Reverse Autoencoder Neural Network for Transport Block Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data communication systems face challenges in achieving spectral efficiency and robustness in the block-wise transmission of data, particularly in cellular communication networks.
Innovation Solution
The use of a reverse autoencoder neural network, comprising an expander neural network and a compressor neural network, to determine expanded and compressed representations of data in transport blocks, thereby optimizing their size and transmission efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If data is transmitted in fixed-size blocks with padding, then transmission structure is simplified, but spectral efficiency deteriorates due to wasted bandwidth on padding data
Solution Approach 1:
The patent changes the parameter of data block size from fixed to variable by using neural network-based expansion and compression. The expander neural network dynamically adjusts the expanded transport block size based on the original transport block size and channel conditions, while the compressor neural network reconstructs the original data. This eliminates the need for fixed-size padding while maintaining structured transmission, thereby improving spectral efficiency without sacrificing transmission structure simplicity.
2Reliability
If traditional error correction coding is used, then channel error resistance is improved, but spectral efficiency deteriorates due to redundant coding bits
Solution Approach 1:
The patent substitutes traditional mechanical error correction coding systems with a neural network-based expansion-compression system. The expander neural network learns to expand data in a way that inherently provides error resilience, and the compressor neural network learns to compress and reconstruct the original data. This substitution eliminates the need for separate redundant coding bits while maintaining or improving error correction capabilities, thereby improving spectral efficiency while preserving channel error resistance.
3Loss of energy
If transport block size is dynamically adjusted, then spectral efficiency is improved, but device complexity increases due to neural network training and deployment
Solution Approach 1:
The patent applies preliminary action by training the expander and compressor neural networks in advance using training assistance data exchanged between the first and second apparatuses. The training phase occurs before actual data transmission, allowing the neural networks to learn optimal expansion and compression strategies for various channel conditions. During operation, the trained neural networks can be deployed with pre-configured parameters, reducing the complexity of real-time decision-making while maintaining dynamic adaptability for spectral efficiency optimization.
4Loss of energy
If padding is eliminated to improve spectral efficiency, then bandwidth usage is optimized, but transmission robustness deteriorates due to variable block sizes
Solution Approach 1:
The patent implements feedback mechanisms where the first apparatus transmits training assistance data to the second apparatus, and both apparatuses exchange information about their neural network capabilities and configurations. The expander neural network at the first apparatus expands the transport block based on learned patterns from training data, and the compressor neural network at the second apparatus reconstructs the original data by comparing the compressed representation with the received expanded block. This feedback-driven approach ensures that variable block sizes maintain robustness through learned optimization patterns while eliminating wasted padding bandwidth.
Data Source
Figure 1~2
Figure 3~4
Figure 5~7
AI summary
Various example embodiments relate to expansion or compression of data of a transport block. An apparatus may comprise means for receiving, from another apparatus, training assistance data for training at least one part of a reverse autoencoder neural network, the reverse autoencoder neural network comprising: an expander neural network configured to determine an expanded representation of data of a transport block to cause the data of the transport block to have a designated size; and a compressor neural network configured to determine, based on the expanded representation of the data of the transport block, a compressed representation of the data of the transport block to reconstruct the data of the transport block, wherein the expander neural network is an encoder of the reverse autoencoder neural network and the compressor neural network is a decoder of the reverse autoencoder neural network.