Mudge
The upsampler has no idea what it's upsampling which is part of the reason for the low pass filter. Upsampling is basically educated mathematical guess work where a few conditions are known. Nyquist theorem suggests that the highest frequencies present in the original signal are half the sample rate. When you calculate an intermediate sample you will almost inevitably introduce high frequency spuriae into the data. Therefore you can be pretty sure that everything in your new 88.2/96KHz data stream above 22KHz is noise, so you filter it out with a low pass filter.
The catch-22 is that with a limited amount of DSP power (the operations occur in real time) you can either use an algorithm that produces less noise and a less severe noise filter or a noisier algorithm and a severe noise filter. Or you can hope and pray that the losses from your tradeoff between the two are compensated by the gains from running your Delta-Sigma DAC at a higher rate.
Apologies if I've over-simplified upsampling, I'm aware that the reality is even worse than I've made out.
I understand basically how upsampling works which is why I asked why you said:
The G68 rolls off the mid-highs and emphasize the bass. This is to be expected, since most upsamplers don't have a flat frequency response.
By "rolls off the mid-highs", "emphasize the bass" and "don't have a flat frequency response" do you mean that it filters out frequencies above 22.05kHz? If not what do you mean by those phrases? Filtering out content above 22.05kHz occurs whether you upsample or not. I am also curious how you were able to determine that the upsampling algorithm in the G68 falls in the group of "most upsamplers" which don't have a flat frequency response. Which upsamplers fall within the group that do have a flat frequency response?
I'm with Frank on this one...
Are you saying that M are unlikely to be using FFT? I would be surprised as I thought that FFT operations were built native into many DSP processors so M would probably make use of them if FFT allowed M to make a better sounding upsampler.
Without access to their code, you can only infer certain things from the behaviour. It would appear that Meridian are using a low pass filter. Why this is used we cannot know. In the case of the apodizing filter the use of a minimum phase low pass causes a dramatic drop in pre-echo.
Sorry to be pedantic, but earlier you were making quite a bold claim "The G68 rolls off the mid-highs and emphasize the bass." but did not explain it other than to later claim that Meridian were one of "most upsamplers" and upsampling in an inferior way to what would be optimal ie not using fft. Yet you are now saying that you don't know for sure what they are doing, except that they need to use a low pass filter so must be doing it wrong.
How do you know this is all related to upsampling? For anyone who has owned M kit, it is pretty clear that M are applying a 'special sauce' to the signal with or without upsampling. Its clearly not a case of 'bits out = bits in' with many of the signal paths as we know that dither is playing its part with many input signals, as was so clearly demonstrated at HH2.
I can equally guess as much as you that M need to filter for other reasons too yet do not use this to make the claim that "The G68 rolls off the mid-highs and emphasize the bass". I know that a number of us want to know what evidence you have of the G68 doing this in a way that is psychoacoustically (sp?) detrimental to the sound.
Your 'theory' to me seems to me that: M are using DSP's therefore can't be doing it right so have bundled it in the "most upsamplers" category due to the lack of power that you have witnessed from your programming of DSP chips.