ALLPCB
Overview
Over the past two decades, mobile wireless technology has evolved from 3G to 4G and then to 5G, enabling advanced applications such as Industrial 4.0. Wireless system design has grown more complex. To address these challenges, many engineers are applying AI (artificial intelligence) capabilities to modern wireless systems.
AI Use Cases in Wireless Systems
The evolution to 5G has been driven by three main requirements: improved broadband speed and quality, ultra-reliable low-latency communications, and time-sensitive communications between devices in industrial environments.
Growing user populations and an increasing number of devices competing for network resources have made wireless systems more complex. Traditional rule-based linear design approaches and classical data processing methods are increasingly inadequate. AI techniques can efficiently model nonlinear problems automatically, addressing problems beyond the reach of manual methods.
Machine learning and deep learning models can enable wireless systems to learn channel models. Systems that incorporate these AI models can improve performance by optimizing link resource allocation. Without AI-based strategies, modern networks with many applications competing for the same resources become difficult or impossible to manage effectively.
AI also accelerates development through model reduction and similar techniques. Engineers can estimate and simulate network environment behavior and directly incorporate it into algorithm models, enabling fast exploration of the factors that most affect system performance while using minimal compute resources. This reduces iteration time and can lower development time and cost.
Best Practices for Deploying AI in Wireless Systems
Data quality is critical for successful AI deployment. AI models require comprehensive data for training to handle the range of real-world scenarios. Tools such as MathWorks 5G Toolbox can generate synthetic data from data primitives or collect data over the air, providing the diversity needed to train robust AI models. If engineers iterate algorithms only on limited datasets, trained models are more likely to converge to local optima rather than a desirable global optimum, reducing reliability in real environments.
Robust and thorough testing of AI in realistic scenarios is equally important. If test data cover only a narrow set of situations, a lack of diversity can produce negative effects during design and optimization. Without comprehensive field data, AI models may fail in edge cases, degrading wireless system performance.
AI Adoption Across Industries
Digital transformation across industries, from telecommunications to automotive, is increasing the need for AI and driving adoption. Connected applications such as smart homes, telecom networks, and autonomous vehicles generate large volumes of sensor data. This data supports AI development and accelerates industry digitization, while also creating additional strain on network resources.
Telecommunications
In telecommunications, AI is being applied from the physical layer upward. Improving call performance typically involves physical-layer operations. Physical-layer AI applications include digital predistortion, channel estimation, and channel resource optimization, as well as automated encoder design that adjusts transceiver parameters during calls.
Channel optimization improves links between devices such as base stations and user equipment. Through fingerprinting and channel state information (CSI) compression, AI can help mitigate local variations in signal propagation.
Fingerprinting techniques are used to improve wireless network localization. In a stable indoor electromagnetic environment, the presence of a target at different locations causes characteristic perturbations in propagation. AI can estimate a target location from collected personalized signal variations. Compared with classical approaches that match received signal strength to a database, AI can overcome many factors that reduce localization accuracy. AI-based CSI compression helps keep feedback from user equipment to base stations within available bandwidth, preventing feedback from exceeding capacity and impacting calls.
Above the physical layer, AI focuses on resource allocation and network management. As the number of users and use cases grows, network engineers are using AI to respond to resource allocation requests in real time. For competing users and applications, AI-based beam management and spectrum allocation functions can optimize core resource management.
Automotive
In the automotive sector, wireless connectivity enabled by AI helps enable safer autonomous driving. Autonomous driving and V2V/V2X rely on multiple data sources, including LiDAR, radar, and wireless sensors, to interpret the environment. Vehicle hardware must manage competing sensor data to operate effectively. AI sensor fusion merges these inputs to provide a comprehensive understanding of the vehicle and its surroundings and to determine how they interact. This gives vehicles 360-degree awareness, enabling detection of other vehicles and potential collision threats. Whether alerting a driver or making autonomous control decisions, AI participation reduces accidents and improves road safety.
Conclusion
As wireless use cases continue to expand, the need to deploy AI in wireless systems is also growing. From 5G to autonomous vehicles and the Internet of Things, many modern applications would struggle to operate effectively without AI. AI's role in wireless system design has risen rapidly in recent years, and this trend is expected to continue as network use cases and user counts increase.
