Difference between revisions of "CELT TODO"

From XiphWiki
Jump to: navigation, search
Line 10: Line 10:
 
** Bit allocation
 
** Bit allocation
 
** Spreading/folding parameter
 
** Spreading/folding parameter
 +
** Stereo bit allocation
  
 
== Other decisions to make ==
 
== Other decisions to make ==
Line 17: Line 18:
 
** encode bands backward?
 
** encode bands backward?
 
** allow "fractional pulses"
 
** allow "fractional pulses"
 +
* Include a configuration packet for more flexibility? (probably not)
  
  
==Old TODO list==
+
== "Research topics" ==
* raw bits
+
* Dynamic allocation
* <strike>Use the last bits after PVQ (for fine energy?)</strike>
+
* Better stereo coupling
* adaptive folding
 
** finalize folding details
 
** Decide on folding gain
 
* dynamic allocation
 
* <strike>Using PVQ for coarse energy?</strike>
 
* configuration packet
 
* revisit stereo coupling
 
* stereo allocation
 
* tune everything
 
* <strike>increase max pulses</strike>
 
** <strike>more pulses for N>5</strike>
 
** <strike>Fast CWRS for N=3</strike>
 
** <strike>change if(_n<=6) to if(_n<=6 || _m>255) in ncwrs_u32</strike>
 
** <strike>Use real limits in fits_in32()</strike>
 
* Tune for lower sampling rates
 
* <strike>Improve dynamic range by changing the energy floor</strike>
 
* <strike>Laplace encoder</strike>
 
** <strike>Make ft=32768 to save some divisions</strike>
 
** <strike>Use the last symbols to have a wider range</strike>
 
 
* Investigate other transient algorithms
 
* Investigate other transient algorithms
** <strike>energy weight</strike>
+
** Energy weighting pre- or post-normalization
 +
** Include energy shape

Revision as of 17:21, 14 April 2010

Open issues

  • Spreading/folding
    • Uniform quantisation
    • Control over the amount of spreading/folding introduced
    • Make it have the same shape as the rest of the spectrum
  • Tuning
    • Band layout (spacing)
    • Bit allocation
    • Spreading/folding parameter
    • Stereo bit allocation

Other decisions to make

  • Should we use raw bits or now?
  • What should we do about remaining bits after the last PVQ?
    • more fine energy?
    • encode bands backward?
    • allow "fractional pulses"
  • Include a configuration packet for more flexibility? (probably not)


"Research topics"

  • Dynamic allocation
  • Better stereo coupling
  • Investigate other transient algorithms
    • Energy weighting pre- or post-normalization
    • Include energy shape