Comment: Revolutionising wireless systems design with AI

With the advent of Industry 4.0 and as mobile wireless technology has evolved from 3G and 4G to 5G and beyond, the complexity of wireless systems design has grown significantly. In turn, wireless networks have become increasingly difficult to manage due to the need for optimal resource sharing among a growing number of users. These challenges have forced engineers to think beyond traditional rules-based approaches, leading many to turn to artificial intelligence (AI) as the preferred solution to tackle the demand introduced by modern systems.

With its ability to manage communication between autonomous vehicles and optimise resource allocation in mobile calls, AI has become essential for modern wireless applications. As the number and range of devices connected to networks expands, the role of AI in wireless is set to grow further. Engineers must be ready to integrate it into increasingly complex systems. Understanding the advantages and current applications of AI in wireless systems, as well as best practice for optimal implementation, will be crucial for the future success of this technology.

Advantages of AI for Wireless

The shift to 5G demands ultra-reliable, low-latency and massive machine-type communication between Industry 4.0 devices. As wireless systems handle more users and apps, linear design patterns become insufficient. AI can automatically extract and solve non-linear problems beyond human-based approaches, optimising resources for various use-cases.

By using machine learning and deep learning systems to recognise patterns within communications channels, AI offers project management benefits, including the ability to quickly study a system’s dominant effect using minimal computational resources.

Incorporating simulated environments into an algorithmic model helps engineers to explore, design and carry out more iterations faster, reducing costs and development time. Without AI, running a network for disparate use cases becomes near impossible.

Mastering AI with a robust approach

The effectiveness of AI models depends on data size and quality. Synthesising new data using primitives or extracting them from over-the-air signals provides the data variability for training robust models that can handle real-world scenarios. Failure to explore a large training data set and iterate on different algorithms could lead to a narrow local optimisation.

Variability in signals for testing AI models in the field is critical. Narrow localised geography may adversely impact a design’s quality. Without field iterations, engineers cannot use the parameters of individual cases to optimise AI for specific locations, negatively impacting call performance.

Exploring potential in the wireless world – AI use cases  

Digital transformation in areas like telecommunications and automotive necessitates the use of AI and is the primary driver for its application. Placing electronic communications in areas once mechanical orientated generates large amounts of data as applications like smart cities, telecommunication networks and autonomous vehicles (AV) connect. As they do so, the resources of the network joining them become stretched.

In telecommunications, AI is deployed at two levels – at the physical layer (PHY) and above PHY. The application of AI for improving performance in a line connecting two users is referred to as operating at PHY. Applications of AI techniques to physical layers includes digital pre-distortion, channel estimation and channel resource optimisation, as well as automatic adjustments to transceiver parameters during a call otherwise known as autoencoder design.

Channel optimisation is the enhancement of the connection between two devices, notably network infrastructure and user equipment. Often, this means using AI to overcome signal variability in localised environments through techniques such as fingerprinting and channel state information compression.

AI can use fingerprinting, to optimise wireless positioning and mapping disruptions to propagation patterns in indoor environments, caused by individuals entering them. AI then estimates the user’s position based on these individualised 5G signals. Channel state information compression can also be used to compress feedback data and prevent bandwidth overload that could cause a dropped call.

Above-PHY uses involve network management and resource allocation such as scheduling, beam management and spectrum allocation. As the number of network users and use cases increase, network designers are turning to AI techniques to respond to allocation demands in real-time.

AI is also used in the automotive industry to enable safe autonomous driving. AVs interpret data from multiple sources, including LiDAR, radar, and wireless sensors, to understand their surroundings. AI enables sensor fusion, fusing competing signals to help the vehicle determine its location and how it will interact with its environment.

Overall, AI is increasingly necessary for the functionality of wireless systems such as 5G, autonomous vehicles and IoT. With expanding use cases, its importance in engineering and wireless system design will continue to rise.

Dr Houman Zarrinkoub is Principal Wireless Product Manager at MathWorks