Sony a9 eats stars

This is the fourth in a series of posts on the Sony a9. The series starts here.

I’ve been asked to look into the digital lowpass filtering in the Sony a9 as a function of exposure time, to see it it has the same “star-eater” filtering beginning as 4 second exposures that was introduced to the a7RII with firmware 3.30.

First, I lo0ked at read noise in general as it relates to exposure time, with electronic shutter on, at ISO 1000, and no lens attached:

You can see a suspicious drop right at the transition from 3.2 to 4 seconds. By the way, the gap on the left is because the a9 has no exposure time settings between 1/32000 second and 1/16000 second.

Let’s take a look in the spatial frequency domain at the 4 second exposure to see if we observe the low-pass behavior that is a telltale sign of star munching:

Yes, indeed. The a9 eats stars. But what’s going on with the ripples in the horizontal direction? Let’s look at 3.2 seconds: By the way, f is the frequency, fs is the sampling frequency, so f/fs equal to a half is the Nyquist frequency.

Yes. Even though the vertical direction shows only a small droop with increasing frequency, 

At 1 second:

At 1/30 second:

At 1/125 second:

At 1/1000 second:

Looks like the a9 does that all the time, at least in electronic shutter mode. I’ll have to test the mechanical shutter.

The noise is low, so should the patterning indicated by the frequency response plots. What’s it look like? here’s a tight crop of the red channel (which was used for the above graphs), show at about 200% and brutally treated with curves in Photoshop:

There is some patterning visible, but it’s not particularly objectionable to my eyes.