Neural Digital Predistortion for PA Linearity With Fewer Coefficients
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current digital predistortion technologies face challenges in achieving high radio frequency performance and linearity in power amplifiers, particularly with newer wireless technologies that require more stringent specifications, leading to increased nonlinearity and power consumption.
Innovation Solution
A multistate, multi-rate digital predistortion system utilizing a series of compensators with neural networks and machine learning-based models to adaptively distort the input signal to the power amplifier, ensuring a linear total system response by compensating for nonlinearity, while reducing the number of coefficients and circuit area.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If digital predistortion is applied to improve power amplifier linearity, then radio frequency performance is improved, but device complexity increases
Solution Approach 1:
The digital predistortion system is divided into multiple processing circuit blocks, each implementing a specific transformation function. These blocks process different aspects of the signal separately (complex to real transformation, memory effects compensation, etc.), making the overall complex system manageable and implementable through modular components
Solution Approach 2:
The patent transforms the complex predistortion problem into a real-valued domain by converting complex intermediate signals to real signals through neural network processing. This dimensional transformation simplifies the computational complexity while maintaining the ability to compensate for power amplifier nonlinearity and memory effects
2Manufacturing precision
If more complex predistortion models are used to meet stringent specifications, then manufacturing precision is improved, but device complexity increases
Solution Approach 1:
The system uses adaptive neural networks that can dynamically adjust their parameters and structure based on the operating conditions. This allows the predistortion model to adapt to different power amplifier states and operating points, achieving high precision linearity compensation without requiring a static overly-complex model structure
Solution Approach 2:
The patent changes the parameters of the neural network (weights, biases, activation functions) based on feedback from the power amplifier output. This parameter adaptation allows the system to achieve high manufacturing precision for linearity specifications by tuning the model to match the specific characteristics of the power amplifier
3Device complexity
If traditional digital predistortion methods are used, then implementation is simpler, but power consumption increases
Solution Approach 1:
The patent replaces traditional mathematical predistortion methods with neural network-based processing. The neural networks can be implemented using efficient hardware architectures that consume less power than conventional digital signal processing approaches, while providing comparable or superior linearity compensation performance
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An adaptable, generalized digital predistortion system that can increase the accuracy of the distortion applied to a power amplifier input signal is disclosed. Further, the digital predistortion system can be implemented using a reduced circuit area by reducing a number of coefficients used by the models applied to generate the distortion compensation signal. Further, the system can implement an improved digital predistortion adaptation engine that can improve one or more of the following metrics of a complete radio frequency (RF) analog front-end radio signal chain: error vector magnitude (EVM), adjacent channel leakage ratio (ACLR), spectrum emission mask (SEM), and power consumption of the circuits and systems. An example of circuits and systems is a radio frequency power amplifier (PA).