Adaptive Predistortion Learning Rates for Wireless Power Amplifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional adaptive predistortion systems for wireless transmitters face challenges in adapting to changing power amplifier characteristics due to their reliance on memory-less models and constant learning factors, leading to compromised convergence speed and increased noise interference, which degrades Bit Error Rate and Carrier to Interference Plus Noise Ratio.
Innovation Solution
The system employs a plurality of time-varying adaptation factors for each predistortion gain coefficient, using Recursive Least Squares updates and a forgetting factor to adjust learning rates, allowing for faster convergence and improved noise rejection, while avoiding instability through exponential decay and sign-based adaptation factor updates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If a constant learning factor is used in LMS engine for adaptive predistortion, then noise rejection is improved, but convergence speed deteriorates
Solution Approach 1:
The patent applies dynamics by transitioning from a constant learning factor to a time-varying learning factor that adapts during the predistortion process. The learning factor starts at a higher value to enable fast convergence during initial training, then decays to a lower value to provide noise rejection during steady-state operation. This dynamic adjustment resolves the contradiction between convergence speed and noise rejection.
Solution Approach 2:
The patent changes the parameter of the learning factor from constant to time-varying. By implementing exponential decay or other time-dependent functions for the learning factor, the system can achieve both fast initial convergence (when learning factor is high) and good noise rejection (when learning factor is low), thereby resolving the technical contradiction.
2Object-affected harmful factors
If learning factor is reduced to reject noise, then noise rejection is improved, but convergence speed drops and key specifications fail
Solution Approach 1:
The dynamic learning factor allows the system to maintain high convergence speed during critical phases (such as switching events) when reliability is most important, while providing noise rejection during steady-state operation. This resolves the contradiction between noise rejection and specifications compliance.
Solution Approach 2:
The system performs preliminary action by using a higher learning factor during initial training and critical transitions to ensure fast convergence and meet key specifications before noise becomes a dominant concern. This preliminary fast convergence ensures specifications are met before steady-state noise rejection becomes the primary goal.
3Device complexity
If memory-less models are used for PA and PD, then device complexity is reduced, but ability to handle PA memory effects deteriorates
Solution Approach 1:
The patent introduces dynamics by using time-varying learning factors that adapt to changing PA characteristics, including memory effects. Instead of requiring complex memory-based models, the dynamic learning factor allows the simple memory-less model to adapt its behavior over time to compensate for PA memory effects, resolving the contradiction between model simplicity and adaptability.
Data Source
AI summary
A method of adaptive predistortion of a power amplifier, characterized in that the method comprises the steps of: storing values of a plurality of corresponding first and second coefficients; selecting one of the stored first coefficients; processing a first signal with the first coefficient to produce an input signal for the power amplifier; amplifying the input signal in the power amplifier to produce an output signal; calculating an error value from the output signal and a previously selected first coefficient; selecting a stored second coefficient corresponding with the previously selected first coefficient; updating the previously selected first coefficient with a value calculated from the error value and the second coefficient; updating the second coefficient; and replacing the previously selected first coefficient and corresponding second coefficient with the updated first and second coefficients respectively.