Testing Black Boxes: Verification and Validation of AI Models in Wireless Communication Systems

Fifth generation (5G) cellular communication system provides a flexible system architecture that enables wireless communication technology to penetrate different verticals and connect various devices. Applying machine learning to develop AI models applicable to various application scenarios becomes a promising direction for designing 5G systems. However, developing AI models for wireless communication algorithms brings new challenges to system design and verification. This paper provides an overview of studies on the verification and testing of wireless communication algorithms developed by machine learning methodology. Due to the black-box nature and scenario-specified data-based model development of the machine learning approach, standardization of testing and verification procedures becomes a new challenge when the device under test implements AI models. We explain the general issues in the context of use cases studied by 3GPP standardization group to illustrate how the application of machine learning to wireless communication systems design emerging as a unique problem from the testing and verification perspective. We then summarize the challenges to develop effective testing and verification procedures. Through explaining the issues and challenges, this paper aims to provide a framework for future studies on resolutions to enable wireless communication AI model testing and verification.