Order-Statistic Spectral Filtering for Transient Signal Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional linear filters fail to discriminate between transient signals and long-duration signals, leading to false detections due to variations in background noise and noise floor, and their detection thresholds are influenced by the signals they are trying to detect.
Innovation Solution
A non-linear, order-statistic filter that determines a transient signal detection threshold based on a power spectral estimate, using a preselected range of magnitude values within a predetermined bandwidth portion, excluding transient signals from influencing the threshold setting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If linear averaging filters are used to set detection thresholds, then the threshold is based on average signal power, but transient signals and long-duration signals cannot be discriminated leading to false detections
Solution Approach 1:
The patent changes the filtering parameter from linear averaging to order-statistic filtering. Instead of computing the mean of signal powers, the system uses the median or other order statistics of the sorted power values. This parameter change enables discrimination between transient signals (which appear as outliers in the sorted sequence) and continuous signals (which populate the middle values), thereby improving detection accuracy while reducing false detections from long-duration signals
Solution Approach 2:
The patent inverts the conventional approach by having the filter threshold adapt to exclude transient signals rather than including them. Conventional linear filters are influenced by all signals including transients; this invention inverts the behavior so that the order-statistic filter specifically excludes transient signals from threshold calculation by selecting values from the middle portion of the sorted power spectrum, making the threshold insensitive to transient energy
2Reliability
If the detection threshold is set above the noise level to avoid false detections from noise, then transient signals can be detected, but long-duration signals are also excluded from detection
Solution Approach 1:
The patent changes the threshold setting parameter from a fixed offset above noise level to a dynamic value derived from order statistics of the power spectrum. By computing the threshold based on the median or specific percentile of sorted power values, the system automatically adapts to the noise floor while maintaining the ability to detect transients, as transient signals appear as high-value outliers that do not influence the median-based threshold
3Measurement precision
If linear filters are used, then the detection threshold is influenced by all detected signals including transients, but this causes the threshold to rise and miss transient signals
Solution Approach 1:
The patent extracts transient signals from the threshold calculation process by using order-statistic filtering. Instead of averaging all signal powers (which includes transients), the system sorts the power values and selects the median or a specific percentile from the sorted sequence. This extraction removes the influence of transient signals (which appear as extreme values at the ends of the sorted sequence) from the threshold computation, keeping the threshold stable and sensitive to transient detection
Solution Approach 2:
The patent changes the aggregation parameter from arithmetic mean to order statistic (median or percentile). This parameter change fundamentally alters how the threshold is computed: instead of being pulled upward by transient energy in the mean calculation, the threshold is determined by the central tendency of the power distribution, making it insensitive to transient signals while maintaining accuracy in representing the background signal level
4Reliability
If the cutoff frequency of the low-pass filter is lowered to reduce transient signal influence, then false detections from transients decrease, but long settling times are required to set the threshold
Solution Approach 1:
The patent substitutes the mechanical low-pass filtering mechanism with a computational order-statistic approach. Instead of using a low-pass filter with a specific cutoff frequency that requires time to settle, the system directly computes the median or percentile of the sorted power values. This substitution eliminates the need for temporal filtering and settling, providing instantaneous threshold adaptation without the time delays inherent in low-pass filter responses
Data Source
AI summary
A non-linear, order-statistic filter that is able to detect transients within an ambient signal environment without the transients affecting the setting of the transient detection threshold. The filter obtains a power spectral estimate (PSE) signal for each one of a plurality of predetermined bandwidth portions of the ambient input signal. Specific PSE magnitude values making up the PSE signal are sorted and ordered from smallest to largest. An average PSE magnitude value is obtained for a preselected center range of the specific PSE magnitude values, and this value is used to set the transient signal detection threshold for the filter.