The non-integer (second) epochs for your observations might be a problem.
AFAIK all the RINEXes we deal with (for time-transfer, from expensive duaö-freq receivers) produce observations at 0.0 s and 30.0 s past the top-of-minute (GPS time-scale).
So some tests would be:
- try to find a RINEX with non-integer second epochs that does pass through csrs-ppp successfully - this would prove the service can handle it
- artificially force your observations to 0.0 and 30.0, and see what happens. Unless you also propagate the pseudoranges and carrier-phases when you change the epoch the results might be all wrong..
- ask/investigate the available online ppp-enginges (csrs-ppp, or the other ones, jpl?) if they can handle non-integer epochs
- ask/investigate the offline codes (rtklib, ESA gLAB).
Anders