Unevenly sampled time series


Hello all,

I’m trying to decide if SciDB is a good fit for seismic/geophysical data. I have collections of sensors sampling at different sampling rates (e.g. 1, 20, 40, 100 Hz) and, of course, starting/stopping at different times. Can anyone tell me if/how SciDB can do the following on these data:

  1. pair-wise cross correlation
  2. FFTs and other transforms
  3. filtering
  4. unaliased downsampling

I’ve looked through all I can find of SciDB’s documentation and presentations, but I haven’t found answers to these questions.

I realize that SciDB is new, and has a basic linear computation engine at this point, but I think much of my work with geophysical time series falls under that category. I just can’t see how that might happen in SciDB yet from looking through the docs.

Thanks for any help!


  1. pair-wise cross correlation

See either github.com/Paradigm4/SciDBR/wik … ix-example for pairwise correlation of vectors with completely specified values (faster).

  1. FFTs and other transforms

SciDB has no FFT capability.

  1. filtering

No native spectral or Kalman filtering or linear convolution capabilities.

  1. unaliased downsampling

SciDB’s native regrid function trivially decimates along coordinate axes independently of any data distribution, so not really. However you could unbias results a few ways manually, either by applying a custom binning variable sensitive to the data distribution and then aggregating along that variable, or maybe by imputing missing data first then using regrid.