In today's ultra-connected world, mobile networks are indispensable to both our professional and personal lives. With continuous technological innovations, such as those ushered in the latest generation of mobile communications, the 5th Generation (5G), a multitude of new services and applications are continuously emerging. However, as the demand for mobile networks soars, networks are becoming increasingly complex, posing growing challenges to operators in managing their networks. Left unmanaged, this could entail rising costs and greater efforts in managing network performance, ultimately impacting user experience and customer satisfaction.
Researchers and industry professionals are therefore turning to the transformative power of artificial intelligence (AI) to address this challenge. By leveraging AI, novel approaches are being developed to navigate the complexities of mobile networks and unlock higher levels of automation and innovation. This blog explores compelling examples of how AI can (and will) revolutionize mobile networks.
Path loss modeling in mobile networks
Path loss refers to wireless signal strength reduction along the path from the transmitter to the receiver. Factors such as distance, antenna height, obstacles, and radio environment conditions all contribute to path loss. Accurate path loss modeling is crucial for planning and optimizing mobile networks, as it helps to predict radio coverage, interference levels, and overall network performance.
Traditionally, path loss modeling in mobile networks has relied on empirical methods based on extensive field measurements. The research department of Celfinet, a Cyient company, which has been doing applied research on path loss modeling and antenna models, published its findings in the article Analysis and Optimization of 5G Coverage Predictions Using a Beamforming Antenna Model and Real Drive Test Measurements in the international journal IEEE Access. Empirical path loss models are often calibrated with new measurements for specific areas to increase their path loss prediction accuracy. However, this approach involves a trade-off between the model's geographical applicability and prediction accuracy. While it can provide higher path loss prediction accuracy within a specific area, the model may not be suitable for other geographical regions, limiting its broader application.
The power of satellite images and deep neural networks
To tackle the challenges associated with geographical generalization and prediction accuracy in path loss modeling, Celfinet conducted research to combine the use of satellite images and deep neural networks. A noteworthy aspect of this research involved adopting a self-supervised learning approach, enabling the autonomous extraction of relevant features from satellite images. This extracted data was then combined with drive test measurements, ultimately leading to the development of a novel model: the Ubiquitous Satellite Aided Radio Propagation (USARP).
The evaluation of the USARP model focused on its accuracy in predicting path loss and its ability to generalize geographically. It was compared with empirical models (calibrated and uncalibrated) and with machine learning-based models. The results demonstrated that the USARP model surpassed all baseline models by achieving higher accuracy in path loss prediction and displaying improved generalization capabilities. This novel approach not only enhances geographical generalization but also exhibits versatility across diverse scenarios, even surpassing environment-specific empirical models in further experiments. By combining satellite images, self-supervised learning, and deep neural networks, the USARP model represents an advancement in path loss modeling, offering significant potential for enhancing network planning and optimization in mobile networks. The research leading to the development of the USARP model has been published in the journal IEEE Access as An Ubiquitous 2.6 GHz Radio Propagation Model for Wireless Networks Using Self-Supervised Learning From Satellite Images.
Maximizing AI effectiveness and performance improvement in mobile networks
The notion of causality is considered essential for portraying general intelligence as it involves comprehending cause-and-effect relationships between events or variables. It enables outcome predictions, informed decision-making, and intervention for desired results—critical skills indeed in any domain! Therefore, an interest in integrating causality into AI has been growing, aiming at incorporating concepts and methods that allow for understanding and reasoning about cause-and-effect relationships within AI systems. By embodying a causal foundation, AI-based systems have the potential to enhance their prediction robustness, enabling them to maintain accuracy across contexts and scenarios. This paradigm will facilitate the development of AI systems with reusable mechanisms, allowing for adaptation to new environments or tasks. In fact, the development of such capabilities, particularly in generalization and scalability, are fundamental challenges for an AI-native generation of wireless networks such as 6G.
In the context of mobile networks, exploring causality in network performance and configuration becomes extremely appealing for developing robust automation mechanisms for network operation, management, and optimization. In this regard, devising data-driven and causal-based approaches that leverage AI techniques to identify well-grounded network performance optimizations holds great promise for mobile operators. Cyient’s Celfinet has been actively exploring these topics through its research. Causal-based approaches for network performance optimization should evaluate existing network performance patterns and determine possible causal associations with explanatory factors for the observed performance patterns. The identified causal associations enable the implementation of targeted network optimization actions to enhance overall network performance.
The way forward
The increasing demand for mobile networks, combined with their growing complexity, necessitates innovative solutions to manage and optimize network performance. AI emerges as a transformative tool in this context, empowering network operators with advanced automation and innovation capabilities. AI holds significant importance, whether by providing novel approaches to well-known problems like path loss modeling or establishing the groundwork for innovative approaches in network optimization. Celfinet, a Cyient company, seeks to continue exploring these advancements, paving the way for a future of highly efficient and optimized mobile networks. As AI continues to evolve, its integration with mobile wireless networks holds tremendous potential for maximizing effectiveness and driving significant performance improvements.
About the author
Marco Sousa is a research lead at Celfinet, a Cyient-owned company, working for the past eight years in its research department. He specializes in AI and its application to mobile networks. With a deep understanding of mobile communications and a strong background in AI, Marco leverages his expertise to drive innovation in the field. He recently earned his Ph.D. at Instituto Superior Técnico (IST), Lisbon, with research work focused on AI and mobile networks.