Neural Digital Predistortion for Real-Time Power Amplifier Linearization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current digital predistortion solutions for power amplifiers in wireless communication systems are inefficient due to high data requirements and inability to adapt in real-time to variations in usage, leading to suboptimal energy efficiency and increased carbon footprint.
Innovation Solution
A digital predistortion module using indirect neural networks with meta-learning to adaptively correct amplitude and phase distortions, employing two neural networks optimized through meta-learning for reduced training data and computational complexity, allowing real-time adaptation to changing power amplifier conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional digital predistortion solutions are used, then power amplifier linearisation is achieved, but high data requirements and long training times result
Solution Approach 1:
The patent segments the training process into two distinct phases: meta-initialisation using a first subset of operating data to establish baseline parameters, and meta-matching using a second subset to fine-tune for specific use cases. This segmentation reduces the computational burden and time required for complete training while maintaining linearisation quality.
Solution Approach 2:
The patent performs preliminary action by conducting meta-initialisation training before meta-matching. The first subset of operating data is used to pre-train the digital predistortion module, establishing initial parameters that are then refined with the second subset. This preliminary training reduces the overall training time required for deployment.
2Reliability
If traditional digital predistortion solutions are used, then power amplifier linearisation is achieved, but complex design and high computational complexity result
Solution Approach 1:
The patent divides the operating data into distinct subsets (first subset for meta-initialisation, second subset for meta-matching) and processes them through separate training stages. This segmentation simplifies the computational complexity by breaking down the complex training process into manageable, less computationally intensive steps.
Solution Approach 2:
The patent changes parameters by using different subsets of operating data for different training stages, with each stage optimized for specific learning objectives. The meta-initialisation stage uses broader operating data to establish general parameters, while meta-matching uses specific use case data to fine-tune parameters, reducing overall computational requirements.
3Adaptability or versatility
If neural networks are re-trained for usage variations, then adaptation to new conditions is achieved, but significant data and time resources are consumed
Solution Approach 1:
The patent segments the adaptation process into meta-initialisation (using first subset of data) and meta-matching (using second subset of data). This allows the system to adapt to new usage conditions with reduced data requirements, as the meta-initialisation provides a robust baseline that requires minimal additional data for fine-tuning.
Solution Approach 2:
The patent creates a generalized model through meta-initialisation that can be quickly adapted to specific use cases through meta-matching. Instead of training completely new neural networks for each usage variation, the system copies and adapts the pre-trained parameters, significantly reducing data requirements while maintaining adaptability.
4Use of energy by moving object
If power amplifier operates near saturation for energy efficiency, then energy efficiency is improved, but non-linear distortions in amplitude and phase increase
Solution Approach 1:
The patent applies preliminary anti-action by using digital predistortion to pre-compensate for the non-linear distortions that will occur when the power amplifier operates near saturation. The neural network-based predistortion module introduces opposite distortions in advance, canceling out the amplifier's non-linearities and maintaining signal linearity while enabling energy-efficient saturation operation.
Data Source
AI summary
The invention relates to a device for linearising a power amplifier by employing digital predistortion, comprising: a digital predistortion module, configured to infer a polar domain predistortion to be applied to a signal, and comprising a first neural network and a second neural network respectively configured to correct amplitude and phase distortion produced by the amplifier; an optimisation module of each of said neural networks configured to implement meta-learning, using: a meta-initialisation providing a prior initialisation of the initial weights of each of said neural networks; a meta-matching of the initial weights into optimal weights of each of said neural networks.