The Emotiva Lounge - FAQ: Oversampling vs Upsampling

« FAQ: Oversampling vs Upsampling »

Welcome Guest. Please Login or Register.
May 18, 2013, 11:51pm

The Emotiva Lounge :: General :: Stupid Question Zone :: FAQ: Oversampling vs Upsampling

Author

Topic: FAQ: Oversampling vs Upsampling (Read 1,064 times)

tatentoby
Minor Hero

member is offline

Joined: May 2011
Gender: Female

Posts: 10

FAQ: Oversampling vs Upsampling
« Thread Started on Aug 3, 2011, 11:59pm »

First of all I want to say that I love this section of the Lounge. It helps greatly. My question is, which is the better of the two technologies, up-sampling or oversampling. I read an article lately that was in favor of up-sampling. Thanks again.

« Last Edit: Aug 10, 2011, 1:45am by hemster »

Link to Post - Back to Top IP:

Logged

Nemesis.ie
Emo VIPs

member is offline

[avatar]

Rock on!

Joined: Oct 2008
Gender: Male

Posts: 5,832
Location: Ireland

Re: oversampling vs upsampling
« Reply #1 on Aug 4, 2011, 4:37am »

Perhaps you can define exactly what the two mean in this context?

Are these what you mean:

http://en.wikipedia.org/wiki/Oversampling

http://en.wikipedia.org/wiki/Upsampling

It think there are other factors at play, in the case of oversampling, what over-sampling factor is used?

For example, if you take an analogue source with a highest frequency of 20kHz and oversample it by a 10x factor you would be sampling at 200kHz (or 192kHz) and have a very nice, accurate digital reproduction.

If you took a standard CD sample of 44.1kHz, of the same source and then upsampled it to 200kHz (192kHz being the closest standard number) you would have a file the same size but I would think the oversampled one would be much more true to the original.

I think a little context is needed, they can likely be used for different things and the one used may depend on what you are trying to achieve.

Or have I missed something?

Link to Post - Back to Top IP:

Logged

7.3 HT (7.1 via 4 subs really):

1 x UMC-1 (pre/pro) - FW 7.04.00.19
1 x MiniDSP for sub EQ
8 x UPA-1

8 x ERM-6.3 (LCR Vertical, surrounds horizontal) 2 for the centre
2 x Rythmik F15HP (front, L+R)
2 x Rythmik F15 (rear, stacked)

Screen Excellence AT 115" (16:9)
Sony VPL-HW50ES projector

Popcornhour C-200 Media/BD player
Cisco HD PVR/CTV

DYohn
Emo VIPs

member is offline

[avatar]

Smile!

Joined: May 2008
Gender: Male

Posts: 10,311
Location: Valley of the Sun

Re: oversampling vs upsampling
« Reply #2 on Aug 4, 2011, 9:21am »

Oversampling is a technique used during an A-D process that helps reduce errors because more data is captured during the quantization (digital conversion) process. Up-sampling means you start with a signal that is digital already and add bytes to it in order to convert it to a different bit rate. In general, oversampling has a direct impact on the quality of the digitized signal while up-sampling does not, you end up with the same signal you started with only encoded at a different bit rate.

Link to Post - Back to Top IP:

Logged

Inaudible is inaudible and you don't get extra brownie points for making it twice as inaudible as another solution. - Kevin Haskins
Double-secret reverse-osmosis speaker-cone-induced high-level interference distortion. Its a killer.

Nemesis.ie
Emo VIPs

member is offline

[avatar]

Rock on!

Joined: Oct 2008
Gender: Male

Posts: 5,832
Location: Ireland

Re: oversampling vs upsampling
« Reply #3 on Aug 4, 2011, 12:26pm »

Which is another way of putting what I said I think.

i.e. oversampling the original analogue is likely to produce the best result.

Link to Post - Back to Top IP:

Logged

chris_L
Minor Hero

member is offline

[avatar]

Joined: Aug 2011
Gender: Male

Posts: 56
Location: Vista, Ca

Re: FAQ: Oversampling vs Upsampling
« Reply #4 on Sept 22, 2011, 4:58pm »

Nemesis.ie is correct, over sampling will always provide the best results.

if you sample at a lower rate, which means to space out your samples in time you loose some information. you can NEVER get that information back. So if you then up sample your doing nothing more than adding more bits in from the already down sampled source which will not recover the lost samples.

Will call this our over sampled version
(sample it at 1sec intervals)
(voltage amplitudes)
|..........*....*
|........*........*..............*
|......*........... * ........*.... *
|....*...............*......*...........*
|..*........................................*
|*_______________________.........*________________
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 (time in seconds)

Then you sample at 2 second intervals the same song
(our under sampled version which we want to "upsample later")
|...............*
|...........*
|.......................*.....*......*
|.....*
|...........................................*
|*_________________________....*________________
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 (time in seconds)

See how its sorta the same picture but missing samples (or information), its not quite as smooth in its transitions. If you attempt to up sample the second graph there you must determine an algorithm to fill in the missing samples. So you may do something like, look at a sample then its proceeding sample if its more, split the difference and add a sample in, if its less, split the difference and add a sample.

This is a simple example and may provide satisfactory results but as the amplitudes are switching faster or the sampling rate increases the ability to reconstruct the recording by up sampling becomes more and more difficult.

