reboost 0.3.0__py3-none-any.whl → 0.4.2__py3-none-any.whl

reboost/hpge/psd.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import logging
+from math import erf, exp
 
 import awkward as ak
 import numba
@@ -236,3 +237,259 @@ def _drift_time_heuristic_impl(
         dt_heu[i] = max_id_metric
 
     return dt_heu
+
+
+@numba.njit(cache=True)
+def _vectorized_erf(x: ArrayLike) -> NDArray:
+    """Error function that can take in a numpy array."""
+    out = np.empty_like(x)
+    for i in range(x.size):
+        out[i] = erf(x[i])
+    return out
+
+
+@numba.njit(cache=True)
+def _current_pulse_model(
+    times: ArrayLike, Amax: float, mu: float, sigma: float, tail_fraction: float, tau: float
+) -> NDArray:
+    r"""Analytic model for the current pulse in a Germanium detector.
+
+    Consists of a Gaussian and an exponential tail:
+
+     .. math::
+
+       A(t) = A_{max}\times (1-p)\times \text{Gauss}(t,\mu,\sigma)+ A \times p (1-\text{Erf}((t-\mu)/sigma))\times
+        \frac{e^{(t/\tau)}}{2e^{\mu/\tau}}
+
+    Parameters
+    ----------
+    times
+        Array of times to compute current for
+    Amax
+        Maximum current
+    mu
+        Time of the maximum current.
+    sigma
+        Width of the current pulse
+    tail_fraction
+        Fraction of the tail in the pulse.
+    tau
+        Time constant of the low time tail.
+
+    Returns
+    -------
+    The predicted current waveform for this energy deposit.
+    """
+    norm = 2 * exp(mu / tau)
+
+    dx = times - mu
+    term1 = Amax * (1 - tail_fraction) * np.exp(-(dx * dx) / (2 * sigma * sigma))
+    term2 = Amax * tail_fraction * (1 - _vectorized_erf(dx / sigma)) * np.exp(times / tau) / norm
+
+    return term1 + term2
+
+
+def convolve_surface_response(surf_current: np.ndarray, bulk_pulse: np.ndarray) -> NDArray:
+    """Convolve the surface response pulse with the bulk current pulse.
+
+    This combines the current induced on the edge of the FCCD region with the bulk response
+    on the p+ contact.
+
+    Parameters
+    ----------
+    surf_current
+        array of the current induced via diffusion against time.
+    bulk_pulse
+        the pulse template to convolve the surface current with.
+
+    Returns
+    -------
+    the current waveform after convolution.
+    """
+    return np.convolve(surf_current, bulk_pulse, mode="full")[: len(surf_current)]
+
+
+@numba.njit(cache=True)
+def get_current_waveform(
+    edep: ak.Array,
+    drift_time: ak.Array,
+    template: ArrayLike,
+    start: float,
+    dt: float,
+    range_t: tuple,
+) -> tuple(NDArray, NDArray):
+    r"""Estimate the current waveform.
+
+    Based on modelling the current as a sum over the current pulse model defined by
+    the template.
+
+    .. math::
+        A(t) = \sum_i E_i \times N f(t,dt_i,\vector{\theta})
+
+    Where:
+        - $f(t)$ is the template
+        - $\vector{\theta}$ are the parameters (sigma, p, tau)
+        - $E_i$ and $dt_i$ are the deposited energy and drift time.
+        - N is a normalisation term
+
+    Parameters
+    ----------
+    edep
+        Array of energies for each step
+    drift_time
+        Array of drift times for each step
+    template
+        array of the template for the current waveforms, with 1 ns binning.
+    start
+        first time value of the template
+    dt
+        timestep (in ns) for the template.
+    range_t
+        a range of times to search around
+
+    Returns
+    -------
+    A tuple of the time and current for the current waveform for this event.
+    """
+    n = len(template)
+
+    times = np.arange(n) * dt + start
+    y = np.zeros_like(times)
+
+    for i in range(len(edep)):
+        E = edep[i]
+        mu = drift_time[i]
+        shift = int(mu / dt)
+
