Comparison of methods for spectral estimation with interrupted data

The problem considered is to match the periodogram (spectrum) of a real sampled data sequence when only the samples outside a gap are available: that is, when the samples in the gap are missing or corrupted. Different arguments lead to three reasonable estimation algorithms. Tests with contrived data records indicate that two of these algorithms are preferable, one when the gap length is less than 15% of the record, and the other for 20%-50% gaps. An algorithm based on an autoregressive model is found to have an estimate performance that is relatively independent of gap length