Characterizing Frequency Stability Measurements Having Multiple Data Gaps

Time series measurements with data gaps (dead times) prevent accurate computations of frequency stability variances such as the Allan variance (AVAR) and its square-root the Allan deviation (ADEV). To extract frequency distributions, time-series data must be sequentially ordered and equally spaced. Data gaps, particularly large ones, make ADEV estimates unreliable. Gap imputation by interpolation, zero-padding, or adjoining live segments, all fail in various ways. We have devised an algorithm that fills gaps by imputing an extension of preceding live data and explaining its advantages. To demonstrate the effectiveness of the algorithm, we have implemented it on 513-length original datasets and have removed 30% (150 values). The resulting data is consistent with the original in all three major criteria: the noise characteristic, the distribution, and the ADEV levels and slopes. Of special importance is that all ADEV measurements on the imputed dataset lie within 90% confidence of the statistic for the original dataset.

Automated summary

This article introduces an algorithm for filling data gaps in time series measurements and demonstrates its effectiveness through experimentation, maintaining consistency in the data.