search

tomojitakasu / RTKLIB_bin

301 stars 220 forks source link

sampling interval confuses PPP #10

Open fgnievinski opened 4 years ago

fgnievinski commented 4 years ago

when I change the sampling interval from ten seconds to one second the vertical coordinate is scaled by a factor of four -- for example, from 5 cm at 4h to 20cm at 4h (as per plots below). I'm also attaching the configuration file, renamed as a text file.

RS4427235643201909210000d-ppp0.conf.txt

image image

luca-nicolini commented 4 years ago

I often do PPP with RTKLIB, so I would like to suggest to you:

  1. If your antennna is fix, use pos1-posmode=ppp-static
  2. It is always a good idea to exclude low SNR observations, I use: pos1-snrmask_r=on pos1-snrmask_L1[L2;L5]=35,35,35,35,35,35,35,35,35
  3. In PPP you need to consider the PCV: to do this you have to insert the antenna type: ant1-anttype=[antenna type]

Finally, and probably most important, check the precise products you are using. The satellites clock bias is not continous so if you want to use high rate observations you need high rate clock products. As far as I know the highest rate is 5 s, which is enough for 10 s observation rate but not optimal for 1 s.

fgnievinski commented 4 years ago

Thanks for your suggestions.

  1. I'm trying to detect small (decimeter level) displacements in a fishing platform, so I'd need to leave it kinematic instead of forcing it to static.

  2. I've applied the SNR mask, it didn't seem to help much, perhaps because that site has 360-degree clear view of the sky.

  3. I've enabled the (maybe auto should be the default), although it should create mostly a constant bias, not a scaling factor.

  4. now I've used high-rate (1-second) clock products from CDDIS: https://cddis.nasa.gov/Data_and_Derived_Products/GNSS/gnss_igsrtclk.html Unfortunately, the problem persists:

    • here is the solution for one-second observations: image
    • here is the solution for ten-second observations: image

I was expecting that one would be just a higher resolution version the other, perhaps noisier, but with similar trajectories after low-pass filtering.

luca-nicolini commented 4 years ago

Mmh this is a strange behaviour... You have an observation file with sampling rate at 1 Hz and you simply select a different interval for the analysis, is it correct? I do not have any other suggestions for now...