So RI is much faster than DMP. Figure 6 shows the cumulative distribution function (CDF) of the distance from the estimator to the actual source under DMP and RI.
To account for the extra variability due to sampling the source population from the (infinite) superpopulation, we propose to use the Lin variance estimator [ 13] that turned out to consist in adding the naïve variance estimator to the robust variance estimator V ^ 1 β ^.
The source localization problem has been studied, e.g., in [1 3], [[4], p. 4089 ff], [[5], p. 746 ff] and [6], all of which use a sequential Bayesian estimator [7] to infer the source position states from observations using multiple sensors.
Note that the proposed CWLS estimator (14) is different from the estimators in [4, 10].
In a classical setting, a shrinkage estimator is a weighted average between the estimator from the concerning feature and a pooled estimator from all features.
The conversion function is a (piecewise) linear estimator that estimates the target features from the source features.
A simple alternative to the estimator we are suggesting when having data from one source is to assume that the samples come from a zero-truncated distribution and use all, or parts of the data to estimate the unobserved zero bin (e.g. [ 10]).
However, aside from resonance, the peak-bias-free estimator proves to be the worst of the three estimators.
If there was clear evidence that this reduction in precision was necessary to remove an important source of bias from the estimates, the ADR/DLR estimators might be preferred.
From Fig. 1 a, it can be seen that the LV-MUSIC estimator can only correctly find the fourth sources, because we fix the polarization parameters to the fourth sources and perform a 2D search.
The temporal based estimator has another expected source of imprecision: the temporal distance between samples.
