Exposure and line_length settings for IMX-678

[alex466]

The doc for line_length is as follows:

Pixel line width horizontal timing size for the sensor mode. Used to calibrate the features in the camera stack. The value must be greater than or equal to active_w.

Unfortunately, the IMX-678 software reference datasheet defines frame line length differently (register is VMAX in MASTER mode) and that value is significantly less than active_w for 4k@30Hz. Is that OK?

Next, the IMX678 has a calculated integration time based on the shutter sweep speed. I have made changes to both the device tree and driver to adhere to formula outlined in the spec.

However, when I am testing it, I am seeing weird behavior between gst/nvarguscamerasrc and the driver. Namely, the value passed to my imx678_set_exposure function:

GST_ARGUS: 3856 x 2180 FR = 29.999999 fps Duration = 33333334 ; Analog Gain range min 0.000000, max 72.000000; Exposure Range min 21459000, max 16073194000;

No matter what values I put in for exposuretimerange with aelock=True I always see the value passed into the driver as 21459.

Device tree settings are as follows:

max_exp_time "14612"
min_exp_time "6"
default_exp_time "6"
line_length "2250"
step_exp_time "1"
exposure_factor "1"

For example, if I pass exposuretimerange 20000 20000 aelock=True within my record pipeline, my driver sees for an exposure value 3333 when I would expect 20 (I assume it converts ns to us) and then the correct value for the shutter sweep speed is wrong. If put in 14612000 I get a value passed to my function of 13507.

This is on a Jetson Nanon (JP 32.5.1) and one IMX-678 camera.

I noticed that “25” seems to be the floor.

EDIT: Another question I have is the minimum and maximum course integration time is in line units right, not microseconds? (same as version 1.0)

[alex466]

I think this maybe a precision issue with my exposure_factor being 1. Can you confirm?

Also, can you answer the question regarding `minimum/maximum course integration time" - I assume even in version 2 that is in lines not microseconds?

[JerryChang]

hello alex466,

@alex466

max_exp_time "14612"
min_exp_time "6"

these should be in micro seconds (us)
and… please try setting exposure_factor = "1000000"; which as same as reference driver.

@alex466

that is in lines not microseconds?

correct, coarse integration time is the exposure intervals. the unit in lines.

[alex466]

Thank you! That worked!

Follow-up, does line_length have to be greater than active_w because in the spec sheet line_length is not (unless NVIDIA’s definition and the sensor’s manufacturer are different).

As you can see above, in fact, it’s static for all supported frame settings.

[alex466]

It also seems that the lowest exposure time is 25us. Is that hard coded somewhere?

[jdiegodelgado]

Hi alex466,

Here you can check more information about the line_length:
https://docs.nvidia.com/jetson/archives/r36.2/DeveloperGuide/SD/CameraDevelopment/SensorSoftwareDriverProgramming.html#device-properties
line_length

Pixel line width horizontal timing size for the sensor mode. Used to calibrate the features in the camera stack.
The value must be greater than or equal to active_w

And about the exposure. You can change it in your device-tree and you have to create an exposure control in your driver implementation.

Best,

Best,

Juan Diego Delgado
Embedded SW Engineer at RidgeRun
Contact us: support@ridgerun.com
Developers wiki: https://developer.ridgerun.com
Website: www.ridgerun.com

[alex466]

Right, I see that. But the software reference manual clearly shows the pixel line width < active_w for the sensor in all supported modes (see “Operating Modes”) - unless somehow I am not reading it correctly.

Exposure is working now as before it was way off (the numbers being sent to the register made absolutely no sense).

In fact, 1 frame line length or VMAX .

VMAX is defined as:

Sets the number of lines per frame

That I believe is what NVIDIA calls line_length. @JerryChang, is that correct?

I think the doc is wrong. I think it means line_length >= active_h.

[JerryChang]

hello alex466,

there’re line length and frame height settings from sensor side usually.
line stands for horizontal direction (active_w), and frame for vertical direction (active_h).

[alex466]

Ok. Then that’s HWIDTH which basically the same as active_w. Things appear to be working and we are now passing the conformance tests which is good news (we were failing before).