Far-field signal calculation

[jaafar-chakrani]

Overview

This PR introduces far-field dipole signal simulation into the charge readout pipeline, enabling complete electromagnetic response calculation for detector characterization. Far-field signals from voxelized ionization charge are now computed and added to the pixel response alongside near-field contributions.

Details

Architecture

The far-field simulation is integrated into the pixel batch processing loop as follows:

For each pixel batch:
  1. Calculate near-field signals (existing)
  2. Accumulate pixel signals from all contributing tracks
  3. [NEW] Add far-field dipole contribution per TPC
  4. [NEW] Handle induction-only pixels (far-field only)
  5. Simulate electronics response (ADC digitization)

Key Components

1. Voxelization (Per-Event-Batch)

• Tracks are voxelized once per event batch to avoid cross-TPC contamination
• Voxel positions: $(x_v, y_v, z_v)$ in detector coordinates
• Voxel charges: summed ionization per voxel
• Separate caches maintained per TPC

for tpc_idx in active_tpc_indices_all:
    voxel_indices, voxel_charges, ... = gpu_voxelize(tpc_tracks)
    voxel_cache[tpc_idx] = {x, y, z, q}  # GPU cupy arrays

2. Pixel Classification

• Pixels categorized as collection, neighbor, or induction-only (right now, collection and neighbor are treated in the same way)
• Classification cached per TPC to avoid recomputation
• Used to identify pixels requiring only far-field treatment

classification_cache[tpc_idx] = pixel_classifier.classify_pixels(
    tracks, plane_id=tpc_idx
)

3. Far-Field Signal Calculation

Mixed Pixels (Near + Far-field):

• Computed per TPC for pixels with track contributions
• Dipole signals added after near-field accumulation
• Handled in per-pixel-batch loop

Induction-Only Pixels (Far-field only):

• Processed in separate loop after pixel batches
• Uses classifier's pre-computed pixel coordinates
• Avoids redundant classification per batch

4. Time Sampling & Digitization

• Far-field signals computed on same time grid as near-field

Validation

See presentation here.