Frank Klemm's Technical Page

Time resolution of Analyse + Synthese

Linear Plot:

[Time resolution, linear plot]

The subband encoders MP1, MP2 and MusePack/MPEGplus have the best time resolution.
MP3 has the worst time resolution.
(MPEG-2) AAC something between.

The time resolution of MP3 and AAC can be switched (you see the 2 graphs for both, one solid, one dotted). You have one for tonal signals, one for transient signals.

But note that for MP3 and AAC the time resolution is not constant and jitters depending on the position of the sample in the transformed block. The diagram shows the average. If the coder is not well tuned this non constant time resolution generates quick changing artefacts.
A critical testcase is Short_Block_Test_2.wav.bz2.

MPEG-4 AAC Main Profile provides an additional feature called TNS: Temporal Noise Shaping. This provides additional time resolution.

A logarithmic plot over 110 dB of dynamics:

[Time resolution, logarithmic plot over 110 dB]

More interesting is the area from -25 . . . 0 dB:

[Time resolution, logarithmic plot over 28 dB]

Why have MP3 and AAC less time resolution?

High quality encodings with 32 kHz.
Time resolution for 32 kHz PCM source is 38% less than at 44.1 kHz and 50% less than 48 kHz. If you have high quality PCM input with 32 kHz (DAT longplay, DSR) and you want to encode it as MP3 or MPEG 2-AAC it is necessary to resample it to 44.1 kHz (441:320) or 48 kHz (3:2), depending on the target.
[Time resolution, 32 kHz, linear plot]

[Time resolution, 44.1 kHz, linear plot]

[Time resolution, 48 kHz, linear plot]

Last modified:  2001-11-28                                Visitors:  ???
[eMail]      [Addr]