Understanding the Limits of Dolby Atmos in TrueHD: A Simple Math Exercise

[JediJoker]

The maximum audio bandwidth for Dolby TrueHD, the lossless codec used for Atmos on Blu-ray, is 18Mbps. At the supposedly maximum 35 channels (24.1.10), a discrete encoding would have only 0.514Mbps available per channel vs. the uncompressed 24-bit/192Khz bitrate of 4.608Mbps per channel. That would require an absolutely unthinkable compression ratio of about 9:1, whereas TrueHD's impressive lossless algorithm manages only about 3:1 (compare that to FLAC at 2:1). The bitrate of a single channel at 24-bit/48kHz uncompressed is only 1.152Mbps, which would fit into 0.514Mbps at a comfortable ~2.25:1 compression ratio. Now, Atmos is not discrete, but this simple math exercise still demonstrates how difficult it would be to use higher sampling rates for immersive formats with current technological and standards limitations.

 

[jimfisheye]

The objects are internal. I mean the delivery between the encoded Atmos source and the Atmos decoder codec is internal. However many objects are separated out and delivered to the Atmos decoder to be rendered into the available output system, the final output will be 7.1.4 or 9.1.6. 12 or 16 channels total using that output bandwidth.

The 8 channel fold down is an 8 channel data stream. The mlp compressed difference signals in the metadata are still a tiny data stream. I don't know how much the metadata stream grows if using the full 128 objects though.

Most I've done is maybe 8 objects. The height channels are always treated as objects and need to be sent to the Dolby renderer separately. So you'll always have at least 4. I mix in 7.1.4 but I've separated a couple things out before thinking about 9.1.6 listeners. This will be rare for me because I can't truly verify the mix in 9.1.6 when I do that.

But lets forget about math and go straight for caveman thinking!
You've probably decoded an encoded render, pulled it back into the DAW, and A/B'd against the original master. I didn't hear a difference when I did that. Yeah, I caught the lossy thing! Tried to null it against the original and it does not! (TrueHD up to 8 channels DOES null against the source, FYI.) But it DOES sound the same. Right, so how many objects do you have to start sending through before that collapses and you start hearing a difference when you A/B check it? Anyone done that?

That would be the way to test the theory, right? I think it shouldn't matter because the objects are not being decoded as separate audio tracks and transmitted across HDMI. The decoder is nulling the objects out of the 8 channel bed using the metadata delivered difference signals internally and only sending the rendered mix out.