MDCT
in [DSP]
|
Prev: ANNOUNCE:-- TimingAnalyzer Free Version -- Draw timing diagrams
Next: Single Sensor Active Noise cancellation
From: aamer on 21 Jun 2008 18:58 Hi Friends, Any reasons on why MDCT is preferred over FFT in audio codecs(eg. Ogg Vorbis). Thanks and regards aamer
From: dbd on 21 Jun 2008 19:53 On Jun 21, 3:58 pm, "aamer" <raqeeb...(a)yahoo.com> wrote: > Hi Friends, > > Any reasons on why MDCT is preferred over FFT in audio codecs(eg. Ogg > Vorbis). > > Thanks and regards > aamer Because of the potential for TDAC, see: http://www-ccrma.stanford.edu/~bosse/proj/node27.html http://en.wikipedia.org/wiki/Modified_discrete_cosine_transform Dale B. Dalrymple http://dbdimages.com
From: Chris Felton on 22 Jun 2008 23:14 > Any reasons on why MDCT is preferred over FFT in audio codecs(eg. Ogg > Vorbis). Couple reasons, first question is why is the DCT used instead of the DFT (fast version of both transforms exist, image compression literature is a good resource for DCT vs. DFT). This is mainly due to the property of energy compaction. On average the DCT represents the energy of a signal in less bins than the DFT, this leads to easier/ better compression. Also as mentioned, TDAC (time domain aliasing cancellation) is where the MDCT comes into play. Also, version of the "TDAC" for the DFT, usually use the general overlap and add algorithms (similar to the MDCT algorithm). Later compression standards (AAC) doesn't use the poly-phase filter bank that MP3 does, I am pretty sure Ogg Vorbis is the same. The audio signal is sent directly to the MDCT and the outputs of the MDCT is the data that is compressed. The perceptual coding (decision what to keep) usually uses a standard FFT.
From: glen herrmannsfeldt on 23 Jun 2008 15:11
Chris Felton wrote: > Couple reasons, first question is why is the DCT used instead of the > DFT (fast version of both transforms exist, image compression > literature is a good resource for DCT vs. DFT). This is mainly due to > the property of energy compaction. On average the DCT represents the > energy of a signal in less bins than the DFT, this leads to easier/ > better compression. The reason should be that DCT has appropriate boundary conditions, but then one result is better energy compaction. Note that DFT has periodic boundary conditions, so that the output of the inverse transform is the same on both ends. (Both sides of the unit cell in video terms.) With enough high frequency terms it can approximate a gradient, but DCT does it much better. -- glen |