Discrete Time Signal Localization Accuracy in Gaussian Noise at Low Signal to Noise Ratios

Abstract

Convolution and matched filtering are often used to detect a known signal in the presence of noise. The probability of detection and probability of missed detection are well known and widely used statistics. Oftentimes we are not only interested in the probability of detecting a signal but also accurately estimating when the signal occurred and the error statistics associated with that time measurement. Accurately representing the timing error is important for geolocation schemes, such as Time of Arrival (TOA) and Time Difference of Arrival (TDOA), as well as other applications. The Cramér Rao Lower Bound (CRLB) and other, tighter, bounds have been calculated for the error variance on Time of Arrival estimators. However, achieving these bounds requires an amount of interpolation be performed on the signal of interest that may be greater than computational constraints allow. Furthermore, at low Signal to Noise Ratios (SNRs), the probability distribution for the error is non-Gaussian and depends on the shape of the signal of interest. In this paper we introduce a method of characterizing the localization accuracy of the matched filtering operation when used to detect a discrete signal in Additive White Gaussian Noise (AWGN) without additional interpolation. The actual localization accuracy depends on the shape of the signal that is being detected. We develop a statistical method for analyzing the localization error probability mass function for arbitrary signal shapes at any SNR. Finally, we use our proposed analysis method to calculate the error probability mass functions for a few signals commonly used in detection scenarios. These illustrative results serve as examples of the kinds of statistical results that can be generated using our proposed method. The illustrative results demonstrate our method’s unique ability to calculate the non-Gaussian, and signal shape dependent, error distribution at low Signal to Noise Ratios. The error variance calculated using the proposed method is shown to closely track simulation results, deviating from the Cramér Rao Lower Bound at low Signal to Noise Ratios.

Keywords

• Covariance matrices
• delay estimation
• digital measurements