# Can someone explain up sampling please ?

Discussion in 'Archived Threads 2001-2004' started by LiorM, Apr 19, 2002.

1. ### LiorM Agent

Joined:
Apr 7, 2002
Messages:
25
0
The one thing I just can understand is,

The up sampling process , takes the digital signal and doubles it’s sampling frequency

For grater details and clarity.

Now, how the hell you create something that was never exists at the beginning.

I worked as sound engineer at the 90’s and the first thing I’ve learn was that you can

Create something that does not exist.

If you don’t have a bass track, you can add as much as you want 60Hz, you still Won’t get it .

So if you have 48Khz of data , you can double it , tripple it , or do what ever you like with it , and you will still come out with the same data sampled at the beginning ...

Am I wrong here or what ?

Shark

2. ### Bob McElfresh Producer

Joined:
May 22, 1999
Messages:
5,182
0
For a simple example, say you have 48 thousand numbers like this:

500

800

400

600

...

This represents 1 second worth of digital samples at 48 Khz (48,000 times per second).

Now you want a 96 Khz up-sample. So you simply do this:

500

650

800

600

400

500

600

...

So you can simply "fill in" a number to smooth things out.

Then you send this set of 96,000 numbers per second to a D/A converter that expects this many samples per second.

In reality, I think they do some other formula's than simple average between the numbers. But the concept is the same.

Does this help?

Does that help?

3. ### LiorM Agent

Joined:
Apr 7, 2002
Messages:
25
0
Well, that clears things up, but I still

Can’t get use to the idea that some algorithm

Adds bits to the original material.

Who knows, I need to hear then I decide what's

Better.

Joined:
Sep 3, 2001
Messages:
426
0

5. ### ling_w Second Unit

Joined:
Sep 3, 2001
Messages:
426
0
A linear interpolation would produce a floawed result.

e.g.

100

200

250

250

200

100

upsampled 2X would produce

100

150

200

225

250

250

250

225

200

150

100

The green interpolation is in error because it would simulate a clipped signal, in addition to a connect the dots look by the red interpolations.

Various upsamplers uses psychoaccoustic models to do their interpolation, while even some real old units, like the ancient Wadia, used their French Curve or spline algorithm to do their 64x oversampling.

A properly done 2x upsampling would probably look more like:

100

156

200

231

250

256

250

231

200

156

100

A higher order upsampling would become much more complicated. 64x requires so much computational power because in order to fill the 63 empty spots between 2 samples, one would need to string up a 64th order polynomial equation consisting 32 samples before and 32 after the current sample.

Joined:
May 22, 1999
Messages:
5,182