Time-Varying Linear Filters for Low-Complexity RF Predistortion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing digital predistortion (DPD) methods for radio frequency (RF) power amplifiers (PAs) require high computational complexity and power consumption due to the need for significant multiplications in nonlinear filters, particularly with the generalized memory polynomial (GMP) model, which is inefficient and power-intensive.
Innovation Solution
Implementing a nonlinear signal processing device that uses a time-varying linear filter with coefficients derived from the input signal, decoupling linear and nonlinear memory, and employing lookup tables to reduce logic gate count, allowing for a more efficient implementation of GMP filters with complexity comparable to linear FIR filters, thereby reducing power consumption and computational load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If generalized memory polynomial (GMP) filters are used for digital predistortion of RF-PAs, then signal quality and linearity are improved, but computational complexity and power consumption increase significantly
Solution Approach 1:
The patent segments the GMP filter into multiple parallel sub-filters, each processing a specific term of the polynomial expansion. This allows independent optimization of each sub-filter and enables selective activation based on operating conditions, reducing overall computational complexity while maintaining signal quality compensation.
Solution Approach 2:
The patent implements a truncated GMP model that uses only the most significant terms of the polynomial expansion. By selecting an optimal truncation point, the system achieves sufficient distortion compensation for high linearity performance while dramatically reducing the number of multiplications required compared to the full infinite series.
2Measurement precision
If GMP filters with memory effects are implemented, then distortion compensation accuracy is improved, but the number of multiplications scales with signal taps multiplied by envelope taps
Solution Approach 1:
The patent separates the handling of signal taps and envelope taps into distinct processing paths. The signal path processes complex baseband samples while the envelope path processes magnitude information, with results combined through efficient multiplication strategies that reduce redundant operations.
Solution Approach 2:
The patent employs dynamic coefficient selection where the number of active filter taps is adjusted based on the current operating point of the PA. During low-power operation, fewer taps are activated, reducing computational load while maintaining adequate compensation accuracy for the current signal conditions.
3Measurement precision
If high-order GMP models are used to meet increasing linearity requirements, then PA linearity is improved, but heat dissipation becomes problematic
Solution Approach 1:
The patent uses a truncated GMP model that includes only the necessary number of polynomial terms to achieve the required linearity performance. By carefully selecting the truncation order based on measured PA characteristics, the system avoids implementing higher-order terms that would increase computational complexity and power consumption without providing meaningful performance improvement.
Solution Approach 2:
The patent dynamically adjusts the model order and number of active taps based on operating conditions such as output power level and signal characteristics. This adaptive approach ensures sufficient linearity compensation is applied when needed while reducing computational effort and power consumption during operation modes where full compensation is not required.
Data Source
AI summary
The present disclosure relates to a concept of nonlinear signal processing which may be used for predistortion for RF power amplifiers. The concept includes generating time variant filter coefficients for a linear filter circuit based on a nonlinear mapping of an input signal, and filtering the input signal with the linear filter circuit using the time variant filter coefficients in order to generate a filtered output signal. Thus, it is proposed to implement a non-linear filter by a time-varying linear filter where the time-varying coefficients are derived from the input signal.