Installation, operation and technical reference for the PathGridData Generator (CPSQDSIM) – a site-specific quasi-deterministic channel model simulator developed at the Radio Signal Processing Laboratory, Niigata University.
As wireless technology evolves, new communication systems must keep pace with the demands of autonomous driving, drones, IoT and Beyond 5G. The frequency band below 6 GHz is becoming increasingly congested, so improving spectrum utilisation and increasing signal bandwidth are necessary to raise transmission speed and data volume. This has led to renewed interest in millimeter-wave communication, which uses the 30–300 GHz band and can enable gigabit transmission rates. As the number of millimeter-wave devices grows, systems must be developed quickly and designed to respond to a wide variety of scenarios, which in turn demands accurate evaluation of the propagation channel.
Because of its short wavelength, the millimeter-wave band has quasi-optical properties and strong rectilinearity, and it can carry large amounts of information – but free-space propagation loss and attenuation due to transmission, diffraction and scattering are significant drawbacks. Where a line-of-sight (LoS) path exists, specular reflection components are the primary propagation paths alongside the LoS. Geometric optical methods (ray tracing) can accurately predict the main propagation paths in a simple environment, and reproducing an environment-specific model more accurately improves prediction accuracy. In complex environments, however, prediction becomes increasingly challenging as processing time grows prohibitive. In addition, the short wavelength means that surface irregularities of reflective materials play a significant role in causing diffuse scattering, which is difficult to simulate accurately.
In this situation a quasi-deterministic (Q-D) model is used to represent environment-specific propagation paths. It combines deterministic components – LoS and one- to two-bounce reflections obtained by simple ray tracing – with probabilistic components such as random scatterers and diffuse scattering clusters associated with specular reflection, modelled through statistical distributions. A pure Q-D model is limited in how accurately it represents an individual environment because it relies on statistical distributions derived from measurements in typical scenarios. A hybrid channel model combining deterministic and probabilistic components is therefore necessary, and developing one requires a large volume of channel data and a clear understanding of the propagation characteristics.
To gather environment-specific channel data in an urban macrocell setting, a channel sounder capable of measuring bi-angular propagation channels in both the 24 and 60 GHz bands was used, allowing simultaneous measurement in both bands. Elementary waves were extracted from the collected data with an ultra-high-resolution channel estimation algorithm and classified into clusters by similar delay and angle. Combining the 24 GHz and power-compensated 60 GHz multipaths made it possible to cluster both frequencies at the same time, identify the propagation path of each cluster, and obtain statistical characteristics of power difference and relative delay for clusters common to both frequencies.
The hybrid quasi-deterministic propagation model calculates the dominant propagation paths using simple ray tracing for direct waves and large structures, while using probabilistic calculations for small objects such as trees and signboards. The inputs are the rays generated by ray tracing and the statistical parameters (LSPs, SSPs) used to generate propagation paths probabilistically. LSP values matching the actual environment can also be derived by feeding 3D map data into a machine-learning model.
The output, PathGridData, contains path power, delay and departure/arrival angles. It is calculated and stored for each grid point at a fixed spatial interval, and is then handed to the wireless emulator so that the channel response at an arbitrary point can be obtained. The generation flow applies probabilistic compensation to the deterministic clusters computed by ray tracing, based on the cluster delay–power characteristics obtained by measurement, and introduces site-specific LSPs/SSPs to obtain more accurate propagation characteristics for the target environment.
