none
[MS-RDPRFX] Dequantization before InverseDWT RRS feed

  • Question

  • Hi all,

        I have a question in dequantization phase in RemoteFX decoding.

        In my decoded Cr/Cb data, after reconstructed these 4096 coefficients,
        I found that the first coefficient is obviously different then others in LL3 sub-band.

        I think that this is strange, because it also happened in tiles of single color.
        For example, first 8 coefficients in a single color tile should be [64 64 64 64 64 64 64 64], but I usually got [64 -1 0 -2 1 0 0 0].

        My question is, is there anything i missed in the dequantization phase?
        Or I have to check my entropy codec again? :P

        Thank you.

    Friday, February 11, 2011 10:48 AM

Answers

  • We believe that that your entropy decoding and dequantization is working fine.

     

    At [MS-RDPRFX] 3.1.8.1.6 Linearization, as part of Figure 10: “Sub-band traversal order” there is the comment ““The coefficients from LL3 also undergo differential encoding. Except for the first coefficient, every raster-scanned LL3 coefficient is subtracted from its previous neighbor.”

     

    After quantization, we do differential encoding for LL3 band, where by the first coefficient is encoded as is and the rest are subtracted from their previous value.  So [64 64 64 64 64 64 64 64] is encoded as [64 0 0 0 0 0 0 0].  [64 32 42 54 50 60 40 70] would be encoded as [64 -32 10 12 -4 10 -20 30].  In your case, after dequantization,  [64 -1 0 -2  1 0 0 0] would become [64 63 63 61 62 62 62 62].


    Bryan S. Burgin Senior Escalation Engineer Microsoft Protocol Open Specifications Team
    Friday, February 18, 2011 5:12 PM
    Moderator

All replies

  • Hi, CSaint,

     

    I can help you with this.  Can you contact me directly at "dochelp (at) microsoft (dot) com"?

     


    Bryan S. Burgin Senior Escalation Engineer Microsoft Protocol Open Specifications Team
    Friday, February 11, 2011 5:18 PM
    Moderator
  • We believe that that your entropy decoding and dequantization is working fine.

     

    At [MS-RDPRFX] 3.1.8.1.6 Linearization, as part of Figure 10: “Sub-band traversal order” there is the comment ““The coefficients from LL3 also undergo differential encoding. Except for the first coefficient, every raster-scanned LL3 coefficient is subtracted from its previous neighbor.”

     

    After quantization, we do differential encoding for LL3 band, where by the first coefficient is encoded as is and the rest are subtracted from their previous value.  So [64 64 64 64 64 64 64 64] is encoded as [64 0 0 0 0 0 0 0].  [64 32 42 54 50 60 40 70] would be encoded as [64 -32 10 12 -4 10 -20 30].  In your case, after dequantization,  [64 -1 0 -2  1 0 0 0] would become [64 63 63 61 62 62 62 62].


    Bryan S. Burgin Senior Escalation Engineer Microsoft Protocol Open Specifications Team
    Friday, February 18, 2011 5:12 PM
    Moderator