2
0
mirror of https://github.com/m-labs/artiq.git synced 2024-12-26 11:48:27 +08:00
artiq/examples/flopping_f_simulation.py

57 lines
1.4 KiB
Python
Raw Normal View History

2015-01-13 19:12:35 +08:00
from math import sqrt, cos, pi
import time
import random
import numpy as np
from scipy.optimize import curve_fit
2015-01-13 19:12:35 +08:00
from artiq import *
def model(x, F0):
t = 0.02
tpi = 0.03
A = 80
B = 40
2015-01-13 19:12:35 +08:00
return A+(B-A)/2/(4*tpi**2*(x-F0)**2+1)*(1-cos(pi*t/tpi*sqrt(4*tpi**2*(x-F0)**2+1)))
class FloppingF(AutoDB):
class DBKeys:
implicit_core = False
npoints = Argument(100)
min_freq = Argument(1000)
max_freq = Argument(2000)
F0 = Argument(1500)
noise_amplitude = Argument(0.1)
2015-01-13 19:12:35 +08:00
frequency = Result()
brightness = Result()
flopping_freq = Parameter()
@staticmethod
def realtime_results():
return {
("frequency", "brightness"): "xy"
}
def run(self):
for i in range(self.npoints):
frequency = (self.max_freq-self.min_freq)*i/(self.npoints - 1) + self.min_freq
brightness = model(frequency, self.F0) + self.noise_amplitude*random.random()
2015-01-13 19:12:35 +08:00
self.frequency.append(frequency)
self.brightness.append(brightness)
time.sleep(0.1)
self.analyze()
def analyze(self):
popt, pcov = curve_fit(lambda xdata, F0: [model(x, F0) for x in xdata],
self.frequency.read, self.brightness.read,
p0=[self.flopping_freq])
perr = np.sqrt(np.diag(pcov))
if perr < 0.1:
self.flopping_freq = float(popt)