+        # Add scaled template starting at index `shift`
+        for j in range(n):
+            if (
+                (shift + j) >= n
+                or (times[shift + j] < range_t[0])
+                or (times[shift + j] > range_t[1])
+            ):
+                continue
+            y[shift + j] += E * template[j]
+
+    return times, y
+
+
+@numba.njit(cache=True)
+def _estimate_current_impl(
+    edep: ak.Array,
+    dt: ak.Array,
+    sigma: float,
+    tail_fraction: float,
+    tau: float,
+    mean_AoE: float = 0,
+) -> tuple[NDArray, NDArray]:
+    """Estimate the maximum current that would be measured in the HPGe detector.
+
+    This is based on extracting a waveform with :func:`get_current_waveform` and finding the maxima of it.
+
+    Parameters
+    ----------
+    edep
+        Array of energies for each step.
+    drift_time
+        Array of drift times for each step.
+    sigma
+        Sigma parameter of the current pulse model.
+    tail_fraction
+        Tail-fraction parameter of the current pulse.
+    tau
+        Tail parameter of the current pulse
+    mean_AoE
+        The mean AoE value for this detector (to normalise current pulses).
+    """
+    A = np.zeros(len(dt))
+    maximum_t = np.zeros(len(dt))
+
+    # get normalisation factor
+    x_coarse = np.linspace(-1000, 3000, 201)
+    x_fine = np.linspace(-1000, 3000, 4001)
+
+    # make a template with 1 ns binning so
+    # template[(i-start)/dt] = _current_pulse_model(x,1,i,...)
+
+    template_coarse = _current_pulse_model(x_coarse, 1, 0, sigma, tail_fraction, tau)
+    template_coarse /= np.max(template_coarse)
+    template_coarse *= mean_AoE
+
+    template_fine = _current_pulse_model(x_fine, 1, 0, sigma, tail_fraction, tau)
+    template_fine /= np.max(template_fine)
+    template_fine *= mean_AoE
+
+    for i in range(len(dt)):
+        t = np.asarray(dt[i])
+        e = np.asarray(edep[i])
+
+        # first pass
+        times_coarse, W = get_current_waveform(
+            e, t, template=template_coarse, start=-1000, dt=20, range_t=(-1000, 3000)
+        )
+
+        max_t = times_coarse[np.argmax(W)]
+
+        # fine scan
+        times, W = get_current_waveform(
+            e, t, template=template_fine, start=-1000, dt=1, range_t=(max_t - 50, max_t + 50)
+        )
+
+        A[i] = np.max(W)
+        maximum_t[i] = times[np.argmax(W)]
+
+    return A, maximum_t
+
+
+def maximum_current(
+    edep: ArrayLike,
+    drift_time: ArrayLike,
+    *,
+    sigma: float,
+    tail_fraction: float,
+    tau: float,
+    mean_AoE: float = 0,
+    get_timepoint: bool = False,
+) -> Array:
+    """Estimate the maximum current in the HPGe detector based on :func:`_estimate_current_impl`.
+
+    Parameters
+    ----------
+    edep
+        Array of energies for each step.
+    drift_time
+        Array of drift times for each step.
+    sigma
+        Sigma parameter of the current pulse model.
+    tail_fraction
+        Tail-fraction parameter of the current pulse.
+    tau
+        Tail parameter of the current pulse
+    mean_AoE
+        The mean AoE value for this detector (to normalise current pulses).
+    get_timepoint
+        Flag to return the time of the maximum current (relative to t0) instead of the current.
+
+    Returns
+    -------
+    An Array of the maximum current for each hit.
+    """
+    # extract LGDO data and units
+    drift_time, _ = units.unwrap_lgdo(drift_time)
+
+    edep, _ = units.unwrap_lgdo(edep)
+
+    curr, time = _estimate_current_impl(
+        ak.Array(edep),
+        ak.Array(drift_time),
+        sigma=sigma,
+        tail_fraction=tail_fraction,
+        tau=tau,
+        mean_AoE=mean_AoE,
+    )
+
+    # return
+    if get_timepoint:
+        return Array(time)
+    return Array(curr)

