Known Payloads for Neural Network Training in Wireless Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current wireless communication technologies, such as LTE and New Radio (NR), face challenges in efficiently training artificial neural networks for improved mobile broadband access, as offline training does not account for the dynamic real-world environment, and existing reference signals are insufficient for certain machine learning models.
Innovation Solution
Implementing a method where both base stations and user equipment communicate to agree on known payloads for physical channel transmissions, which can be used as ground truth labels for online training of neural networks, allowing for fine-tuning with over-the-air signaling reflective of the wireless environment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If offline training is used for neural networks, then training can be performed with static datasets, but the model does not adapt to dynamic real-world wireless environments
Solution Approach 1:
The patent applies preliminary action by pre-agreeng known payloads between transmitting and receiving devices before actual data transmission. These pre-established known payloads serve as ready-to-use training data for online neural network training, eliminating the need to collect and process training data in real-time, thus enabling rapid adaptation to dynamic wireless environments without significant training time loss.
Solution Approach 2:
The system implements self-service by having the wireless communication devices themselves generate and exchange known payloads during normal operation. The transmitting device automatically provides known payload information to the receiving device, which then uses this data for its own neural network training, creating a self-sustaining adaptation mechanism without external intervention.
2Quantity of substance
If existing reference signals are used for training, then signaling overhead is maintained at current levels, but the data is insufficient for certain machine learning models like MIMO demapping
Solution Approach 1:
The patent applies universality by designing known payloads that serve multiple functions: they act as both normal communication data and as training data for neural networks. The same payload structure is used for both information transmission and machine learning training, eliminating the need for separate dedicated training signals and thereby increasing the quantity of usable training data without proportionally increasing system complexity.
Solution Approach 2:
The system merges the functions of data transmission and training data provision by combining known payloads with regular communication traffic. Instead of separating training signals from data channels, the patent integrates both purposes into a unified transmission framework where the known payload simultaneously carries information and provides training examples for MIMO demapping and other ML tasks.
3Measurement precision
If full payload decoding is performed for training, then complete data is available for training, but processing time and computational resources are significantly increased
Solution Approach 1:
The patent applies the extraction principle by removing the need for full payload decoding when training data is needed. Instead of decoding entire payloads, the system extracts only the known payload portions that are already identified and agreed upon between devices. This selective extraction provides sufficient training data with high accuracy while dramatically reducing computational power requirements compared to complete decoding.
Data Source
AI summary
A method of wireless communications by a receiving device includes communicating about content of a known payload with a transmitting device. The method also includes requesting, from the transmitting device, the known payload for training an artificial neural network. The method also receives the known payload in response to the request. The method further performs online training of the artificial neural network with the known payload. A method of wireless communications by a transmitting device includes communicating about content of a known payload with a receiving device and then transmitting an indication informing the receiving device that the known payload will be transmitted. The transmitting device unicasts the known payload to the receiving device for online training of a neural network.