CLAS Positioning with Septentrio mosaic-G5¶

This guide describes how to achieve centimetre-level CLAS PPP-RTK positioning using MRTKLIB with a Septentrio mosaic-G5 P3 receiver. We present two approaches: using the receiver's built-in RTK engine with VRS corrections generated by mrtk cssr2rtcm3, and using MRTKLIB's own PPP-RTK engine via mrtk run.

Overview¶

mosaic-G5 P3¶

The mosaic-G5 P3 is one of the few commercially available receivers that can directly track the QZSS L6 band and output raw L6D data. In this tutorial we used the mosaic-go G5 P3 evaluation kit (mosaic-go G5), which provides a convenient out-of-the-box setup for evaluating the mosaic-G5 P3 module.

Supported constellations and bands:

Two Approaches: VRS vs. MRTKLIB Engine¶

Approach 1 — VRS (mrtk cssr2rtcm3)¶

In this approach, MRTKLIB converts CLAS corrections into standard RTCM3 MSM7 messages and feeds them back to the mosaic-G5, which then computes a VRS-based RTK position using its built-in engine.

1. mrtk relay bridges the serial connection to the mosaic-G5, forwarding SBF data to mrtk cssr2rtcm3 and returning RTCM3 corrections
2. mrtk cssr2rtcm3 decodes L6D CSSR, computes OSR via clas_ssr2osr(), and encodes RTCM3 MSM7
3. The mosaic-G5 receives the RTCM3 corrections and computes a VRS-RTK position
4. mrtk relay also outputs the positioning result (SBF/NMEA)

flowchart LR
    A["mosaic-G5 P3"] -- "Serial(SBF)" --> B["mrtk relay"]
    subgraph "Host (PC / SBC)"
    B -- "TCP/IP(SBF)" --> C["mrtk cssr2rtcm3"]
    B -- "SBF" --> D@{ shape: stadium, label: "Output" }
    end
    C -- "Serial(RTCM3)" --> A

Approach 2 — MRTKLIB Engine (mrtk run)¶

In this approach, MRTKLIB performs the PPP-RTK positioning directly. The mosaic-G5 serves only as an observation and correction source; all positioning computation happens on the host.

1. mrtk run reads the SBF stream from the mosaic-G5 (observations, L6D corrections, and broadcast NAV)
2. MRTKLIB decodes CLAS CSSR and computes the PPP-RTK position internally
3. The positioning result is output directly from mrtk run

flowchart LR
    A["mosaic-G5 P3"] -- "Serial (SBF)" --> B["mrtk run"]
    subgraph "Host (PC / SBC)"
    B -- "Positioning Result" --> C@{ shape: stadium, label: "Output" }
    end

Equipment¶

Setting Up the Receiver¶

RxTools is a GNSS receiver control and analysis software suite by Septentrio. Configure the mosaic-G5 using RxTools (the mosaic-G5 module does not have a Web UI).

1. Download and install RxTools: Download the latest version of RxTools from the Septentrio website and install it on your computer.

2. Connect the receiver: Connect the mosaic-go G5 via USB. The board will be powered over USB and the LED will turn on.

3. Launch RxControl: Open RxControl and configure the connection (first time only).

   • Select Serial Connection > Create New...
   
   • Choose the USB COM port, set a Connection Name, and click Finish.
   
   • RxControl will launch after the connection is established.
   

4. Set SBF output: Go to Communication > Output Settings > SBF Output and configure Stream 1 as follows:

   • Ports: USB2
   • Off: unchecked
   • Support: checked
   • Interval: 1 sec (CLAS corrections update every 5 s and PVT can be paced at 1 Hz; 1 Hz is sufficient for static and most kinematic use cases. For high-rate dynamics, set 100 msec instead.)

   Click Apply, then OK to close.

   
   Required SBF blocks (reference)

   Selecting Support enables a curated set of blocks that includes everything needed by mrtk cssr2rtcm3 and mrtk run. The principal blocks consumed are listed below.

   SBF Block	ID	Purpose
   MeasEpoch	4027	Raw GNSS observations (required)
   QZSRawL6D	4270	QZSS L6D raw data (CLAS CSSR)
   GPSNav	4017	GPS broadcast ephemeris
   GALNav	4022	Galileo broadcast ephemeris
   QZSNav	4030	QZSS broadcast ephemeris
   GLONav	4004	GLONASS broadcast ephemeris (optional)
   BDSNav	4081	BeiDou broadcast ephemeris (optional)
   PVTGeodetic	4007	Receiver position (latched by mrtk cssr2rtcm3 as the VRS reference)

   Note

   Block IDs are listed for reference. Septentrio occasionally renumbers blocks across firmware revisions; consult the SBF Reference Guide that ships with your firmware version if you need exact IDs. Enabling Support on Stream 1 is the recommended way to ensure all required blocks are output regardless of firmware version.

