TDA4VM: Can RYYCY data tuning be optimized？

Hi Wei,

Wei Wang said:

Due to the limitation of TDA4 CCM within ± 1024,

Where do you get this [-1024, 1023] limitation and what is max value of your floating point coefficients?

CCM coefficients should be in 12-bit format [-2048, 2047], i.e., [-8.0, 8.0), if I remember correctly.

Wei Wang said:

And the image is affected by color shading, resulting in a greenish background.

What are the sensor companding knee points and pedestal of your shared raw image?

The "result.bmp" does not have consistent gray color and I suppose you would need to get that right before working on color.

Hi Gang,

The limitation of [-1024, 1023] is in manual entry of color correction plugin in DCC software. And where can I find the limitation requirements for the CCM coefficient of [-2048, 2047]?

The companding knee points of the sensor are as follows.

0    0
328    1024
741    4096
1261    16384
1818    65536
2267    196608
2551    393216
2836    786432
3046    1310720
3239    2097152
3470    3670016
3573    4718592
3691    6291456
3810    8388608
3977    12582912
4095    16777215
65535    16777215

We found that the original RAW image had more uniform colors and did not have color shading that caused inconsistent gray blocks, but the YUV image after ISP pipelining showed obvious color shading on the upper and lower edges. The original RAW is shown in the first image below, and the second image is the YUV image.

Which module do you think is causing this problem?

Wei Wang said:

The limitation of [-1024, 1023] is in manual entry of color correction plugin in DCC software. And where can I find the limitation requirements for the CCM coefficient of [-2048, 2047]?

Do you mean tuning tool output will limit the parameters to S11 (signed 11-bit)?

I believe the h/w registers are in S12Q8 as you can see in the TRM "Color Conversion (CCM-1)" section.

Wei Wang said:

We found that the original RAW image had more uniform colors and did not have color shading that caused inconsistent gray blocks,

Could you please check if the knee points are correct?
I don't see uniform gray after decompanding.
Darker grays are more and more greenish.

What is the sensor pedestal value we need to remove after decompanding?
I suppose this sensor adds pedestal before companding.