Neural-Network-Assisted DPD for PA Gain Oscillation Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-power RF systems, particularly those used in 5G wireless communications, face challenges in achieving linearity and efficiency due to nonlinear distortions caused by power amplifier drain inductor resonance, which results in unwanted out-of-band emissions and fails to meet regulatory standards.
Innovation Solution
A neural-network-assisted physical model is employed to mitigate PA gain oscillations by precompensating the input signal using an envelope regulator circuit and DPD actuator circuit, updating DPD coefficients based on feedback to reduce the effects of drain inductor resonance, thereby improving DPD performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If digital predistortion is applied to enhance linearity of power amplifier, then linearity is improved, but device complexity increases due to need for accurate PA modeling and continuous updates
Solution Approach 1:
The DPD system is segmented into distinct functional blocks: envelope detection unit, delay elements, multiplication units, and subtraction units. Each block performs a specific operation on the signal or model parameters, allowing independent optimization and simplifying the overall complex system design while maintaining linearity performance
Solution Approach 2:
The system performs preliminary action by pre-calculating and storing PA model parameters (alpha coefficients and delay values) in lookup tables before actual predistortion operation. This pre-computation approach simplifies real-time processing complexity while maintaining accurate linearity compensation
2Measurement precision
If higher sampling rates are used in DPD to improve accuracy, then measurement precision is improved, but processing speed requirements and device complexity increase
Solution Approach 1:
PA model parameters including alpha coefficients and delay values are pre-calculated and stored in lookup tables during an offline training phase. This preliminary action allows the system to use higher effective sampling rates during operation without the full computational burden, as the complex modeling work is already completed beforehand
Solution Approach 2:
The signal processing is segmented into discrete operational stages: envelope detection at full sampling rate, followed by processed operations at reduced rates using stored parameters. This segmentation allows high precision where needed while reducing overall processing complexity through rate reduction in subsequent stages
3Reliability
If feedback from PA output is used to update PA model, then linearity is improved, but loss of time occurs due to continuous model updating and processing delays
Solution Approach 1:
Comprehensive PA model parameters (multiple alpha coefficients and delay values) are pre-calculated and stored in lookup tables during an offline training phase using feedback data. This preliminary action allows rapid online updates by simply retrieving pre-computed values, minimizing time loss while maintaining accurate linearity adaptation
Solution Approach 2:
The system implements dynamic operation by allowing flexible selection between different update modes: frequent updates using pre-stored parameters for speed, or less frequent updates with full recalculation for maximum accuracy. This dynamic approach balances linearity improvement needs against time loss constraints based on operating conditions
Data Source
Figure 1A
Figure 1B~1C
Figure 2~3
AI summary
Systems, devices, and methods related to envelope regulated, digital predistortion (DPD) are provided. An example apparatus includes an envelope regulator circuit to process, based on a parameterized model, an input signal to generate an envelope regulated signal; a digital predistortion (DPD) actuator circuit to process the envelope regulated signal and the input signal based on DPD coefficients associated with a nonlinearity characteristic of a nonlinear component; and a DPD adaptation circuit to update the DPD coefficients based on a feedback signal indicative of an output of the nonlinear component.