We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
A Deconvolution Network with Squared Activation Function for Decomposition of Monotonic Multi-Component Signals.
- Authors
Shtrauss, V.
- Abstract
The article is devoted to improving quality of decomposition of monotonic multi-component time- and frequency-domain signals having narrow distributions of time constants (DTC). To solve the problem, the inverse functional filters operating with geometrically sampled data (equispaced data on a logarithmic scale) [V. Shtrauss. Signal Processing, 45, 293-312 (1995)] are combined into a processing unit consisting from several functional filters with common inputs, which outputs are nonlinearly activated by the square activation function, multiplied by weights and summed. It is demonstrated that the square activation function provides some useful features for the decomposition problem under consideration. First, contrary to conventional activation functions (sigmoid, radial basis functions) the square activation function enhances the spikyness of DTC. Second, it ensures physically justified nonnegativity for the recovered DTC. Third, the square activation function transforms the Gaussian input noise into the nonnegative output noise with specific probability distribution having the standard deviation proportional to the variance of input noise, which increases radically the noise immunity of the proposed nonlinear algorithms. Simulations are provided demonstrating improvement of quality of decomposition for a frequency-domain multi-component signal
- Subjects
SIGNAL processing; SIGNAL detection; INFORMATION measurement; GAUSSIAN distribution; RANDOM noise theory; ANALYSIS of variance; DISTRIBUTION (Probability theory); SYSTEMS engineering; SIMULATION methods &; models; MATHEMATICAL models
- Publication
Telecommunications & Electronics, 2008, p6
- ISSN
1407-8880
- Publication type
Article