This tool calculates the estimated noise amplitude spectral density of a time series.
From the white noise part the rms value of a desired bandwidth can be calculated: rms = value * bandwidth^0.5
E.g. it is used in electronic component datasheets to show 1/f (1/f^n, n:real) corner and white noise part (flat) of voltage, but same can be applied to current noise.
Based on Welch's method.
Parameters:
-----------
ts_values : array_like
Time series of the signal (amplitude)
sample_frequency : int/float
The sample frequency in Hz (SPS - Samples per second)
nsd_bins : int, optional
number of NSD bins (points) to be calculated
needs to be lower than the ts_values count
default: 1/4 count of ts_values
window_function : function(length: int), optional
returns: [int, array_like]
first index is the optimum overlapping in samples for the window
second index are the values of the window function
default: window_HFT90D
crop : slice, optional
default: the last and the first 3 NSD values are dropped as they are not reliable
Returns
-------
out : [array, array]
The frequencies (index 0) and the corresponding NSD values (index 1)
If the acquisition time of the input signal (aperture) is lower than the time between samples of the output (ts_values), than the NSD is incorrect (e.g. AZ on ADCs/DMMs)!
Nsd using acquisition time (aperture) as sample_frequency only shows correct white noise part.
Nsd using time between samples as sample_frequency only shows correct 1/f (1/f^n, n:real) part, the white noise part is too high, being (Tsps/Tacq)^0.5, e.g. for AZ: (40ms/20ms)^0.5, so ~1.4
Neither will give the correct 1/f (1/f^n, n:real) noise corner!
nsd_example generates white & brownian noise with a corner frequency of 0.1Hz/1nV:
nsd_example_sps shows how a difference between acquisition time and time between samples effects the NSD:
%20noise%20-%20corner%200.1Hz%20x%201nV.png?raw=true "nsd_example_sps")
nsd_example_csv imports data from csv:
nsd_example_smooth shows different filters:
