fix(indoor): sample d_2D-in as min(U1,U2) per TR 38.901 §7.4.3.1

[Vinu2111]

Summary

Fixes #1108

The indoor 2D distance d_2D-in in _sample_indoor_distance() was computed from a single uniform sample U(min_2d_in, max_2d_in). Per TR 38.901 §7.4.3.1, it must be min(d_2D-in-1, d_2D-in-2) where both are independent uniform samples over the same range.

Root Cause

The previous implementation produced a flat uniform distribution. The correct min(U1, U2) of two i.i.d. uniforms follows a distribution skewed toward the lower bound — meaning indoor UTs are statistically placed closer to the building wall, which materially affects pathloss and O2I penetration loss calculations.

Fix

Draw two independent rand() samples and take torch.minimum(u1, u2) before applying indoor_mask:

u1 = rand(...) * (self.max_2d_in - self.min_2d_in) + self.min_2d_in
u2 = rand(...) * (self.max_2d_in - self.min_2d_in) + self.min_2d_in
distance_2d_in = torch.minimum(u1, u2) * indoor_mask

The fix applies to all three subclasses (UMa, UMi, RMa) since _sample_indoor_distance() lives in the base class SystemLevelScenario.

Changes

• src/sionna/phy/channel/tr38901/system_level_scenario.py — replace single uniform sample with min(U1, U2)
• test/unit/channel/test_3gpp_channel_scenario.py — two new tests added to TestRMaScenario:
  • test_d2d_in_range_and_outdoor_zero — verifies d_2D-in is within [0, max_2d_in] for all-indoor topology
  • test_d2d_in_outdoor_uts_are_zero — verifies d_2D-in is exactly 0 for all-outdoor topology

Testing

New tests cover range bounds and the indoor_mask zeroing logic. The existing test_distance_calculation continues to verify additive consistency (d_2D-in + d_2D-out == d_2D).