Changeset 11212 in ntrip for trunk/BNC/src/bnchelp.html


Ignore:
Timestamp:
Nov 18, 2026, 12:25:58 PM (98 minutes ago)
Author:
stuerze
Message:

SSR QC and clock combination improved

File:
1 edited

Legend:

Unmodified
Added
Removed
  • trunk/BNC/src/bnchelp.html

    r11205 r11212  
    53875387      The latest code biases of each stream are used,</li>
    53885388    <li>the phase bias signals provided,</li>
     5389    <li>the phase bias indicators: the SSR format, for IGS-SSR and old RTCM-SSR the dispersive bias consistency
     5390      indicator and the Melbourne-Wübbena (MW) consistency indicator (new RTCM-SSR has none), per signal the integer
     5391      and the wide-lane integer indicator (with percentages if they change), and the satellites with changes of the
     5392      discontinuity counter (PPP-AR users have to reset their ambiguities),</li>
     5393    <li>the Melbourne-Wübbena check of the phase biases: the indicators only declare the integer property; it is
     5394      checked with the biases themselves. Every 300 seconds, the MW combination of the phase and code biases of the
     5395      reference signals is formed per satellite in wide-lane cycles (phase bias of another signal of the frequency if
     5396      the reference signal has none, e.g. L1C instead of L1W), and differenced with that of each stream of another
     5397      AC. If the wide-lane biases of both streams are consistent (integer wide-lane ambiguities can be fixed with
     5398      either), the fractional parts of these differences are the same for all satellites of a system: a common
     5399      offset, e.g. from different code datums. The RMS of the fractional parts about this offset is reported per
     5400      other stream (at least 4 satellites; GLONASS not checked, FDMA): below 0.10 cycles consistent, above 0.15
     5401      cycles not consistent (0.29 cycles for random fractional parts), with the satellites above 0.15 cycles. This
     5402      checks what a combination of the phase biases (PPP-AR) of several ACs requires, e.g.
     5403      <pre>
     5404  G phase bias format and indicators: IGS-SSR, dispersive bias consistency indicator 0, MW consistency indicator 1
     5405  G phase bias integer / wide-lane integer indicators: 1C integer 1 / wide-lane 2; 2W integer 1 / wide-lane 2; 5I integer 1 / wide-lane 2
     5406  G MW check: phase biases L1C L2W with the code biases C1W C2W of the reference signals, wide-lane cycles (0.862 m)
     5407  G MW biases vs. SSRA00CHC1 (1 epochs, 30 satellites): fractional parts about the common offset -0.12 cycles: RMS 0.265 cycles, NOT consistent
     5408  G MW biases vs. SSRA00NRC0 (1 epochs, 30 satellites): fractional parts about the common offset -0.12 cycles: RMS 0.139 cycles, not clearly consistent
     5409  G MW biases vs. SSRA00NRC0: satellites with RMS &gt; 0.15 cycles: G01(0.21) G04(0.26) G11(0.24) G21(0.15) G23(0.29) G28(0.30) G29(0.17)
     5410</pre></li>
    53895411    <li>the availability of the satellite yaw angle with the phase biases: 'not provided', or the number of
    53905412      satellites and the percentage of phase bias records with yaw angle, listing satellites without or only partly
    … …  
    83398361  10 seconds). The solution is regularized by minimal constraints: per epoch the combined clocks sum up to zero, the
    83408362  satellite specific offsets of all ACs providing a satellite sum up to zero, and the satellite specific offsets of
    8341   each AC sum up to zero. The last condition makes the AC specific offset the mean clock difference of the AC; without
     8363  each AC sum up to zero (offsets of ACs differing persistently from the other ACs are left out, see 'Robust clock
     8364  datum' below). The last condition makes the AC specific offset the mean clock difference of the AC; without
    83428365  it, the combined clock of a satellite not provided by all ACs would depend on the history of the filter (e.g. after a
    83438366  restart). In case of a change of the 'SSR Provider ID',
    … …  
    85278550    AC is used for the clock datum again, which is reported in both logfiles. A single passing check is not
    85288551    sufficient, because the offsets are reset with the exclusion and not yet converged; a borderline AC would
    8529     otherwise change the clock datum every few minutes. A smaller weight factor of the AC would not
     8552    otherwise change the clock datum every few minutes. Since each change of the datum shifts the combined clocks of
     8553    the system, the number of passing checks required is doubled each time the same AC is excluded again (2, 4, 8 and