The propagation channel was examined in the vicinity of Yokohama World Porters, a six-storey commercial complex in Kanagawa Prefecture. The base station (BS) was fixed and the mobile station (MS) was moved to 25 locations (#001–#025) along an L-shaped path around the outer perimeter of the facility. Measurements used two frequencies simultaneously and lasted approximately five minutes at each location.
| Item | Value |
|---|---|
| Site | Yokohama World Porters, Kanagawa Prefecture, Japan |
| Frequencies | 24 GHz and 60 GHz (simultaneous) |
| MS positions | 25 locations (#001–#025) along an L-shaped route |
| MS spacing | ≈1 m for #009–#023; ≈10 m for the remaining positions |
| Tx–Rx distance | 24–109 m |
| Antenna heights | Tx (BS) 3 m, Rx (MS) 1.5 m |
| Link conditions | Downlink (BS: Tx, MS: Rx), line-of-sight |
| Scenario | Millimeter-wave urban microcell (UMi) – BS on a streetlight, MS a mobile terminal |
| Duration | ≈5 minutes per location |
After noise removal is applied to the measured DDADPS data, the sub-grid CLEAN algorithm extracts the multipath components (MPCs). The algorithm measures a power image (replica) of an MPC, calculated from the continuous horizontal-plane antenna pattern and the autocorrelation function of the sounding signal (a sinc function for rectangular spectra such as multitone). It then applies successive interference cancellation (SIC) to perform maximum-likelihood estimation, subtracting power sequentially in order of magnitude. This yields finer-resolution MPC estimates that are not influenced by the horizontal-plane antenna pattern or the signal bandwidth. An angular resolution of 0.1° and a delay resolution of 0.01 ns were achieved.
The extracted MPCs are clustered into groups by angle and delay. The composite datasets of MPCs obtained at each frequency are used for clustering. To account for the increased propagation loss caused by the frequency difference, the power of each 60 GHz MPC is increased by 20 log10(58.32 / 24.15) = 7.66 dB. After clustering, the MPCs are separated back into their respective frequency datasets and the 60 GHz powers are restored to their original values. This identifies clusters common to both frequencies as well as clusters unique to one of them. Clustering uses the K-PowerMeans algorithm; the number of clusters K is determined manually by visual inspection so that the results retain physical meaning.
The channel parameters consist of LSPs and SSPs. LSPs represent the spatio-temporal power spread of the clusters, while SSPs represent the spatio-temporal power spread of the MPCs within a cluster.
Computed over the clusters of a link: delay spread (DS), azimuth spread of departure (ASD) and of arrival (ASA), zenith spread of departure (ZSD) and of arrival (ZSA), and the Ricean K-factor. DS is the power-weighted RMS spread of the cluster delays; the angular spreads use the circular (complex exponential) definition.
The intra-cluster delay spread (cDS) and intra-cluster angular spreads (cASD / cASA), derived from the individual MPCs inside each cluster using the same power-weighted definitions applied within a cluster.
Parameters of the exponential-decay model used to calibrate deterministic cluster power, refining the 3GPP map-based model where detailed building orientation, dimension and dielectric data are not available.
Obtaining detailed data on the orientation, dimensions and dielectric properties of every building object in the evaluation area is essential for a truly accurate depiction of deterministic clusters, but this is impractical in complex urban settings with a large separation between transmission and reception points. An exponential decay model is therefore used to calibrate deterministic cluster power. For the power delay characteristics of NLoS clusters it takes the form:
where P0 is the initial path gain, τ0 the delay decay factor, and SF the standard deviation of the cluster shadowing variation, which is treated as a zero-mean Gaussian shadowing term.
The deterministic NLoS cluster power is obtained from the decay model for each BS–MS link. Compensation is applied using the z-score of the power deviation from the model, in three steps:
CPSQDSIM is distributed as a standalone Windows desktop application: a GUI for generating channel data, exploring maps and multipath components, and exporting the data as CSV files. The application and the site data are distributed separately.
| Component | File | Source |
|---|---|---|
| Windows installer (Ver. 1.0) | CPSQDSIM-Installer.zip (7.0 MB) |
radio.eng.niigata-u.ac.jp |
| Site-data package (per frequency) | 28350MHz.zip (33.2 MB)4850MHz.zip (400.8 MB)LSP_SSP_Table.csv (4.1 kB) |
Zenodo – 10.5281/zenodo.22644137 (CC BY 4.0) |
| Site-specific statistical parameters | sscp_mmwave.csv, sscp_sub6ghz.csv |
CPSQDSIM dataset page |
The installer sets up a demo dataset in the input fields automatically, so the installation can be verified before any site data is prepared. The demo consists of a small sample of receiver grids selected near the base station; pressing Run generates the outputs for them. The information bar states that file saving is off, and that the input files can be replaced with your own site data when you are ready.
