Unified AI-DPD Hardware for Ultra-Wideband PA Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial-intelligence-digital-predistortion (AI-DPD) technologies face challenges in adapting to the complex dynamic nonlinear characteristics of ultra-wideband power amplifiers (UWB PAs) due to their reliance on specific neural network structures, which limits their effectiveness across varying conditions such as power, voltage, frequency, temperature, and material variations.
Innovation Solution
A unified digital predistortion hardware structure is implemented using flexible software configuration to enable AI-DPD schemes with multiple neural-networks, allowing for adaptable predistortion with minimal hardware overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a particular neural network structure is used for AI-DPD, then the hardware structure can be simplified, but the adaptability to various PA characteristics deteriorates
Solution Approach 1:
The patent implements a unified DPD hardware structure that can perform multiple neural network computations (FC layers, Conv layers, ResNet blocks) through software configuration. The hardware includes configurable elements such as fully connected layer units, convolutional layer units, and residual network units that can be activated based on the selected neural network model, enabling one hardware structure to serve multiple AI-DPD algorithm requirements.
Solution Approach 2:
The patent introduces dynamic configurability to the hardware structure through software control. The hardware structure can dynamically switch between different neural network architectures (e.g., from simple FC networks to complex ResNet structures) by activating or deactivating specific computational units. This dynamic reconfiguration allows the same hardware to adapt to different PA characteristics without physical modification.
2Adaptability or versatility
If multiple neural network structures are provided for different PA characteristics, then the adaptability improves, but the hardware overhead increases
Solution Approach 1:
The patent merges multiple neural network computational units into a single unified hardware structure. Instead of having separate hardware for FC networks, Conv networks, and ResNet architectures, the patent combines all necessary computational elements (fully connected layer units, convolutional layer units, residual network units with skip connections) into one integrated structure that can be configured to implement any of these network types through software control.
Solution Approach 2:
The unified hardware structure is designed to be universal, capable of implementing multiple AI-DPD algorithms through software configuration. The hardware includes configurable computational units that can be activated based on the selected neural network model, eliminating the need for multiple dedicated hardware structures and reducing overall hardware overhead while maintaining full adaptability.
3Device complexity
If traditional DPD methods are used, then the hardware requirements are low, but the effectiveness in handling complex nonlinear characteristics deteriorates
Solution Approach 1:
The patent replaces traditional mathematical DPD methods with AI-based neural network approaches implemented in hardware. Instead of using conventional signal processing algorithms, the patent employs neural network computational units (FC layers, Conv layers, ResNet blocks) that can learn and adapt to complex nonlinear PA characteristics, providing superior compensation effectiveness while being implemented in a unified hardware structure.
Data Source
AI summary
Provided is a method for implementing a digital predistortion scheme. The method includes obtaining (S210) artificial-intelligence-digital-predistortion (AI-DPD) schemes of eight neural-networks by using a unified digital predistortion (DPD) hardware structure and using software configuration.