Eyevinn / moqlivemock / 25937594530

53%

Build Type

push

github

Committed by tobbee

Commit Message

fix: compute LOCMAF wire bitrate separately from CMAF for catalog Track.bitrate

Catalog generation was calling calcCmafBitrate for every track regardless
of packaging, so locmaf-packaged tracks reported the CMAF wire bitrate
(e.g. ~171.5 kbps for 128 kbps AAC at one-sample-per-object). That's
wrong — LOCMAF delta moofs are ~2 B in steady state, not ~100 B.

New calcLocmafBitrate measures one full + one delta chunk with a fresh
MoofDeltaCompressor, then amortises the per-group full moof over a
1-second group (MoqGroupDurMS=1000) of subsequent delta moofs:

  overhead/sec = full_overhead * (1 / group_dur)
               + delta_overhead * (objects/sec - 1 / group_dur)
  bitrate = sample_bitrate + 8 * overhead/sec

GenCMAFCatalogEntry now picks calcLocmafBitrate when packaging ==
'locmaf' and calcCmafBitrate otherwise. New regression test
TestLocmafBitrateIsLowerThanCmaf asserts the contract for every bundled
track: LOCMAF bitrate must be both (a) lower than the CMAF bitrate for
the same track and (b) higher than the raw sample bitrate.

docs/LOCMAF.md gains a 'Catalog bitrate impact on the bundled assets'
subsection with measured numbers for the test10s tracks (AAC 128 kbps
drops from 171.5 to 131.9 kbps; constant 19.9 kbps per-object saving
across all 25 fps video tracks).

Coverage Stats

40 of 59 new or added lines in 1 file covered. (67.8%)

42 existing lines in 3 files now uncovered.

3311 of 6154 relevant lines covered (53.8%)

1283.99 hits per line