Also the term "oversampled" is tricky in its meaning. There is no actual way to sample too much per say. There are rules of thumb i.e. nyquist rate in which its determined that if you sample at twice the bandwidth of a signal you should be able to accurately reconstruct that signal. So if you sampled at say 3*Bandwidth you have effectively "oversampled"

Wanted to add on to the above, everything is over sampled in order to make the low pass filter required to reconstruct the signal more realistic and cheaper. So for example, the telephone line has a bandwidth of 3.2Khz and they sample at 8Khz to give themselves some margin. The typical rule of thumb is 10%-20% more.

Sorry for the long winded response, my engineer side takes over from time to time

Funny thing, my digital communications class was all about sampling and rates last night.

If you want to know more in detail or see the notes from that lecture i would be happy to scan them and email them over to you.

-Chris

« Last Edit: Sept 23, 2011, 8:38am by chris_L »

Link to Post - Back to Top IP:

Logged

Receiver: UMC-1
Amp: XPA-3
Fronts: Studio 100's
Center: CC-490
Sub: Ultra 12
TV: UN55D8000

tatentoby
Minor Hero

member is offline

Joined: May 2011
Gender: Female

Posts: 10

Re: FAQ: Oversampling vs Upsampling
« Reply #5 on Nov 29, 2011, 6:49pm »

I have been meaning to get back to this thread and thank you for your reply, it helped alot. It was the best explanation that I have heard.

Link to Post - Back to Top IP:

Logged

KeithL
Administrator

member is offline

[avatar]

Joined: Jun 2012
Gender: Male

Posts: 590
Location: Brentwood, Tennessee

Re: FAQ: Oversampling vs Upsampling
« Reply #6 on Dec 18, 2012, 3:29pm »

I was wandering past, and sensed the need for some serious clarification here...

Oversampling and upsampling are BOTH ways of changing the sample rate of a digital audio file (or stream) to a higher sample rate. When this is done inside the DAC itself (and usually in even multiples; ie 8x) it is usually called oversampling. When it's done somewhere else before the DAC (could be in a player, or a computer, or a separate box), it's usually called upsampling. Usually upsampling is done to some sample rate that is not an even multiple, but is instead a standard value (96k or 192k). Neither process is more accurate, or "better", and often both are used in one device.

Neither process, no matter how well it is done, can "create information", so the resulting digital audio CANNOT be more accurate than the original. The extra samples are interpolated from the original data, and, if the math is done well, they will not adversely affect the accuracy, but they cannot improve it. Oversampling does NOT mean "upsampling too much", and either process, if done correctly, is equally accurate. (Oversampling, to an even multiple, uses easier math, so is easier to do.)

So, then, why bother to do it?

The answer is simple. The highest frequency that a particular digital signal can contain is limited by the sample frequency (specifically, the limit is 1/2 the sample frequency); this is called the Nyquist Frequency. So, for a CD, with a sample rate of 44,100 , the highest frequency it can contain is 22,050 Hz (actually slightly lower). But, even more importantly, the conversion process results in all sorts of nasty noise and "byproducts" at frequencies above that 22,050 Hz. Without going into a lot of math.... you MUST use a high-cut filter to filter out EVERYTHING above that 22,050 in order to get back your original signal (and to prevent a lot of nasty noise and distortion).

Unfortunately, designing and building this filter can be a real problem. Audio extends up to 20 kHz, so we need a filter that passes everything up to 20 kHz without messing it up, but cuts off EVERYTHING above 20 kHz. Ideally, it should be down about 100 dB at 22 kHz. This is referred to as "a brick wall filter", and is impossible to actually make. In real life, you're stuck with a compromise that cuts off most of the stuff past 20 kHz, yet doesn't do too much damage to the audio band. [These filter compromises were why the early CD players often didn't sound very good.]

Now, let's try upsampling our signal to 192k. By upsampling, we have "magically" changed our filter requirement to one that is easy to implement. The audio information stays the same but, because we have increased the sample rate, the Nyquist frequency is much higher. Instead of needing a brick wall filter, now all we need is a filter that passes everything up to 20 kHz without messing it up (that part doesn't change), yet is down a lot at our NEW Nyquist frequency (96 kHz). This filter is a lot easier to design (it's actually possible and practical), and we can even build it with cheaper components and still get excellent results.

There you have it.....

The short answer is that upsampling and oversampling don't do anything to improve the audio quality; what they do is make it possible to design the (required) filter circuitry in such a way that it doesn't make a mess of the converted audio.... What they do is to make it possible for the DAC to do its job properly (which is virtually impossible to do without upsampling). And, finally, since most modern DACs do oversampling internally (it's referred to as an oversampling filter), upsampling outside the DAC as well is really more or less redundant.

Link to Post - Back to Top IP:

Logged

Keith
Emotiva Audio

paradigmE
Sensei

member is offline

Joined: Sept 2012
Gender: Male

Posts: 182
Location: Bay Area, California

Re: FAQ: Oversampling vs Upsampling
« Reply #7 on Dec 18, 2012, 5:15pm »

Keiths answer should be a "sticky" on the Internet as far as this question goes

Link to Post - Back to Top IP:

Logged

XPA-200
USP-1
ADSL990's/Magneplanar MMG's
Carver C1 Pre
Denon AVR-2801
Sources:
Itunes on MAC - Jriver PC
TT - Lovely Oak - 1950's Rek-O-Kut Rondine Jr L-34 w/JICO SAS M55E
TT - Dual 1257 as my ROK has no 45 RPM
Reel to Reel - Teac A-3340
Alesis - Masterlink
Alesis ADATs 20/48's
Various forms of mid END DAC/ADC/soundcards/digital mixers

Click Here To Make This Board Ad-Free
Terms of Service | Privacy Policy | Report Abuse | Mobile