Demo, Grid selection
enabled, Fixed randomness and File save off – a receiver route
close to the transmitter, at 4.85 GHz, UMa, hBS 31 m, hUT 4 m.
Enable File save and set an output directory when you want the results written to
disk.
The steps below are the shortest route from a fresh installation to an exported dataset. Each step links to the detailed reference later in this manual.
| # | Step | What to do |
|---|---|---|
| 1 | Install | Run the installer, install the MATLAB Runtime, extract the site-data package to a writable folder. §3 |
| 2 | Try the demo run | The installer pre-loads a demo dataset. Press Run to confirm the installation works before preparing your own site data. §3.1 |
| 3 | Prepare ray-tracing input | If path_info.mat and buildings.mat are already available, use them
directly. Otherwise generate them from the Wireless InSite output with the WI Exporter
sub-app. §5 |
| 4 | Choose PathGridData Generator | Select path_info and buildings, then check the displayed frequency,
scenario and Tx/Rx heights. §6.2 |
| 5 | Set model and inputs | Choose the path generation engine and the parameter sources; add matching user PL or LSP CSV files when available. §6.2 |
| 6 | Select receiver grids | Enable Grid selection. For a route, pick points and press connect; for an area, choose a shape, press Draw and close the region of interest. Check the preview and apply it to the main app. §6.2 |
| 7 | Save and run | Enable File save, choose an output folder and project name, and press Run. Choose Fixed randomness to repeat a run with the same setup. Wait for the completion message. §6.2 |
| 8 | Read the results | Inspect the LSP, Individual LSP, SSP & RP and Path tabs. §7 |
| 9 | Export | Open csv File generation and press File Generate to write the emulator CSV files into the project folder. §8 |
In the main application window, the ray-tracing (Wireless InSite) input is supplied by two files:
path_info.mat and buildings.mat. A sub-application, opened with
open WI Exporter, extracts the required information from the Wireless InSite output and
generates those two files.
path_info.mat and buildings.mat already exist they can be used directly in
the main window. The sub-app must be used the first time, whenever a different WI project is used,
or whenever the WI files have changed.
The exporter requires six inputs before it can produce the .mat files:
The name assigned to the project when the Wireless InSite software was run. It can be identified
from the folder name created when the WI project completed – for example Tx38.
The p2m_folder is created under the main project folder when the WI project completes.
It contains many files, of which only a few are needed for import. Confirm that the required
*.paths and *.pl files with the .p2m extension are present; their
exact names vary with the project name and the order in which Wireless InSite was executed.
The city file is generated when the Wireless InSite project completes. Ensure that a
building.city file exists in the folder; if the file has a different name, rename the
*.city file to building.city and select that folder.
The p2m folder may contain several files with the same extension distinguished by a
number in the file name – for example xxxxx.r007.p2m and xxxxx.r008.p2m.
All *.paths and *.pl files with the .p2m extension are required, so
in that example the ID range must be specified as 7 to 8.
Specify the scenario, the frequency and the BS and UT heights exactly as they were set during the Wireless InSite run. The frequency must be given in GHz.
Specify the folder in which the generated .mat files will be stored for further
processing.
If all the information above is correct, the sub-app generates building.mat and
path_info.mat in the specified output folder, together with the final map view of the
exported area.
All user input for a run is provided through the left panel of the main application window: the environment, the scenario and the generation options are set there before generation starts. The application accepts site-specific measured path loss and large-scale parameter values as user input, which gives the most accurate result for the target environment.
Two modes are available:
Upload the two input files – path_info.mat and buildings.mat. Once they are
uploaded, the setup information (frequency, scenario, BS and UT heights) is displayed for
confirmation.
