AI-Driven Channel State Information (CSI) Extrapolation for 6G: Current Situations, Challenges, and Future Research

The acquisition of channel state information (CSI) plays a vital role in enhancing the performance of sixth-generation (6G) wireless communication systems. Conventional channel estimation approaches encounter significant scalability limitations in emerging scenarios, such as high-mobility environments, extremely large-scale multiple-input multiple-output (XL-MIMO) configurations, and multi-band operations, where pilot overhead grows dramatically. CSI extrapolation offers an effective solution to these issues by leveraging limited or partial CSI measurements to reconstruct or predict the full CSI, thereby substantially lowering the required overhead without compromising accuracy. Artificial intelligence (AI) has emerged as a powerful tool to advance CSI extrapolation, enabling more accurate and efficient inference across diverse channel conditions. Although research in this area is expanding rapidly, the literature still lacks a thorough and unified survey that synthesizes the latest developments in AI-based CSI extrapolation methods. This paper aims to bride this gap by providing the first comprehensive review of AI-driven CSI extrapolation techniques, covering their current state, key limitations, and promising research avenues. We begin by outlining the foundational aspects of AI-driven CSI extrapolation. This includes essential wireless channel properties that influence extrapolation performance and an overview of the most commonly employed AI architectures suited to this task. Building on these basics, we systematically examine the major categories of extrapolation approaches, both traditional model-based and modern AI-enhanced ones, across the primary domains: time, frequency, antenna, and multi-domain scenarios. For each category, we highlight representative techniques, their underlying principles, strengths, and limitations, along with distilled insights from comparative studies. Recognizing the strong potential of AI-based methods to satisfy the demanding performance targets of future systems, we also review publicly available open channel datasets and channel simulators that support the development and benchmarking of robust AI-driven extrapolation models. Finally, we identify persistent challenges in the field, and outline forward-looking research directions to guide future progress toward practical deployment in 6G networks.