I just performed a rather detailed analysis of the overscan pattern that HDNet runs as part of its test patterns every Tuesday morning. This pattern is inaccurate and you will not set your overscan properly if you use its numbers!

My overscan was calibrated with an Accupel HDG-3000 test pattern generator and an overlay grid. I am convinced that my geometry and linearity are very accurate.

So, what does the HDNet pattern show me?

The circle in the middle is exactly 47cm high and 47cm wide. Good.

Anywhere I measure 3 of the squares on the grid pattern, they are 20cm high (or across). Good.

Overscan is just over 4% on the top, 6.5% on the bottom. OK, slightly offset but my grid may have been slightly offset that way, and vertical positioning often varies a bit for different DVD players/STBs.

Overscan on the left is 10%, overscan on the right is just barely under 10%. That can't be right, can it?

This confuses me. The circles are round. All the squares are square. Why is the geometry off. I measured my screen and it is 16/9.04. That shoudn't make it different by more than 1%.

Here is a nice high quality scan from "feldon23" on AVS Forum of the image that you can refer to.

So, where is the problem? Counting the grid they use, it is 27 squares by ~15.2 squares.

They appear to be marking the "numbers" as 1/8 of a square. Um, but look at the bottom ones? They go 14, 12, 10, then a line (so add 2 to the numbers 10 and above as the line is really 10), then 8, 6 , 4... Whoops, they skipped one! Sure enough, the last one before the edge is 4, not 2. But, since the image extends past the bottom square those numbers are pretty close. [I used to suggest subtracting two, but in light of the extra image below the bottom line the numbers from 8 down are OK.]

Now if one looks at the very left and right edges, you can see that the distance between the 2 and the edge is about half the distance between 2 and 4. This leads me to believe you need to subtract 1 from the HDNet left and right edge numbers.

Well, we normally consider overscan in terms of percentage of 100% of the picture. 1% cuts off 1%. This pattern does not represent that!

Horizontally, there are 27 squares. So their numbers actually represent 1/216ths not 1/100ths. So, multiply the left and right numbers by 100/216 (after subtracting 1) to get an actual percentage. My "10" is now 4.2%.

Vertically there are 15.2 squares. So their numbers actually represent 1/122ths, not percentages. Multiply my 4% and 4.5% by 100/122 to get 3.3% and 3.7%.

OK folks, here is the way to get actual overscan numbers from the numbers on the HDNet grid.

Top and Bottom: HDNet * .82 = overscan %
Left and Right: (HDNet - 1) * .46 = overscan %

If you don't want to drag out a calculator and are just trying to get things pretty close, you would do well to divide the horizontal number by 2 (actually 1.8). For most RPTV owners, shooting for around 4%, that would mean 5 on the top and bottom, and 10 on the edges.

The ratio is most likely exactly 1.78:1 or, ta da!, 16:9. The numbers on the HDNet pattern are absolute, representing (scaled of course) how many pixels are chopped off each each edge. When setting overscan we are concerned with what PERCENTAGE of the image is chopped off which we want to be constant around the screen. Since the screen is not square, we have to factor the HDNet pattern's numbers by the screen dimensions of 16:9.

Update: HDNet has responded...

"The scales on the side of the test pattern are simply there to give a scale for comparison. Someday soon, I will have our graphic artist change these over to the proper %. I will e-mail you the image to make sure we got it right.

BTW, we are trying to promote the use of HD video from edge to edge. Safe title area is a left-over from tube deflection from the early age of TV. Hopefully someday there will not be overscan!"

Hopefully we'll see an updated pattern soon!