REMB - fix serialisation

[theimowski]

I've found a bug in REMB packet serialisation.

For reference, here's the layout:

// 0                   1                   2                   3
// 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
//+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
//|  Unique identifier 'R' 'E' 'M' 'B'                            |
//+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
//|  Num SSRC     | BR Exp    |  BR Mantissa                      |
//+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
//|   SSRC feedback                                               |

Based on sample Wireshark dissect:

current code reads mantissa as ..00 0110 0100 0010 0010 (first two bits are always 0)

Made following changes:

• Mantissa is using 2 least significant bits from first byte, so it needs to be AND-ed with 3 (0000 0011 instead of 4 0000 0100) - fixed this both ways
• Also added a roundtrip unit test - pretty much copied and adjusted the one above it.
• The unit test showed another issue - when serialising REMB, the exponent and mantissa need to be OR-ed (instead of AND-ed) for the shared byte - otherwise that byte would always be 0, so fixed that as well

[sipsorcery]

Thanks for the PR.

[theimowski]

Cheers! Any chance of getting a new version including the fix released any time soon?