Frequency analysis of fast changing data.

Rolling FFT

import numpy as np
from collections import deque
import time
import threading
import matplotlib.pyplot as plt


def rollingFFT(s, n, dt):
    fy = np.fft.fft(s)
    # Frequencies associated with each samples
    freqs = np.fft.fftfreq(n, d=dt)

    hz = np.fft.fftshift(freqs)
    ampl = np.fft.fftshift(abs(fy))

    index = hz > 0

    hz = hz[index]
    ampl = ampl[index]

    return hz, ampl


def plotFFT(fig, n, dt, data, title="Frequencies Observed", show_data=True):
    """A thread to run the fft as fast as it can.

    data is a shared dequeue - filled from somewhere else
    """
    if show_data:
        ax = fig.add_subplot(211)
        ax2 = fig.add_subplot(212)
        line2, = ax2.plot(data)
    else:
        ax = fig.add_subplot(111)

    f, a = rollingFFT(data, n, dt)
    line1, = ax.plot(f, a, 'r--')
    ax.set_xlabel("Hz")
    ax.set_xlim([0, f[-1]])
    ax.set_ylim([-500, 10000])
    plt.title(title)

    while True:
        f, a = rollingFFT(data, n, dt)

        line1.set_ydata(a)

        if show_data:
            line2.set_ydata(data)

        plt.draw()


if __name__ == "__main__":

    # simulate generation
    def sim(fig, l=20):

        n = 200
        dt = 0.003

        t = np.arange(start=0, stop=l, step=dt)
        noise = 25*np.random.normal(0, 2, len(t))
        motor_speed = 4000 * np.ones_like(t)
        ripple = 70 * np.sin(2*np.pi*(5*t)*t) + \
                (40 * np.cos(2*np.pi*t*30)) + \
                (80 * (t/10) * np.cos(2*np.pi*t*60))

        signal = noise + motor_speed + ripple

        signal_queue = deque(maxlen=n)
        for i in range(n+1):
            signal_queue.append(signal[i])

        t = threading.Thread(target=plotFFT, args=(fig, n, dt, signal_queue))
        t.start()

        for new_value in signal:
            signal_queue.append(new_value)
            time.sleep(dt)

        t.join()


    fig = plt.figure()
    t = threading.Thread(target=sim, args=(fig,))
    t.start()

    plt.show()
    t.join()