2
\$\begingroup\$

I currently have this code to generate a discrete signal with random amplitudes, but a fixed frequency:

import matplotlib.pyplot as plt
import numpy as np

# constants
t_len = 0.5
freq = 10.0
dt = 0.001

# set the see for reproducible results
rng = np.random.RandomState(0)

# create the random values to interpolate between
assert dt < freq
white_size = int(t_len * freq)
white_vals = rng.uniform(low=-1, high=1, size=white_size)

# setup the interpolation
white_noise = np.zeros(int(t_len / dt))
assert white_size < white_noise.shape[0]

# how can I avoid using this for-loop?
for w_i in xrange(white_size):
    white_noise[int(w_i/freq/dt):int((w_i+1)/freq/dt)] = white_vals[w_i]

# plot the result for visual verificaition
plt.plot(white_noise)
plt.show()

It generates something like this:

sig_plots

Like it says in the code comments, how do I get rid of this for-loop and vectorize the operation? I feel like I should be using reshape to accomplish this, but I'm not sure how.

\$\endgroup\$

1 Answer 1

3
\$\begingroup\$

I figured out that this assignment could be rephrased as a vectorized multiplication which eliminates the need for a for-loop and costly indexing of individual groups of elements.

import matplotlib.pyplot as plt
import numpy as np
import ipdb

# constants
t_len = 0.5
freq = 10.0
dt = 0.001

# set the see for reproducible results
rng = np.random.RandomState(0)

# create the random values to interpolate between
assert dt < freq
white_size = int(t_len * freq)
white_vals = np.random.uniform(low=-1, high=1, size=white_size)

# do the interpolation
tot_size = int(t_len / dt)
step_size = int(1 / freq / dt)
n_shape = (white_size, step_size)
white_noise = (white_vals[:, None] * np.ones(n_shape)).reshape(tot_size)

assert white_vals.shape[0] < white_noise.shape[0]

# plot the result for visual verificaition
plt.plot(white_noise)
plt.show()
\$\endgroup\$
0

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.