Generative AI, powered by large language models (LLMs), enables devices and applications to generate content such as text, images, and videos based on user input. As AI models, computing power, and applications advance, generative AI adoption is expanding across connected devices like smartphones, smart glasses, and robots. However, reliable testing methods, performance metrics, and standardized frameworks for measuring user experience are still developing. Industry groups including 3rd Generation Partnership Project and Global TD-LTE Initiative are working to address these challenges through ongoing research.
The generative AI user experience testbed, jointly demonstrated at MWC Shanghai 2026, is designed to address these challenges. In the test system architecture, the large language model is deployed on the server side, or in the cloud, while the CMX500 radio communication testerestablishes a radio connection with a smart terminal running a client-side test application. The CMX500 flexibly configures network parameters, channel fading models, latency, jitter, RF impairments and other variables, enabling realistic simulation of a wide range of wireless environments, from ideal coverage to weak-signal edge conditions. By entering prompts through the terminal-side application and receiving inference responses from the large language model, users can measure key performance indicators such as Time to First Token (TTFT) and Tokens Per Second (TPS). The resulting data is collected and analyzed using host PC software, which generates corresponding charts. This approach enables quantitative evaluation and comparison of the user experience of on-device generative AI applications under different network conditions. It can support optimization of on-device generative AI application design and provide valuable reference points for future network optimization.
At the core of this test solution is the tried and tested CMX500 5G one-box signaling tester from Rohde & Schwarz. This all-in-one test platform supports both SA and NSA architectures, covers frequency ranges from FR1 to FR3 (up to 18 GHz) and features a built-in IMS server, enabling one-stop testing for RF conformance, protocol validation, VoNR acoustics and carrier aggregation. With these capabilities, the CMX500 serves as a key test platform for 5G-Advanced and 6G research and development.
Liu Guangyi, Chief Expert of China Mobile and Chief Scientist of ZGC-XNET, said: “To realize the vision of 6G-enabled ubiquitous intelligent connectivity, mobile networks must evolve beyond connectivity alone. By harnessing native network computing capabilities, they will deliver pervasive, ultra-low-latency intelligent services. This shift also calls for new AI service metrics and testing methodologies. Together, Rohde & Schwarz and the China Mobile Research Institute have established a testbed that enables generative AI user experience testing across diverse wireless environments. Looking ahead, both parties plan to expand test metrics and application scenarios to support the evolution, validation and standardization of 6G-enabled AI technologies.”
Jin Hailiang, Senior Director of Products & Systems Division, Rohde & Schwarz China, said: “At a time of rapid advancement in large language models, the testbed jointly developed by Rohde & Schwarz and the China Mobile Research Institute sets an important benchmark for generative AI user experience testing. We are pleased to showcase this achievement together at MWC Shanghai 2026. Building on this testing architecture, both parties aim to expand test scenarios and industry use cases, helping advance the standardization of generative AI user experience testing and laying a strong technical foundation for the 6G AI era.”
Visitors to MWC Shanghai 2026 were able to experience this joint demonstration in China Mobile’s 6G Zone (Hall 3, N3E60), where experts from both organizations were available to discuss how this innovative solution can help optimize generative AI applications for future networks.
