Modulation Format Estimation Using Frequency-Shifted Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In unlicensed radio wave monitoring and cognitive radio, accurately estimating the modulation format of unknown signals with unknown center frequencies is challenging, especially when the center frequency cannot be accurately grasped, leading to difficulties in identifying the modulation format, particularly for signals with short signal lengths.
Innovation Solution
A modulation format estimation device and method that includes a frequency shift correction unit, frequency error generation unit, and a machine learning model to estimate and correct frequency shifts, generating learning baseband signals with introduced frequency errors, and using these signals to accurately estimate the modulation format of received signals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional modulation format estimation methods are used for signals with unknown center frequencies, then the estimation process becomes extremely complex and time-consuming, but the patent achieves high-speed and high-accuracy estimation by using machine learning models trained with frequency-shifted learning data
Solution Approach 1:
The patent applies preliminary action by pre-training the machine learning model with learning data that has been frequency-shifted by various error amounts. This preparation in advance allows the model to automatically compensate for frequency estimation errors during actual operation, achieving both high speed and high accuracy without complex real-time processing
Solution Approach 2:
The patent changes the parameter of frequency shift amount in the learning data to create multiple versions with different frequency errors. By training the model with these varied frequency conditions, the model learns to accurately estimate modulation formats even when the center frequency is not precisely known, resolving the contradiction between speed and accuracy
2Measurement precision
If the center frequency of the received signal is not accurately matched with the baseband conversion frequency, then the constellation characteristics become distorted and modulation format identification becomes difficult, but the patent overcomes this by using machine learning models trained with frequency error variations
Solution Approach 1:
The patent applies self-service by enabling the machine learning model to automatically compensate for frequency mismatch errors. The model, trained with frequency-shifted data, self-corrects for center frequency inaccuracies without requiring complex external frequency correction circuits or manual adjustment, simplifying the overall system while maintaining high precision
3Adaptability or versatility
If a wide frequency band is monitored for unlicensed radio wave monitoring and cognitive radio, then more unknown signals need to be analyzed, but the patent enables efficient analysis by reducing the required signal data length through machine learning
Solution Approach 1:
The patent applies preliminary action by pre-training the machine learning model with diverse frequency-shifted data covering wide frequency bands. This advance preparation enables the model to quickly analyze unknown signals from any frequency band within the training range, reducing the signal data length required for accurate modulation format estimation and enabling efficient wideband monitoring
Data Source
AI summary
A modulation format estimation device 100 includes: a frequency shift correction unit 112 configured to estimate the amount of a frequency shift using a baseband signal acquired from a received signal and correct the baseband signal based on an estimation result; a frequency error generation unit 122 configured to generate a plurality of frequency errors from a range set based on an error occurring in the estimation of the frequency shift amount; a frequency error introduction unit 123 configured to acquire learning baseband signals in which each of a plurality of source signals modulated by different modulation formats is frequency-shifted by each frequency error; and a modulation format estimation unit 113 configured to input a corrected baseband signal to a first machine learning model created by machine learning using learning data including the plurality of learning baseband signals and a label, and estimate a modulation format of the received signal.