reboost/hpge/surface.py

@@ -4,10 +4,12 @@ import logging
 
 import awkward as ak
 import legendhpges
+import numba
 import numpy as np
 from lgdo import VectorOfVectors
 from lgdo.types import LGDO
 from numpy.typing import ArrayLike
+from scipy import stats
 
 log = logging.getLogger(__name__)
 
@@ -104,4 +106,146 @@ def distance_to_surface(
             local_positions[indices], surface_indices=surface_indices
         )
 
-    return VectorOfVectors(ak.unflatten(distances, size))
+    return VectorOfVectors(ak.unflatten(distances, size), dtype=np.float32)
+
+
+@numba.njit(cache=True)
+def _advance_diffusion(
+    charge: np.ndarray,
+    factor: float,
+    recomb: float = 0,
+    recomb_depth: float = 600,
+    delta_x: float = 10,
+):
+    """Make a step of diffusion using explicit Euler scheme.
+
+    Parameters
+    ----------
+    charge
+        charge in each space bin up to the FCCD
+    factor
+        the factor of diffusion for the Euler scheme
+    recomb
+        the recomination probability.
+    recomb_depth
+        the depth of the recombination region.
+    delta_x
+        the width of each spatial bin.
+
+    Returns
+    -------
+    a tuple of the charge distribution at the next time step and the collected charge.
+    """
+    charge_xp1 = np.append(charge[1:], [0])
+    charge_xm1 = np.append([0], charge[:-1])
+
+    # collected charge
+    collected = factor * charge[-1]
+
+    # charge at the next step
+    charge_new = charge_xp1 * factor + charge_xm1 * factor + charge * (1 - 2 * factor)
+
+    # correction for recombination
+    charge_new[0 : int(recomb_depth / delta_x)] = (1 - recomb) * charge_new[
+        0 : int(recomb_depth / delta_x)
+    ]
+
+    return charge_new, collected
+
+
+@numba.njit(cache=True)
+def _compute_diffusion_impl(
+    init_charge: np.ndarray,
+    nsteps: int,
+    factor: float,
+    recomb: float = 0,
+    recomb_depth: float = 600,
+    delta_x: float = 10,
+):
+    """Compute the charge collected as a function of time.
+
+    Parameters
+    ----------
+    init_charge
+        Initial charge distribution.
+    nsteps
+        Number of time steps to take.
+    kwargs
+        Keyword arguments to pass to :func:`_advance_diffusion`
+    """
+    charge = init_charge
+    collected_charge = np.zeros(nsteps)
+
+    for i in range(nsteps):
+        charge, collected = _advance_diffusion(
+            charge, factor=factor, recomb=recomb, recomb_depth=recomb_depth, delta_x=delta_x
+        )
+        collected_charge[i] = collected
+
+    return collected_charge
+
+
+def get_surface_response(
+    fccd: float,
+    recomb_depth: float,
+    init: float = 0,
+    recomb: float = 0.002,
+    init_size: float = 0.0,
+    factor: float = 0.29,
+    nsteps: int = 10000,
+    delta_x: float = 10,
+):
+    """Extract the surface response current pulse based on diffusion.
+
+    Parameters
+    ----------
+    fccd
+        the full charge collection depth (in um)
+    recomb_depth
+        the depth of the recombination region (in um)
+    init
+        the initial position of the charge (in um)
+    recomb
+        the recombination rate
+    init_size
+        the initial size of the charge cloud (in um)
+    factor
+        the factor for the explicit Euler scheme (the probability of charge diffusuion)
+    nsteps
+        the number of time steps.
+    delta_x
+        the width of each position bin.
+    """
+    # number of position steps
+    nx = int(fccd / delta_x)
+
+    # initial charge
+    charge = np.zeros(nx)
+
+    # generate initial conditions
+    x = (fccd / nx) * np.arange(nx)
+    x_full = (fccd / nx) * np.arange(2 * nx)
+
+    # generate initial conditions
+    if init_size != 0:
+        charge = stats.norm.pdf(x, loc=init, scale=init_size)
+        charge_full = stats.norm.pdf(x_full, loc=init, scale=init_size)
+        charge_col = [(np.sum(charge_full) - np.sum(charge)) / np.sum(charge_full)]
+        charge = charge / np.sum(charge_full)
+    elif int(init * nx / fccd) < len(charge):
+        charge[int(init * nx / fccd)] = 1
+        charge_col = np.array([])
+    else:
+        charge_col = np.array([1])
+
+    # run the simulation
+    charge_collected = _compute_diffusion_impl(
+        charge,
+        nsteps=nsteps,
+        factor=factor,
+        recomb=recomb,
+        recomb_depth=recomb_depth,
+        delta_x=delta_x,
+    )
+
+    return np.cumsum(np.concatenate((charge_col, charge_collected)))

reboost/iterator.py

@@ -8,7 +8,7 @@ import awkward as ak
 from lgdo.lh5 import LH5Store
 from lgdo.types import LGDO, Table
 
