sumedian.su.main

su 的源代码库

 SUMEDIAN - MEDIAN filter about a user-defined polygonal curve with	
	   the distance along the curve specified by key header word 	

 sumedian <stdin >stdout xshift= tshift= [optional parameters]		

 Required parameters:							
 xshift=               array of position values as specified by	
                       the `key' parameter				
 tshift=               array of corresponding time values (sec)	
  ... or input via files:						
 nshift=               number of x,t values defining median times	
 xfile=                file containing position values as specified by	
                       the `key' parameter				
 tfile=                file containing corresponding time values (sec)	

 Optional parameters:							
 key=tracl             Key header word specifying trace number 	
                       =offset  use trace offset instead		

 mix=.6,1,1,1,.6       array of weights for mix (weighted moving average)
 median=0              =0  for mix					
                       =1  for median filter				
 nmed=5                odd no. of traces to median filter		
 sign=-1               =-1  for upward shift				
                       =+1  for downward shift				
 subtract=1            =1  subtract filtered data from input		
                       =0  don't subtract				
 verbose=0             =1  echoes information				

 tmpdir= 	 if non-empty, use the value as a directory path	
		 prefix for storing temporary files; else if the	
	         the CWP_TMPDIR environment variable is set use		
	         its value for the path; else use tmpfile()		

 Notes: 								
 ------								
 Median filtering is a process for suppressing a particular moveout on 
 seismic sections. Its advantage over traditional dip filtering is that
 events with arbitrary moveout may be suppressed. Median filtering is	
 commonly used in up/down wavefield separation of VSP data.		

 The process generally consists of 3 steps:				
 1. A copy of the data panel is shifted such that the polygon in x,t	
    specifying moveout is flattened to horizontal. The x,t pairs are 	
    specified either by the vector xshift,tshift or by the values in	
    the datafiles xfile,tfile.	For fractional shift, the shifted data	
    is interpolated.							
 2. Then a mix (weighted moving average) is performed over the shifted	
    panel to emphasize events with the specified moveout and suppress	
    events with other moveouts.					
 3. The panel is then shifted back (and interpolated) to its original	
    moveout, and subtracted from the original data. Thus all events	
    with the user-specified moveout are removed.			

 For VSP data the following modifications are provided:		
 1. The moveout polygon in x,t is usually the first break times for	
    each trace. The parameter sign allows for downward shift in order	
    align upgoing events.						
 2. Alternative to a mix, a median filter can be applied by setting	
    the parameter median=1 and nmed= to the number of traces filtered.	
 3. By setting subtract=0 the filtered panel is only shifted back but	
    not subtracted from the original data.				

 The values of tshift are linearly interpolated for traces falling	
 between given xshift values. The tshift interpolant is extrapolated	
 to the left by the smallest time sample on the trace and to the right	
 by the last value given in the tshift array. 				

 The files tfile and xfile are files of binary (C-style) floats.	

 The number of values defined by mix=val1,val2,... determines the	
 number of traces to be averaged, the values determine the weights.	

 Caveat:								
 The median filter may perform poorly on the edges of a section.	
 Choosing larger beginning and ending mix values may help, but may	
 also introduce additional artifacts.					

 Examples:								



 Credits:

 CWP: John Stockwell, based in part on sumute, sureduce, sumix
 CENPET: Werner M. Heigl - fixed various errors, added VSP functionality

 ideas for improvement:
  *	a versatile median filter needs to do:
	shift traces by fractional amounts -> needs sinc interpolation
	positive and negative shifts similar to SUSTATIC
	make subtraction of filtered events a user choice
	provide a median stack as well as a weighted average stack

 Trace header fields accessed: ns, dt, delrt, key=keyword