The Parameter Input Files section takes two CSV inputs:
| Box | Status | Contents |
|---|---|---|
| LSP | Optional – LSP and PL data | Site-specific measured path loss and large-scale parameter values for each grid, in a single CSV file. Path loss and LSPs share this one input; there is no separate PL box. If it is left empty, PL and LSPs are generated from the parameter sources selected in (c). |
| SSP | Required – SSP parameter table | The small-scale parameter table used by the SSP generator. This input is mandatory;
the site-data package on Zenodo supplies LSP_SSP_Table.csv for this
purpose. |
Grid ID, PL, DS, ASD, ASA, ZSD, ZSA, K. A different order will be interpreted
incorrectly and will produce an incorrect result.
| Engine | Description |
|---|---|
| NU map-based model (Q-D) | Follows the exponential decay model described in §2.4, with recipe-parameter calibration of the deterministic cluster power. |
| 3GPP map-based model (Q-D) | Follows the channel modelling approach of 3GPP TR 38.901 version 16.1.0 Release 16. |
Select the parameter sources used to generate the PathGridData:
The recipe parameter implements the NU exponential-decay model. It is assigned automatically when the NU path generation engine is selected. If the 3GPP engine is chosen it becomes inactive, because the 3GPP model has no recipe-like statistical parameter.
Check File Save to save the output files, then specify the directory in which they will
be written. The Project Name text box creates a corresponding folder under the output
directory. If it is left blank, the default folder name is untitled – or
untitled_01 and so on when several projects already exist – and the output files are
saved there.
This opens a new window showing the entire map with the buildings, so that a particular route or area can be selected and PathGridData generated only for the chosen grids. A route can be created in two ways:
For an area rather than a route, choose a shape, press Draw and close the region of interest. The selection is shown in the preview area on the right-hand side; check it, then apply it to the main application.
This panel controls how the statistical values are drawn. Select Fixed when the same values must be reproduced across several runs, for example for testing; use Shuffle to draw new random values on each execution.
Run starts the generation of the PathGridData and Exit closes the application window. Wait for the completion message before moving on.
The loader is used to view channel behaviour quickly for each grid in the form of path loss, LSPs and SSPs, when the folder and files generated by CPSQDSIM already exist. Select the Project dir in which the previously generated files are stored and press Run; the information bar reports when loading has completed. In the SSP & RP tab, use the load button to read the summary.
The right panel displays the generated results. Four tabs are available.
Displays the generated LoS/NLoS condition, path loss and LSPs in separate plots. A control box in the lower-right corner of the window sets the minimum and maximum range of the LSP values shown on the map, so that the colour limits can be matched across parameters.
Displays one selected LSP on the map over a larger area. Use the select button below the plot to choose the desired LSP or PL from the drop-down box.
Presents the stochastic small-scale parameter values used to create the SSP values for each grid, together with the general information for the run. Recipe parameter values are also displayed if the NU map-based model was used, or if the recipe was chosen in the left panel during parameter selection. Otherwise the tab is left empty.
If a route was selected in the grid-selection window during generation, this tab shows the current properties in four plot areas – path loss, PDP, angular power spectrum at the Tx (APS@Tx) and at the Rx (APS@Rx) – at each Rx point as the MS moves along the selected route. The playback buttons below the plots (play, next, pause, previous and back) move the current MS position forwards and backwards along the route. On the map (PL plot) the red circle indicates the current Rx location and the yellow circles represent the route. The Display Setting button sets the minimum and maximum display values.
The CSV file generation window creates the files corresponding to the PathGridData that the
emulator requires. Specify the location of the project folder that has already been created, then
click File Generate: the necessary .mat files are loaded, converted to CSV
and saved in the same project folder. Depending on the size of the .mat files this may
take a few minutes. Three files are produced.
