End-to-end learning in multi-carrier communication systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current communication systems, particularly in multi-carrier transmission systems like OFDM, face challenges in optimizing the entire system as a single entity, leading to suboptimal performance due to separate design and optimization of transmitters and receivers.
Innovation Solution
Organizing transmitter and receiver neural networks into pairs for each subcarrier frequency band, mapping symbols into transmit blocks, and training weights using a loss function with stochastic gradient descent, while incorporating channel equalization and carrier frequency offset correction using neural networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If separate design and optimization of transmitter and receiver are used, then device complexity is reduced, but system performance deteriorates
Solution Approach 1:
The patent merges the separate design of transmitter and receiver into a unified end-to-end learning framework where both components are jointly optimized as a single system. The transmitter neural network and receiver neural network are trained together using a shared loss function that measures the overall system performance, allowing the system to achieve optimal performance while managing complexity through integrated training rather than separate optimization.
2Reliability
If end-to-end learning is implemented, then system performance is improved, but device complexity increases
Solution Approach 1:
The patent segments the end-to-end learning system into distinct transmitter and receiver neural network components that can be independently analyzed and implemented, while still being jointly trained. This segmentation allows the complex end-to-end optimization problem to be broken down into manageable parts that can be separately designed and then integrated through shared training data and loss functions.
3Reliability
If joint training of transmitter and receiver neural networks is performed, then block error rate is reduced, but training time increases
Solution Approach 1:
The patent implements continuous joint training of transmitter and receiver neural networks using stochastic gradient descent, where both networks are updated simultaneously throughout the training process rather than alternately. This continuous joint optimization ensures that both components adapt together to minimize the block error rate, maintaining the beneficial interactions between transmitter and receiver designs throughout the entire training duration.
Data Source
AI summary
This specification relates to end-to-end learning in communication systems and describes: organising a plurality of transmitter neutral networks and a plurality of receiver neural networks into a plurality of transmitter-receiver neural network pairs, wherein a transmitter-receiver neural network pair is defined for each of a plurality of subcarrier frequency bands of a multi-carrier transmission system; arranging a plurality of symbols of the multi-carrier transmission system into a plurality of transmit blocks; mapping each of said transmit blocks to one of the transmitter-receiver neural network pairs; transmitting each symbol using the mapped transmitter-receiver neural network pair; and training at least some weights of the transmit and receive neural networks using a loss function for each transmitter-receiver neural network pair.