AI in Wireless
Artificial Intelligence and Machine Learning (AI/ML) have upended the technology scene over the last two decades, with astonishing progress being achieved in fields such as pattern recognition and computer vision. The wireless field, with its ever-evolving complexity and technical challenges, is also a strong candidate domain for the application of AI/ML techniques.
Future networks will require robust intelligent algorithms to adapt network protocols and resource management for different services in different scenarios. Artificial intelligence is providing a solution for the emerging complex communication system design. The recent advances in deep learning, convolutional neural networks, and reinforcement learning hold significant promise for solving complex problems considered intractable until now. It is now appropriate to apply AI technology to 5G wireless communications to tackle optimized physical layer design, complicated decision-making, network management, and resource optimization tasks in such networks. Moreover, the emerging big data technology has brought us an excellent opportunity to study the essential characteristics of wireless networks and to help us to obtain more clear and in-depth knowledge of the behaviour of 5G wireless networks.
HSC's Offerings in Wireless AI Space
HSC is working on AI/ML techniques for 5G cellular networks. Hughes Systique’s work is focused on three specific areas:
- AI/ML Deployment: AI/ML has become a key enabler of extreme automation to manage the increasing complexity of 5G networks, including the operationalization of Open RAN and vRAN. It has been recognized that customer experience, network efficiency, and service flexibility can be improved by applying AI/ML at RIC. The near real-time RIC (near-RT RIC) provides closed-loop control at very tight latencies, typically focusing on local, high-speed control. The non-real-time RIC (Non-RT-RIC), on the other hand, provides slower control using a relatively higher latency link. It has substantially larger computing and memory resources at its disposal and can afford to take a global view of the network. Our focus is on using edge cloud as a platform and associated integration with near-RT RIC and Non-RT-RIC based on the various use cases and applications.
- AI/ML Architecture: The application of AI would be useful in solving multiple use cases related to wireless communication. However, the challenge of using AI in wireless networks is that the data is distributed across different entities- as a result, traditional centralized learning methodology will not be practical in these use cases.
To handle such scenarios, the way ahead is to implement Federated Learning architecture. It enables collaborative learning, keeping all the data distributed on different devices. Federated Learning architecture is a good balance between deployability and adaptability. We have developed an end-to-end framework in PyTorch for Federated Learning.
- AI/ML Applications: One of the use cases where Al/ML can contribute significantly to wireless communications is Initial Beam Selection. In 5G mmWave, the process of Initial Beam Selection i.e. finding of appropriate beam pair between transmitter and receiver, is time-consuming. ML/AI can play a significant role in reducing the Beam Selection time during initial access. Our work is focused on AI/ML-based applications for optimal beam selection for 5G and beyond networks. The data related to millimeter wave (mmWave) multiple-input multiple-output (MIMO) channels, paired with data from sensors such as LIDAR and the vehicle’s GPS positions, are used to train the ML/AI model and predict the Top K beam pair.