For a sample LRO test case ('17 Sep 2016 00:00:00.000'), we solved for a RangeRate bias for one of the stations. At the start of the run, we had data for the station of interest but eventually all the data for that station was edited out.
This caused the normal matrix to have a zero row which generated a matrix inversion error.
Implement the GTDS or similar algorithm to address this issue.
See MJN email dated '17 Sep 2016 00:00:00.000' for a discussion on how GTDS handles this issue.