

An implicit relationship between user's coordinates information and the codebook from the perspective of signal radiation mechanisms is formed, and a novel learning-based method, implicit neural representations (INRs), is introduced to solve this implicit coordinates-to-codebook mapping problem.

Abstract

Reconfigurable Intelligent Surface (RIS) is envisioned to be an enabling technique in 6G wireless communications. By configuring the reflection beamforming codebook, RIS focuses signals on target receivers to enhance signal strength. In this paper, we investigate the codebook configuration for RIS-aided communication systems. We formulate an implicit relationship between user's coordinates information and the codebook from the perspective of signal radiation mechanisms, and introduce a novel learning-based method, implicit neural representations (INRs), to solve this implicit coordinates-to-codebook mapping problem. Our approach requires only user's coordinates, avoiding reliance on channel models. Additionally, given the significant practical applications of the 1-bit RIS, we formulate the 1-bit codebook configuration as a multi-label classification problem, and propose an encoding strategy for 1-bit RIS to reduce the codebook dimension, thereby improving learning efficiency. Experimental results from simulations and measured data demonstrate significant advantages of our method.

Figures (6)

• Figure 1: A practical scenario of RIS-aided system. Our objective is to configure the optimal codebook solely based on the user coordinates.
• Figure 2: (a) The similarity between the NVS and the codebook configuration. (b) The architecture of INR. It includes six layers: the input layer, four hidden layers, and the output layer, with dimensions of $6L$, 128, 128, 128, 128 and $M+N-1$, respectively.
• Figure 3: An illustration of the proposed encoding scheme: a $3 \times 3$ 1-bit RIS example. Specifically, we employ diagonal matrices $diag\left ( I_{1}, I_{2}, I_{3}\right )$ and $diag\left ( I_{4}, I_{5}, I_{6}\right )$ to represent row and column traversal, respectively. We find that both encoding mode(1): $[+1,-1,+1,-1,+1,+1]$ and mode(2): $[-1,+1,-1,+1,-1,-1]$ yield the same optimal codebooks. To further reduce the dimension, XOR operation is applied between the first bit of "+1" and the remaining 5 bits of $[-1,+1,-1,+1,+1]$, the same operation applies to the encode mode(2). In the end, encoding result of $[+1,-1,+1,-1,-1]$ can uniquely represent the optimal codebooks for $3 \times 3$ 1-bit RIS. That is, the predicted codebook dimension decreases from $M\times N$ to $M+N-1$ and finally to $M+N-1$.
• Figure 4: The framework includes two parts: the training step and the inference step. In the training step, INR model is trained on a dataset comprising user coordinates and the encoded codebook pairs. In the reference step, input solely the user coordinates to produce the optimal codebook.
• Figure 5: (a) The simulation scenario. (b) The experimental scenario.