Nokia and Qualcomm Research AI-Interoperability Technology

Nokia and Qualcomm Research AI-Interoperability Technology

Qualcomm and Nokia developed a prototype technology that could solve the problem of interoperability, one of the fundamental challenges facing AI’s use in wireless systems. By enabling multi-vendor interoperability between the different AI models used in networks and devices, future wireless systems would support the use of AI to achieve greater capacities and better performance while potentially reducing their energy consumption.

Nokia Bell Labs and Qualcomm have implemented this technology into an AI proof-of-concept using a Nokia Bell Labs test network and a Snapdragon 5G modem-RF system. The two companies will be demonstrating this proof-of-concept at Mobile World Congress 2024 in Barcelona.

AI and machine learning will have a significant impact on wireless communications in the future as AI models can optimize a radio’s performance to its specific surroundings. But for AI to reach its full potential in the radio network, different AI systems utilized by networks and devices need to coordinate their activities. Otherwise, these AI systems could work at cross purposes.

Qualcomm and Nokia have taken a new approach to AI interoperability that utilizes a technique called sequential learning. Using sequential learning, the network and device share data about the connection relevant to AI training but do not have to share the actual AI model. Then independent encoders and decoders use that data to train the network AI and device AI separately. These independent encoders and decoders effectively act as translators, allowing multiple vastly different AI systems to share their knowledge. Through sequential training, these various AI models would reinforce one another, regardless of vendor. Additionally, sequential training can also enable wireless handsets or devices from multiple vendors, each with their own encoder implementation, to interwork with the same base station with its common decoder in the network. This saves time in training and helps scalability for future real-world deployments.