Add option to cluster detector hits by time

[fhagemann]

The current clustering algorithm in https://github.com/JuliaPhysics/SolidStateDetectors.jl/blob/main/src/ChargeClustering/ChargeClustering.jl only clusters by position, but does not consider timing information.

Similar to a cluster_radius, it might also make sense to add a cluster_time.

Also, when clustering a Geant4 output table, the times thit are not clustered but remain untouched.

SolidStateDetectors.jl/src/ChargeClustering/ChargeClustering.jl

Lines 59 to 74 in e68da8b

	function cluster_detector_hits(table::TypedTables.Table, cluster_radius::RealQuantity)
	@assert :pos in TypedTables.columnnames(table) "Table has no column `pos`"
	@assert :edep in TypedTables.columnnames(table) "Table has no column `edep`"
	@assert :detno in TypedTables.columnnames(table) "Table has no column `detno`"
	clustered_nt = map(
	evt -> cluster_detector_hits(evt.detno, evt.edep, evt.pos, cluster_radius),
	table
	)
	TypedTables.Table(merge(
	TypedTables.columns(table),
	map(
	VectorOfVectors,
	TypedTables.columns(clustered_nt)
	)
	))
	end

It might make sense to add (optional) clustering of thit also here.

[sm2909]

Hi! I was going through this issue and had a small clarification.
The time-based clustering mentioned here — is it conceptually independent of the radius-based clustering currently implemented, or is there some coupling between the two clustering?

I’m trying to understand this better because I’d like to contribute to this feature. Thanks!

[fhagemann]

Thanks for reaching out,
maybe I can try to explain this issue in a bit more detail using an example:

using SolidStateDetectors, Geant4, Unitful
sim = Simulation{Float32}(SSD_examples[:InvertedCoax])
source = IsotopeSource(90, 228, 0.0, 0.0, CartesianPoint(0.05,0,0.05))
app = G4JLApplication(sim, source)
t = run_geant4_simulation(app, 1_000)

Here, we get a lot of entries in pos (position), edep (deposited energy) and thit (time of hit).

t.edep[1]

7-element view(::Vector{Quantity{Float32, 𝐋^2 𝐌 𝐓^-2, Unitful.FreeUnits{(keV,), 𝐋^2 𝐌 𝐓^-2, nothing}}}, 1:7) with eltype Quantity{Float32, 𝐋^2 𝐌 𝐓^-2, Unitful.FreeUnits{(keV,), 𝐋^2 𝐌 𝐓^-2, nothing}}:
  132.01038f0 keV
 5685.4146f0 keV
  103.50545f0 keV
 6288.1274f0 keV
  116.61237f0 keV
 6778.213f0 keV
    5.398038f0 keV

t.thit[1]

7-element view(::Vector{Quantity{Float32, 𝐓, Unitful.FreeUnits{(s,), 𝐓, nothing}}}, 1:7) with eltype Quantity{Float32, 𝐓, Unitful.FreeUnits{(s,), 𝐓, nothing}}:
 1.7250684f6 s
 1.7991591f6 s
 1.7991591f6 s
 1.7992252f6 s
 1.7992252f6 s
 1.7992252f6 s
 1.7992252f6 s

When clustering using a cluster_radius, we add all energies within that given radius together and determine a combined position for all of those hits.

tc = SolidStateDetectors.cluster_detector_hits(t, 20u"μm")
tc.edep[1]

2-element view(::Vector{Quantity{Float32, 𝐋^2 𝐌 𝐓^-2, Unitful.FreeUnits{(keV,), 𝐋^2 𝐌 𝐓^-2, nothing}}}, 1:2) with eltype Quantity{Float32, 𝐋^2 𝐌 𝐓^-2, Unitful.FreeUnits{(keV,), 𝐋^2 𝐌 𝐓^-2, nothing}}:
   357.52625f0 keV
 18751.754f0 keV

tc.thit[1]

7-element view(::Vector{Quantity{Float32, 𝐓, Unitful.FreeUnits{(s,), 𝐓, nothing}}}, 1:7) with eltype Quantity{Float32, 𝐓, Unitful.FreeUnits{(s,), 𝐓, nothing}}:
 1.7250684f6 s
 1.7991591f6 s
 1.7991591f6 s
 1.7992252f6 s
 1.7992252f6 s
 1.7992252f6 s
 1.7992252f6 s

The issue here is that

1. the column thit is just completely ignored by the clustering, no there is no way of inferring what hit happened at what time, after having clustered the hits.
2. We might not want to cluster hits that are separated in time, so we would need to make sure we ONLY cluster events that are also within a given time window (and maybe consider splitting events with hits at very different times into different rows in the clustered table).

The time-based clustering mentioned here — is it conceptually independent of the radius-based clustering currently implemented, or is there some coupling between the two clustering?

I would say, there should be coupling, and we should only cluster events together, if they are BOTH within the given cluster_radius and the given cluster_time.