Neural Network Uplink RAN Compression for Fronthaul Bottlenecks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In 5G and beyond radio access networks, the radio head apparatus and central processing apparatus face challenges in efficiently processing a high number of transmission layers due to limited fronthaul capacity, leading to severe signal degradation when using traditional quantization schemes.
Innovation Solution
The implementation of neural networks at the radio head apparatus for compressing data signals, selected based on compression model information from the central processing apparatus, to reduce signal dimensionality and mitigate quantization noise, while allowing joint training and optimization with the central processing apparatus.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional quantization schemes are used to process high-dimensional received signals at the radio head apparatus, then the processing capability is maintained within existing hardware constraints, but severe signal degradation occurs due to the large dimension of the received signal
Solution Approach 1:
The patent extracts and separates the compression function from the traditional quantization process by introducing a dedicated neural network compression module. This module pre-processes the high-dimensional received signal by extracting essential features and compressing the signal representation before quantization, thereby reducing the effective signal dimension that needs to be processed while maintaining signal quality.
Solution Approach 2:
The neural network compression module acts as an intermediary between the received signal and the quantization process. It transforms the high-dimensional input signal into a compressed intermediate representation that is then passed to the quantization module, mediating the transition from high-dimensional to low-dimensional processing while preserving critical signal information.
2Loss of energy
If more functions are performed at the radio head apparatus to reduce fronthaul strain, then the processing load on the fronthaul is reduced, but the processing and memory capabilities required at the radio head apparatus increase
Solution Approach 1:
The patent changes the parameter representation of the signal by transforming it from its original high-dimensional form into a compressed low-dimensional representation through the neural network. This parameter transformation enables the radio head apparatus to perform advanced processing functions while reducing the actual data volume that needs to be transmitted over the fronthaul, thereby reducing bandwidth consumption without requiring proportionally increased processing capabilities.
3Quantity of substance
If efficient compression is applied at the radio head apparatus to reduce signal dimension, then the fronthaul capacity requirements are reduced, but the processing complexity at the radio head apparatus increases due to the need for neural network implementation
Solution Approach 1:
The patent segments the overall signal processing function into distinct modules: a neural network compression module for dimensionality reduction, a quantization module for precision control, and a transmission module for fronthaul communication. This segmentation allows each module to be optimized independently, with the compression module specifically designed to reduce data volume while the other modules handle their respective functions efficiently.
Data Source
AI summary
In a base station of a radio access network, a distribution unit is configured to receive, through a radio head apparatus of the base station, a channel information signal transmitted by a user equipment over a radio channel, obtain based on the channel information signal compression model information indicating a neural network to be used for compression by the radio head apparatus amongst a set of neural networks, and sending the compression model information to the radio head apparatus. The radio head apparatus is configured to receive the compression model information from the central processing apparatus, receive a data signal from a user equipment over a radio channel, pre-process the data signal, including compress the data signal by using a neural network, the neural network being selected based on the compression model information sent by the distribution unit, and transmit the pre-processed data signal to the central processing apparatus.