Computation flow and speed

Computation is begun by a change in the parameter settings. The forward wavelet transform was computed at startup time. The appropriate coefficient filtering is applied, then the inverse wavelet transform computed. Finally the data is converted into an image and displayed.

The wavelet transform performed in floating point. I tried byte integers to avoid image conversion, but the precision is too low. Longer integers should work alright and be faster on workstations with slow floating point.

The overall speed is configuration dependent. The number-crunching can be done on a fast computer server and displayed on a workstation or entirely on a workstation. (I use a fast workstation for development and computing and a slow workstation in my office for display.) Second times for numerical and display are printed at the base of the control window and image window respectively. The transform and bandpass is less than a second on the Convex supercomputer but the display of the image on a workstation consumes a couple seconds. The reverse is true on a SPARC.