     8554    at most 16 hours). A smaller weight factor of the AC would not
    85308555    help instead: the weights act on the epoch-wise clock variations, while the satellite-specific constant part is
    85318556    fixed by the zero-sum conditions, independent of the weights.
    … …  
    85378562  </p>
    85388563  <p>
    8539     <b>Persistent outliers:</b>
     8564    <b>Robust clock datum and persistent outliers:</b>
    85408565    A clock correction whose residual exceeds 'Maximal Residuum' is rejected as outlier. If an AC's clock for a single
    8541     satellite differs persistently from those of the other ACs, its satellite offset cannot absorb the difference, since
    8542     the offsets of all ACs of a satellite sum up to zero, and the correction is rejected again in every epoch. This
    8543     protects the combined clock, but it is worth to be reported to the AC. Therefore, once per hour and system, BNC
    8544     reports all AC satellites rejected in at least half of their epochs of the last hour (minimum 10 epochs), in the
    8545     combination logfile and in the BNC logfile, e.g.
     8566    satellite differs persistently from those of the other ACs, its satellite offset cannot absorb the difference as
     8567    long as it is part of the zero-sum condition of the satellite: depending on the history of the filter, the
     8568    difference either shifts the combined clock (e.g. accepted at the start of the filter) or the correction is
     8569    rejected again in every epoch. Therefore, for each satellite, BNC compares the satellite specific clock difference
     8570    of each AC referred to the reference signals (AC clock minus AC specific offset minus combined clock, rejected
     8571    corrections included) with the median of the ACs. A satellite offset rejected as outlier for 60 seconds is taken
     8572    out of the zero-sum conditions (of the satellite and of the AC) and re-initialized: it then absorbs the constant
     8573    difference, and the AC only contributes to the variation of the combined clock. This applies to any number of ACs,
     8574    as long as another offset of the satellite stays in the conditions; a rejected offset in the conditions would
     8575    otherwise shift the combined clock. For satellites provided by at least three such ACs, an offset deviating by more
     8576    than 'Maximal Residuum' (at most 0.5 m) for 60 seconds leaves the conditions as well: the offsets with the largest
     8577    deviations leave first, and a majority of the ACs of the satellite (at least two) stays in the conditions, e.g. if
     8578    the ACs form two groups with a clock difference. An offset rejoins the zero-sum conditions after its deviation has
     8579    stayed below half of this threshold for 10 minutes without being rejected. The zero-sum condition of a satellite
     8580    refers to the values of its offsets only if all offsets in the datum of the satellite (provided within the last
     8581    hour) are present in the epoch. During a data gap of such an AC, or if its correction is rejected, the sum of the
     8582    present offsets is kept instead, so that the gap does not shift the combined clock; offsets out of the datum
     8583    never define it, even if no offset of the datum is present. Both changes are reported in the combination logfile and in the BNC
     8584    logfile, e.g.
     8585  </p>
     8586  <pre>
     8587bncComb: SHA G26 out of the clock datum: deviation -3.33 m from the median of 6 ACs, rejected as outlier for 60 s
     8588bncComb: CAS01 G13 back in the clock datum: deviation 0.08 m from the median of 3 ACs (below 0.10 m for 600 s)
     8589</pre>
     8590  <p>
     8591    Corrections can still be rejected persistently, e.g. for satellites provided by fewer than three ACs or for noisy
     8592    clocks. This protects the combined clock, but it is worth to be reported to the AC. Therefore, once per hour and
     8593    system, BNC reports all AC satellites rejected in at least half of their epochs of the last hour (minimum 10
     8594    epochs) and the satellite offsets out of the clock datum, in the combination logfile and in the BNC logfile, e.g.
    85468595  </p>
    85478596  <pre>
    85488597bncComb: SHA G26 rejected as outlier in 99% of 720 epochs in the last hour (median residual 1.08 m)
     8598bncComb: G satellite offsets out of the clock datum: SHA G26 (-3.33 m, since 2026-10-02 11:42:15)
    85498599</pre>
    85508600  <p>
Note: See TracChangeset for help on using the changeset viewer.