How the autonomous vehicle shapes our future wireless networks

Over the past decade, engineers across the automotive ecosystem have invested countless hours and capital in trying to teach cars to drive. Achieving this goal could mean an end to traffic accidents and fatalities and recouping the wasted time we spend in traffic. But teaching a machine to operate with greater safety, precision, and experience than human drivers in an endless number of road, road and weather conditions is an incredibly difficult problem.

Artificial intelligence (AI) technologies, leveraging vast improvements in the cost and availability of massive computing and data storage and processing, make possible incredibly difficult challenges like driving a car. Today, as the complexity of our wireless networks faces a dramatic increase in complexity, engineers designing future wireless networks, like 6G, are rushing to apply AI technologies and techniques. similar design.

  • Design requirements and results: For today’s classically designed systems, the software in our networks encodes the functionality described in today’s wireless standards. For ML-driven cellular networks, we will need to move from a design that conforms to a set of human-written rules to a set of defined outcomes. In the automotive world, software and AI engineers define a set of constraints for their vehicles, called operational domain definitions (ODDs). For example, an autonomous vehicle might only travel on US highways during the day. They then train algorithms to optimize a specific goal within that SDG, such as minimizing the likelihood of an accident for a variety of different scenarios the vehicle might encounter, which might require the trade-off of breaking traffic rules or passenger comfort to optimize safety. As part of research into using AI for our cellular networks, we must have a conceptually similar set of domain goals and outcomes. For example, do we want our AI algorithms to optimize total throughput? Or the fastest response times? Or the reliability of the network? Or minimal power consumption?
  • Training data: To create AI algorithms, engineers use large data sets from a variety of real-world scenarios to teach or train the network – much like how humans learn tasks. In each data set, for a set of input conditions, we tell the algorithm what we think is the correct result. By introducing millions of different scenarios into the training phase, each with a correct result, we hope to teach the algorithm how to react. Today, automakers have spent hundreds of millions of dollars to acquire and label data sets for vehicles. Ultimately, the quality of the AI ​​algorithm strongly correlates with the quality of the datasets used to train it. The data sets needed to form wireless networks are still in their infancy. Cellular infrastructure companies with large deployed facilities can use their existing equipment to collect datasets for their own use. For small businesses, the National Science Foundation (NSF) is funding initiatives like RFDataFactory to create tools that will automate the generation and management of new community datasets for wireless research.
  • Open network software: To develop application-specific machine learning technology, it is essential to have open software platforms to experiment with. In the automotive industry, new research often begins with software like the open source Robot Operating System (ROS) or more focused, but more comprehensive, proprietary software stacks like the NVIDIA DRIVE AGX Pegasus platform. In wireless networks, the software ecosystem is, again, at an early stage. Open source tools like OpenAirInterface (OAI) software are evolving rapidly to gain the performance needed to be useful in real testing. As more proprietary and comprehensive software stacks begin to become available, tools are still fragmented and will need to evolve before achieving the ability to prototype and test the equivalent of a Layer 3 or higher stand-alone network.
  • Validation / test methods: The test of algorithms and ML systems is more complex than the test of conventionally programmed systems (conformity test). For autonomous vehicles, rapid advancements in software-in-the-loop, model-in-the-loop and hardware-in-the-loop methods have become crucial for testing these vehicles in the laboratory for the infinite set of road conditions. which they’ll face before these vehicles see a real-world test drive. Similar SIL, MIL and HIL technologies will also need to evolve for telecommunications testing. For example, Northeastern University recently demonstrated the use of the world’s largest RF network emulator, originally developed for DARPA, to test a new AI-based cellular network. Technology.

As our cellular networks continue to grow in their capabilities and complexities, engineers will increasingly turn to AI technologies. In many ways, network engineers will face design and validation challenges similar to those encountered by automotive engineers over the past decade. And while automotive engineers typically meet at different forums and conferences than wireless engineers, there is a substantial mutual benefit in these groups working together as they develop, validate, and implement. the potential of AI systems.

Comments are closed.