好久没有更新了，汇报下最近的进度吧。
First repeat the data processing steps:
1. Collect rover's ubx data and base's rtcm data;
2. Use rtkconv to convert to rinex format;
3. Select configuration parameters, put them in rtkpost for processing, and use rtkplot to view.
The test receiver is a consumer-grade receiver of a certain brand—single-frequency dual-mode GPS/BDS. The test environment is a strong signal on the roof, and the motion state is static. The data used is placed in Baidu cloud, link : extraction code: y71q.
Step1 : Let's start my first processing, config and result are shown in the following figure: 
From the figure above, we can see that the ambiguity fixed rate is 25.4% this time, and the internal and external accuracy of STD and RMS are not very ideal. In response to the above two questions, I first eliminated GEO satellites. Why do I eliminate GEO satellites? To be sure, the quality of the observation data is not good. Use plot to view the number of observation epochs of specific satellite data in the observation arc. This method is relatively rough. To be more specific, you can print out the satellite observations in the rtklib source code, and use matlab to plot. The epoch of the problem is convenient for checking the correlation between the epochs before and after (for example, I found that there will be a certain epoch satellite data loss in the rtcm of Chihiro, but the signal-to-noise ratio of the epoch before and after is very high, 40+).
Printed the carrier observations and Doppler of satellites G27 and satellites C01\C02. The reason for this is based on G27. Let’s see the difference between GEO and G27. I diff the carrier observation epoch and compare it with The Doppler (inverted sign) of the current epoch is the difference, and the formula is (CP2-CP1)+D2. In addition, a high-order difference method (3 times or 4 times is acceptable) was also done on C01\C02, which verified that the GEO satellite is not effective. The results of the difference between G27 and C01 are as follows:
According to the above figure, the amplitude of C01 is greater than that of G27, and the difference changes more frequently. C02 is worse than C01. Due to space limitations, I didn't put it up.
Step2 : With the above analysis, I changed the configuration and disabled the GEO satellite. The configuration and result diagram are as follows:
It can be seen from the above figure that the fixed rate of ambiguity in the observation arc is 94.1%, which is a great improvement over step 1, but there are still floating point solutions and flying points (fixed solutions but incorrect solution results). By observing the ratio value of the result, you will find that every time you go to the star, there will be a sudden change in the ratio (ratio value of the previous epoch>3.0, ratio of the current epoch<3.0). This is a phenomenon we have discovered. We are going to To verify this phenomenon, by modifying the rtk_t structure and adding codes in relpos, adding the changes in the common visual star and the available star of the previous epoch and the current epoch respectively, the specific codes are as follows:
In the above 4 figures, Figure 3 must be memset when saving the change of the common visual star and the available star of the current epoch, because otherwise, if the current epoch is one more star than the previous epoch, it will cause the Star repeats are redundant; Figure 4 shows only the changes in available stars during the intermediate analysis.
When we add the code according to the above figure, it will strongly prove that the phenomenon we found is correct. We need to find out the epoch before and after the problem epoch by epoch in Figure 4, and see what changes in the satellite data in question, we will find that the LLI is set up when the satellite is fixed after the drop, then it will cause the udbias The bias is reset, and when we re-track the satellite, it is best to let it run for 30 epochs first, then step 3 will follow.
Step3 : I want to set the minimum lock epoch 30 in the configuration and see how the result is. The configuration and result are as shown in the figure below:
Then we can find from the result that the ambiguity fixed rate has an increase of 99.7%, which is higher than step2 Improve, but we don’t want to be satisfied with this, don’t cover up the problem. There are still floating point solutions and flying points. After debugging, it is found that it is caused by the star. A check will find that the Beidou accounted for the majority. Do not at this time There is a preconceived notion that Beidou is inferior to GPS. We must seek truth from facts to find out what is wrong with the epoch satellite in question. We will find that the SNR is not high. Then I set 35 first, which is an empirical value. This is our step4.
Step4: Careful students should notice that I have configured the snr mask in step1, and I set both rover and base to 35, then the configuration and results are shown in the figure below:
From the figure above, the ambiguity fixed rate has reached 100%. But there are still flying spots, which is extremely unsightly, right? Check again and find that the signal-to-noise ratio of the satellite (on the satellite) in question always fluctuates around 35. The signal-to-noise ratio and the altitude angle are strongly correlated. After all, we The random model can use the signal-to-noise ratio and the height angle to determine the weights beforehand, so I raise the snr threshold below the height angle of 35° to 40, and then let’s see how the result is.
Step5 : The previous step has been analyzed, just put the configuration and result graph directly.
As can be seen from the above figure, the positioning accuracy is very nice, and the ambiguity fixed rate is nice.
Summary : In the process of debugging, I tried to use the partial ambiguity algorithm of rtklibexplorer in the ambiguity resolution part, and found that it only covered up the problem, and the epoch was delayed by nb occurrences (the number of ambiguities in the epoch where the problem occurred) ), I went back to the most essential question, the quality of observation. In fact, the most important thing we need to do is to ensure that the quality of the injected data is ok.
If there is time, the next section may write about how to detect and repair the clock jump in the parameter field when there is a problem with the carrier measurement value of the chip found during debugging.