Coherence is a property of information-carrying signals. There are two types of coherence: Fully-coherent and differentially-coherent.

Fully-coherent signals are signals where the information mapped directly onto a property of the carrier signal. For instance, amplitude modulation is almost always fully-coherent; a high amplitude corresponds to the bit '1', a low amplitude corresponds to a bit '0' (or vice-versa).

Differentially-coherent signals are signals where the information is encoded onto changes in the carrier signal. Phase modulation more often uses differential coherence, because two systems with different clocks may not be able to easily agree (long-term) on what a reference phase is, but either or both should be able to recognise a change in phase when it happens.

For example, imagine there are two states for a transmitter to be in: A and B. Fully-coherent transmission of data might map the value '0' to state A, and the value '1' to the state B (of course, it can be done the other way: B for '0' and A for '1'). Differentially-coherent transmission of data might leave the transmitter in the same state whenever there's a '0' in the data stream, and flip state when there's a '1' (this can also be done the other way, but it's nicer in some ways when '1' is associated with change and '0' with no change). An example follows.

Fully-coherent example:
000011110101010011
AAAABBBBABABABAABB

Differentially-coherent example (starting in state A):
000011110101010011
AAAABABAABBAABBBAB

I've seen another nifty instance of differential coherence where the modulated signal had a period of three times a base clock. For a '0' bit, the phase of the transmitted signal was retarded by one-third (i.e. one tick of the base clock), for a '1' bit, the phase was advanced by one-third. This had the extra advantage of being self-clocking; the transmitter never stays in the same state for more than a bit period, so drift between the transmitter's clock and the receiver's clock never becomes a problem.

The relationship between fully-coherent signals and differentially-coherent signals is the same as the relationship between binary numbers and gray code. Binary data transmitted with differentially-coherent (change-on-1) frequency modulation, for example, is identical to the same data transmitted in gray code with fully-coherent frequency modulation.

Coherence is not -
I cannot always say -
sometimes I can't express...

this jumble of words
tumbled in a tumbril
waiting to be executed
won't walk up the steps
won't set themselves down to be
guillotined into
neat sentences.

It's not that I don't -
it's just -
they're meaningful, really
as full of meaning
as veins are full of blood
but
it takes a cut to spill them
and flesh shrinks
from the blade.

If you asked a question
like a hypodermic needle
I might haemorrhage
words enough
to transfuse a thousand
awkward silences;

yet, until then
coherence is not.

I'm sorry.

Coherence in Computer Graphics

Look around an electronics store and you will find dozens of extremely powerful 3d-graphics cards that will allow your computer to play games rapidly approaching photorealism at frame rates much higher than motion pictures. For less than $100 your computer can do a real-time animation that would have taken weeks to render just two decades ago. If you do the math, graphical processing power has increased well ahead of the curve of general computer processors. How is this possible?

The answer is simply Coherence, a word bandied about by computer graphics educators everywhere. Essentially almost anything you would like to render with a computer will be made up of some set of mostly-continuous data. Consider the fact that with 256 possible colors an 800x600 pixel screen can display 256^480,000 possible pictures. The vast majority of those would simply look like noise, but the ones that look interesting would all have one thing in common: continuous areas of related color. The result is the calculating the value of a pixel in a computer-generated image can often be simplified by using the value of a previously computed pixel.

In modern consumer hardware the assumption is that the most computationally expensive graphics to render are ones based on 3d models and lighting. Then they use the coherence inherent in these models to specifically design hardware to solve a handful of optimized equations. Generally these ideas are tested in software then added into hardware for the next generation of graphics cards.

The advanced algorithms used today are so optimized as to require months or years of study to fully understand. Simpler algorithms from decades ago are a much better illustration of the principle of coherence.

Examples:

Bresenham's Algorithm

Bresenham's Algorithm draws a straight line on a 2d raster. This is one of the most basic algorithms in computer graphics, yet also one of the most optimized. The naive method of drawing a line might be to use the parametric representation and the location of each pixel one by one.

y ^
  |  OOO
  |     OOO
  |        OOO
  |           OOO
  |              OOO
  +--------------------->
                       x

Bresenham's Algorithm first picks the x or y axis so that the slope is always between 1 and -1, then takes advantage of the fact that each pixel x_n will be equal to x_n-1+1, and that the y value will either vary by 1 pixel or not at all depending on the slope.

So this algorithm takes advantage of the fact that each pixel in a line is very-closely related (ie. coherent with) the previous pixel. The result is an order-of-magnitude increase in performance without any special hardware.

Z-Buffer Algorithm

The Z-Buffer Algorithm is a method of hidden-surface removal from 3d scenes. It has the advantage of being able to work with polygons regardless of what order they are passed in, and handling intersecting gracefully without weird holes or artifacts being generated. On the downside it takes a lot of memory, and has to be done as the final step in the rendering process after a lot of unnecessary work may already have been done.

The Z-Buffer Algorithm works with 3d coordinates, the x and y being screen coordinates, and the z being depth. For each pixel on the screen, the algorithm tracks a last-drawn depth, and for each pixel of a polygon to render, it calculates whether that pixel is shallower than the last-drawn depth for the pixel. If so then it is rendered, otherwise it is not.

If only triangles are passed to the algorithm, then it turns out there is an extremely efficient way to calculate subsequent pixel depths along the lines of Bresenham's algorithm.

Co*her"ence (?), Co*her"en*cy (?), n. [L. cohaerentia: cf. F. coh'erence.]

1.

A sticking or cleaving together; union of parts of the same body; cohesion.

2.

Connection or dependence, proceeding from the subordination of the parts of a thing to one principle or purpose, as in the parts of a discourse, or of a system of philosophy; consecutiveness.

Coherence of discourse, and a direct tendency of all the parts of it to the argument in hand, are most eminently to be found in him. Locke.

 

© Webster 1913.