rank deficient when computing covariance of ppp_estimator

[222233tommy]

Hi, Dr. chi, thanks for your amazing work. I try to test the PPP-AR and meet the rank deficient problem of the jacobian matrix. I guess the cause may be related to the constant parameter while the sliding window is not full (jacobian rank = jacobian column -1). It seems that the problem will not occur on GPS-only measurements. Could you please help me with that? Thanks a lot!

[chichengcn]

We did not extensively test the PPP-AR module. Could you provide the configuration file and the number of datasets?

[222233tommy]

Configuration file:

GICI offline pseudo-real-time PPP estimation. stream: streamers:

• streamer: tag: str_gnss_rov_file output_tags: [fmt_gnss_rov] type: file path: /home/aae/projects/gici-open/data/1.2/gnss_rover.bin
• streamer: tag: str_gnss_eph output_tags: [fmt_gnss_eph] type: file path: /home/aae/projects/gici-open/data/1.2/gnss_ephemeris.bin
• streamer: tag: str_gnss_ssr output_tags: [fmt_gnss_ssr] type: file path: /home/aae/projects/gici-open/data/1.2/gnss_ssr.bin
• streamer: tag: str_gnss_atx output_tags: [fmt_gnss_atx] type: file path: /home/aae/projects/gici-open/option/igs14.atx enable_time_tag: false
• streamer: tag: str_dcb_file output_tags: [fmt_dcb_file] type: file path: /home/aae/projects/gici-open/option/CAS0MGXRAP_20221580000_01D_01D_DCB.BSX enable_time_tag: false

• streamer: tag: str_solution_file input_tags: [fmt_solution_file] type: file path: /home/aae/projects/gici-open/data/ppp_solution.txt enable_time_tag: false

  formators:

• formator: io: input tag: fmt_gnss_rov type: gnss-raw sub_type: tersus
• formator: io: input tag: fmt_gnss_eph type: gnss-rtcm-3 start_time: 2023.03.27
• formator: io: input tag: fmt_gnss_ssr type: gnss-rtcm-3 start_time: 2023.03.27
• formator: io: input tag: fmt_gnss_atx type: atx-file
• formator: io: input tag: fmt_dcb_file type: dcb-file

• formator: io: output tag: fmt_solution_file type: nmea use_esa: true talker_id: GP

  replay: enable: false speed: 1.0 start_offset: 0.0

estimate:

• estimator: tag: est_ppp type: ppp input_tags: [fmt_gnss_rov, fmt_gnss_eph, fmt_gnss_ssr, fmt_dcb_file, fmt_gnss_atx] fmt_gnss_rov_roles: [rover] fmt_gnss_eph_roles: [ephemeris] fmt_gnss_ssr_roles: [ssr_ephemeris] fmt_dcb_file_roles: [code_bias] fmt_gnss_atx_roles: [phase_center] output_tags: [fmt_solution_file] output_align_tag: fmt_gnss_rov compute_covariance: true ppp_options: max_window_length: 3 use_ambiguity_resolution: true estimate_velocity: false gnss_estimator_base_options: use_outlier_rejection: true max_pesudorange_error: 4.0 max_phaserange_error: 0.03 gnss_error_parameter: code_to_phase_ratio: 1000.0 gnss_common: min_elevation: 12.0 min_SNR: [30.0, 30.0]

  system_exclude: [R]

  estimator_base_options: max_iteration: 5 num_threads: 2 max_solver_time: 0.02 verbose_output: true solver_type: dense_schur trust_region_strategy_type: dogleg log_intermediate_data: false log_intermediate_data_directory: /home/aae/projects/gici-open/data

logging: enable: true min_log_level: 0 log_to_stderr: true file_directory: /home/aae/projects/gici-open/data

dataset: 1.2

[222233tommy]

There is another issue when I test the dual-frequency PPP with BDS. I guess it may be related to the rank-deficient problem.

Error: W20231108 21:33:17.284116 7313 gnss_estimator_base.cpp:774] IFB for code 27 for system C has not been initialized!

I think the IFB size should be zero for dual-frequency PPP, but the result shows that it only optimizes with single-frequency observation since it fails to add all the dual-frequency observations.

[chichengcn]

Thanks for pointing out. We have fixed this BUG and uploaded the codes. This BUG is caused by improper strategy of adding initial constraints for ambiguity and ionosphere parameters. Please note that phase bias product is demanded for PPP-AR algorithm, but we did not provide this product in our dataset. Forcely applying PPP-AR algorithm without phase bias corrections may cause large errors in solutions.

As for the IFB issue, code 27 (BDS B2bI) is not one of the base frequencies (B1I and B3I), however our dual-frequency PPP algorithm was designed for the base frequencies. So perhaps this BUG is foreseeable. But anyway, you can provide the configuration file and the number of dataset. We will focus on this issue in future development.