In the regression test of LDPC Encoder/Decoder, a single block is incorrectly decoded in one of the tests.
The failure is limited to one block of incorrect data output.
This is a known issue in LDPC Encoder/Decoder v1.0 of Vivado 2017.3, which has been fixed in the 2017.4 release.
A particular sequence of events (block scheduling, early termination) has hit a condition of the scheduler that results in an LDPC code layer length being read incorrectly for one layer of the block.
This results in the particular block output being incorrect (along with the associated parity check). The circuit recovers and subsequent blocks are decoded correctly.
This behavior has only been seen in one test, after several months of randomized testing, so it is a low probability event.
For other LDPC Encoder/Decoder known issues, please refer to (Xilinx Answer 69399)