• State Resolved
  • Locked Locked
  • Replies 2 replies
  • Subscribers 99 subscribers
  • Views 511 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

Preview module with 8bit RAW data

Matthew Knutson
Prodigy 90 points

Other Parts Discussed in Thread: DM3730

I have an 8-bit RAW (Bayer GRBG) camera that I'm trying to connect to D0-D7 on the DM3730 Camera ISP parallel camera interface (CPI).  I get an image with the correct colors, but the image is somewhat dark (even when the RAW values from the sensor are saturated to 0xFF).  I'm guessing that the issue is that my 8-bit data is not being routed correctly into the 10-bit Preview module.

What is the correct way to route 8-bit RAW data to the Preview module?  I have had a number of ideas:

1) I tried increasing the brightness/contrast (PRV_CNT_BRT) in the Preview module, but the images are washed out (as would be expected since it only operates on the Y component).

2) I know that I could connect the camera up to D2-D9 (and ground D0-D1), but a board re-spin isn't an option right now.

3) Have the CCDC convert the 8-bit data to 10-bit data for use by the Preview module.  I know that the lane-shifter can right-shift the data to reduce data width (e.g. 12-bit to 10-bit), but it doesn't appear to offer support for left-shifting, so it appears that this is a dead end.

4) Have the Preview module convert the 8-bit data to 10-bit data.  Section 6.4.7.1.5 of the TRM says, "Even if the inverse A-Law block is not enabled, but the input is still 8 bits (PRV_PCR [4] WIDTH), the data is-left shifted by 2 to make it 10-bit data."  However, my impression from other reading is that the WIDTH bit only applies if the data is coming from RAM.  Also, section 6.4.7.1.2 of the TRM says, "When the source of input data is the CCDC, the input data is always 10 bits wide."  So it doesn't seem like this is an option either.

5) Apply a multiplier to my data instead of a left-shift.  For example, it appears that PRV_WB_DGAIN defaults to 0x100 (1x), so it appears that I could set it to 0x3FF and it would essentially perform a 4x gain (equivalent of a left-shift by 2).  This seems to work, but it seems like kind of a "hack."  I would have thought that there would be a better solution since Table 6-35 seems to imply that the "normal" way to connect an 8-bit sensor is to D0-D7.  However, this also appears to be what is described in the YUV Image Capture section here: https://gitorious.org/cm-t3530-camera-kernel/pages/Home

If anyone has any feedback on the above approaches, or other ideas that I have not considered, I would be glad to hear them.  Thanks in advance!

  • 0
    TI__Guru 64820 points

    Although a delayed response, I'm curious what approach have you taken?

  • 0 in reply to BrandonAzbell
    Prodigy 90 points

    I ended up using solution #5 (apply a 4X multiplier using the DGAIN in the Previewer).  I'm using the omap3isp driver in Linux 3.0, and I modified the Previewer driver to accept 8 or 10 bit data (instead of only accepting 10-bit data), and set the DGAIN to 0x100 or 0x3FF accordingly.  Of course, I had to disable the option to adjust the DGAIN via IOCTL when 8-bit data is used.  It seems to be working fine, but still feels kind of like a "hack" to work around a hardware limitation.  This change is scheduled to be included in Logic PD's upcoming DM3730 BSP release.