Trellis-Based Postprocessing for Short Delay Measurement Using NDFT

Resumo

For digital measurement of short delays between fast analogues signals can be used phase shift method with non-uniform sampling and non-uniform Fourier transform (NDFT), but if the signals are short, non-uniform sampling causes random spurious peaks in the spectrum. In case we have sequence of spectrograms with slow frequency change, correct peaks could be found using evaluation of such sequence. Our new Trellis-based postprocessing selects several greatest peaks from each spectrogram in the sequence and interprets them as nodes of an acyclic directed graph. These nodes are weighted and correct ones are found by graph algorithm. Experiments showed that our method provides correct results in cases when ordinary peak detection is not able to distinguish correct and spurious peaks.