Modulation Scheme Identification With Unknown Signal Rejection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional modulation scheme identification methods face challenges in supporting a wide range of modulation schemes, particularly those like amplitude and phase-shift keying (APSK), and often misidentify unknown modulation schemes due to training data limitations.
Innovation Solution
A modulation scheme identification apparatus using deep learning to automatically extract features and output probability distributions, coupled with a matching rate calculator and determination circuit to accurately identify known schemes and flag unknown schemes based on a threshold.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If a convolutional neural network is trained to classify signal waveforms into known modulation schemes, then the identification speed and automation are improved, but the reliability deteriorates when unknown modulation schemes are input because the system erroneously identifies them as known schemes
Solution Approach 1:
The patent introduces an intermediary verification mechanism (matching rate calculator) between the neural network classification and the final identification result. This intermediary calculates how well the input signal matches the learned probability distribution of each modulation scheme, and only accepts the neural network's classification when the matching rate exceeds a threshold. This resolves the contradiction by maintaining automation while adding a reliability check that prevents erroneous identification of unknown schemes.
2Adaptability or versatility
If conventional feature calculation methods are used to support many modulation schemes, then the adaptability is improved, but the device complexity increases due to requiring different features for each modulation scheme
Solution Approach 1:
The patent applies self-service by using the neural network to automatically learn and extract features from raw signal waveforms without human intervention. The system trains the neural network on labeled data, allowing it to autonomously develop the feature extraction capabilities needed to support multiple modulation schemes. This eliminates the need for manual feature design for each scheme while maintaining high adaptability.
3Ease of operation
If deep learning is used to automatically learn features from signal waveforms, then the ease of operation is improved by supporting many schemes without manual feature design, but the loss of information occurs when the system cannot distinguish unknown schemes from known ones
Solution Approach 1:
The patent implements feedback by using the matching rate calculation to verify whether the neural network's classification is reliable. The system compares the input signal against the learned probability distributions and feeds back this matching information to determine whether to accept the classification or declare the scheme unknown. This feedback mechanism prevents information loss by ensuring that unknown schemes are not mistakenly classified as known ones.
Data Source
AI summary
A modulation scheme identification apparatus includes: a modulation scheme specification circuitry that outputs a first probability distribution for a plurality of modulation schemes for a signal waveform input; a matching rate calculator that calculates a matching rate between the first probability distribution and a second probability distribution defined for a first modulation scheme that is a modulation scheme having a highest probability in the first probability distribution; and a determination circuitry that determines whether the modulation scheme of the signal waveform is the first modulation scheme or an unknown modulation scheme, based on the matching rate.