CNN Digital Pre-Distortion for Multi-PA Linearity Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional digital pre-distortion techniques for power amplifiers in wireless communication systems face challenges with high operating frequency bands and multiple power amplifiers, requiring a large number of coefficients, which leads to insufficient performance and impracticality for deployment in embedded systems like FPGAs and ASICs.
Innovation Solution
A deep learning-based approach using a neural network architecture comprising multiple convolutional neural network layers and dilated convolutional neural network layers processes in-phase and quadrature signals to generate a pre-distorted signal for power amplifiers, with a linear transformation block adapting coefficients using the least mean square method for efficient digital pre-distortion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional DPD techniques use memory polynomial models with many coefficients to handle high frequency bands and multiple power amplifiers, then the modeling capability improves, but the device complexity and computational resources increase significantly
Solution Approach 1:
The patent replaces the traditional memory polynomial mathematical model with a neural network-based model. The neural network learns the non-linear characteristics of the power amplifier through training data, substituting the explicit mathematical coefficient-based approach with a data-driven learning approach that requires fewer parameters to achieve the same or better modeling accuracy.
Solution Approach 2:
The patent changes the fundamental parameters of the DPD system by using neural network weights and biases instead of memory polynomial coefficients. This parameter transformation allows the system to achieve equivalent or superior modeling performance with significantly fewer adjustable parameters, reducing device complexity while maintaining reliability.
2Reliability
If deep learning models like RNN or LSTM are used to capture memory effects and non-linearity, then the performance improves, but the model complexity becomes too high for embedded systems
Solution Approach 1:
The patent extracts and separates the memory effect compensation function from the main neural network. By using dilated convolutional layers specifically designed to capture temporal dependencies and memory effects, the system isolates this function into a dedicated module that can be efficiently implemented in embedded systems without requiring the full complexity of RNN or LSTM architectures.
Solution Approach 2:
The patent segments the DPD system into distinct functional blocks: a main neural network for non-linearity compensation and a separate dilated convolutional network for memory effect compensation. This segmentation allows each module to be optimized independently, reducing overall model complexity while maintaining the performance benefits of capturing both non-linear and memory effects.
3Adaptability or versatility
If a large number of coefficients are used to model distortion in current and future wireless communication systems, then the modeling coverage improves, but the ease of operation and deployment in embedded systems deteriorates
Solution Approach 1:
The patent creates a universal DPD model using neural networks that can adapt to different power amplifier types, frequency bands, and communication standards through training on diverse datasets. The neural network architecture serves multiple functions simultaneously: non-linearity compensation, memory effect modeling, and adaptation to different operating conditions, eliminating the need for separate models for each scenario and simplifying deployment across diverse embedded systems.
Data Source
AI summary
This specification relates to an apparatus which stores a machine-learned model including a first neural network block including multiple convolutional neural network layers, a second neural network block including at least one dilated convolutional neural network layer, and a linear transformation block. The apparatus is configured to receive input data representing in-phase and quadrature signals of an input signal that is to be amplified by a power amplifier, to process the received input data using the first neural network block of the machine learned model to generate a first neural network block output, to process the received input data using the second neural network block of the machine learned model to generate a second neural network block output, and to combine, using the linear transformation block, the first neural network block output and the second neural network block output to generate a pre-distorted signal for amplification by the power amplifier.