5. Enable L6 tracking: Go to Navigation > Advanced User Settings > Tracking, select the Signal Tracking tab, and enable QZSSL6. Click Apply, then OK to close.

   L6 tracking is disabled by default

   The mosaic-G5 does not track QZSS L6 signals out of the box. Without this step, the SBF stream will contain no QZSRawL6D blocks and cssr2rtcm3 will receive no CLAS data.

   

6. Set Positioning Mode: Go to Navigation > Positioning Mode:

   • In the PVT Mode tab:
     • Enable RTK Float solutions: In the PVT Mode panel, set Rover mode to all (this allows the receiver to output Float solutions in addition to Fix).
     • Change Solution Selectivity: In the Solution Selectivity panel, set Level to Loose (this relaxes the strictness with which the PVT engine filters satellite signals and transitions between PVT modes).
   • In the PPP and Differential Corrections tab:
     • Change Max Age of Differential Corrections: In the Max Age of Differential Corrections panel, set Maximum age of RTK data to 60.0 s (aligned with the CLAS correction update interval).

   Click Apply, then OK to close.

7. Configure the RTCM3 input port (Approach 1 / VRS only): Identify the COM port that the host will use to send RTCM3 corrections to the receiver (e.g. COM1, COM2, USB1), and set its Input Type explicitly to RTCMv3.

   Via RxControl GUI: Communication > Input/Output Selection, choose the target port, set Input Type to RTCMv3, click Apply.

   Via ASCII commands (USB1 terminal):

   Text Only

   setDataInOut, COM1, RTCMv3
   

   Verify with gdio:

   Text Only

   DataInOut, COM1, RTCMv3, SBF+NMEA+ASCIIDisplay, (on)
   

   auto does not reliably detect RTCMv3

   The firmware reference states that Input Type = auto auto-detects RTCMv3, but at least on firmware 20250611b this is not the case in practice — the receiver silently ignores incoming RTCM3 and stays in SPP mode forever, even when mrtk cssr2rtcm3 is producing valid corrections. Always set the InputType explicitly to RTCMv3 for the port that receives corrections.

   Diagnostic symptom: PVT stays at mode = 1 (SPP / Stand-alone) indefinitely, with no transition to DGPS (mode = 2) / Float (mode = 5) / Fix (mode = 4), despite mrtk cssr2rtcm3 running and emitting RTCM3 bytes. At a clear-sky stationary point a 24 h SPP-only session can still cluster within ±15 cm — looks plausible at a glance, so check the PVT mode field in PVTGeodetic, not just the scatter plot.

   Also confirm that the COM port baud rate matches the host side (mrtk cssr2rtcm3 -out serial://...:115200 ⇒ setComSettings, COM1, baud115200):

   Text Only

   gcs
   

8. Save configuration to the receiver: Go to File > Copy Configuration and set:

   Field	Value
   Source	Current
   Target	Boot

   Click Apply, then OK to close.

   Equivalent ASCII command:

   Text Only

   exeCopyConfigFile, Current, Boot
   

Running — Approach 1: VRS¶

The VRS approach requires two processes running simultaneously: mrtk relay to bridge the serial connection, and mrtk cssr2rtcm3 to convert CLAS corrections to RTCM3.

Step 1: Start mrtk relay¶

mrtk relay connects to the mosaic-G5 serial port, exposes the SBF stream on a TCP server port, and relays RTCM3 corrections back to the receiver using the -b (relay-back) option.

# Terminal 1: Bridge serial <-> TCP
mrtk relay \
  -in serial://tty.usbmodem01000124303:115200#sbf \
  -out tcpsvr://:9000#sbf \
  -out file://mosaic-g5.sbf#sbf \

• -in serial://tty.usbmodem01000124303:115200#sbf — read SBF from the mosaic-G5 USB2 serial port
• -out tcpsvr://:9000#sbf — serve the SBF stream on TCP port 9000 (for mrtk cssr2rtcm3)
• -out file://mosaic-g5.sbf#sbf — log raw SBF data to file (optional, for post-analysis)

Step 2: Start mrtk cssr2rtcm3¶

mrtk cssr2rtcm3 connects to the relay's TCP port, decodes CLAS CSSR from the SBF stream, and sends RTCM3 MSM7 corrections back through the same TCP connection.

