## Tuesday, April 6, 2010

### How to fit a sine wave – An example in Python

If the frequency of a signal is known, the amplitude, phase, and bias on the signal can be estimated using least-squares regression. The key concept that makes this possible is the fact that a sine wave of arbitrary phase can be represented by the sum of a sin wave and a cosine wave.

The regression problem to find the amplitude and phase is an optimization problem. However,  it is not easily solved when using the amplitude and phase directly. This is because the problem is nonconvex; it has multiple minima.  By applying trigonometric identities, an equivalent problem, which is convex, is formed.

Once the regression problem is in this form, the solution is found by forming linear least squares problem. The python function illustrates how to do this. Since python’s function work in radians but most people prefer Hertz and degrees, this script performs those conversions.

from pylab import *
from math import atan2

def fitSine(tList,yList,freq):
'''
freq in Hz
tList in sec
returns
phase in degrees
'''
b = matrix(yList).T
rows = [ [sin(freq*2*pi*t), cos(freq*2*pi*t), 1] for t in tList]
A = matrix(rows)
(w,residuals,rank,sing_vals) = lstsq(A,b)
phase = atan2(w[1,0],w[0,0])*180/pi
amplitude = norm([w[0,0],w[1,0]],2)
bias = w[2,0]
return (phase,amplitude,bias)

if __name__=='__main__':
import random

tList = arange(0.0,1.0,0.001)
tSamples = arange(0.0,1.0,0.05)
random.seed(0.0)
phase = 65
amplitude = 3
bias = -0.3
frequency = 4
yList = amplitude*sin(tList*frequency*2*pi+phase*pi/180.0)+bias
ySamples = amplitude*sin(tSamples*frequency*2*pi+phase*pi/180.0)+bias
yMeasured = [y+random.normalvariate(0,2) for y in ySamples]
#print yList
(phaseEst,amplitudeEst,biasEst) = fitSine(tSamples,yMeasured,frequency)
print ('Phase estimate = %f, Amplitude estimate = %f, Bias estimate = %f'
% (phaseEst,amplitudeEst,biasEst))

yEst = amplitudeEst*sin(tList*frequency*2*pi+phaseEst*pi/180.0)+biasEst

figure(1)
plot(tList,yList,'b')
plot(tSamples,yMeasured,'+r',markersize=12,markeredgewidth=2)
plot(tList,yEst,'-g')
xlabel('seconds')
legend(['True value','Measured values','Estimated value'])
grid(True)
show()

All text is copyright © 2010, Ed Tate, All Rights Reserved.

All software and example codes are subject to the MIT License

Copyright (c) 2010, Ed Tate, Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

#### 4 comments:

1. Awesome! Thanks, this is just what I was looking for. Nice explanation!

2. Thanks! Is there a formal name for this approach or examples in the scientific literature for me to cite?

3. Good. But requires known frequency. What if I don't know the freq?

4. Thanks for this. A minor typo in the latex:
atan2(a,b) should be atan2(b,a)
the code looks correct though.
Again, thanks!