By oversampling, we mean printing on the same row more than once. There are two reasons for oversampling: to increase the horizontal resolution of the printout and to reduce banding.
Oversampling to increase horizontal resolution is necessary because, although the printer might be able to position an ink drop to, for example, 1/1440" horizontally, it may not be able to lay down two such drops 1/1440" apart. If it can print two drops 1/720" apart, 2x oversampling will be necessary to get a 1/1440" horizontal resolution. If it can only print two drops 1/360" apart, 4x oversampling will be necessary for a 1/1440" horizontal resolution. The printer enforces this "drop spacing" by only accepting raster passes with a horizontal resolution matching the spacing with which it can print dots, so we must print passes at different horizontal positions if we are to obtain a higher horizontal resolution. (Another reason it does this may be to reduce the amount of memory needed in the printer.)
Oversampling can also be done to decrease the banding apparent in an image. By splitting a row into two or more sets of dots ("lines") and printing each line on the same row, but with a different nozzle for each line, we can get a smoother print.
To quantify these two kinds of oversampling, we'll introduce two new constants: @math{H} shows how many different horizontal offsets we want to print at (the "horizontal oversampling") while @math{O} shows how many times we want to print each row, over and above the number of times necessary for horizontal oversampling (the "extra oversampling").
It is necessary for all the lines printed by a given pass to have the same horizontal offset, but there need not be any relation between them in terms of extra oversampling. For the moment, however, we will treat all oversampling as potentially requiring this alignment; all lines in one pass must be derived from the original row data in the same way. Thus, we'll assume @math{O=1} for now.
So, how do we do this oversampling? In fact, it can be done easily: advance the paper by a factor of @math{H} less between each pass. We'll define a new variable, @math{A}, to show how much we advance the paper between passes. Previously, we'd have defined @math{A=J}; we now let @math{A=J/H}. This also affects our pass blocks. Printing one pass block used to involve advancing the paper @math{S*J} rows; it now advances the paper @math{S*J/H} rows. We therefore name a group of @math{H} pass blocks a "band". Printing one band involves advancing the paper @math{S*J} rows, as a pass block did before.
To keep our weave pattern working correctly, so that overprinting does not occur within a pass block, we also have to redefine @math{G} as @math{GCD(S,A)}. Here's an example of an oversampled weave pattern:
@math{S=4}, @math{J=10}, @math{H=2}, @math{A=J/H=10/2=5},
@math{G=GCD(4,5)=1},
passesperblock=@math{S}=4,
passespersubblock=@math{S/G}=4/1=4:
0 *---*---*---*---*---*---*---*---*---* 1 *---*---*---*---*---*---*---*---*---* 2 *---*---*---*---*---*---*---*---*---* 3 *---*---*---*---*---*---*---*---*---* 4 *---*---*---*---*---*---*---*---*---* 5 *---*---*---*---*---*---*---*---*---* 6 *---*---*---*---*---*---*---*---*---* 7 *---*---*---*---*---*---*---*---*---* 8 *---*---*---*---*---*---*---*---*---* 9 *---*---*---*---*---*---*---*---* 10 *---*---*---*---*---*---*--- 11 *---*---*---*---*---*-- 12 *---*---*---*---*- 13 *---*---*---* 14 *---*--- 15 *--
Now we have to determine which line is printed by each jet on each pass. If we number each line generated as we split up a row, we can use these numbers. We'll number the lines in our diagram by replacing the `*'s with integers in the range [0...@math{H-1}].
