Transfer Learning Beam Weight Adaptation for mmWave Blockage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Millimeter wave (mmW) communication systems face challenges with hand and other body part blockages, which degrade signal performance and require inefficient processing time to determine effective beam weights for mitigation.
Innovation Solution
The implementation of transfer learning (TL) and federated learning (FL) techniques to train neural networks for determining beam weights associated with blockages, allowing for the storage of this information in a machine learning database and communication with an ML server to generate signals that mitigate blockages efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If traditional methods are used to determine beam weights for blockage mitigation, then processing time and power consumption increase, but beam weight determination can be achieved
Solution Approach 1:
The system performs preliminary training of neural networks using transfer learning and federated learning techniques before actual blockage mitigation is needed. Learned parameters and beam weight patterns are pre-computed and stored in ML databases, allowing the system to quickly adapt to blockages without performing full training in real-time, thus reducing processing time while maintaining determination capability
Solution Approach 2:
The system creates and stores copies of learned parameters and neural network models in ML databases across multiple devices and servers. These copied models can be rapidly deployed to new situations without retraining, enabling fast adaptation to blockages while preserving the ability to determine appropriate beam weights through model selection and fine-tuning
2Use of energy by moving object
If traditional methods are used to determine beam weights for blockage mitigation, then power consumption increases, but beam weight determination can be achieved
Solution Approach 1:
The system performs computationally intensive neural network training in advance using transfer learning and federated learning, storing the results in ML databases. When blockages occur, the system only needs to retrieve and apply pre-trained models rather than performing full training computations, dramatically reducing real-time power consumption while maintaining beam weight determination capability
Solution Approach 2:
The system distributes copies of trained neural network models across multiple devices and ML servers. Instead of each device performing independent training (which would consume significant power), devices can share and reuse copied models from the ML database, reducing overall power consumption while preserving the ability to determine beam weights for blockage mitigation
Data Source
AI summary
A UE may train a NN, based on a blockage of a beam transmission, to indicate one or more beam weights in association with the blockage of the beam transmission. The UE may store, in an ML database, information indicative of at least one of the trained NN or the one or more beam weights indicated via the trained NN, such that the UE may communicate, to an ML server, the information via the trained NN. The ML server may train the NN, based on a TL/FL procedure for the one or more beam weights associated with the at least one blockage, to indicate one or more TL/FL beam weights in association with the at least one blockage, and communicate, to at least one UE, information indicative of at least one of the trained NN or the one or more TL/FL beam weights indicated via the trained NN.