- At MWC Shanghai 2026, Rohde & Schwarz and the China Mobile Research Institute showcased a joint test setup dedicated to generative artificial intelligence (AI) applications.
- This solution enables the quantitative measurement of end-user experience when using generative AI applications across various wireless network conditions.
The use of generative AI services—which rely on Large Language Models (LLMs) to create text, images, or video—is growing across connected devices such as smartphones, smart glasses, and robots. As the telecommunications industry currently lacks standardized testing frameworks or mature metrics to evaluate the user experience of these applications, organizations such as 3GPP and GTI are actively conducting research in this area.
The test system presented in Shanghai addresses this gap by measuring specific Key Performance Indicators (KPIs), notably Time to First Token (TTFT)—the delay before the first unit of generated text appears—and Tokens Per Second (TPS), which measures the generation speed of the text data stream.
Testbed Technical Architecture
In this testbed configuration, the Large Language Model is hosted on a server or in the cloud, while a device runs a test application on the client side. The core of the solution is the Rohde & Schwarz CMX500 one-box signaling tester, which establishes the radio link with the device.
This equipment allows for the configuration of variable network parameters, such as channel fading models, latency, jitter, and radio frequency (RF) impairments. This approach enables the simulation of real-world wireless communication environments, ranging from nominal coverage to degraded conditions at the cell edge. Prompts are entered via the terminal, and the LLM’s inference responses are collected and analyzed by software on a host PC to generate comparative graphs.
Hardware Platform Evolution
In terms of hardware, the CMX500 series test platform supports both 5G Standalone (SA) and Non-Standalone (NSA) communication architectures. Its frequency spectrum spans from FR1 to FR3 bands, reaching up to 18 GHz. Equipped with an integrated IMS server, the platform centralizes RF conformance testing, protocol validation, VoNR acoustics, and carrier aggregation, serving as a foundation for 5G-Advanced technology development and 6G research.
Looking ahead, the two partners plan to expand the range of metrics and application scenarios to support the validation and standardization of future communication technologies that natively integrate artificial intelligence capabilities.
