Browse Source

calc.py solved warnings

Egor 2 years ago
parent
commit
b365ff7854
1 changed files with 47 additions and 46 deletions
  1. 47 46
      source/backend/calc.py

+ 47 - 46
source/backend/calc.py

@@ -1,4 +1,4 @@
-from cmath import atan
+# from cmath import atan
 import numpy as np
 
 
@@ -14,12 +14,13 @@ def open_file(path):
             re.append(float(temp[1]))
             im.append(float(temp[2]))
     return freq, re, im
- 
+
+
 def prepare_data(freq, re, im, fl=None):
     """the function takes raw data and gives vectors of eq (8)"""
     # finding fl from the point with smallest magnitude if argument not provided
     if fl is None:
-        s = abs(np.array(re) + np.array(im)*1j)
+        s = abs(np.array(re) + np.array(im) * 1j)
         # frequency of loaded resonance
         fl = freq[list(abs(s)).index(min(abs(s)))]
 
@@ -39,6 +40,7 @@ def prepare_data(freq, re, im, fl=None):
     data = np.array([e1, e2, e3, gamma, p], dtype=np.cdouble)
     return data, fl
 
+
 def solution(data):
     """ takes projections of equation (8) on vectors e1, e2, e3 and solves the equations.
      It is also gives matrix of equation"""
@@ -68,27 +70,39 @@ def q_factor(a):
 def recalculation_of_data(data, a, c, d, error=False):
     """preparation data for the next iteration of solving system"""
     # data = np.array([e1, e2, e3, gamma, p], dtype=complex), t = e1, 1 = e2
-    eps = np.array(a[0]*data[0] + a[1]*data[1] - a[2]*data[0]*data[3] - data[3], dtype=complex)
+    eps = np.array(a[0] * data[0] + a[1] * data[1] - a[2] * data[0] * data[3] - data[3], dtype=complex)
     # eps is eq(7) line's errors
-    S2 = np.dot(abs(data[4]), abs(eps)*abs(eps))  # the weighted squared sum of errors
+    S2 = np.dot(abs(data[4]), abs(eps) * abs(eps))  # the weighted squared sum of errors
     sigma2A = []  # the square of standart deviation coefficients a
-    temp = c[0][0]*d[0][0] + c[1][1]*d[1][1] + c[2][2]*d[2][2]
+    temp = c[0][0] * d[0][0] + c[1][1] * d[1][1] + c[2][2] * d[2][2]
     for i in range(3):
         sigma2A.append(d[i][i] * S2 / temp)
     for i in range(len(data[4])):  # recalculation of weight coefficients P
-        data[4][i] = 1/(data[0][i]**2 * sigma2A[0] + sigma2A[1] + data[0][i]**2 * sigma2A[2] * (abs(data[3][i])**2))
+        data[4][i] = 1 / (
+                    data[0][i] ** 2 * sigma2A[0] + sigma2A[1] + data[0][i] ** 2 * sigma2A[2] * (abs(data[3][i]) ** 2))
     if error:
         return abs(np.array(sigma2A))
     else:
         return data
 
 
+def recalculating(data, a, c, d, n, printing=False):
+    for i in range(2, n):
+        data = recalculation_of_data(data, a, c, d)
+        a, c, d = solution(data)
+        Ql, diam, k, Q = q_factor(a)
+        sigma2A = recalculation_of_data(data, a, c, d, error=True)
+        sigmaQ0, sigmaQl = random_deviation(a, sigma2A, diam, k, Ql)
+        if printing:
+            print(f"Q = {Q} +- {sigmaQ0}, if i == {i}")
+
+
 def random_deviation(a, sigma2A, diam, k, Ql):
     """defines standart deviations of values"""
-    sigmaQl = sigma2A[2]**0.5
-    sigmaDiam = (sigma2A[0]/(abs(a[2])**2) + sigma2A[1] + abs(a[0]/a[2]/a[2])**2 * sigma2A[2])**0.5
-    sigmaK = 2*sigmaDiam/((2-diam)**2)
-    sigmaQ0 = ((1 + k)**2 * sigma2A[2] + Ql**2 * sigmaK**2)**0.5
+    sigmaQl = sigma2A[2] ** 0.5
+    sigmaDiam = (sigma2A[0] / (abs(a[2]) ** 2) + sigma2A[1] + abs(a[0] / a[2] / a[2]) ** 2 * sigma2A[2]) ** 0.5
+    sigmaK = 2 * sigmaDiam / ((2 - diam) ** 2)
+    sigmaQ0 = ((1 + k) ** 2 * sigma2A[2] + Ql ** 2 * sigmaK ** 2) ** 0.5
     return sigmaQ0, sigmaQl
 
