Sub-Band Time Delay Estimation for Noisy Reverberant Signals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing time delay estimation methods, such as GCC-PHAT, are ineffective in high background noise and moderate reverberation conditions, limiting their accuracy in applications like source localization and beamforming.
Innovation Solution
A signal processing system that employs a filter bank to divide signals into sub-bands, followed by cross-correlation analysis and normalization, integrated across frequency sub-bands to estimate time delay, mimicking human cochlear processing for robustness against noise and reverberation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If GCC-PHAT method is used for time delay estimation, then the method performs satisfactorily with low and moderate levels of background noise, but the method does not do well with larger levels of background noise or moderate reverberation
Solution Approach 1:
The patent divides the frequency spectrum into multiple sub-bands using a filter bank, processing signals in separate frequency regions. This segmentation allows the system to handle noisy environments more effectively by analyzing correlations in individual sub-bands and combining results, rather than processing the entire frequency spectrum as a single unit, which is vulnerable to noise contamination.
Solution Approach 2:
The patent transforms the time-domain signals into frequency-domain representations through Fourier transforms and further decomposes them into sub-bands. By changing the domain from time to frequency and introducing sub-band decomposition, the system alters the parameters of signal representation to improve robustness against noise and reverberation in time delay estimation.
2Measurement precision
If traditional time delay estimation methods are used, then the processing is simpler, but the accuracy deteriorates in noisy and reverberant environments
Solution Approach 1:
The patent applies filter bank segmentation to divide the signal processing into multiple parallel sub-band channels. Each sub-band can be processed independently through cross-correlation, and the results are integrated to produce the final time delay estimate. This segmented approach improves measurement precision in noisy environments while distributing the computational complexity across multiple simpler parallel operations.
Solution Approach 2:
The patent transitions from one-dimensional time-domain analysis to two-dimensional frequency-time analysis by introducing frequency sub-band decomposition. This dimensional expansion allows the system to exploit frequency-domain properties and sub-band correlations that are not visible in the time domain, improving estimation accuracy at the cost of increased processing complexity.
Data Source
AI summary
A method for time delay estimation performed by a physical computing system includes passing a first input signal obtained by a first sensor through a filter bank to form a first set of sub-band output signals, passing a second input signal obtained by a second sensor through the filter bank to form a second set of sub-band output signals, the second sensor placed a distance from the first sensor, computing cross-correlation data between the first set of sub-band output signals and the second set of sub-band output signals, and applying a time delay determination function to the cross-correlation to determine a time delay estimation.