The Band Played On

The number one question in my In Box seems to be: does the Z7 have banding in its images?

This is a result of a report on Bill Claff's Web site, where he does sensor heat maps (here's the the Z7/D850 comparison).

bythom heatmap

Visually, you should see that the Z7 example (left) has some clear horizontally-oriented information in it, and is not the much more random noise that the D850 example (right) has. This was picked up by dpreview and articulated in their first impressions

The most likely cause of this very low-level artifact is almost certainly due to data offload from the sensor. 

But you want to know if it's a problem or not with images. 

My answer would be not.

This is a little different than the problem with the D7100's sensor. On the D7100 you could fairly easily trigger visibility to banding in out-of-camera JPEGs by simply shooting at a high ISO value with Active D-Lighting enabled on Extra High. The combination of Active D-Lighting's underexposure (to preserve highlights) and shadow boost (to reveal shadow detail) often triggered some visibility to the underlying sensor issue on the D7100. 

Not so with the Z7. I've done a number of experiments trying to promote any underlying banding into visibility for anything approaching "normal" photography, and I'll just say this: if you manage to do it, you've probably done something else very wrong. I even ran post processing D-Lighting on an Active D-Lighting image to see if I could force visibility. Nope. 

The one case I and others can trigger banding is with the electronic shutter when there's a huge exposure differential—generally blown out highlights—in the frame and you perform a huge push of the image. Jim McKasson has an excellent description of that test.

One thing that keeps getting mentioned is that because of this "banding" the Z7 has less dynamic range than the D850.

Dynamic range should be defined as the range between the acceptable noise floor and highlight saturation. And actually, it should even be refined from that, as there is a noticeable shoulder to highlight information in recent sensors. But my definition, and the one I promote in my books, is the usable data set (range) the sensor can provide from which will result in black to white in my image output. 

Take a look at both heat sensor maps you see above. Those noise patterns are deep in black. By "deep in black" I mean they are not generally visible because they're recorded as black (a DN with a center of ~1000 for output off the Z7 sensor). You'd have to be promoting black to a middle gray or higher to see those patterns (and then, what would your blacks be in your image? You'd probably have none). dpreview used a six stop push (not sure how they did that with ACR; and note that their examples aren't correctly matched in output). 

They also write "if you're trying to brighten the foreground of a sunset image" as one of their examples. Again, if you're pushing the few of the lower bits recording data as far as they suggest, you're going to have problems with most cameras, as you don't have enough bit differentiation to truly hold detail. Even if you don't get banding, you get what I call a muddy presentation of "detail": mush (though today's mush is far better than early DSLR mush, as the linearity of the data is better). Your exposure needed to be different in the first place, or you needed to use HDR techniques if your scene truly is that much broader than the camera's abilities. 

Thus, can I trigger visibility of banding? Sure. By doing something I normally wouldn't do to an image. Or by missing exposure by so much that all my image data would be in only four or five bits to start with. Moreover, this seems to be a silent shutter issue—all electronic shutter—so there's a clear workaround.

text and images © 2020 Thom Hogan
portions Copyright 1999-2019 Thom Hogan-- All Rights Reserved
Follow us on Twitter: @bythom, hashtags #bythom, #sansmirror