| File | Contents |
|---|---|
dcdl_condition_…MHz_Rx…m.csv | Frequency and transmitter position |
dcdl_area_data_…MHz.csv | Receiver coordinates |
dcdl_path_…MHz_Rx…m.csv | Path delays, angles and gains |
dcdl_condition_4850MHz_Rx4m.csv| condition_id | frequency | tx_point_x | tx_point_y | tx_point_z |
|---|---|---|---|---|
| 1 | 28350 | -818.418 | -270.731 | 32.89 |
dcdl_area_data_4850MHz.csv| condition_id | rx_id | rx_point_x | rx_point_y | rx_point_z |
|---|---|---|---|---|
| 1 | 1 | -1414 | -864 | 4 |
| 1 | 2 | -1413 | -864 | 4 |
| 1 | 3 | -1412 | -864 | 4 |
| 1 | 4 | -1411 | -864 | 4 |
dcdl_path_28350MHz_Rx4m.csv| condition_id | rx_id | path_id | path_delay | path_aod | path_zod | path_aoa | path_zoa | path_distance | path_gain |
|---|---|---|---|---|---|---|---|---|---|
| 1 | 13966 | 1 | 6.40E-07 | -132.7 | 98.7 | 47.3 | 81.3 | 10 | -124.7 |
| 1 | 13966 | 2 | 6.40E-07 | -130.8 | 99.1 | 49.7 | 82.5 | 10 | -143.7 |
| 1 | 13966 | 3 | 6.40E-07 | -138.3 | 99.9 | 14.3 | 89.2 | 10 | -168.0 |
| 1 | 13966 | 4 | 6.40E-07 | -150.4 | 98.7 | 30.1 | 86.1 | 10 | -158.5 |
Angles are given in degrees, delay in seconds and path gain in dB.
The generated PathGridData was verified against measured data by comparing the CDFs of the delay and angular spreads over all grid points in the evaluation area. The simulation environment used for the verification was:
| Setting | Value |
|---|---|
| Scenario | UMa (urban macrocell) |
| hBS | 31 m |
| hUT | 4 m |
| LSP | DS, ASA, ASD, ZSA, ZSD – user input (KDDI) |
| K-factor | 3GPP TR 38.901 Table 7.5-6, UMa |
| SSP | 3GPP TR 38.901 Table 7.5-6, UMa |
| PL | User input (KDDI) |
| Frequencies verified | 28350 MHz, 4850 MHz, 2462 MHz, 922 MHz |
For each frequency, the CDFs of the delay and angle spreads obtained with the PathGridData Generator following the site-specific NU map-based model were compared with the measured data, taking all grid points in the entire area.
The CDFs indicate that the results closely match the measured data at all four frequencies, which supports the validity and accuracy of the simulation model. Comparing the CDF plots against both the existing 3GPP map-based model and the measured data shows that the NU model aligns more closely with the measurements. Some discrepancies remain: the CDFs of ASA, particularly at the sub-6 GHz frequencies, show a significant gap between the measured data and the simulated result, and further analysis is needed to improve how the model reflects site-specific characteristics.
| Item | Description |
|---|---|
| Project title | R&D for the realization of a high-precision radio wave emulator in cyberspace (JPJ000254) |
| Main objective (Group A) | To model a site-specific channel model (propagation layer) following the NU map-based model, producing PathGridData that contains detailed information at each grid position (Rx point) – delay, power and angle of departure/arrival – for use in channel emulators and for network performance analysis and optimisation. |
| Development objective | To design a customised channel model and simulation software that can rapidly and effectively produce PathGridData for the channel emulator (managed by a separate team), enabling comprehensive testing of channel behaviour ahead of field testing or real-world implementation. |
| Specific objective | To enhance the accuracy of predicting channel characteristics in the mmWave and sub-6 GHz bands, reflecting site-specific radio-wave characteristics more accurately than existing channel models. |
| Environment and scenario | Frequencies 28350, 4850, 2462 and 922 MHz; scenario Urban Macro (UMa) |
| Necessary input | (a) The R-layer, providing the ray-tracing files for each frequency and scenario; (b) actual measured data for those frequencies and scenarios. |
| Outputs | PathGridData for each frequency and scenario as .mat files and corresponding
.csv files – eight output sets in total, for four frequencies at two MS
heights (4 m and 9 m). |