Neural Volterra Digital Predistortion With Lower Model Complexity

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing digital predistortion (DPD) solutions for transceivers face challenges with high complexity, power consumption, and impracticality due to Volterra-based models or artificial neural networks, which are either dimensionally cursed or inefficient, especially in high-performance wireless systems.

Innovation Solution

A hybrid approach combining a Neural Volterra actuator with an artificial neural network and non-linear gain blocks, featuring a connection circuit to adjust electrical connections and a combiner to generate a digitally predistorted signal, which includes a first processing block and a second processing block with multipliers, to reduce non-linearity in power amplifiers.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If Volterra-based models are used for digital predistortion, then non-linearity compensation performance is improved, but device complexity increases due to dimensional curse

Engineering Contradiction:
Improvenon-linearity compensation performanceVSAvoidmodel complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent merges Volterra-based models with artificial neural networks into a hybrid Neural Volterra actuator architecture. This combination allows the system to leverage the mathematical rigor of Volterra series for non-linearity compensation while using neural networks to automatically learn optimal kernel parameters, thereby reducing manual configuration complexity and mitigating the dimensional curse through adaptive learning.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The artificial neural network acts as an intermediary between the input signal and the Volterra series computation. The neural network processes the input signal to generate optimized kernel parameters that are then fed into the Volterra-based predistortion function, enabling the system to achieve high compensation performance with reduced complexity by delegating parameter optimization to the neural network.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Adaptability or versatility

If artificial neural networks are used for digital predistortion, then adaptability is improved, but power consumption increases

Engineering Contradiction:
Improveadaptability to non-linearityVSAvoidpower consumption
Core Design Contradiction:
Adaptability or versatilityVSUse of energy by moving object

Solution Approach 1:

The patent implements a dynamic hybrid architecture where the artificial neural network and Volterra-based model work together with adaptive switching mechanisms. The system can dynamically adjust the contribution of each component based on operating conditions, allowing high adaptability when needed while reducing power consumption by relying more on the computationally lighter Volterra model during stable operating conditions.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The predistortion function is segmented into multiple components: the artificial neural network handles complex adaptive parameter learning, while the Volterra-based model handles real-time predistortion computation. This segmentation allows each component to operate at its optimal efficiency point, reducing overall power consumption while maintaining adaptability through the neural network's parameter optimization capabilities.

Inventive Principle:
Principle #1Segmentation

3Reliability

If higher performance DPD is implemented to meet demanding specifications, then non-linearity compensation is improved, but device complexity and power consumption increase

Engineering Contradiction:
Improvenon-linearity compensation performanceVSAvoidDPD system complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent employs parameter optimization through artificial neural networks to automatically adjust the Volterra series kernel parameters based on measured non-linearity characteristics. This adaptive parameter tuning enables the system to achieve high compensation performance for demanding specifications without manually increasing system complexity, as the neural network learns optimal parameters through training data.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The system incorporates feedback mechanisms where the output of the hybrid Neural Volterra actuator is monitored and used to refine the neural network's parameter learning. This feedback loop enables the system to continuously optimize its performance for high-demanding specifications while maintaining manageable complexity through automated adaptation rather than manual system redesign.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP4668578A1Neural volterra system for digital predistortion
Publication Date: 2025.12.24 ANALOG DEVICES INT UNLTD CO
  • EP4668578A1 patent drawingFigure 1
  • EP4668578A1 patent drawingFigure 2
  • EP4668578A1 patent drawingFigure 3

AI summary

Aspects of this disclosure relate to Neural Volterra actuators. In an embodiment, a Neural Volterra actuator includes a first processing block that includes an artificial neural network, a second processing block that includes non-linear gain blocks, multipliers, a connection circuit configured to adjust an electrical connection to an input of a multiplier of the multipliers, and a combiner that generates a combined output signal based on output signals from the multipliers. The combined output signal can be a digitally predistorted version of an input signal received by the Neural Volterra actuator. Related methods and systems are also disclosed.