- Posts: 1
- Joined: Tue Jun 10, 2025 11:22 am
- Real Name: Tom Naber
Tue Jun 10, 2025 12:37 pm
While using the HDMI/SDI Bidirectional Micro Converter 12G as an HDR LUT-box (i.e., using HDMI In to HDMI Out only), I've come across some serious issues in the color reproduction.
Whenever you enable any LUT (even if it doesn't perform any remapping) while sending YCbCr 4:2:0/4:2:2/4:4:4, the chroma gets misaligned to the left by half a pixel. Furthermore, toggling any LUT on causes worsened banding in gradients (even 10b gradients) and a vertical red line that's a single pixel wide to appear on the rightmost part of the display. The images below are taken 5 seconds apart, with the only difference being that the one on the right has the LUT checkbox checked, using a 33-point 'empty' LUT that does not alter any of the RGB values:
The chroma misalignment is happening because the micro converter box performs a poor nearest neighbor upsampling on the chroma channels. When using YCbCr 4:2:0/4:2:2, it needs to upsample the CbCr channels to be 1:1 with luma because it uses RGB values internally. However, the same issues occur when outputting YCbCr 4:4:4, which suggests the device first subsamples the CbCr channels before using the same (bad) upsampling. Importantly, these issues do not occur when the source device (in my case, an Android TV device) outputs RGB. Unfortunately, many devices either cannot output RGB in HDR (e.g., Apple TV 4K), also use nearest neighbor for upsampling the chroma channels of 4:2:0 content (Google streamer), or perform a bad YcBcR to RGB color space conversion, wrecking accuracy (Homatics box).
I am a little bit surprised and disappointed that a device marketed as production-quality does not at least use bilinear interpolation for chroma upsampling. I tried to live with it for a couple of weeks, but despite my HDR accuracy being excellent with the LUT (mean dE ITP of ~1.5), artifacts regularly frustrate me, such as color banding being more common in gradients (even with an 'empty' LUT). Another frustrating issue is that colored jackets over a white shirt seem to 'glow' due to the color information on the jacket being shifted to the left, so that the neighboring pixel gets the luma of a white T-shirt but the chroma of the colored jacket. These artifacts are almost always noticeable with any hard transition between two even surfaces of color.
Besides a software fix, one possible solution may be to use a different device to perform the YcBcR to RGB conversion, such as the HDFury Vertex². Unfortunately, it's expensive and I haven't seen anyone try this. Moreover, HDFury support themselves don't know which chroma upsampling algorithm the device uses.
I've already had another user of one of these boxes confirm the issue and I've sent an email to BMD support one and a half months ago, but I don't know whether they're working on a fix. They seemed not to be aware of this issue, so I hope BMD values providing top-tier fidelity to their customers and come with a fix.
I'm sharing this in case anyone else has run into the same issues, but hasn't identified the cause. If you're using a LUT and your source device can output RGB properly, you may be able to circumvent these issues. If anyone has any suggestions for troubleshooting or has tried using a Vertex² for color space conversion, I'd be very happy to hear about it.
Whenever you enable any LUT (even if it doesn't perform any remapping) while sending YCbCr 4:2:0/4:2:2/4:4:4, the chroma gets misaligned to the left by half a pixel. Furthermore, toggling any LUT on causes worsened banding in gradients (even 10b gradients) and a vertical red line that's a single pixel wide to appear on the rightmost part of the display. The images below are taken 5 seconds apart, with the only difference being that the one on the right has the LUT checkbox checked, using a 33-point 'empty' LUT that does not alter any of the RGB values:
The chroma misalignment is happening because the micro converter box performs a poor nearest neighbor upsampling on the chroma channels. When using YCbCr 4:2:0/4:2:2, it needs to upsample the CbCr channels to be 1:1 with luma because it uses RGB values internally. However, the same issues occur when outputting YCbCr 4:4:4, which suggests the device first subsamples the CbCr channels before using the same (bad) upsampling. Importantly, these issues do not occur when the source device (in my case, an Android TV device) outputs RGB. Unfortunately, many devices either cannot output RGB in HDR (e.g., Apple TV 4K), also use nearest neighbor for upsampling the chroma channels of 4:2:0 content (Google streamer), or perform a bad YcBcR to RGB color space conversion, wrecking accuracy (Homatics box).
I am a little bit surprised and disappointed that a device marketed as production-quality does not at least use bilinear interpolation for chroma upsampling. I tried to live with it for a couple of weeks, but despite my HDR accuracy being excellent with the LUT (mean dE ITP of ~1.5), artifacts regularly frustrate me, such as color banding being more common in gradients (even with an 'empty' LUT). Another frustrating issue is that colored jackets over a white shirt seem to 'glow' due to the color information on the jacket being shifted to the left, so that the neighboring pixel gets the luma of a white T-shirt but the chroma of the colored jacket. These artifacts are almost always noticeable with any hard transition between two even surfaces of color.
Besides a software fix, one possible solution may be to use a different device to perform the YcBcR to RGB conversion, such as the HDFury Vertex². Unfortunately, it's expensive and I haven't seen anyone try this. Moreover, HDFury support themselves don't know which chroma upsampling algorithm the device uses.
I've already had another user of one of these boxes confirm the issue and I've sent an email to BMD support one and a half months ago, but I don't know whether they're working on a fix. They seemed not to be aware of this issue, so I hope BMD values providing top-tier fidelity to their customers and come with a fix.
I'm sharing this in case anyone else has run into the same issues, but hasn't identified the cause. If you're using a LUT and your source device can output RGB properly, you may be able to circumvent these issues. If anyone has any suggestions for troubleshooting or has tried using a Vertex² for color space conversion, I'd be very happy to hear about it.