Overprinting occurs once per pass block, so we can simply print pass block 0 with line 0, pass block 1 with line 1, pass block 2 with line 2, etc, wrapping to 0 when we've run out of lines:
0 0--0---0--0---0--0---0--0---0--0 1 0--0---0--0---0--0---0--0---0--0 2 0--0---0--0---0--0---0--0---0--0 3 0--0---0--0---0--0---0--0---0--0 4 1--1---1--1---1--1---1--1---1--1 5 1--1---1--1---1--1---1--1---1--1 6 1--1---1--1---1--1---1--1---1--1 7 1--1---1--1---1--1---1--1---1--1 8 0--0---0--0---0--0---0--0---0--0 9 0--0---0--0---0--0---0--0---0 10 0--0---0--0---0--0---0--- 11 0--0---0--0---0--0-- 12 1--1---1--1---1- 13 1--1---1--1 14 1--1--- 15 1--
@math{S=4}, @math{J=12}, @math{H=2}, @math{A=J/H=12/2=6}, @math{G=GCD(4,6)=2},
passesperblock=@math{S}=4,
passespersubblock=@math{S/G}=4/2=2:
0 0--0---0--0---0--0---0--0---0--0---0--0 1 0--0---0--0---0--0---0--0---0--0---0--0 2 0--0---0--0---0--0---0--0---0--0---0--0 3 0--0---0--0---0--0---0--0---0--0---0--0 4 1--1---1--1---1--1---1--1---1--1---1--1 5 1--1---1--1---1--1---1--1---1--1---1--1 6 1--1---1--1---1--1---1--1---1--1---1 7 1--1---1--1---1--1---1--1---1-- 8 0--0---0--0---0--0---0--0- 9 0--0---0--0---0--0--- 10 0--0---0--0---0 11 0--0---0-- 12 1--1-
But what do we do if @math{J} is not an exact multiple of @math{H}? This is a difficult problem, which I struggled with for quite a few days before giving in and taking the easy (but less elegant) way out. The easy solution is to round @math{J/H} down, then add on the accumulated error at the end of each band.
@math{S=4}, @math{J=11}, @math{H=2} @math{A=floor(J/H)=floor(11/2)=5}, @math{G=GCD(4,5)},
passesperblock=@math{S}=4,
passespersubblock=@math{S/G}=4/1=4
Band 0: 0 0--0---0--0---0--0---0--0---0--0---0 1 0--0---0--0---0--0---0--0---0--0---0 2 0--0---0--0---0--0---0--0---0--0---0 3 0--0---0--0---0--0---0--0---0--0---0 4 1--1---1--1---1--1---1--1---1--1---1 5 1--1---1--1---1--1---1--1---1--1---1 6 1--1---1--1---1--1---1--1---1--1---1 7 1--1---1--1---1--1---1--1---1--1--- Band 1: 8 | 0--0---0--0---0--0---0--0---0- 9 \-----------------------------------------/ 0--0---0--0---0--0---0--0 10 S*J rows 0--0---0--0---0--0--- 11 0--0---0--0---0-- 12 1--1---1--1- 13 1--1---1 14 1---
We can calculate the starting row and subpass number of a given pass in this scheme as follows:
A = floor(J / H)
subblocksperblock = gcd(S, A)
subpassblock = floor((p % S) * subblocksperblock / S)
if subpassblock * 2 < subblocksperblock
subblockoffset = 2*subpassblock
else
subblockoffset = 2*(subblocksperblock-subpassblock)-1
band = floor(P / (S * H))
passinband = P % (S * H)
startingrow = band * S * J + passinband * A + subblockoffset
subpass = passinband / S
So the row number of jet @math{j} of pass @math{p} is
A = floor(J / H)
subblocksperblock = gcd(S, A)
subblockoffset(p)
= 2*subpassblock , if subpassblock * 2 < subblocksperblock
= 2*(subblocksperblock-subpassblock)-1 , otherwise
where
subpassblock = floor((p % S) * subblocksperblock / S)
band(p) = floor(p / (S * H))
passinband(p) = p % (S * H)
row(j, p) = band(p) * S * J + passinband(p) * A + subblockoffset(p) + j * S
row(j, p) = p * J + subblockoffset(p) + j * S
To be continued@enddots{}
Go to the first, previous, next, last section, table of contents.