Federated Learning for mm Wave Sector Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Fast sector-steering in the mm wave band for vehicular mobility scenarios is challenging due to the time-consuming exhaustive search over predefined antenna sectors, which cannot be assuredly completed within short contact times.
Innovation Solution
The implementation of federated learning using multimodal deep learning architectures that fuse data from LiDAR, GPS, and camera images to locally predict the best mm wave network sectors for alignment, combined with a multimodal federated learning framework that reduces individual training times and shares information privately.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If exhaustive search over predefined antenna sectors is performed, then sector selection accuracy is improved, but sector selection time increases
Solution Approach 1:
The system performs preliminary actions by pre-defining a set of antenna sectors and their corresponding beamforming vectors before actual communication begins. This allows the exhaustive search to be performed offline or in advance, so that during time-critical communication, the system can directly select from pre-computed sectors without performing real-time exhaustive search, thus reducing sector selection time while maintaining accuracy.
Solution Approach 2:
The patent segments the antenna array into multiple predefined sectors, each associated with specific beamforming vectors. This segmentation allows the system to divide the complex task of finding the optimal beam into discrete, manageable sectors. By segmenting the search space into predefined sectors rather than performing continuous search, the system achieves faster selection while maintaining the ability to evaluate all possible directions for optimal accuracy.
2Productivity
If machine learning-based sector prediction is used, then sector selection speed is improved, but system complexity increases
Solution Approach 1:
The patent introduces machine learning models as intermediary components that mediate between the raw sensor data (LiDAR, camera, GPS) and the sector selection decision. These ML models act as a bridge, processing complex sensor inputs and translating them into predicted optimal sectors. This intermediary approach enables fast prediction by leveraging pre-trained models that capture spatial relationships and communication patterns, thereby increasing selection speed without requiring real-time complex computations during communication.
Solution Approach 2:
The system performs preliminary training of machine learning models offline using historical communication data and sensor information. During actual communication, the pre-trained models are deployed for rapid inference to predict optimal sectors. This preliminary action separates the complex training process from the real-time operation, allowing the system to achieve high prediction speed during communication while concentrating the computational complexity in the offline training phase, thus managing overall system complexity.
3Reliability
If centralized system with control channel is used, then coordination is improved, but bandwidth consumption increases
Solution Approach 1:
The patent implements self-service by enabling each vehicle to autonomously perform sector selection using its own sensor data and pre-trained machine learning models. Each vehicle independently processes LiDAR, camera, and GPS information to predict the optimal sector without requiring continuous coordination or data exchange with a centralized system. This self-service approach eliminates the need for high-bandwidth control channels while maintaining reliable coordination through distributed decision-making, directly reducing bandwidth consumption while preserving coordination effectiveness.
Data Source
AI summary
Provided herein are systems and methods for selecting a mm wave network sector for use by a vehicle operating within a network environment, the method including collecting data from a plurality of non-RF sensors on the vehicle, training, by the collected data in a deep learning inference (DL) engine, a locally trained model analyzing, using the trained model and the DL engine, the collected data to predict a sector of the mm wave network having a best alignment at a position of the vehicle; and probing the predicted sector of the mm wave network.