Spectrogram Noise Reduction for Low-SNR Signal Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing High Probability of Intercept (HPOI) receivers are limited by high noise levels, causing signals of interest to be masked below the noise floor, and aggressive noise reduction algorithms degrade signal identification and classification processes.
Innovation Solution
A method involving a neural network-based noise reduction system that transforms captured data into a spectrogram, applies thresholding algorithms, and optimizes noise reduction using wavelet basis functions and convolution filters to enhance signal-to-noise ratio while preserving signal integrity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a wideband High Probability of Intercept (HPOI) receiver is used to detect signals in real-time, then the coverage of signal detection is improved, but the noise floor increases causing signals of interest to be masked below the noise floor
Solution Approach 1:
The patent segments the wide bandwidth into multiple narrowband channels, each processed independently through parallel FFT operations. This segmentation allows the system to maintain wideband coverage while reducing the noise floor in each individual channel, as noise power is distributed across multiple narrower frequency bins rather than concentrated in a single wide bin.
Solution Approach 2:
The patent transforms the one-dimensional time-domain signal into a two-dimensional time-frequency representation using parallel FFT operations at different frequency offsets. This dimensional transformation allows signals to be identified in the frequency domain where they stand out from the noise, effectively adding a frequency dimension to the detection process.
2Measurement precision
If aggressive noise reduction algorithms are applied to detect signals below the noise floor, then the signal detection capability is improved, but signal degradation occurs that interferes with identification and classification processes
Solution Approach 1:
The patent applies partial noise reduction by only processing portions of the spectrogram data through noise reduction algorithms, rather than aggressively processing the entire dataset. The system identifies candidate signal regions first, then applies targeted noise reduction only in those regions, preserving overall signal integrity while improving detection capability for specific signals of interest.
Solution Approach 2:
The patent performs preliminary signal identification by detecting peaks and candidate signals in the spectrogram before applying noise reduction algorithms. This preliminary action allows the system to target noise reduction efforts only where needed, preventing degradation of signals that are already clearly visible while enhancing detection of weaker signals buried in noise.
Data Source
AI summary
A method for assessing a noise reduction algorithm and subsequent training of a neural network including the steps of: (1) providing a spectrogram containing noise and at least one signal of interest to the neural network; (2) analyzing the spectrogram to create a transfer function for a noise reduction algorithm; (3) applying the noise reduction algorithm to the spectrogram to produce a noise reduced spectrogram; (4) identifying a signal of interest within the noise reduced spectrogram; (5) calculating an assessment criterion for the noise reduced spectrogram or the signal of interest; and (6) iteratively creating an updated transfer function based on a value of the assessment criteria.