[Discuss-gnuradio] Correlation Estimator Interpolation Technique

[Paul Garver]

I’ve been testing the Correlation Estimator block (corr_est_cc) and it’s a really nice block overall.  However, I’ve noticed the time_est tag for estimating the fractional (sub-sample) delay doesn’t appear to be very accurate. I’ve attached my example configuration (pnseq_sim_corrsync.grc) although it won’t run without the filter taps file.

I’m not familiar with the “center of mass” interpolation method (paper reference?), so I implemented parabolic interpolation [1] which appears to be more accurate. It is still a biased estimator [2]  but the optimal ones are much more complicated [3]. I’m proposing replacing the existing interpolation technique with parabolic interpolation.

[1] https://github.com/garverp/gnuradio/tree/corr_est_parabolicinterp

[2] “Analysis of Discrete Implementation of Generalized Cross Correlator” R. Boucher and J. Hassab iEEE Transactions on Acoustics, Speech, and Signal Processing 

[3] "Optimum estimation of time delay by a generalized correlator” J. Hassab and R. Boucher iEEE Transactions on Acoustics, Speech, and Signal Processing 

pnseq_sim_corrsync.grc
Description: Binary data