Pulse Signal Detection Using CWT Scalogram Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional demodulation techniques struggle to accurately detect pulsed communication signals in noisy environments due to varying pulse shapes and minimal distinguishing features, making it difficult to differentiate valid pulses from interference.
Innovation Solution
A deep learning approach utilizing continuous wavelet transformation (CWT) and scalogram processes to convert time-frequency representations of signals into images, which are then classified by a trained deep learning architecture, enabling the detection and classification of pulsed communication signals amidst noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional demodulation techniques are used to detect pulsed communication signals, then the detection process is simple, but the accuracy of signal detection deteriorates in noisy environments due to varying pulse shapes and minimal distinguishing features
Solution Approach 1:
The patent transforms the one-dimensional time-domain signal into a two-dimensional time-frequency representation using Continuous Wavelet Transformation (CWT). This dimensional transformation allows the deep learning model to analyze signals in both time and frequency domains simultaneously, capturing varying pulse shapes and characteristics that are invisible in the time domain alone, thereby improving detection accuracy in noisy environments
Solution Approach 2:
The patent introduces an intermediary deep learning architecture that acts as a mediator between the raw noisy signal and the final detection result. The CWT scalogram serves as an intermediary representation that enhances distinguishing features, allowing the neural network to learn robust patterns and differentiate valid pulses from interference even when traditional direct detection fails
2Measurement precision
If deep learning approaches with CWT and scalogram processes are used, then signal detection accuracy improves in noisy environments, but the system complexity increases
Solution Approach 1:
The patent applies preliminary signal processing steps (CWT and scalogram generation) before the main deep learning detection task. By pre-transforming the signal into an enhanced time-frequency representation, the system prepares the data in a form that makes distinguishing features more prominent, allowing the subsequent neural network to focus on classification rather than feature extraction, thus managing complexity more effectively
Solution Approach 2:
The deep learning architecture serves multiple functions: it performs feature extraction from the scalogram, classifies pulse patterns, and detects signals in varying noise conditions. This multi-functionality consolidates what would otherwise require separate processing stages into a single unified model, improving accuracy while containing system complexity through functional integration
3Measurement precision
If deep learning architecture is trained with plurality of training images, then classification accuracy of bit information improves, but the training time and computational resources increase
Solution Approach 1:
The patent generates a comprehensive set of training images covering various noise conditions, pulse shapes, and bit states (excessive action in terms of data quantity), but uses efficient deep learning training techniques to process this large dataset. The CWT scalogram method efficiently extracts relevant features, allowing the model to achieve high accuracy with reasonable training time by focusing computational resources on the most informative features
Data Source
AI summary
A method and system for detecting pulsed communication signals, the method and system includes obtaining a time-frequency representation of a received training data signal, the received training data signal representative of a known/training bitstream associated with a transmitted pulse signal; using a scalogram process to convert the time-frequency representation of the received training data signal into a plurality of training images, and training a deep learning architecture platform with the plurality of training images to generate a classification model representative of a plurality of high bit states and a plurality of low bit states included in the plurality of training images. The disclosed method and system uses the CWT process to obtain a time-frequency representation of a target data signal, and uses the scalogram process to convert the time-frequency representation of the received target data signal into a plurality of respective target images representative of each of the bit states in the incoming target data signal bitstream; and using the trained deep learning architecture platform, classifies each of the plurality of respective target images as one of a high bit state and a low bit state; and generates an output bitstream based on the plurality of respective target images associated with the incoming data signal, wherein the incoming target data signal is one or both of turbo encoded and multipulse pulse position modulation (MPPM) encoded.