[time-nuts] GPS message jitter (preliminary results)

Mark Sims holrum at hotmail.com
Tue Jul 19 02:20:13 UTC 2016


Yes,  turning on the display filter only gives an indication that one could gain some some benefit if they were to use the message arrival timing to implement some sort of NTP-ish algorithm to their 1PPS-less clock.

I have added some options to Lady Heather calculate the adevs of the message arrival time,  and also a flag to calculate the message times from the first byte of the message instead of the last byte.   That could be useful if you are dealing with variable length timing messages,  but a lot more difficult to make use of in a simple implementation of a clock.
Here the some xDEVs of the Z3801A :PTIM:TCOD? timing message arriving over a hardware serial port:
##  ADEV over 20292 points - sample period=1.0 secs#      1.000 tau  4.5444e-001 (n=20290)#      2.000 tau  2.2716e-001 (n=20288)#      5.000 tau  9.0872e-002 (n=20282)#     10.000 tau  4.5410e-002 (n=20272)#     20.000 tau  2.2708e-002 (n=20252)#     50.000 tau  9.0837e-003 (n=20192)#    100.000 tau  4.5402e-003 (n=20092)#    200.000 tau  2.2713e-003 (n=19892)#    500.000 tau  9.0882e-004 (n=19292)#   1000.000 tau  4.5393e-004 (n=18292)#   2000.000 tau  2.2708e-004 (n=16292)#   5000.000 tau  9.0686e-005 (n=10292)#  10000.000 tau  4.5325e-005 (n=292)##  HDEV over 20292 points - sample period=1.0 secs#      1.000 tau  4.7906e-001 (n=20289)#      2.000 tau  2.3946e-001 (n=20286)#      5.000 tau  9.5752e-002 (n=20277)#     10.000 tau  4.7872e-002 (n=20262)#     20.000 tau  2.3941e-002 (n=20232)#     50.000 tau  9.5741e-003 (n=20142)#    100.000 tau  4.7861e-003 (n=19992)#    200.000 tau  2.3940e-003 (n=19692)#    500.000 tau  9.5769e-004 (n=18792)#   1000.000 tau  4.7845e-004 (n=17292)#   2000.000 tau  2.3930e-004 (n=14292)#   5000.000 tau  9.5558e-005 (n=5292)##  MDEV over 20292 points - sample period=1.0 secs#      1.000 tau  4.5444e-001 (n=20290)#      2.000 tau  1.6062e-001 (n=20287)#      5.000 tau  4.0620e-002 (n=20278)#     10.000 tau  1.4358e-002 (n=20263)#     20.000 tau  5.1625e-003 (n=20233)#     50.000 tau  5.1755e-004 (n=20143)#    100.000 tau  1.8317e-004 (n=19993)#    200.000 tau  6.5388e-005 (n=19693)#    500.000 tau  9.3230e-006 (n=18793)#   1000.000 tau  1.8958e-006 (n=17293)#   2000.000 tau  7.5843e-007 (n=14293)#   5000.000 tau  1.0835e-007 (n=5293)##  TDEV over 20292 points - sample period=1.0 secs#      1.000 tau  2.6237e-001 (n=20290)#      2.000 tau  1.8547e-001 (n=20287)#      5.000 tau  1.1726e-001 (n=20278)#     10.000 tau  8.2898e-002 (n=20263)#     20.000 tau  5.9611e-002 (n=20233)#     50.000 tau  1.4940e-002 (n=20143)#    100.000 tau  1.0575e-002 (n=19993)#    200.000 tau  7.5503e-003 (n=19693)#    500.000 tau  2.6913e-003 (n=18793)#   1000.000 tau  1.0945e-003 (n=17293)#   2000.000 tau  8.7576e-004 (n=14293)#   5000.000 tau  3.1279e-004 (n=5293)#
------------------
> At this point you're much better off not to use raw or filtered std dev at all -- and just go with ADEV on the raw data, where the tau of the sigma is explicit. 		 	   		  


More information about the Time-nuts_lists.febo.com mailing list