

In [1]:

    
%matplotlib inline

import numpy as np
import matplotlib.pyplot as plt
from scipy.signal import hilbert
import scipy.io.wavfile as wavfile

# Larger figure size
fig_size = [14, 8]
plt.rcParams['figure.figsize'] = fig_size

Generate a WAV IQ file at 48ksps containing pulses with 10Hz repetition rate. The duration of each pulse is 10ms.



In [2]:

    
samp_rate = 48000
pulse_on = samp_rate//100
pulse_off = samp_rate//10 - pulse_on
duration = 600 # seconds
n_pulses = duration * 10
amplitude = 0.01

pulse = np.array([1]*pulse_on + [0]*pulse_off, dtype='float32')
i = amplitude*np.tile(pulse, n_pulses)

iq = np.zeros((len(i),2), dtype='float32')
iq[:,0] = i

wavfile.write('/home/daniel/pulses.wav', rate=samp_rate, data=iq)
del iq
del i

Function to stack up the pulses and plot.



In [3]:

    
def plot_audio(file):
    audio = np.memmap(file, offset=0x28, dtype=np.int16)
    #audio = np.abs(hilbert(audio)) # improves display, but takes computation time
    pulse_len = 4800
    lines = len(audio)//pulse_len
    plt.imshow(np.abs(audio)[:lines*pulse_len].reshape(lines, pulse_len), cmap='viridis', vmin = 0, vmax=1e4)
    del audio

We open the WAV IQ file in Linrad using adwav. The output soundcard is a snd-aloop. The pulses are tuned in SSB mode and the audio output is recorded at 48ksps in Linrad using 'W'.

The pulses are stacked and plotted. They are slanted, which means that the resampling rate is constant but not correct (not 1:1 from input to output). The results are the same regardless of whether ntpd is running.



In [4]:

    
plot_audio('/tmp/output-file-input.wav')

Now we set up the audio input in Linrad to use a snd-aloop device. The WAV IQ file is played into Linrad through snd-aloop using aplay -D plughw:1,0 ~/pulses.wav The results seem to depend on whether ntpd is running, but now we always see a line with changing slope, meaning that the resampling rate is being adjusted.

This is with ntpd not running.



In [5]:

    
plot_audio('/tmp/output-alsa-ntpd-off.wav')

This is with ntpd running. The resampling rate variations sometimes seem much greater.



In [6]:

    
plot_audio('/tmp/output-alsa-ntpd-on.wav')