Assessing Air-Interface Dataset Similarity and Diversity for AI-Enabled Wireless Communications

Ahstract- The integration of artificial intelligence (AI) into wireless communication systems is set to profoundly transform the design and optimization of emerging sixth-generation (6G) networks. The success of AI-driven wireless systems hinges on the quality of the air-interface data, which is fundamental to the performance of AI algorithms. Within data quality assessment (DQA), the measurement of similarity and diversity stands as crucial. Similarity assesses the consistency of datasets in mirroring their intrinsic statistical properties, which is essential for AI model accuracy. In contrast, diversity relates to the models' ability to generalize across various contexts. This paper concentrates on these aspects of DQA and proposes a comprehensive framework for analyzing similarity and diversity in wireless air-interface data. Catering to various data types, such as channel state information (CSI), signal-to-interference-plus-noise ratio (SINR), and reference signal received power (RSRP), the framework is validated using CSI data. Through this validation, we demonstrate the framework's efficacy in improving CSI compression and recovery in Massive Multiple-Input Multiple-Output (MIMO) systems, highlighting its significance and versatility in complex wireless network environments.