Wellcome to our project PathFinder.

PathFinder: Advancing Path Loss Prediction for Single-to-Multi-Transmitter Scenario.

Code

A. Comparison on S2MT-RPP task

Visualization examples of different models in the DS-RPP task

Fig. 7: Visualization of performance of different models when the number of transmitters is 4.

When the number of transmitters reaches four, as shown in Fig. 7, the generalization capabilities of all baseline models have nearly failed, particularly for AE and RadioUNet. PMNet and REM-Net perform relatively better, but even in straightforward building distribution scenarios, they still struggle to predict clear path loss boundaries. In contrast, PathFinder retains a certain clarity in path loss boundaries across various situations, especially when the building distribution is simpler (as in Sample 1), with an MSE of only 0.0006, indicating that the model's predictions closely align with the true targets.

B. Comparison on DS-RPP task

Fig. S1 displays the prediction results of all baseline methods alongside our model across different samples.

Visualization examples of different models in the DS-RPP task

Fig. S1: Visualization examples of different models in the DS-RPP task.

C. Coverage Analysis

Figs. S2, S3, S4, S5, and S6 visualize the results of different models when the importance of the coverage area is 40%, 30%, 20%, 10%, and 5%, respectively.

Visualization cases of different models when dealing with the most important 40% of the coverage area

Fig. S2: Visualization cases of different models when dealing with the most important 40% of the coverage area.

Visualization cases of different models when dealing with the most important 30% of the coverage area

Fig. S3: Visualization cases of different models when dealing with the most important 30% of the coverage area.

Visualization cases of different models when dealing with the most important 20% of the coverage area

Fig. S4: Visualization cases of different models when dealing with the most important 20% of the coverage area.

Visualization cases of different models when dealing with the most important 10% of the coverage area

Fig. S5: Visualization cases of different models when dealing with the most important 10% of the coverage area.

Visualization cases of different models when dealing with the most important 5% of the coverage area

Fig. S6: Visualization cases of different models when dealing with the most important 5% of the coverage area.

D. Comparison of Performance under distribution shift.

Table S1 shows the comparison of performance under distribution shift of different models when facing varying numbers of transmitters.

TABLE S1: Comparison of performance under distribution shift of different models when facing varying numbers of transmitters.
Tx-num/Model Metric AE RadioUNet PMNet REM-Net Ours Improvement
2 MSE 0.008311 0.009541 0.008171 0.008204 0.00095 88.43%
RMSE 0.091157 0.097661 0.090341 0.090515 0.03075 65.97%
NMSE 0.133632 0.15345 0.131345 0.131917 0.0152 88.43%
3 MSE 0.01307 0.025124 0.008654 0.006909 0.000825 88.05%
RMSE 0.114313 0.158482 0.092957 0.083091 0.028727 65.43%
NMSE 0.220913 0.424658 0.146268 0.116794 0.013955 88.05%
4 MSE 0.016445 0.031034 0.00786 0.005737 0.000805 85.97%
RMSE 0.128223 0.176148 0.088597 0.075713 0.028369 62.53%
NMSE 0.28556 0.538926 0.136471 0.099624 0.013983 85.96%
5 MSE 0.018037 0.032217 0.00685 0.005018 0.00083 83.48%
RMSE 0.134286 0.179481 0.082717 0.070813 0.02879 59.35%
NMSE 0.318152 0.568353 0.120816 0.088533 0.01463 83.48%