NRT Encoding claims to save 40% bandwidth retain the same quality of h264 broadcasts?

[kalehrl]

I've come across this article in which a company claims to have developed a solution to reduce bandwidth consumption by up to 40% whilst retaining the same quality of h264 broadcast. Does anyone have any idea how this is done?

http://svgeurope.org/blog/headlines/ntt-data-provides-encoding-solution-to-sky-italia/

[LoRd_MuldeR]

Unless they provide a sample that shows 40% bandwidth reduction at the same quality, compared to other top-notch H.264 encoders (like x264), they can claim a lot ;)

Also this seems to be targeted explicitly to VoD (Video on Demand) services. Furthermore, it seems that the 40% reduction is achieved for transmitting various encodes of the same movie for different devices...

The basic idea is very simple: in fact, NTT Data had in the past developed and provided to Sky Video On Demand a system of encoding/transcoding offline input files supplied on different formats, generating an encoded version in various forms for various devices [...] The system core is the same, based on AVC H264, with two different codes that share a large part of the initial encoding process, then depending on the device that requests them there are slight differences in the end-result.

[kalehrl]

No, in the past it was aimed at VoD and some other services but now it is also used on some HD channels transmitted via satellite in their DTH bouquet on the Hotbird satellite.
The main part of that article seems to be this:
The new encoding solution, meanwhile, us able to handle a continuous channel entry in linear mode and create an encoded version in output.
The linear flow input is segmented into small segments, each encoded in single mode, while a final component of the software makes a ‘merge’ of these segments in the correct way, then manages the output signal.

[benwaggoner]

No, in the past it was aimed at VoD and some other services but now it is also used on some HD channels transmitted via satellite in their DTH bouquet on the Hotbird satellite.
The main part of that article seems to be this:
That sounds like they're doing VBR or something in the end, reducing bitrate in easier to encode segments.

Getting a 40% ABR improvement by using CRF instead of CBR wouldn't be remarkable.

[kieranrk]

The basis of this system of linear encoding by NTT Data uses very advanced techniques for encoding in H264, which also integrates open source libraries mixed with advanced techniques of adaptive encoding, rather than psycho-visual techniques.

Not even sure what to say about that

[Ghitulescu]

How is quality measured? ;)

[Blue_MiSfit]

Perhaps some compressed domain transcoding? I looked at a company doing things like this. Sadly, it broke HRD which is a big no-no for STBs.

[benwaggoner]

Not even sure what to say about that
Yeah, encoding is pretty much all psycho-visual techniques.


Gamma 2.2 is a psychovisual optimization
8-bit luma range is a psychovisual optimization
Default quant tables are nothing but a psychovisual optimization
DCT-like transforms are a psychovisual optimization
Rec. 709 is a psychovisual optimization


Just because something doesn't show up in PSNR doesn't mean it's not psychovisual. Tons of tuning for the human visual system happened before the source frame even gets provided to the codec!

[Ghitulescu]

IIRC MP3 also was claimed to preserve quality, even at 128kbps :)