Keysight Technologies and MediaTek collaborate to enhance AI-powered uplink performance, improving network efficiency, latency, and next-generation 5G connectivity
Keysight Technologies, Inc. and MediaTek have unveiled a working prototype that marks a significant step forward in AI-driven uplink optimization and model life cycle management for next-generation Radio Access Networks (RAN). The joint innovation, which will be showcased at Mobile World Congress 2026, demonstrates how artificial intelligence can be embedded more deeply and intelligently into wireless infrastructure to enhance real-time uplink performance while ensuring long-term adaptability through structured retraining and over-the-air (OTA) updates.
As mobile networks continue to evolve toward AI-native architectures, the importance of uplink performance has grown substantially. Historically, much of the focus in cellular development centered on downlink speeds to support content consumption. However, the rise of AI-driven applications, augmented reality (AR), immersive communications, cloud collaboration tools, and user-generated content platforms has shifted the balance. Modern networks must now support higher uplink throughput, lower latency, and more consistent reliability across diverse operating environments.
Maintaining consistent uplink performance presents considerable technical challenges. Coverage conditions vary dramatically across urban high-density centers, suburban neighborhoods, rural expanses, indoor enterprise spaces, industrial facilities, and other hard-to-reach areas. Signal reflections, building materials, interference patterns, mobility conditions, and device orientation all influence transmission quality. Traditional transmitter diversity techniques—often based on static configurations—struggle to dynamically adapt to these constantly changing variables. As a result, uplink efficiency and reliability can degrade when real-world conditions deviate from assumed design scenarios.
Compounding this challenge is the growing use of AI models within network systems. AI-driven transmit optimization can significantly improve performance, but models trained in one environment may not generalize effectively to others. A model developed using urban training data, for example, might not perform optimally in rural or indoor industrial deployments. Updating AI models in live networks also introduces operational complexity. Operators must ensure accuracy, avoid service disruptions, validate performance gains, and maintain compliance with evolving standards—all while managing large-scale deployments across multiple geographies.
To address these challenges, Keysight and MediaTek have combined their respective expertise in wireless test solutions and advanced chipset design to create a prototype that integrates RAN-assisted AI decision-making, site-specific model retraining, and OTA model updates within a controlled yet realistic validation environment.
At the core of the solution is a tightly coordinated system that allows the RAN to assist AI-based decision processes at the device level. This approach enables uplink transmit diversity decisions to adapt dynamically to real-time channel conditions. Instead of relying on fixed transmitter configurations, the AI-enhanced model can analyze contextual data—such as channel state information, interference conditions, and propagation characteristics—and adjust transmission parameters accordingly. The result is measurable improvement in uplink throughput, spectral efficiency, and reliability.
To ensure that these gains are sustainable over time, the prototype incorporates practical model life cycle management. Site-specific retraining allows operators to refine AI models based on localized deployment conditions. By capturing environment-specific data and feeding it back into the training pipeline, the system ensures that models remain accurate and optimized for their intended operating context. Once improved models are validated, OTA updates enable seamless deployment across devices in the field without requiring physical intervention.
Keysight’s advanced test and emulation infrastructure plays a central role in validating this process. The prototype leverages Channel Studio RaySim and its Network and Channel Emulation solutions to create realistic, repeatable scenarios that closely mimic real-world network behavior. These tools incorporate high-fidelity RF digital twin capabilities, including 3D ray tracing, to simulate complex propagation environments. By generating precise training datasets and enabling controlled testing conditions, the system ensures that AI models are evaluated rigorously before live deployment.
This methodology allows measurable performance improvements to be demonstrated in a controlled environment while minimizing the operational risk associated with AI model updates. Engineers can validate uplink enhancements under varied conditions, compare baseline and AI-enhanced performance, and quantify improvements in spectral efficiency and reliability. Such repeatable validation is essential for operator confidence, particularly as AI becomes increasingly embedded in critical network functions.
The collaboration aligns closely with the broader goals of the AI-RAN Alliance, which promotes scalable and interoperable AI integration into radio access networks. By combining open infrastructure principles with coordinated device-network intelligence, the prototype demonstrates how AI-native RAN deployments can be implemented practically rather than theoretically. It underscores the importance of ecosystem cooperation, standardized interfaces, and test-driven validation in accelerating AI adoption across global mobile networks.
Mingxi Fan, Senior General Manager of Wireless Technology Group at MediaTek, emphasized the growing importance of uplink enhancement in scaling next-generation mobile networks. As AI-era applications place increasing demands on real-time data transmission from devices to the cloud, transmit diversity techniques implemented at the device level become increasingly valuable. Through collaboration with Keysight, MediaTek demonstrates how open infrastructure and coordinated device-network cooperation can transform advanced AI concepts into deployable, scalable solutions.
Peng Cao, Vice President and General Manager of the Wireless Test Group at Keysight, highlighted the novelty of the approach, describing it as “AI-enabling-AI.” By training and testing AI-enhanced transmitter models within the user equipment (UE) using high-fidelity datasets generated from RF digital twin channel emulation, the system ensures both precision and adaptability. The ability to generate realistic training data with 3D ray tracing provides a robust foundation for model accuracy and consistency across varied deployment scenarios.
Beyond the immediate performance gains, the broader significance of this development lies in its contribution to AI-native network transformation. As 5G evolves and research into 6G accelerates, networks are expected to become increasingly software-defined, intelligent, and autonomous. AI will not simply optimize isolated functions but will orchestrate network behavior holistically. For this vision to materialize, robust validation frameworks, retraining pipelines, and lifecycle management strategies are essential.
By demonstrating real-time uplink optimization combined with practical model maintenance and OTA updates, Keysight and MediaTek provide a blueprint for how AI can be operationalized in RAN environments at scale. The prototype illustrates that AI-driven optimization does not need to introduce operational instability. Instead, when paired with rigorous test environments and structured update mechanisms, it can deliver continuous improvement without sacrificing reliability.
As operators worldwide prepare for the next phase of network evolution, solutions that address both performance enhancement and lifecycle sustainability will be critical. The collaboration between Keysight and MediaTek signals a meaningful advancement toward AI-native RAN deployments that are accurate, site-aware, scalable, and manageable. Their demonstration at Mobile World Congress 2026 offers a practical glimpse into how future mobile networks can intelligently adapt in real time while continuously learning and improving—ultimately delivering stronger, more consistent uplink performance for the AI-driven applications of tomorrow.