# Terminal 2: CSSR -> RTCM3 conversion
mrtk cssr2rtcm3 \
  -k conf/cssr2rtcm3.toml \
  -in sbf://tcpcli://localhost:9000 \
  -out serial://cu.usbmodem01000124301

• -in sbf://tcpcli://localhost:9000 — connect to relay and read SBF (single-stream mode: L6D, NAV, and PVT are all extracted from the same SBF stream)
• -out serial://cu.usbmodem01000124301 — send RTCM3 MSM7 corrections to the mosaic-G5 via USB1

Once CLAS corrections converge (typically 1–2 minutes after startup), the mosaic-G5 receives the RTCM3 corrections and performs VRS-RTK positioning internally. The positioning result is available in the SBF output from the receiver (forwarded by mrtk relay).

Monitoring with sbf_plot¶

sbf_plot.py connects to the relay's TCP port and plots the receiver's position and fix quality in real time.

# First time only: set up the Python virtual environment
# cd scripts && uv sync && source .venv/bin/activate && cd ..

# Terminal 3: Real-time position plot
python scripts/plotting/sbf_plot.py --port 9000

• Points are colored by fix quality: green = RTK Fix, orange = RTK Float, red = SPP
• The first received position is used as the reference origin (ENU in meters)
• To use an explicit reference: --ref 35.3231,139.5221
• Fix rate and satellite count are shown in the title bar

Debug Trace¶

Add -d 3 to mrtk cssr2rtcm3 for detailed processing logs:

mrtk cssr2rtcm3 \
  -k conf/cssr2rtcm3.toml \
  -in sbf://tcpcli://localhost:9000 \
  -out serial://cu.usbmodem01000124301 \
  -d 3

Running — Approach 2: MRTKLIB Engine¶

Instead of converting CLAS to RTCM3 and relying on the receiver's RTK engine, you can run MRTKLIB's own CLAS PPP-RTK engine directly.

Step 1: Start mrtk relay¶

Same as Approach 1:

mrtk relay \
  -in serial://tty.usbmodem01000124303:115200#sbf \
  -out tcpsvr://:9000#sbf

Step 2: Start mrtk run¶

mrtk run -k conf/claslib/rtkrcv_mosaic_g5.toml

The MRTKLIB engine reads the SBF stream from the relay, automatically extracts L6D (CLAS) data from QZSRawL6D blocks, and performs PPP-RTK positioning. The NMEA solution is written to ./clas_rt.nmea by default.

To output the solution to a TCP server (e.g., for downstream applications):

mrtk run -k conf/claslib/rtkrcv_mosaic_g5.toml -out tcpsvr://:9002

Configuration¶

The default configuration conf/cssr2rtcm3.toml is suitable for most use cases:

# MRTKLIB Configuration (TOML v1.0.0)
# cssr2rtcm3: CSSR to RTCM3 MSM converter

[positioning]
mode = "ssr2osr"
elevation_mask = 0.0
satellite_ephemeris = "brdc+ssrapc"
systems = ["GPS", "Galileo", "QZSS"]

[positioning.corrections]
tidal_correction = "solid+otl-clasgrid+pole"
phase_windup = "on"
exclude_eclipse = true
receiver_antenna = false

[positioning.atmosphere]
ionosphere = "est-adaptive"
troposphere = "off"

[files]
# CLAS compact network grid definition (shared across all CLAS configs).
# Path is relative to the working directory where mrtk is invoked.
cssr_grid = "./tests/data/claslib/clas_grid.def"

# Signal code remapping: CLAS code → receiver tracking code.
# Key format: {G|R|E|J|C}{RINEX obs code}  (e.g. G2X = GPS L2X)
# Value: target RINEX obs code string       (e.g. "2W" = L2W)
#
# Uncomment and adjust for your receiver. Example for Septentrio mosaic-G5:
[signal_remap]
G2X = "2W" # GPS     L2C(M+L) → L2W  (needed: CLAS smode[1]=2X for most GPS sats)
#E1X = "1C" # Galileo E1(B+C)  → E1C
#E5X = "5Q" # Galileo E5a(I+Q) → E5aQ
J2X = "2L" # QZS     L2C(M+L) → L2L

