a7x, a9 PDAF stripe noise technical analysis, part deux

Started 9 months ago | Discussions thread
OP JimKasson Forum Pro • Posts: 21,621
Re: Regarding that OLPF -

Antisthenes wrote:

JimKasson wrote:

Tons o Glass 0 Class wrote:

Tinfoil hat conspiracy theory - Sony implemented a simulated/electronic horizontal AA filter into the sensor row reading architecture, and it is partly to blame for the line artifacts.

All the a7x OLPFs that I know anything about work only in one direction.

Such an OLPF structure would be quite unusual, and your post thus prompted me to conduct an experiment myself (^^;

If the OLPF works only in one direction, then, the aliasing / moiré in the perpendicular direction must be (visibly ?) stronger.

To test that hypothesis, we can simply use a test chart with vertical and horizontal stripes.

I used the screen of a laptop computer (a MacBook Pro Retina 15.4", with a LCD resolution of 2880x1800 pixels) to display a test chart I quickly put together, containing black/white stripe patterns of various pixel widths.

One convenient aspect of glossy LCD laptop displays is that they act a bit like a mirror, allowing one to accurately align the lens perpendicularly to the screen, just by centering in the viewfinder the camera with its own reflection on the glossy laptop's display.

The test chart, created using a simple C program, is divided into 12 zones of roughly similar size.
The 12 zones contain vertical black and white stripes of the following thickness:

10 pixels / 8 pixels / 9 pixels / 11 pixels
6 pixels / 4 pixels / 5 pixels / 7 pixels
14 pixels / 12 pixels / 13 pixels / 16 pixels

A copy of that 12-zone chart, rotated 90 degrees, and thus consisting entirely of horizontal black and white stripes, was also created.

The aggregated vertical / horizontal test chart 1 looks like this:

This first test chart was photographed using a Sony A7ii and a Sony/Zeiss FE Planar 50/1.4 lens, stopped down to F/4.0. The center resolution of that lens is said to exceed even the Otus 55mm's.

The capture format was uncompressed RAW, developed with ACR's default parameters. As an aside, I suspect that Sony's compressed RAW format might generate some noise in the horizontal direction at the contrasty transitions one encounters e.g. with slanted edge / razor edge measurements, and this anisotropic noise may, in turn, introduce some unforeseen variation e.g. in the Imatest MTF measurements conducted using orthogonal slanted edges.

The distance between the camera and the MacbookPro was adjusted to about 5 meters, so that a 5-pixel wide stripe, displayed on the MBP's LCD, would map almost exactly to the width of one pixel — i.e. 6µ — on the A7ii's sensor.

Note the blue/yellow color moirés triggered by that nearly exact map to the sensor's Nyquist frequency.

Photographed chart 1, MacBook slanted towards the left

Photographed chart 1, MacBook slanted towards the right

The amplitude and frequencies of the aliased signals look substantially identical in the four charts, regardless of their slant direction, and of whether they consist of vertical or horizontal lines.

This, IMHO, indicates that the OLPF has equal strength in two perpendicular directions.

A second test chart, containing only two large patches of horizontal and vertical stripes that are all 5-pixel wide, has also been created.

At 5m and with a 50mm lens, this test chart projects white and black lines that are almost exactly 6µ-wide on the sensor — i.e. at the A7ii sensor's Nyquist frequency.

Test chart 2 looks like this:

Photographed chart 2, MacBook slanted towards the left

Photographed chart 2, MacBook slanted towards the right

Again, the amplitude and frequencies of the aliased signal look substantially identical on the four charts, regardless of their slant direction.

We can thus see that, if an OLPF is present,

  • its strength looks substantially the same in two perpendicular (presumably horizontal and vertical) directions
  • it's a quite weak OLPF, because obvious aliasing / color moiré is still present at the A7ii's Nyquist frequency.

It’ll be a few days before I can do a test aimed at measuring this with the a9, which is the only Sony camera I have with an AA filter, but in the meantime, take a look at the a7 shots here and nearby


I’ll also ask Jack Hogan about this. He’s the expert.


-- hide signature --
 JimKasson's gear list:JimKasson's gear list
Nikon D5 Fujifilm GFX 50S Sony a9 Sony a7R III Sony a7 III +5 more
Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow