如何更改 numpy.array 采样率？

Question

Say, I've got a signal which is one dimensional numpy array lasting for one second with sample rate equal to 16 kHz. How can I resample this array for instance to 1024 Hz without loosing information about "peaks" present in this signal as shown below? I add only that intervals between "peaks" are not less, than 40 ms.

UPDATE

I've found an interesting approach based on scipy.signal.resample function, but the problem with this function is that it return absolute value for each y value of the given vector.

Answer 1

If you can use scipy , I suggest scipy.interpolate.interp1d .

However also numpy has a 1D interpolation in numpy.interp , so that might just be what you want.

For what you are asking (not loosing the peaks), that is a bit of an ill posed problem, that can't be solved by a simple and naîve interpolation.

My 2 cents solution: Since you are downsampling, and you know the downsampling ratio, you can just extend the peaks so to not lose them after downsampling, ie so that the peak duration in the downsampled signal is at least one sample, that is, is at least ceil(SR1/SR2) samples before downsampling, where SR1 is the original sample rate, and SR2 is the downsampled rate..

Answer 2

Plain resampling will always loose some information. If the height of the peaks is important, but not so much width, you could just take 1 every 16 samples (given that the peaks are at least 16 samples wide at 16kHz).

If you have few peaks, you could consider a far better algorithm: Just record the time instant and delta, for each change. In your diagram (estimated):

21050 +2000
21060 -2000
24030 -500
24040 +500
26030 +1000
26040 -1000

This will surely occupy much less than sampling. In just 6 values (which could be stored in binary) I compressed your data.