# cssr2rtcm3-specific settings
[cssr2rtcm3]
# MSM message type: 4 (compact), 5 (+ Doppler), 7 (full resolution + Doppler)
msm_type = 7
# Constellations to include in RTCM3 MSM output (default: all CLAS systems)
# Recognised: GPS, Galileo, QZSS, GLONASS
systems = ["GPS", "Galileo", "QZSS"]
# SNR model for VRS pseudo-observations:
#   0 or omitted = elevation-dependent (25 + 20·sin(el) dB-Hz)
#   >0           = fixed value in dB-Hz (mosaic-CLAS uses 50)
snr_fixed = 50.0
# L6D satellite auto-selection: minimum elevation angle (degrees).
# cssr2rtcm3 watches all QZS satellites broadcasting L6D and selects the one
# with the highest elevation. When the selected satellite drops below this
# threshold, stops transmitting for >30 s, or another satellite reaches a
# clearly higher elevation, it fails over to the next best candidate.
# Default 10 degrees keeps GEO satellites (PRN >= 199) eligible.
l6d_elmin = 10.0

Key parameters:

Test Results¶

Test Configuration¶

The latest long-term static test was conducted under the following setup.

Block Diagram

flowchart LR
    A["mosaic-G5 P3"] -- "Serial/USB2" --> B["mrtk relay<br/>(SBF)"]
    subgraph "Host (PC / SBC)"
    B -- "SBF" --> C["mrtk cssr2rtcm3<br/>(RTCM Conversion)"]
    B --> E[("Log (SBF)")]
    end
    C -- "Serial/USB1 (RTCM3)" --> A

Long-term stability (24 h)¶

The first 24-hour continuous static test of the mosaic-G5 + mrtk cssr2rtcm3 chain produced the following result.

Solution mode breakdown (87,839 epochs, no data loss):

Hourly Fix rate (GPST hour-of-day):

Highlights:

• 24 h continuous operation with no crashes and no output stalls, confirming long-running daemon stability after the recent actualdist fixes (d028b0c, 7fb497f) and the eph_prev correction-continuity fix (3a8cc51).
• DGPS dips eliminated: the 10-minute periodic DGPS dip seen in earlier builds did not reappear (DGPS reached 0 epochs over 24 h).
• 16 of 24 hours achieved ≥ 90 % Fix rate (8 hours at ≥ 99 %).
• Two windows account for almost all of the Float share: h ≈ 04–05 GPST (~2 h, 23–46 % Fix) and h ≈ 14–18 GPST (~5 h, 0–57 % Fix; one full hour at 0 % Fix). In both windows the tracked-satellite count (ns ≈ 12) and DGPS correction age (~2 s) were healthy, so the cause is neither observation starvation nor a correction-link outage — investigation is ongoing (see Known Issues and #98).

Reference comparison: mosaic-CLAS¶

For context, a 24-hour static run of a Septentrio mosaic-CLAS receiver at a different location in the Kantō region on the same day produced the following result.

The mosaic-CLAS reference is at a different site and uses different hardware, so this is not a like-for-like comparison; nevertheless it indicates that there is still meaningful headroom in the cssr2rtcm3 → external RTK engine path relative to a receiver that consumes CLAS L6D natively. See #98 for the investigation tracking this gap.

Known Issues¶

Fix-rate gap vs. mosaic-CLAS reference — #98¶

In the 24-hour static test summarised above, the cssr2rtcm3 → mosaic-G5 chain achieved a 72 % Fix rate, versus 98 % on a mosaic-CLAS reference at a different Kantō site on the same day — a remaining headroom of roughly 20 %. Two windows (h ≈ 04–05 and h ≈ 14–18 GPST) showed sustained drops in Fix rate while the tracked-satellite count (≈ 12) and the DGPS correction age (≈ 2 s) remained healthy. The corrections were arriving on time and enough satellites were tracked, yet integer ambiguities did not consolidate to Fix in those windows.

Possible causes under investigation include cycle-slip cascades on specific PRNs, ionospheric activity around dawn / late afternoon, the elevation profile of the active L6D satellite, and CLAS network-region transitions. Tracking issue: #98.

Vertical-component dispersion (~30 s sawtooth) — #97¶

When solutions are plotted, the vertical (Up) component shows a roughly 30-second sawtooth pattern of a few centimetres peak-to-peak that is not present in reference implementations processing the same data. Horizontal accuracy is unaffected.

The cause is cssr2rtcm3's extrapolation of CLAS Compact SSR orbit/clock corrections between snapshots: rate terms are currently held at zero, so each ~30 s SSR update introduces a small step in the corrected satellite position/clock that propagates into the vertical component. Estimating the rate by numerical differentiation between consecutive snapshots is being investigated as a remediation.

This is a property of the correction extrapolation logic, not a receiver-specific issue; any RTK engine consuming this RTCM3 stream is expected to show the same pattern.