Multiband DPD Linearization Using RBF Kernel Regression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current multiband radio designs face challenges in mitigating non-linear distortion caused by Power Amplifiers (PA) due to cross-carrier inter-modulations, requiring new algorithms for multiband linearization that can handle increased hardware complexity and computational resources, especially with the exponential growth in complexity of Volterra-based Digital Predistortion (DPD) models for three or more bands.
Innovation Solution
The implementation of kernel regression-based multiband Digital Predistortion (DPD) using Radial Basis Function (RBF) kernels, which transforms input signals into a constructed input vector space and predistorts them based on determined kernel centroid locations, widths, and weights, reducing computational complexity and enabling a one-dimensional Lookup Table (LUT) implementation regardless of the number of bands.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If Volterra-based Digital Predistortion (DPD) models are used for multiband linearization, then linearization accuracy is improved, but device complexity increases exponentially with the number of bands
Solution Approach 1:
The patent transforms the input signal representation by changing parameters from time-domain samples to frequency-domain features (amplitude and phase of DFT coefficients). This parameter transformation reduces the dimensionality of the input space and enables the use of simpler lookup table structures, thereby reducing device complexity while maintaining linearization accuracy across multiple bands
Solution Approach 2:
The patent segments the multiband signal processing into independent frequency bins through Discrete Fourier Transform. Each frequency bin is processed separately using individual lookup tables, avoiding the need for complex multivariate models. This segmentation approach reduces algorithmic complexity from exponential to linear with respect to the number of bands
2Adaptability or versatility
If the whole range of spectrum is linearized, then linearization coverage is improved, but computational resources and sampling rates increase significantly
Solution Approach 1:
The patent divides the wideband spectrum into multiple frequency bins using DFT, allowing selective processing of only those bins that contain active signals. This segmentation enables the system to achieve wideband linearization coverage while computing predistortion parameters only for relevant frequency regions, significantly reducing computational resource requirements
Solution Approach 2:
The patent applies partial linearization by focusing computational efforts only on frequency bins that contain active transmissions rather than processing the entire spectrum uniformly. This partial action approach maintains linearization coverage for all active bands while avoiding unnecessary computational overhead in idle frequency regions
3Ease of manufacture
If multivariate Volterra DPD is implemented using Lookup Table (LUT), then implementation feasibility is improved, but memory requirements increase with the number of bands
Solution Approach 1:
The patent segments the multiband predistortion implementation into independent lookup tables for each frequency bin. Instead of requiring a single large multivariate LUT that scales exponentially with the number of bands, the system uses multiple small univariate LUTs, one per frequency bin. This segmentation dramatically reduces total memory requirements while maintaining implementation feasibility in digital hardware
Solution Approach 2:
The patent transforms the problem from a high-dimensional multivariate lookup table challenge to a series of low-dimensional univariate lookup table problems by introducing the frequency domain as another dimension. The DFT transformation maps time-domain multiband signals to frequency-domain bins, where each bin can be handled by a simple 1D LUT, reducing memory requirements from exponential to linear scaling with the number of bands
Data Source
AI summary
Systems and methods for multiband linearization using kernel regression are provided. In some embodiments, a method includes, for each band of the multiband transmitter: transforming a group of input signals from one or more bands into a constructed input vector space to provide transformed input signals; predistorting the transformed input signals to provide a respective group of predistorted input signals in accordance with a Radial Basis Function (RBF) kernel regression; and transmitting the respective group of predistorted input signals. In this way, some advantages include a semi blind approach as one need not to account for the non-linearity order as in Volterra-based DPD for example, only the memory depth is needed to be incorporated to the input vector space. The computational complexity of DPD is reduced compared to Volterra-based DPD. Implementation complexity is relaxed by means of using a 1D Lookup Table implementation regardless of the number of bands.