at Stanford:  5/95
	Contains much unused code, needs cleanup.
	Seems to work.
	Contents of the parameter file wz.10.p indicates filter windowed to speed up
	I better check the result.  Use otube.

at La Jolla:

	setfree.rt	A humongous mess.
	Steep.rs 	formerly did everything (Lomoplan3). Needs cleanup.

	To experiment with symmetric residual,
		various weights, etc, use:   burg23.rt pef2.rt lopef2.rt

	original wz.10 figure required 5 hours of calculation.  Now 10 min.



---------------------------------------------------------------

Connection Machine 3-D lomoplan code with SEP73/jon/moplan test cases.
and also some new filter shapes for coherency and water waves.

Should contemplate the horrible business of saving the filters
for use in the conjugate operation.

HANDY DEBUG:   if( minval(CM_rr) == 0. ) call erexit('no output')

---------------------------------------------------------------
	theory below is a lot of huey.
---------------------------------------------------------------

	Comprehensive New Theory, (not yet implemented)
	------------------------

If geophone groups were separated by 6 meters, then at 4ms sampling,
the diamond operator below would zap water waves perfectly:

		   1
		-1 0 -1
                   1

Now think about 12 meter spacing, so we need two time levels
for each depth level. (At 48 meter spacing we would have 8 time
points per space point).  Examine the filter below, 5x11. (In
practice it might be about 5x41.

		A B C D E
		. F G H a
		. I J K b
		. . L c d
		. . M e f
		. . 1 g h
		. . . i j
		. . . k l
		. . . . m
		. . . . n
		. . . . .

Study the stair-case above carefully.  All the steps have height
of two, EXCEPT THE STEP CONTAINING THE "1" which has a height of 3.
Notice that because the step containing the "1" is of height 3,
each letter in [A-M][a-n] has a mirroring "." on the opposite side of the "1".
This is exactly what a Burg 2-D spectral representation requires.
So these coefficients absorb the 2-D color in the data making a white output.
Now notice that all the lower case letters are Canales-like predictors
and all the upper case letters are needed to predict waves
steeper than water waves.
Notice also that the [A-M] family are symmetrically placed
with respect to the "1", as are the [a-n] family.
What we do is to solve for all the coefficients at the same time,
[A-M] and [a-n], (minimizing power out, taking "." constrained to zero).
After having solved for these coefficients by least squares,
we apply the following filter to the data and display the result.

		. . . . .
		n . . . a
		m . . . b
		l k . c d
		j i . e f
		h g 0 g h
		f e . i j
		d c . k l
		b . . . m
		a . . . n
		. . . . .

Notice this filter is symmetric through the origin,
which now contains a zero instead of a 1.  Thus the output
is the sum of the left prediction and the right prediction
of the coherent waves.  (Incoherent stuff would require
much higher order filters.)
