ZH: find beacon multiple by reconstructing shower amplitudes

This commit is contained in:
Eric Teunis de Boone 2022-11-28 19:03:14 +01:00
parent 6c0ae17b07
commit 83dafb0cc6

View file

@ -0,0 +1,184 @@
#!/usr/bin/env python3
# vim: fdm=indent ts=4
"""
Find the best integer multiple to shift
antennas to be able to resolve
"""
import h5py
from itertools import combinations, zip_longest, product
import matplotlib.pyplot as plt
import numpy as np
from scipy.interpolate import interp1d
from earsim import REvent
from atmocal import AtmoCal
import aa_generate_beacon as beacon
import lib
from lib import rit
def find_best_sample_shifts_summing_at_location(test_loc, antennas, allowed_sample_shifts, dt=1):
"""
Find the best sample_shift for each antenna by summing the antenna traces
and seeing how to get the best alignment.
"""
a_ = []
t_ = []
t_min = 1e9
t_max = -1e9
# propagate to test location
for i, ant in enumerate(antennas):
aloc = [ant.x, ant.y, ant.z]
delta, dist = atm.light_travel_time(test_loc, aloc)
delta = delta*1e9
t__ = np.subtract(ant.t_AxB, delta)
t_.append(t__)
a_.append(ant.E_AxB)
if t__[0] < t_min:
t_min = t__[0]
if t__[-1] > t_max:
t_max = t__[-1]
# Interpolate and find best sample shift
max_neg_shift = np.min(allowed_sample_shifts) * dt
max_pos_shift = np.max(allowed_sample_shifts) * dt
t_sum = np.arange(t_min+1+max_neg_shift, t_max-1+max_pos_shift, dt)
a_sum = np.zeros(len(t_sum))
best_sample_shifts = np.zeros( (len(antennas)) ,dtype=int)
for i, (t_r, E_) in enumerate(zip(t_, a_)):
t_min = t_r[0]
f = interp1d(t_r, E_, assume_sorted=True, bounds_error=False, fill_value=0)
a_int = f(t_sum)
if i == 0:
a_sum += a_int
best_sample_shifts[i] = 0
continue
shift_maxima = np.zeros( len(allowed_sample_shifts) )
for j, shift in enumerate(allowed_sample_shifts):
augmented_a = np.roll(a_int, shift)
shift_maxima[j] = np.max(augmented_a + a_sum)
# transform maximum into best_sample_shift
best_idx = np.argmax(shift_maxima)
best_sample_shifts[i] = allowed_sample_shifts[best_idx]
a_sum += np.roll(a_int, best_sample_shifts[i])
# Return ks
return best_sample_shifts, np.max(a_sum)
if __name__ == "__main__":
from os import path
import sys
atm = AtmoCal()
fname = "ZH_airshower/mysim.sry"
allowed_ks = np.arange(-2, 3, 1)
Xref = 450
x_coarse = np.linspace(-2e3, 2e3, 11)
y_coarse = np.linspace(-2e3, 2e3, 11)
####
fname_dir = path.dirname(fname)
antennas_fname = path.join(fname_dir, beacon.antennas_fname)
time_diffs_fname = 'time_diffs.hdf5' if not True else antennas_fname
# Read in antennas from file
_, __, antennas = beacon.read_beacon_hdf5(antennas_fname)
# Read original REvent
ev = REvent(fname)
# .. patch in our antennas
ev.antennas = antennas
# For now only implement using one freq_name
freq_names = antennas[0].beacon_info.keys()
if len(freq_names) > 1:
raise NotImplementedError
freq_name = next(iter(freq_names))
f_beacon = ev.antennas[0].beacon_info[freq_name]['freq']
# determine best ks per location
dt = ev.antennas[0].t[1] - ev.antennas[0].t[0]
allowed_sample_shifts = np.ceil(allowed_ks/f_beacon /dt).astype(int)
# Remove time due to true phase
for i, ant in enumerate(ev.antennas):
true_phase = ant.beacon_info[freq_name]['phase']
true_phase_time = true_phase/(2*np.pi*f_beacon)
ev.antennas[i].t -= true_phase_time
# Prepare polarisation and passbands
rit.set_pol_and_bp(ev, low=0.03, high=0.08)
dXref = atm.distance_to_slant_depth(np.deg2rad(ev.zenith),Xref,0)
scale2d = dXref*np.tan(np.deg2rad(2.))
# Setup Plane grid to test
for r in range(6):
xoff, yoff = 0,0
if r == 0:
x = x_coarse
y = y_coarse
else:
best_idx = np.argmax(maxima_per_loc)
xoff = xx[best_idx]
yoff = yy[best_idx]
x /= 4
y /= 4
print(f"Testing grid {r} centered on {xoff}, {yoff}")
ks_per_loc = np.zeros( (len(x)*len(y), len(ev.antennas)) )
maxima_per_loc = np.zeros( (len(x)*len(y)))
## Check each location on grid
xx = []
yy = []
N_loc = len(maxima_per_loc)
for i, (x_, y_) in enumerate(product(x,y)):
if i % 10 ==0:
print(f"Testing location {i} out of {N_loc}")
test_loc = (x_+xoff)* ev.uAxB + (y_+yoff)*ev.uAxAxB + dXref *ev.uA
xx.append(x_+xoff)
yy.append(y_+yoff)
# Find best k for each antenna
shifts, maximum = find_best_sample_shifts_summing_at_location(test_loc, ev.antennas, allowed_sample_shifts, dt)
# Translate sample shifts back into period multiple k
ks = shifts*f_beacon*dt
ks_per_loc[i] = ks
maxima_per_loc[i] = maximum
xx = np.array(xx)
yy = np.array(yy)
best_idx = np.argmax(maxima_per_loc)
np.savetxt(__file__+f'.run{r}.txt')
print(ks_per_loc[best_idx])
if True: #plot maximum at test locations
fig, axs = plt.subplots()
axs.set_title(f"Grid Run {r}")
axs.set_ylabel("vxvxB [km]")
axs.set_xlabel(" vxB [km]")
sc = axs.scatter(xx/1e3, yy/1e3, c=maxima_per_loc, cmap='Spectral_r', alpha=0.6)
fig.colorbar(sc, ax=axs)
fig.savefig(__file__+f'.run{r}.pdf')
plt.show()