08.06.2013 Views

Bernese GPS Software Version 5.0 - Bernese GNSS Software

Bernese GPS Software Version 5.0 - Bernese GNSS Software

Bernese GPS Software Version 5.0 - Bernese GNSS Software

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

14. Clock Estimation<br />

the clock values between two epochs. The purpose of the clock jump detection is to get<br />

a reasonable model for the clock behavior for the clock extrapolation and to update the<br />

statistic summary in the program output.<br />

The procedure is composed of four steps:<br />

(1) In a first step the mean change (absolute value only) of the clock value per second is<br />

computed for all epoch differences:<br />

∆δ = 1<br />

n − 1<br />

n<br />

<br />

<br />

<br />

<br />

δ(ti) − δ(ti−1) <br />

<br />

<br />

ti − <br />

ti−1<br />

i=2<br />

The mean noise level of the clock σ(∆δ) is the standard deviation of the mean<br />

change ∆δ.<br />

A jump hypothesis is setup if the change between two epochs exceeds a specified<br />

threshold which depends on the mean value ∆δ, the noise of the clock σ(∆δ), and an<br />

user-specified confidence interval n (option “Confidence interval”, see Figure 14.7):<br />

<br />

<br />

<br />

<br />

δ(ti) − δ(ti−1) <br />

<br />

<br />

ti − > ∆δ + n · σ(∆δ) .<br />

ti−1<br />

The mean change of the clock per second is computed iteratively without using the<br />

epochs flagged with a jump hypothesis until no further jumps are found.<br />

If the clock is very noisy no jump detection is done because a distinction between<br />

noise and jumps is no longer possible. The limit is hardwired to σ(∆δ) > 100 µs<br />

in subroutine ${LG}/CCJUMP.f90. On the other hand the computed noise level of<br />

excellent clocks can be very small. Because it makes no sense to detect “jumps” on the<br />

few picosecond level in a <strong>GPS</strong> derived solution, the noise σ(∆δ) needs to have a lower<br />

limit which is a user input option (“Minimum RMS for jump detection”, see Figure 14.7).<br />

(2) The next step is to check the list of possible clock jumps for outliers in the clock time<br />

series. If two clock jumps are found for the same clock within a few epochs and with<br />

the same size (but the opposite sign) the clock values between these two events are<br />

marked as outliers:<br />

<br />

<br />

<br />

<br />

δ(tj+1) − δ(tj) δ(ti) − δ(ti−1) <br />

<br />

−<br />

≤ n · σ(∆δ)<br />

tj+1 − tj<br />

ti − ti−1<br />

where |ti − tj| is smaller than the user-specified limit for “few epochs” in option<br />

“Maximum time interval for outlier detection”, see Figure 14.7. The two jumps are deleted<br />

from the list of clock jumps as long as no jump is found just before the first or after<br />

the second jump.<br />

(3) To distinguish a real clock jump from noise the uncertainty of the mean clock rate<br />

has to decrease significantly if the clock change at the jump epoch is not considered<br />

for its computation. For this test only a few clock values around the event are used. If<br />

no significant improvement can be found when introducing the jump the clock event<br />

is removed from the list of clock jumps. The value n from option “Confidence interval”<br />

is used here to define the significance level.<br />

Page 304 AIUB<br />

.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!