The partnership, which is focused on 6G research and development, has developed a method to import algorithms into design and measurement tools to validate their performance and accelerate industry adoption.
“Many network operators and vendors are already using AI and machine learning in their networks. But, so far, there has been no support from wireless standards on how AI and machine learning should be deployed,” said Javier Campos, R&D Engineer at Keysight. “This situation creates a lot of interoperability problems among equipment vendors and having a standard that isn’t optimised for AI / ML technology – this is the challenge we have been working with the university to solve.”
One of the major ways to optimise wireless performance is by providing and using accurate channel state information (CSI) – or the known channel properties and conditions of the communications link.
CSI is used in real-time throughout a wireless system to adapt transmissions as best as possible to current channel conditions with the goal of maintaining the best performance. This is a crucial component to achieving reliable communications with high data rates and multi-antenna systems.
Traditionally, calculating and reporting accurate CSI is computationally and resource intensive making it a good candidate to begin integrating AI and ML into the network.
“While AI has been around for quite a long time, the industry has now been able to identify concrete use cases, like optimising CSI feedback, where AI can deliver huge gains in performance, resource utilisation, and energy efficiency,” said Campos.
Identifying CSI as a use case that would benefit from AI and ML, wireless researchers at UMA developed an AI / ML model for CSI feedback enhancement.
“What we are doing is reducing the information you have to send over the air to provide accurate CSI from the user equipment to the base station,” explained Mari Carmen Aguayo, Professor, Universidad de Málaga and Head of the Institute of Telecommunication Research (TELMA). “We are doing this by using AI algorithms that can squeeze the information into very small quantities, sending the minimum information required to maintain the best performance.”
When it came time to validate that their ML model performed better than traditional digital signal processing (DSP) for CSI feedback, they turned to Keysight for a digital twin platform to emulate the performance of their model in real world conditions.
Building an interface layer that connected to Keysight’s PathWave System Design (SystemVue) modelling tool, UMA researchers were able to evaluate the model under a wide array of fading profiles and other test conditions to prove this method outperforms traditional approaches.
As a result of this new interface, any AI / ML algorithm that adheres to the most common AI / ML APIs and frameworks can now be imported into Keysight’s SystemVue and used by the entire industry. To assist in this process, Keysight and UMA are working together to bring this innovation to the 3GPP RAN-1 standards body.
“Together we are looking forward to progressing in the different areas that 3GPP are studying,” said Campos. “We are working to have better usability and better measurements, so we can get the insights needed to bring this new technology to the industry.”
Following the validation of the AI models, UMA and Keysight presented them and the associated measurement best practices to the 3GPP as part of Release 18, the first 3GPP release studying AI/ML enhancements for the air interface.
This ongoing partnership is continuing its work with plans to make more contributions in the future and extend these findings to even more Keysight tools to enable wireless researchers everywhere.
“What makes this collaboration special is that we were able to connect our researchers with measurement experts in the industry,” said Aguayo. “This kind of intersection yields exciting innovation for wireless communications.”
