Multi-Band DPD Adaptation Using Overlapping Splines
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Concurrent dual-band power amplifier systems face significant computational complexity challenges due to the need for high-order nonlinear distortion modeling and fixed-point precision issues in existing digital pre-distortion (DPD) methods, particularly with the Generalized Memory Polynomial (GMP) and cubic spline-based approaches, which struggle with flexibility and accuracy in real-time adaptation.
Innovation Solution
The proposed solution employs overlapping low-complexity spline functions to model nonlinear distortion, using a piecewise approximation with quadratic or cubic splines that span only two bins, allowing for direct computation and reducing the complexity of tap weight adaptation through a novel closed-loop algorithm that normalizes step sizes based on tap-specific power statistics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Generalized Memory Polynomial (GMP) or cubic spline-based DPD methods are used for concurrent dual-band systems, then nonlinear distortion modeling capability is improved, but computational complexity increases significantly
Solution Approach 1:
The patent segments the continuous nonlinear distortion function into discrete piecewise polynomial segments (splines) defined by control points and knots. This segmentation transforms the complex continuous modeling problem into manageable discrete segments that can be computed efficiently, reducing overall computational complexity while maintaining modeling accuracy.
Solution Approach 2:
The patent changes the parameter representation from high-order polynomial coefficients to spline control points and knot positions. This parameter transformation enables the use of lower-complexity piecewise polynomials that can approximate nonlinear distortion with fewer computational operations, directly addressing the complexity-accuracy tradeoff.
2Measurement precision
If high-order polynomial basis functions are used in GMP DPD, then distortion modeling accuracy is improved, but fixed-point precision issues and computational complexity worsen
Solution Approach 1:
By segmenting the distortion function into piecewise low-order polynomials rather than using a single high-order polynomial, the patent avoids the fixed-point precision problems associated with high-order terms. Each segment uses simpler polynomial expressions that can be accurately represented in fixed-point arithmetic.
Solution Approach 2:
The patent uses computationally inexpensive piecewise polynomial segments that can be evaluated quickly and discarded, rather than relying on expensive high-order polynomial calculations. This approach trades the reuse of complex mathematical objects for simpler, more efficient computations.
3Ease of operation
If conventional DPD adaptation algorithms are used, then tap weight updating is achieved, but convergence speed and real-time adaptation capability are insufficient
Solution Approach 1:
The patent implements feedback mechanisms where the error between predicted and actual distortion is continuously monitored and used to update spline control points and knot positions. This feedback-driven adaptation enables real-time tracking of PA nonlinearities and accelerates convergence compared to conventional open-loop or slowly adapting methods.
Solution Approach 2:
The patent transforms the static spline parameters into dynamic, adaptively adjustable parameters that can change in real-time based on operating conditions. This dynamic adaptation allows the DPD system to track time-varying PA characteristics, improving both convergence speed and real-time performance.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
A technique for updating tap weights in a digital pre-distorter, DPD is presented. The DPD is configured to compensate for a non-linear operation of a power amplifier, and a method aspect comprises computing (140) an average power of each input to a plurality of tap weight calculators over a plurality of samples (S112); computing an approximate logarithm of the average power of each input (S114); and modulating a step size of an adaptation process to update each tap weight, the step size being modulated based on the approximate logarithm of the average power of the input (S116).