To upsample or not to upsample?

I think we all know by now what this section is for.
Post Reply
Cressy Snr
Amstrad Tower of Power
Posts: 10552
Joined: Wed May 30, 2007 12:25 am
Location: South Yorks.

#1 To upsample or not to upsample?

Post by Cressy Snr »

I have been playing around with the Audio Midi setup app in Mac OSX Mountain Lion on my Mac Mini media server and have come away very impressed.

Upsampling CD rips at 16 bit 44.1KHz to 24 bit 88.2KHz I find,makes for a more natural presentation (relaxed yet with tremendous presence).

Upsampling the same files to 24/96 does not give the same benefits at all.

iTunes downloads, which for the uninitiated are actually 24/44.1 encoded, fare even better, showing a greater improvement than the CD rips.


This seems to go against the doctrine that you must output a bit perfect stream from the PC.
I much prefer "power-of-two" upsampled streams to straight bit perfect or 96KHz upsampled from 44.1.

It's simply better. There's no contest.

Out on a limb again
Sgt. Baker started talkin’ with a Bullhorn in his hand.
Cressy Snr
Amstrad Tower of Power
Posts: 10552
Joined: Wed May 30, 2007 12:25 am
Location: South Yorks.

#2

Post by Cressy Snr »

Here's something else.
after having listened to a track, ripped from CD using Max with CD paranoia and comparing it to the identical track ripped using data reduced 256K VBR AAC. I hear absolutely no difference, no matter how hard I listen. If there is any difference at all, it is waaay below my threshold for giving a toss


I don't deserve to be on an audio forum. :( :(
Sgt. Baker started talkin’ with a Bullhorn in his hand.
User avatar
ed
retired
Posts: 5384
Joined: Thu Jun 21, 2007 4:01 pm
Location: yorkshire
Contact:

#3

Post by ed »

SteveTheShadow wrote:If there is any difference at all, it is waaay below my threshold for giving a toss

word!

:)
There's nowhere you can be that isn't where you're meant to be
gazjam
User
Posts: 103
Joined: Sat May 05, 2012 1:21 pm

#4

Post by gazjam »

I always went with the idea of not messing up the original sound, no upsampling, dithering any of that stuff.

A recent wet Sunday afternoon I tried all the upsampling options in my server/Jriver setup, played through my Dac which can do up to 32/384.

Short answer...
As you said Steve, upsampling can improve the music, more air, better resolution of micro detail in my experience.
Simply better. :)

I think you can only go so far with it until it becomes detrimental?
No point upsampling a 16/44 to 32/384..sounds worse than the original!
Also, its no good upsampling by anything other than a round number imo, for instance 44 upsampled to 88 sounds better than it does converted to 96.
I think this may be down to less number crunching when converting?
No doubt theres more to it than that.

I reckon as long as you upsample by a factor of two (as you say Steve) theres improvement to be had.
This applies for any bitrate file, and I have set my playback software to double everyting.

Spent a few hours on it, found what works for me now I can forget about it! :)
User avatar
Nick
Site Admin
Posts: 15706
Joined: Sun May 06, 2007 10:20 am
Location: West Yorkshire

#5

Post by Nick »

But the devil is in the detail.

I would expect it would depend on just what is meant by upsampling.

While its true you cant create more information, you can make the dac's life better (ish).

For example, if you want to upsample from 44.1 to 88.2, the a simple way would be to draw a straight line between the two real points and create a new sample at that point. So say you have values of 110 and 120, you create a new middle value of 115. And thats very simple to do if you just double the date, you just add a new point between the two existing ones. That means the DAC can run at 88.2k, and the filtering is simpler as its twice as far away from the audio band. But you still have corners in the waveform on the real points. So you can use filtering instead of a straight line to create each point, so instead of looking at the values on either side of the new point, you look ate more either side, and calculate the new point so that you can imagine a smooth curve joining all the real point sup. But still no new information.

But if you go from 44.1 to 96, you need to add more that one extra point between existing ones, in fact you need to throw most of the original points away and create new ones. At 44.1k you get a sample every 22.6757ns, at 96k you get a sample every 10.41667ns, so for evey original sample, you create 2.1769.

I cant see any reason why 44.1 could not be converted to 96 without damage, but its a lot more complex, and I suspect there is a lot more scope to get it wrong.
Whenever an honest man discovers that he's mistaken, he will either cease to be mistaken or he will cease to be honest.
Post Reply