AI-Driven Uplink Optimization for Next-Generation RAN

Keysight Technologies and MediaTek have developed a working prototype that advances AI-driven uplink optimization and AI model lifecycle management for next-generation radio access networks (RAN). The system will be demonstrated at Mobile World Congress 2026 and illustrates how RAN-assisted intelligence can improve uplink throughput in real time while maintaining performance through controlled retraining and over-the-air (OTA) model updates.

Addressing Uplink Variability in Evolving Networks
As mobile networks expand across heterogeneous environments—urban, rural, indoor and hard-to-reach areas—uplink performance becomes increasingly difficult to maintain consistently. Variations in propagation conditions, interference levels and user mobility affect spectral efficiency and reliability.

Conventional transmitter diversity and uplink optimization techniques are typically static or semi-static, limiting their ability to adapt dynamically to changing radio conditions. In addition, AI models trained in one deployment scenario may degrade when applied to different site configurations or traffic profiles. Updating models in live networks introduces operational complexity and risk.

Operators require mechanisms that enable site-aware AI optimization while maintaining manageable operational workflows.

RAN-Assisted AI and Model Lifecycle Management
The joint prototype integrates RAN-assisted AI decision-making with site-specific retraining and OTA model updates. In this architecture, the RAN provides contextual network information to support AI-driven uplink optimization decisions, enabling adaptive control of transmission parameters in response to real-time channel conditions.

To address model drift and environment-specific behavior, the solution incorporates controlled retraining processes tailored to individual deployment scenarios. Updated models can then be distributed via OTA mechanisms, reducing manual intervention and enabling scalable model lifecycle management.

This closed-loop approach supports sustained AI effectiveness over time rather than one-time model deployment.

High-Fidelity Emulation for Measurable Validation
The prototype leverages Keysight’s Channel Studio RaySim and Network and Channel Emulation solutions to replicate realistic, repeatable radio conditions in a controlled laboratory environment. By emulating diverse propagation scenarios, interference patterns and mobility conditions, engineers can validate AI performance before field deployment.

The controlled test environment enables quantifiable assessment of uplink throughput, spectral efficiency and reliability improvements. It also allows evaluation of retraining workflows and OTA update processes without impacting live networks.

Alignment with AI-RAN Objectives
The work aligns with the objectives of the AI RAN Alliance, which promotes scalable integration of AI into RAN architectures. By combining AI-assisted radio optimization with practical model management mechanisms, the prototype demonstrates a pathway toward continuous performance optimization in evolving 5G-Advanced and future 6G networks.

Through high-fidelity emulation and lifecycle-aware AI deployment, the collaboration illustrates how operators can accelerate adoption of intelligent RAN technologies while maintaining operational stability and repeatability.