Linear vs. Nonlinear Quantizer Scale

[yup]

In CCE 2.62 and 2.64 in Video buton may choose Linear Quantizer Scale. In 2.66 this option non availible. I try encode DV source with Linear and Nonlinear quantizer scale at 6000Mb/s CBR and show result in BitrateViewer. Result very interesting difference for average Q.level about two times 3.5 for linear and 6.8 for nonlinear. I know that average quantizer very good quantative criteria for video quality. May be usefull choose nonlinear for large bitrate or prefiltered materials? In other encoder TMPGEnc or Canopus Procoder can not choose between nonlinear or linear quantizer scale. If any know about this explain me.
With kind regards yup.

[yup]

paceri thank You for reply.
I read about MPEG-2 quantization and found some formula:
for AC coefficients of intra coded blocks
QF = [(32 x F) + (sign(F) x W x S)]/(2 x Q x W)
for all coefficients of non intra coded blocks
QF = (32 x F)/(2 x S x W)
where F DCT coefficient, S scale nonlinear or linear, W coefficients quantization matrice. Bach wrote that at S<60 nonlinear scaling quantizier better, but we could not forget about constant which equal 32 and nonlinear quantizier will be better for Q.level<2. Many source have Q.level in range 3 to 6 and see very good. I think nonlinear quantizier good for very high bitrate and smooth source.
with kind regards yup.

[yup]

Hi folk!
On holiday I try reencode one dvd source animation (82min) to CVD (352x576 25fps) on 90min CD. I has old version this reencode with CCE 2.66 and nonlinear quantscale and made version with CCE 2.64 with linear quantscale. I compare video in BitrateViewer and in DVD2AVI frame by frame and last on my TV 21". In all cases video with linear quantscale look better. VBR parameter min=0, avg=1294, max=2556. Result very simple if You not have possibility use high bitrate better choice linear quantscale.
With kind regards yup.

[yup]

Now only number. After encoding I compare source and target clip.
For linear scale:

Minimum Average Maximum
Mean Absolute Deviation: 0.2146 1.8364 5.4242
Mean Deviation: -0.1589 +0.0389 +0.3162
PSNR: 27.8639 34.8455 47.6138

For nonlinear scale:

Minimum Average Maximum
Mean Absolute Deviation: 0.2164 1.9235 5.4295
Mean Deviation: -0.2328 +0.0386 +0.3155
PSNR: 27.8541 34.6292 47.5454

Source clip has length 10000 frames, PAL. CCE set vbr 3-pass, bias=0. From number view that linear scale little better (very little but better).
With kind regards yup.