[FFmpeg-devel] [PATCH 5/5] avformat/mxfdec: rework MCA channel layout parsing

[Marton Balint]

On Mon, 29 Nov 2021, Pierre-Anthony Lemieux wrote:

> On Mon, Nov 29, 2021 at 11:20 AM Marton Balint <cus at passwd.hu> wrote:
>>
>>
>>
>> On Sun, 28 Nov 2021, Pierre-Anthony Lemieux wrote:
>>
>>> On Sun, Nov 28, 2021 at 5:00 PM Marton Balint <cus at passwd.hu> wrote:
>>>>
>>>> Setting the channel layout was based on SoundfieldGroupLabelSubDescriptor, but
>>>> this was wrong, because soundfield groups are not necessarily used and a file
>>>> can also contain e.g. a single channel of a multichannel audio.
>>>
>>> For reference, below are MCA scenarios I am aware of:
>>>
>>> - single MXF file containing a single soundfield and all its
>>> constituent channels (specified in the Interoperable Master Format -
>>> IMF)
>>> - single MXF file containing a single soundfield and all its
>>> constituent channels, plus additional auxiliary channels, e.g.
>>> visually-impaired narration (specified in Digital cinema)
>>> - N MXF files, each containing one channel of a single soundfield
>>> (specified in IMF, but I have personally never seen it in practice)
>>>
>>>>
>>>> The new approach is to parse every AudioChannelLabelSubDescriptor and create a
>>>> channel layout based on those. We should also make sure that we only provide a
>>>> channel layout if every channel designation is recognized and is supported. If
>>>> a designation is not supported, or if a channel designation is used more than
>>>> once (e.g. multiple languages) we fall back to the unknown channel layout.
>>>>
>>>> MCA channel ID was also ignored previously, that should have been used to
>>>> determine which AudioChannelLabelSubDescriptor references which audio channel.
>>>> This is also fixed now.
>>>>
>>>> The new parsing code also detects ambigous language or service types, and only
>>>> set these metadata if all channels have the same value.
>>>>
>>>>
>>>> Soundfield groups are now only used for determinig the language of the tracks,
>>>> if it is not provided in the AudioChannelLabelSubDescriptor. Track sub
>>>> descriptors not necessarily contain the references to their soundfield groups,
>>>
>>> ST 377-4 states that "If the Audio Channel is part of a Soundfield
>>> Group then the SoundfieldGroupLinkID shall be present in the
>>> AudioChannelLabelSubDescriptor Set. The SoundfieldGroupLinkID shall be
>>> the MCA Link ID of the SoundfieldGroupLabelSubDescriptor to which this
>>> Audio Channel belongs."
>>>
>>> What do you mean by "Track sub descriptors not necessarily contain the
>>> references to their soundfield groups,"?
>>
>> I mean that the SubDescriptors item of the WavePCMDescriptor not
>> necessarily contains the references to the soundfield group. E.g. if a
>> soundfield group consist of two channels, one channel is in audio track 1,
>> another channel is in audio track 2. As far as I understand only one
>> WavePCMDescriptor can reference the SoundfieldGroup in its SubDescriptors,
>> because it is a strong reference.
>
> As I recall, the intent is for all MCA subdescriptors relevant to a
> track to be contained within the File Descriptor for that track, as
> suggested by Figures 1 and A.2 of the MCA spec (see attached).
>
> For example, if the L channel of a stereo soundfield is in track 1 and
> the R channel of a stereo soundfield is in track 2, then the File
> Descriptor for track 1 would contain one L MCA Audio Channel
> SubDescriptor and one Stereo MCA Soundfield SubDescriptor, while the
> the File Descriptor for track 2 would contain one R MCA Audio Channel
> SubDescriptor and one Stereo MCA Soundfield SubDescriptor. The two MCA
> Soundfield SubDescriptors would have identical MCA Link ID.

I guess duplicating the SoundfieldGroupLabelSubDescriptors as many 
times as it is referenced as a strong reference can work, but having more 
than one SoundfieldGroupLabelSubDescriptor with the same MCA Link ID in 
the same file seems a bit ugly...

>
> The MCA spec states "all MCALabelSubDescriptor instances associated
> with the audio described by a File Package shall be contained within
> the same File Package". This could be construed as allowing a MCA
> Soundfield SubDescriptor in one child descriptor of a
> MultipleDescriptor to be referenced by a MCA Audio Channel
> SubDescriptor from another child descriptor of the same
> MultipleDescriptor. Happy to check with other implementers and report.

The MCA tests of BMXlib applications (raw2bmx, bmxtranswrap) can generate 
such files. Actually raw2bmx can generate both types, either with repeated 
soundfield groups, or non-repeated. See the attached samples and the track 
maps which were provided to raw2bmx.

> In any case, this is somewhat academic at this point unless anyone is
> aware of MCA-enabled MXF files that are not single track and/or that
> have soundfields split across File Packages. [ed.: I am not aware of
> such files.]
>
> It might therefore make sense to focus, for now, on single track MXF
> files that contain complete soundfields, which your patch addresses if
> I am not mistaken.

Yeah, those should work no problem, as long as all the channels in a 
track have different designation, but this is an ffmpeg limitation, and 
not an issue with parsing.

Regards,
Marton
-------------- next part --------------
A non-text attachment was scrubbed...
Name: mxf-mca.zip
Type: application/zip
Size: 10213 bytes
Desc: 
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20211130/b921250f/attachment.zip>