 
@@ -96,13 +110,7 @@ def apply(filename):
     freq, re, im = open_file(filename)
     data = prepare_data(freq, re, im)
     a, c, d = solution(data)
-    for i in range(2, 10):
-        data = recalculation_of_data(data, a, c, d)
-        a, c, d = solution(data)
-        Ql, diam, k, Q = q_factor(a)
-        sigma2A = recalculation_of_data(data, a, c, d, error=True)
-        sigmaQ0, sigmaQl = random_deviation(a, sigma2A, diam, k, Ql)
-        print(f"Q = {Q} +- {sigmaQ0}, if i == {i}")
+    recalculating(data, a, c, d, 10, printing=True)
 
 
 def fl_fitting(freq, re, im):
@@ -110,42 +118,35 @@ def fl_fitting(freq, re, im):
 
     data, fl = prepare_data(freq, re, im)
     a, c, d = solution(data)
-
+    Ql, Q, sigmaQ0, sigmaQl = None, None, None, None
     # Repeated curve fitting
     # 1.189 of Qfactor Matlab 
     # fl2 = 0
     # g_d=0
     # g_c=0
     for x in range(0, 3):
-        g_c = (np.conj(a[2])*a[1]-a[0])/(np.conj(a[2])-a[2])
-        g_d = a[0]/a[2]
-        g_2 = 2*g_c-g_d
-        dt = (a[1]-g_2)/(g_2*a[2]-a[0])
-        fl2 = fl*(1 + np.real(dt)/2)
+        g_c = (np.conj(a[2]) * a[1] - a[0]) / (np.conj(a[2]) - a[2])
+        g_d = a[0] / a[2]
+        g_2 = 2 * g_c - g_d
+        dt = (a[1] - g_2) / (g_2 * a[2] - a[0])
+        fl2 = fl * (1 + np.real(dt) / 2)
         data, fl = prepare_data(freq, re, im, fl2)
         a, c, d = solution(data)
-
-    for i in range(2, 20):
-        data = recalculation_of_data(data, a, c, d)
-        a, c, d = solution(data)
-
-        Ql, diam, k, Q = q_factor(a)
-        sigma2A = recalculation_of_data(data, a, c, d, error=True)
-        sigmaQ0, sigmaQl = random_deviation(a, sigma2A, diam, k, Ql)
+    recalculating(data, a, c, d, 20)
 
     # taking into account coupling losses on page 69 of Qfactor Matlab
     # to get results similar to example program 
-    if False:
-        phi1=np.arctan(np.double(g_d.imag/g_d.real)) # 1.239
-        phi2=np.arctan(np.double((g_c.imag-g_d.imag)/(g_c.real-g_d.real)))
-        phi=-phi1+phi2
-        d_s=(1-np.abs(g_d)**2)/(1-np.abs(g_d)*np.cos(phi))
-        diam = abs(a[1] - a[0] / a[2])
-        qk=1/(d_s/diam-1)
-
-        sigma2A = recalculation_of_data(data, a, c, d, error=True)
-        sigmaQ0 = random_deviation(a, sigma2A, diam, k, Ql)
-        Q = Ql * (1 + qk)  # Q-factor = result
-        print(f"Q0 = {Q} +- {sigmaQ0}")
-    
-    return Q,sigmaQ0, Ql, sigmaQl,a
+    # if False:
+    #     phi1=np.arctan(np.double(g_d.imag/g_d.real)) # 1.239
+    #     phi2=np.arctan(np.double((g_c.imag-g_d.imag)/(g_c.real-g_d.real)))
+    #     phi=-phi1+phi2
+    #     d_s=(1-np.abs(g_d)**2)/(1-np.abs(g_d)*np.cos(phi))
+    #     diam = abs(a[1] - a[0] / a[2])
+    #     qk=1/(d_s/diam-1)
+    #
+    #     sigma2A = recalculation_of_data(data, a, c, d, error=True)
+    #     sigmaQ0 = random_deviation(a, sigma2A, diam, k, Ql)
+    #     Q = Ql * (1 + qk)  # Q-factor = result
+    #     print(f"Q0 = {Q} +- {sigmaQ0}")
+
+    return Q, sigmaQ0, Ql, sigmaQl, a