Streaming Compression Math

Question

Using 4:2:2, the data rate and compression are: 4k -- 11.88 Gb/s -- 339.4x HD -- 2.38 Gb/s -- 237.6x

Assuming 4:2:0 for both, Netflix and Stadia use the same compression for HD. To use the same compression for 4K, Stadia would require 50 Mb/s.

Netflix probably streams using pre-compressed video files. They were probably compressed using HEVC.

According to Vimeo, for live streaming they use something called key frames which are fully loaded, and delta frames which are predicted algorithmically. This can be done using HEVC.

If Stadia uses a key frame every 3 seconds for 1080p, then the bandwidth requirement is 9.9 Mb/s (which is close to 10 Mb/s).

And if it uses a key frame every 4 seconds for 4K, then the bandwidth requirement is 37.125 Mb/s (which is closer to 35 Mb/s than to 40 Mb/s). (If it were every 3 seconds, it would have been 49.5 Mb/s - close to 50 Mb/s - 1 TB cap in less than 45 hrs.)

So, for HD, it is predicting 179 frames, and for 4K, it is predicting about 239 frames.

Could they really be predicting this much?

Answer 1

I believe the definition of “predicted” you are using does not match how the term is used in video compression. A “predicted” frame is just a frame that can be reconstructed using information from previous frames, along with some additional data. The frames are not zero sized, but that are smaller than key frames. You can have as many predicted frames in a row as you want. A key frame is usually periodically sent because of error propagation. If there is a transmission error and the video becomes corrupt, a key frame can fix it.

Also keyframe interval and bitrate have a slight correlation, but they are independent variables and not much information can be assumed about one given the other.