-from reboost import build_glm
+from . import build_glm
 
 log = logging.getLogger(__name__)
 
@@ -25,17 +25,28 @@ class GLMIterator:
         n_rows: int | None,
         *,
         stp_field: str = "stp",
-        read_vertices: bool = False,
         buffer: int = 10000,
         time_dict: dict | None = None,
+        reshaped_files: bool = False,
     ):
-        """Constructor for the glmIterator.
+        """Constructor for the GLMIterator.
+
+        The GLM iterator provides a way to iterate over the
+        simulated geant4 evtids, extracting the number of hits or steps for
+        each range in evtids. This ensures a single simulated event
+        is not split between two iterations and allows to specify a
+        start and an end evtid to extract.
+
+        In case the data is already reshaped and we do not need to
+        read a specific range of evtids this iterator is just loops
+        over the input stp field. Otherwise if the GLM file is not provided
+        this is created in memory.
 
         Parameters
         ----------
         glm_file
             the file containing the event lookup map, if `None` the glm will
-            be created in memory.
+            be created in memory if needed.
         stp_file
             the file containing the steps to read.
         lh5_group
@@ -46,12 +57,12 @@ class GLMIterator:
             the number of rows to read, if `None` read them all.
         stp_field
             name of the group.
-        read_vertices
-            whether to read also the vertices table.
         buffer
             the number of rows to read at once.
         time_dict
             time profiling data structure.
+        reshaped_files
+            flag for whether the files are reshaped.
         """
         # initialise
         self.glm_file = glm_file
@@ -62,18 +73,47 @@ class GLMIterator:
         self.n_rows = n_rows
         self.buffer = buffer
         self.current_i_entry = 0
-        self.read_vertices = read_vertices
         self.stp_field = stp_field
+        self.reshaped_files = reshaped_files
 
         # would be good to replace with an iterator
         self.sto = LH5Store()
         self.n_rows_read = 0
         self.time_dict = time_dict
         self.glm = None
+        self.use_glm = True
+
+        glm_n_rows = 0
 
-        # build the glm in memory
-        if self.glm_file is None:
-            self.glm = build_glm.build_glm(stp_file, None, out_table_name="glm", id_name="evtid")
+        # build the glm in memory if needed
+        if self.glm_file is None and (
+            (self.n_rows is not None) or (self.start_row != 0) or not reshaped_files
+        ):
+            if self.time_dict is not None:
+                time_start = time.time()
+
+            self.glm = build_glm.build_glm(
+                stp_file, None, out_table_name="glm", id_name="evtid", lh5_groups=[lh5_group]
+            )
+
+            if self.time_dict is not None:
+                self.time_dict.update_field("read/glm", time_start)
+
+            glm_n_rows = len(self.glm)
+
+        elif self.glm_file is None:
+            self.use_glm = False
+        else:
+            glm_n_rows = self.sto.read_n_rows(f"glm/{self.lh5_group}", self.glm_file)
+
+        # get the number of stp rows
+        stp_n_rows = self.sto.read_n_rows(f"{self.stp_field}/{self.lh5_group}", self.stp_file)
+
+        # heuristics for a good buffer length
+        if self.use_glm:
+            self.buffer = int(buffer * glm_n_rows / stp_n_rows)
+            msg = f"Number of stp rows {stp_n_rows}, number of glm rows {glm_n_rows} changing buffer from {buffer} to {self.buffer}"
+            log.info(msg)
 
     def __iter__(self) -> typing.Iterator:
         self.current_i_entry = 0
@@ -81,78 +121,99 @@ class GLMIterator:
         self.start_row_tmp = self.start_row
         return self
 
-    def __next__(self) -> tuple[LGDO, LGDO | None, int, int]:
+    def get_n_rows(self):
+        """Get the number of rows to read."""
         # get the number of rows to read
+        if self.time_dict is not None:
+            time_start = time.time()
+
         if self.n_rows is not None:
             rows_left = self.n_rows - self.n_rows_read
             n_rows = self.buffer if (self.buffer > rows_left) else rows_left
         else:
             n_rows = self.buffer
 
-        if self.time_dict is not None:
-            time_start = time.time()
+        glm_rows = None
+        start = 0
+        n = 0
 
