ML-Based Multiband CFR and DPD for Non-Contiguous Radio Bands
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing digital predistortion (DPD) and crest factor reduction (CFR) methods struggle to efficiently handle multiple non-contiguous frequency bands in wideband radio systems, leading to signal distortion and complexity challenges, particularly in wireless communication networks.
Innovation Solution
Implementing machine learning (ML) based CFR and DPD architectures that construct feature spaces from complex-valued signals, incorporating real, imaginary, absolute, and phase components, suitable for non-contiguous multiband scenarios, utilizing neural networks and recurrent neural networks to optimize signal conditioning before power amplification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional DPD and CFR methods are used to handle multiple non-contiguous frequency bands, then signal conditioning is performed, but signal distortion occurs and computational complexity increases
Solution Approach 1:
The patent replaces traditional mechanical/mathematical signal processing methods (DPD and CFR algorithms) with a machine learning-based system. The ML model learns optimal signal conditioning parameters and transformations directly from training data, substituting complex iterative computational methods with a trained neural network that provides both accuracy and efficiency.
Solution Approach 2:
The patent segments the complex multiband signal processing task into distinct frequency band components. Each band is processed independently through the ML model, which handles non-contiguous bands separately rather than as a single complex signal, reducing overall computational complexity while maintaining signal quality.
2Adaptability or versatility
If machine learning based CFR and DPD architectures are implemented, then adaptability to hardware advancements is improved, but initial system complexity increases
Solution Approach 1:
The patent implements a universal ML-based signal conditioning architecture that can adapt to different hardware configurations and frequency band combinations. The same core ML model structure handles various scenarios (different numbers of bands, contiguous or non-contiguous arrangements) by learning from diverse training data, providing hardware-agnostic adaptability.
Solution Approach 2:
The patent utilizes parameter changes in the ML model during training and inference to adapt to different hardware capabilities. By adjusting model parameters, input feature selections, and processing configurations based on available hardware resources, the system achieves adaptability without requiring fundamental architectural changes.
Data Source
AI summary
Method and device(s) for supporting performance of machine learning based CFR and DPD on multiple digital input signals relating to different frequency bands, respectively, in order to signal condition said signals before power amplification and subsequent transmission in said frequency bands by a wireless communication network. The device(s) obtain said multiple digital input signals as complex valued signals. The device(s) perform feature construction that takes said multiple digital input signals as input and provides constructed feature signals according to predefined constructed feature types as output. Said predefined constructed feature types relate to at least the following per complex valued sample of the obtained complex valued multiple digital input signals the real part of the sample, the imaginary part of the sample and at least one of the absolute value of the sample and the phase of the sample.