2

I have only 1 pulse as input and I use qubic-spline to reshape the pulse. The user plot manually the signal inputs from a flow-image and then the signal is imported to Matlab in a ASCII form. That means that the sampling period is not uniform. Then I have a pressure pulse that are sampled at 128 Hz.

If I use same samplig period (128Hz) using qubic spline on the flow signal. If I increase the sampling to 4-8 times what will it effects of the FFT analysis? I can see in Matlab that the amplitude changes at different up-sampling rate.

If I multiply the input signal (pulse train) e.g. 2-6 times it will also effect the FFT analysis. The bins are the same at higher pulse train but not at just one single pulse. One pulse or many should not impact the FFT analysis?

Right know I have no filter on the signal (before and after resampling)

user3699
  • 21
  • 2
  • 2
    @user3699, your question is not completly clear. Most of the issues that come about with up-sampling are how you actually do the upsampling. zero-order hold and an ideal lowpass filter will take care of the job well. Implementing that low pass filter is how you break things. – Kortuk Apr 02 '11 at 13:12

1 Answers1

1

why do you want to upsample? upsampling comes at a cost in the digital domain, and can save expense in the analogue domain.

For example if you have a signal with content that you are interested in the range of 0-3 kHz you can sample using inexpensive a/d at 64 ksps using a one pole passive lower pass anti alias filter. Once in the digital domain you can pass it through a IIR/FIR multipole low pass and perform decimation (drop seven out of eight samples) so you have 8 KS/s with no aliasing issue and ALL the signal information with no spectral overlap.

On the other hand up sampling a 8 KS/s to 64 KS/s signal will not give you any benefit. It will reduce the resolution of information you are interested within the 0-3 kHz range, increase memory requirements, and increase processor usage. Of course you can define smaller frequency bins, but this comes at a computational and memory cost.

stevenvh
  • 145,832
  • 21
  • 457
  • 668
smashtastic
  • 3,336
  • 22
  • 33
  • If one upsamples an 8KS/s signal to 64KS/s, in such a way as to minimally affect signals below 4KHz but maximally attenuate signals in the range 4KHz-32KHz, one can then pass the resulting signal through a low-pass filter whose pass band extends to 4KHz and whose stop band starts at 32KHz. That may be a lot easier than trying to design a purely-analog filter whose pass band extends to 3.5KHz and whose stop band starts at 4.5KHz, while offering better performance in the 3.5-4.5KHz region than would the analog design. – supercat May 10 '11 at 16:02