Benchmark AI Performance in 6G Wireless Systems
Wireless engineers are working to embed artificial intelligence (AI) and machine learning (ML) into 5G and 6G networks. To deploy AI into real world systems, it must have measurable performance gains over traditional methods to justify the investment in new hardware and software. Engineers need to benchmark their AI both in simulation and in real-time testbeds to show its performance under practical deployment conditions. To compare the performance of AI to traditional methods, engineers need:
- A way to collect concrete, measured performance data of AI algorithms
- Access to a modifiable real-time wireless link to show how AI performs at a system level
- A testbed that can easily switch between the traditional and AI algorithm
AI/ML Performance Benchmarking Testbed
- A real-time base station (gNB) emulator and a real-time user equipment (UE) emulator run on standard servers using OpenAirInterface (OAI) 5G NR software
- NI Ettus USRP Software Defined Radio (SDR) Devices act as the RF front end to create an over-the-air testbed, or they can be connected to a channel emulator to change the signal-to-noise and run fading channel models
- The OAI gNB software can switch between running a traditional algorithm or an ML model in TensorFlow™
- Specific validation scenarios can be run with both the traditional and ML models to collect performance data for how each method performs under definable, repeatable channel conditions
- Pairs with the RF Data Recording API to create a complete testing setup for AI research from training models to measuring AI enhanced networks system-level performance
Solution Advantages
-
A cost-effective testbed for AI/ML research that runs an open source 5G NR-compliant software stack
-
Configure the system to run with a channel emulator or over-the-air to prototype a wide variety of realistic network conditions
-
Modifiable and flexible to benchmark AI/ML models in the physical layer as part of a real-time network
Watch How to Measure the Performance of an AI Model
Neural Receiver Example for 6G Networks
NI’s SDR hardware can be used to build a testbed for measuring the performance of an AI model inside a real-time wireless network. Learn how this testbed can be used to measure the performance of a neural receiver compared to traditional methods in this demonstration.
BUILD YOUR SOLUTION WITH THE NI ECOSYSTEM
NI offers a variety of solution integration options customized to your application-specific requirements. You can use your own internal integration teams for full system control or leverage the expertise of NI and our worldwide NI Partner Network to obtain a turnkey solution.
The NI Partner Network is a global community of domain, application, and overall test development experts working closely with NI to meet the needs of the engineering community. NI Partners are trusted solution providers, systems integrators, consultants, product developers, and services and sales channel experts skilled across a wide range of industries and application areas.
NI works with customers throughout the life of an application, delivering training, technical support, consultation and integration services, and maintenance programs. Teams can discover new skills by participating in NI-specific and geographic user groups and build proficiency with online and in-person training.
AI/ML Performance Benchmarking Webinar
Learn more about the AI/ML Performance Benchmarking Testbed. This webinar shows how the testbed is configured and goes through the neural receiver example use case in detail. It discusses the system setup, results, and key learnings from their research.
TensorFlow, the TensorFlow logo, and any related marks are trademarks of Google Inc.