Stochastic Buffer Management for Adaptive Digital Predistortion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional digital predistortion (DPD) adaptation processes are computationally intensive, leading to reduced accuracy and robustness due to limitations in computational resources, which affects the speed and quality of DPD coefficient derivation.
Innovation Solution
A stochastic gradient approach with pseudo-random buffer management is employed, allowing for independent scheduling of buffer updates and sample selection, enabling efficient computation of DPD coefficients by approximating sub-gradient values based on pseudo-randomly selected output samples.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional DPD adaptation processes are used to achieve accurate coefficient derivation, then measurement precision is improved, but device complexity and computational resources are excessively consumed
Solution Approach 1:
The patent changes the parameter of buffer management from deterministic scheduling to stochastic scheduling. By using random buffer management where buffer locations are selected randomly for reading and writing operations, the system reduces computational complexity while maintaining coefficient derivation accuracy. This parameter change transforms the adaptation process into a more computationally efficient operation.
Solution Approach 2:
The patent employs a disposable buffer approach where buffer contents are randomly accessed and then replaced with new samples without requiring complex management of buffer lifecycle. This simplifies the computational overhead by treating buffer management as a simple replace-when-full operation rather than a complex scheduling task.
2Reliability
If conventional DPD adaptation processes are used to ensure robustness, then reliability is improved, but productivity is reduced due to computational limitations
Solution Approach 1:
The patent changes the temporal parameter of buffer management from synchronized deterministic scheduling to independent stochastic scheduling. Buffer writes occur at regular intervals while buffer reads occur at random times, decoupling the two operations. This parameter change increases productivity by allowing faster coefficient derivation without compromising reliability, as the random sampling still provides sufficient statistical information for accurate adaptation.
Solution Approach 2:
The patent uses partial buffer contents for coefficient derivation by randomly selecting a subset of buffer locations rather than processing the entire buffer in a deterministic manner. This partial action approach maintains robustness through statistical sampling while improving productivity by reducing the effective computational workload.
3Manufacturing precision
If deterministic buffer scheduling is used to manage samples, then manufacturing precision is maintained, but device complexity increases due to coordination requirements
Solution Approach 1:
The patent implements a self-service buffer management system where the buffer automatically manages its own read/write operations through random indexing without requiring external coordination. The buffer simply increments its write pointer and randomly selects read pointers, eliminating the need for complex scheduling logic and reducing device complexity while maintaining sample management accuracy.
Solution Approach 2:
Instead of using deterministic scheduling to manage buffer samples, the patent inverts the approach by using random stochastic scheduling. This inversion simplifies the buffer management logic from a complex coordination task to a simple random selection process, reducing device complexity while maintaining the necessary sample management precision through statistical properties of random sampling.
Data Source
AI summary
Disclosed are implementations, including a method comprising selecting a location of a selection buffer (that stores observed output samples produced by a signal chain) according to a pseudo-random buffer selection process, and accessing at least one observed output sample generated in response to corresponding input samples (to the system) and/or intermediary samples (outputs of a digital compensator) produced by a compensator of the system. The method further includes computing an approximation of a sub-gradient value computed according to a stochastic gradient process to derive a sub-gradient of an error function representing a relationship between the observed output sample, and the input and/or intermediary samples, computing updated values for the compensator coefficients based on current values of the coefficients and the approximation of the sub-gradient value, and replacing, at a time instance determined independently of the buffer selection process, content of part or all of the selection buffer.