Discriminator Using Filter Bank and Softmax for Signal Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing signal discrimination methods, particularly those using kernel density functions, face challenges in efficient calculation and scalability when approximating probability distributions and extending to multiple classes, making it difficult to achieve high accuracy within realistic time frames.
Innovation Solution
A discriminator comprising a filter bank of nonlinear matched filters, a softmax function, and a parameter updating unit using cross-entropy loss and extended Kalman filters to adjust parameters, enabling efficient and accurate signal discrimination across multiple classes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a probability distribution is approximated using a kernel density function to improve discrimination accuracy, then the calculation amount increases and it becomes difficult to implement efficient calculation within realistic time
Solution Approach 1:
The patent changes the fundamental parameter approximation method from kernel density function to histogram-based probability distribution. This parameter change maintains discrimination accuracy while dramatically reducing calculation complexity from O(n^2) to O(n), enabling efficient real-time calculation
Solution Approach 2:
The patent uses simple histogram bins as disposable approximation structures instead of complex kernel density functions. Each bin provides a coarse but sufficient probability estimate, sacrificing fine-grained precision for computational efficiency and enabling rapid updates
2Measurement precision
If the discriminator is extended to many classes to improve discrimination accuracy, then the calculation amount increases and extension becomes difficult
Solution Approach 1:
The patent segments the probability distribution calculation into independent histogram bins for each class. This segmentation allows each class to be processed independently and in parallel, reducing overall system complexity while maintaining accuracy for multi-class discrimination
Solution Approach 2:
The histogram-based probability distribution calculation serves as a universal mechanism that works for any number of classes. The same simple binning approach can handle binary, multi-class, or even continuous classification tasks, making the system universally applicable without increasing complexity
Data Source
AI summary
A discriminator includes: a filter bank having a response characteristic to a signal with a specific waveform and including a plurality of matched filters transforming a time-series input signal into a plurality of features in accordance with the response characteristic; a softmax function configured to accept the plurality of features and transform the plurality of features into a probability distribution; and a loss function configured to obtain a cross-entropy loss between the probability distribution and a class label. The parameter of each of the plurality of matched filters is adjusted based on the cross-entropy loss.