-        # read the glm rows]
-        if self.glm_file is not None:
-            glm_rows, n_rows_read = self.sto.read(
-                f"/glm/{self.lh5_group}", self.glm_file, start_row=self.start_row_tmp, n_rows=n_rows
-            )
-        else:
-            # get the maximum row to read
-            max_row = self.start_row_tmp + n_rows
-            max_row = min(len(self.glm[self.lh5_group]), max_row)
+        if self.use_glm:
+            if self.glm_file is not None:
+                glm_rows, n_rows_read = self.sto.read(
+                    f"glm/{self.lh5_group}",
+                    self.glm_file,
+                    start_row=self.start_row_tmp,
+                    n_rows=n_rows,
+                )
+            else:
+                # get the maximum row to read
+                max_row = self.start_row_tmp + n_rows
+                max_row = min(len(self.glm[self.lh5_group]), max_row)
 
-            if max_row != self.start_row_tmp:
-                glm_rows = Table(self.glm[self.lh5_group][self.start_row_tmp : max_row])
+                if max_row != self.start_row_tmp:
+                    glm_rows = Table(self.glm[self.lh5_group][self.start_row_tmp : max_row])
 
-            n_rows_read = max_row - self.start_row_tmp
+                n_rows_read = max_row - self.start_row_tmp
 
-        if self.time_dict is not None:
-            self.time_dict.update_field("read/glm", time_start)
+            if self.time_dict is not None:
+                self.time_dict.update_field("read/glm", time_start)
 
-        self.n_rows_read += n_rows_read
-        self.start_row_tmp += n_rows_read
+            self.n_rows_read += n_rows_read
+            self.start_row_tmp += n_rows_read
 
-        if n_rows_read == 0:
-            raise StopIteration
+            # view our glm as an awkward array
+            if glm_rows is not None:
+                glm_ak = glm_rows.view_as("ak")
 
-        # view our glm as an awkward array
-        glm_ak = glm_rows.view_as("ak")
+                # remove empty rows
+                glm_ak = glm_ak[glm_ak.n_rows > 0]
 
-        # remove empty rows
-        glm_ak = glm_ak[glm_ak.n_rows > 0]
+                if len(glm_ak) > 0:
+                    # extract range of stp rows to read
+                    start = glm_ak.start_row[0]
+                    n = ak.sum(glm_ak.n_rows)
 
-        if len(glm_ak) > 0:
-            # extract range of stp rows to read
-            start = glm_ak.start_row[0]
-            n = ak.sum(glm_ak.n_rows)
+        else:
+            start = self.start_row_tmp
+            n = n_rows
+            n_rows_read = n
+            self.start_row_tmp += n
+
+        return start, n, n_rows_read
+
+    def __next__(self) -> tuple[LGDO, int, int]:
+        """Read one chunk.
+
+        Returns
+        -------
+        a tuple of:
+            - the steps
+            - the chunk index
+            - the number of steps read
+        """
+        # read the glm rows]
+        start, n, n_rows_read = self.get_n_rows()
 
-            if self.time_dict is not None:
-                time_start = time.time()
+        if self.time_dict is not None:
+            time_start = time.time()
 
+        try:
             stp_rows, n_steps = self.sto.read(
-                f"/{self.stp_field}/{self.lh5_group}",
+                f"{self.stp_field}/{self.lh5_group}",
                 self.stp_file,
                 start_row=int(start),
                 n_rows=int(n),
             )
+        except OverflowError:
+            raise StopIteration from None
 
-            # save time
-            if self.time_dict is not None:
-                self.time_dict.update_field("read/stp", time_start)
+        if n_rows_read == 0 or n_steps == 0:
+            raise StopIteration
 
-            self.current_i_entry += 1
+        # save time
+        if self.time_dict is not None:
+            self.time_dict.update_field("read/stp", time_start)
 
-            if self.read_vertices:
-                vert_rows, _ = self.sto.read(
-                    "/vtx",
-                    self.stp_file,
-                    start_row=self.start_row,
-                    n_rows=n_rows,
-                )
-            else:
-                vert_rows = None
-            # vertex table should have same structure as glm
+        self.current_i_entry += 1
 
-            return (stp_rows, vert_rows, self.current_i_entry - 1, n_steps)
-        return (None, None, self.current_i_entry, 0)
+        return (stp_rows, self.current_i_entry - 1, n_steps)