Joint OFDM-RIS optimization in 6G: approaches and challenges

The evolution toward 6G networks poses unprecedented optimization challenges, especially in combining orthogonal frequency division multiplexing (OFDM) with reconfigurable intelligent surfaces (RIS). This type of problem, classified as mixed-integer nonlinear programming (MINLP), involves maximizing the sum rate, energy efficiency, max-min fairness, and controlling the peak-to-average power ratio (PAPR). Although recent literature —with nearly eighty works between 2021 and 2026— has explored various strategies, there is still no standardized benchmark that allows comparing results across research efforts. This lack of uniformity makes it difficult for engineers and companies to transfer academic advances to production environments. From a technological and business perspective, joint OFDM-RIS optimization can be approached through four major paradigms: model-based convex relaxation, heuristic and metaheuristic search, deep reinforcement and unsupervised learning, and emerging methods such as foundation models, diffusion-based generative artificial intelligence, and quantum optimization. Studies report that machine learning techniques (paradigm III) achieve between 95% and 99% of the theoretical spectral efficiency of model-based methods, but with inferences between 100 and 10,000 times faster —provided that pretraining costs are excluded. A particularly relevant finding for real-time system design is that GPU inference with neural networks (DDQN, PPO, GNN, unsupervised deep learning) is invariant to problem size, maintaining the same execution time for N=16 and N=128, while iterative solvers (AO+SCA, PSO) scale polynomially. This property opens the door to custom software solutions that can run on cloud infrastructures without compromising latency. However, the community points out six open challenges that hinder industrial adoption: the absence of a unified benchmark across paradigms, real hardware limitations (phase constraints, quantization, power), the need for joint waveform-RIS optimization in doubly dispersive channels, multi-objective trade-offs with PAPR, the security of large language models (LLMs) in live network control, and the diminishing performance of independent heuristics. To overcome these barriers, standardized simulation platforms that include power models and realistic waveform generators are required. In this context, technology companies can make a difference by combining their knowledge of AI for businesses with custom application development capabilities. Implementing AI agents trained for dynamic resource allocation in 6G networks requires not only efficient algorithms, but also a robust and secure cloud infrastructure. For example, AWS and Azure cloud services allow deploying reinforcement learning models with horizontal scalability, while cybersecurity ensures the integrity of training data and real-time decisions. Likewise, business intelligence, through tools such as Power BI, can monitor key network performance indicators and alert on deviations from spectral or energy efficiency targets. At Q2BSTUDIO, we understand that OFDM-RIS optimization is not just a mathematical problem, but an ecosystem where research, software development, and business strategy converge. Our experience in business intelligence services and process automation allows telecommunications operators to integrate artificial intelligence solutions with their legacy systems, reducing the time-to-market of innovations. Additionally, we offer cybersecurity consulting to protect network control loops against adversarial attacks, a critical aspect when using language models or autonomous agents to manage resources. This holistic approach is what enables academic theory to become tangible value, accelerating the transition toward 6G networks.