Arbitrary error metrics  software
This webpage describes the
Matlab files
used in our work on signal estimation with arbitrary error metrics, described in the following paper,
J. Tan,
D. Carmon,
and D. Baron,
"Optimal Estimation with Arbitrary Error Metrics in Compressed Sensing,"
to appear in IEEE Statistical Signal Processing workshop,
Ann Arbor, MI, August 2012
(pdf).
The software was implemented by
Jin Tan,
based on an earlier implementation by
Sundeep Rangan
for estimation with relaxed belief propagation
(Matlab). The implementation uses Bayesian method to minimize the expected error metric of interest.
Below is a brief description of files used in our implementation. Comments will be appreciated.
Jin Tan, July 2012
Main Files

main_MetricOpt.m: script that runs the relaxed belief propagation algorithm and our metricoptimal algorithm.

Px_Gen.m: function that generates the prior distribution p(x).

Pxr_Gen.m: function that generates the conditional distribution p(xr), where r is the output of the scalar Gaussian channel by decoupling principle.

Dist_Gen.m: function that generates the distortion matrix, used to implement expected error metric integration numerically.
External Files

main (folder): the relaxed belief propagation algorithm implementation by
Sundeep Rangan.
Note that this folder must be in the Matlab path before running main_MetricOpt.m.
Back to my homepage.