SNR Estimation via Statistical Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating signal-to-noise ratio (SNR) in communications systems suffer from low accuracy and fluctuation across SNR ranges, and data-aided methods require additional power consumption and reduce channel capacity.
Innovation Solution
A method utilizing a machine learning agent, such as an artificial neural network, to process extracted features from signal samples, providing a vector of probabilities for SNR estimation, which improves accuracy and maintains consistency across a wide SNR range without the need for additional data transmission.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If data-aided methods are used for SNR estimation, then accuracy is improved, but power consumption increases and channel capacity is reduced
Solution Approach 1:
The patent extracts and utilizes statistical features (mean, variance, skewness, kurtosis) directly from the received signal samples to estimate SNR, eliminating the need for additional pilot data transmission. This extraction approach achieves accurate SNR estimation without the overhead of data-aided methods.
Solution Approach 2:
The received signal itself serves as the source of information for SNR estimation. By computing statistical moments from the received samples, the system uses the signal's own characteristics to determine SNR, without requiring external assisting data or additional power consumption for pilot transmission.
2Ease of manufacture
If traditional estimators (M2M4, SVR) are used, then implementation is simple, but accuracy is low and fluctuates across SNR ranges
Solution Approach 1:
The patent segments the SNR estimation problem into multiple statistical feature extractions (first moment/mean, second moment/variance, third moment/skewness, fourth moment/kurtosis), where each feature captures different aspects of the signal distribution. This segmentation allows the system to maintain accuracy across varying SNR conditions while keeping individual computations simple.
Solution Approach 2:
The patent combines multiple statistical features (mean, variance, skewness, kurtosis) into a composite estimation approach. By integrating information from different statistical moments, the system achieves robust and accurate SNR estimation across the entire SNR range, overcoming the limitations of single-feature estimators.
3Measurement precision
If pilot data is transmitted for SNR estimation, then estimation accuracy is improved, but channel capacity for actual data transmission is reduced
Solution Approach 1:
The patent extracts sufficient SNR estimation information directly from the statistical properties of the received signal samples themselves, eliminating the need to allocate channel resources to pilot data transmission. This extraction method achieves accurate estimation without sacrificing channel capacity for actual data transmission.
Solution Approach 2:
The received signal performs dual functionality: it carries both the information needed for SNR estimation and the actual data transmission. By using the signal's own statistical characteristics for estimation, the system eliminates the separate pilot transmission requirement, maximizing channel capacity utilization.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method (500) of determining an estimate of a characteristic of a set of samples, the method comprising, at a processor: receiving (501) the set of samples; extracting (503), from the samples, a value of at least one feature indicative of the characteristic; providing (505) the at least one extracted feature value as an input to a machine learning agent; and causing (507) the machine learning agent to process the at least one extracted feature value to provide an output indicative of the estimate of the characteristic, wherein: the output of the machine learning agent is a vector comprising elements corresponding to the probabilities associated with different predetermined characteristic values.