Upsampling bad?

I am curious to what people here think of asynchronous upsampling to 192 kHZ/96 kHz. Most DACs already do oversampling, so why use another chip to upsample? Marketing hype? I have found out that most DACs/players that use the Crystal sample rate convertor chip have a 'samey' sound with fuzzy/busy highs and a synthetic sound that is far removed from my non-oversampling DAC. I might hear distortion, but why do acoustic instruments sounds so much more real with my NON OS dac? Also, I think most digital filters do more harm than good.

Any comments? :JPS:

 

NOS is the way of the future.

 

i have switchable 96khz upsampling on my CD player. Upsampling works well- much better detail to cymbals and things without creating sibilance on vocals. Don't ask me what chipset is used. Details are in my sig

 

give me an old fashioned 8x oversampling, multibit dac with a good but simple filter design. couple that to a well designed and beefily psu'ed output stage with lots of power regulation and i don't think you can beat it.
if i can;t have that give me a nos dac.

i've only heard 1 oversampling dac / cd player i could have possibly lived with (the cec cx 51 / dx 71 combo) but that used a proprietory link between the two (not spdif) the others have all given me a headache after a while.
cheers


julian.

 

In my experience upsampling is awful.

 

In my experience upsampling is wonderful.

Now, who gets the prize for the most useful and insightful post on this question?

 

Hi,

Useless gizmo. I am working with a chinese startup on what one might call "less compromised CD Player". We are pretty well going to town, cannot say too much on details right now, but it includes switchable syncronous and asyncronous over/upsampling using pretty much the best available.

The best sound is attained using no oversampling and a analogue filter of my own design, the worst sound is asyncronous upsampling, the best measurements are syncronous upsampling (aka. oversampling).

As these options will remain switchable all owners will be able to make the same experiments.

My own experience of upsampling (and oversampling) match very much what you comment.

Ciao T

 

Whilst I'd broardly agree with you Thorsten, it has to be said that I have heard impressive things from dCS in the past, although all of the best CD players I have heard (including dCS) claim better than everage linearity for their Dac( so it might have just been the ring dac).

 

Hi,

I heard dCS with/without upsampling on several occasions. I PERSONALLY felt that upsampling changed the sound in a qway I found to make it more artificial and less involving. In many technical aspects the sound appeared better, but in the to me crucial areas it was markedly worse, if not as bad as the Cirrus Logic implementations.

I have no simple or conclusive explanation for this, I can merely observe and others have observed the same. I can equally see why the sound of upsampling could appeal to others.

Ciao T

 

I totally agree, artificial and less involving sums it up well

 

Me I'm a 44.1khz man, however there are a few U's which don't sound artifical, but they aint involving

 

WM, Wadia's surely couldn't be described as 44.1khz - given then interpolation filters they use. This in reality is the same as upsampling isn't it?

 

Sm,
Input dear boy, input 44.1khz.

 

Sorry, you've lost me? Input? You mean the CD itself?

 

most asynchronus upsampling i've heard adds sparkle to the top end but tends to pay the price in the bass. rhythmn also seems to be less distinct somehow. as for wadia - i'm not 100% sure but don't they only interpolate to interger multipliers? i.e. oversampling / 88.2 / 176.4?
this to my ears certainyl sounds better possibly due to the fact that the original signal is still present whereas with 192 khz it's only there every few hundred (thousand??) samples. - can;t be bothered working it out.

 

SM,
The digital input to the dac section is stock redbook, a lot of players have an SRC BEFORE the main chipset. so the input frequency is already at 96 or 192khz.
The wadia input at redbook 16 bits (44.1khz), oversampled ( 32 direct multiples of 44100 1.41mhz or 2.82mhz 64 multiples for the top of the line dac, this is NOT done with asynchronous upsampling) but purely in dsp and the dac chipset.

 

Ah thanks for that.

It was this that got me confused. Is the dCS all done in the DSP too?

 

is that straight across the wadia range there wm?

 

Hi,

I think we need to understand what happens with "upsampling" (sorry to those who know this well, I am trying to keep this on Discovery channel level).

The process is actually more precisely called an asyncronous sample rate converter (ASRC). Actually, there are syncronous sample converters as well, they are most commonly seen as "oversampling" filters.

The reason that ASRC's exist is that pro-audio digital recording routinely used 48KHz sample rate (since days long before the CD) and the CD has a sample rate of 44.1KHz. As there is no integer multiple between the two you need to do some fancy DSP to get from one to the other and get there you must, to get a CD from a 48KHz master tape.

It is pretty obvious that we need this process bi-directional as well, so for the longest time ASRC's have existed that could go from 32KHz-48KHz input sample rate to any output sample rate in the same range.

The way this was done is usually analogous to oversampling by a very high factor (1 Million Times +) and then decimating back down to the desired output sample rate using a digital lowpass filter whose performance is set by the output sample rate. The way this is actually done in hardware (as it is obviously problematic to upsample to 44GHz sample rate) is to use various lookup tables and other smart stuff.

What must be understood is that all the original samples are discarded and are replaced by a "guess" what it should have been. The same process of course happens when you use an oversampling filter. You can programm a DSP to the job or use a dedicated chip.

It should by now be obvious that ANY TYPE of sample rate conversion, be it up-, over-, under- and sideways sampling will loose information of the orginal recording and will invariably will add distortion. The very best sample rate converters will loose very little of the original and add very little of their own.

What happened next is that DVD Audio appeared together with 96KHz sample rate (and 192 KHz optional), so the same sample rate conversion process was applied for a range of 32-96 or even 32-192KHz (ratios of 8:1 are now available).

So, someone thinks, what if I feed 44.1KHz into an ASRC and I fix the output rate to 96 (or 192KHz)? The result, it sounds different (it must, even if using dCS). Surely different is better and anyway, we now have a 96KHz signal, surely it's as good as DVD Audio but froma CD Source! Or so the marketing spiel goes.

What you have in reality is a system where you have lost information and have simpley changed one Digital filter for a different one (so it will sound different). Maybe, if your DAC has an abysimally bad Digital Filter then feeding it upsampled data (be it syncronous or asyncronous) might sound better because it actually is better. Maybe your DAC has an excellen t digital filter and the ASRC has a less good one (usually the case), you now get what is objectively clearly a disimprovement.

Equally, ASRC's are claimed to "reject" jitter, what happens is actually that the input jitter is "hardcoded" into the output signal while the output signal now has a clear clock. Compared to a propper re-clocking mechanism this again is a disimprovement.

However, as said elsewhere, there is little point arguing about taste, which is why our upcoming CDP will have the choice left to the user, we ourselfs already know what we take ....

Ciao T

 

To put all that into perspective, a huge amount of this damage is done during the decimation process involved in producing a 44.1khz CD.

You can't get that back regardless of your playback system and by definition, very little of the original recording remains on the disc.

Also you will find that dCS offer upsampling to 88.2 & 176khz, as do Meridian. I have to believe that there is a difference between these implementations along with Wadia's and the cheap fitting of a SRC on the dac board.