Someone explain this resampling thing to me...
May 22, 2005 at 3:25 PM Thread Starter Post #1 of 30

Kameleon

500+ Head-Fier
Joined
Apr 26, 2005
Posts
695
Likes
12
I don't get it. When people complain about Creative cards resampling 44.1KHz audio, what does this mean and what are the effects? And when people rave about foobar2000 having a really good resampler, how is this a good thing whilst the Audigys are committing an horrendous sin?

I've had a search around the forums but everyone seems to understand all this already, can someone explain it for me? Cheers
smily_headphones1.gif


EDIT - Oh, I've got an Audiotrak Prodigy 7.1 LT and am mainly playing flac but some ogg/mp3.
 
May 22, 2005 at 3:41 PM Post #2 of 30
If you have a decent DAC, there's no need to resample. "Resampling" is just a load of rubbish, and after the ridiculous craze, people now seek out for NOS (non oversampling) DACs, and are prepared to pay a fortune for it.

Resampling has one use for music, to override the bad resampling in Audigys or AC97 cards.

Creative have low quality resampling, which is slightly audible and worsens the sound quality
 
May 22, 2005 at 4:51 PM Post #3 of 30
That doesn't really answer his question. I would also like to know what resampleing is. So what if it is moved from 44.1k to 96k? How does that change the sound?
 
May 22, 2005 at 5:14 PM Post #5 of 30
All I know is that resampling can be used for error correction by sampling each piece of data multiple times, or for creating a certain flavor of sound, when it is used to interpolate the "missing data". A dac like the benchmark uses it for the first purpose, some of the cd players out there use it for the second. It all really comes down to personal preference.
 
May 22, 2005 at 5:17 PM Post #6 of 30
Well here very non-formal explanation. To understand resampling, you need to understand what sampling rate is. It's a parameter in the process of Analog to Digital conversion. In CD's the audio signals are represented digitally storing amplitud values of that signal sampled at equal intervals many many times per second, in fact 44.1 thousand times per second. That's therefore the sampling rate used for CD's: 44.1 KHz. Some other media use other sampling rates. DATs (Digital Audio Tapes), for instance, use 48 KHz. Resampling is to convert one sequence of sampled values taken at a certain sampling rate, and produce another sequence in some other target sampling rate, in principle preserving as good as possible a digital encoding of the original analog signal.

Notice the difference between Resampling and Encoding. In Analog-to-Digital Encoding you start from the input analog signal, and generate some digital version at some sampling rate K. In resampling, you start from a digital version already, sampled at say that rate K, and you want to produce another digital version of the same signal, but at rate Q. You don't have the analog signal when resampling, you just work at the digital domain.

A related link:
http://www.mathworks.com/access/help.../specto10.html
 
May 22, 2005 at 5:23 PM Post #7 of 30
Resampling is the process of converting sampled audio data from one sample rate to another on the digital level. Ideally you'd interpolate things to a sample rate that's the lowest common multiple of both the source and destination sample rates and pick your sample values there. Problem is, you need to lowpass filter the signal underway, and when interpolating you'd preferably use sinc interpolation but the sinc function (sinc(t/T)=sin(pi*t/T)/(pi*t/T)) extends to +/- infinity and that pretty much collides with the fixed sets of samples you're typically using when processing things. I hope I caught the main trouble spots; all in all, there are plenty of places for quality-vs-performance tradeoffs. Bad resamplers typically exhibit noticeable harmonic generation and/or bad high-frequency IMD (Creative cards are particularly notorious for the latter).
 
May 22, 2005 at 5:25 PM Post #8 of 30
Quote:

Originally Posted by rsaavedra
Well here very non-formal explanation. To understand resampling, you need to understand what sampling rate is. It's a parameter in the process of Analog to Digital conversion. In CD's the audio signals are represented digitally storing amplitud values of that signal sampled at equal intervals many many times per second, in fact 44.1 thousand times per second. That's therefore the sampling rate used for CD's: 44.1 KHz. Some other media use other sampling rates. DATs (Digital Audio Tapes), for instance, use 48 KHz. If you need to convert data from media that use different sampling rates, you need resampling. Resampling is to convert one sequence of amplitude values taken at a certain sampling rate, and produce another sequence in some other target sampling rate, in principle preserving as good as possible a digital encoding of the original analog signal.

Notice the difference between Resampling and Encoding. In Analog-to-Digital Encoding you start from the input analog signal, and generate some digital version at some sampling rate K. In resampling, you start from a digital version already, sampled at say that rate K, and you want to produce another digital version of the same signal, but at rate Q..

A related link:
http://www.mathworks.com/access/help.../specto10.html



That's more what I was looking for, thanks
wink.gif
So if you're converting from 44.1KHz to 48KHz (as I believe the Creative cards do), does it work out what goes in the missing 4KHz by waveform? Or something more like how a TFT screen displays resolutions other than their native, by dropping or duplicating bits?
 
May 22, 2005 at 5:30 PM Post #9 of 30
Quote:

Originally Posted by Firam
That doesn't really answer his question. I would also like to know what resampleing is. So what if it is moved from 44.1k to 96k? How does that change the sound?


Resampling is when you change the sampling rate of the digital music stream. CD's are sampled at 44.1KHz and we can resample that to other sampling rates, like 48 KHz or 96 KHz. There are different ways to resample, the basic way to upsample is to simply place a bunch of zeros inbetween the original samples. This will preserve the original signal fine but it gets a little messy when you want to resample by non-integer multiples. They will sometimes use interpolation to fill the zero samples as well.

The idea behind resampling for the Creative soundcards is due to the fact that the native sampling rate of the card's processors is 48KHz. So, when the soundcard plays back a CD, it resamples it to 48KHz, processes it (even if it is just doing pass through), and then outputs the audio (I don't know if it resamples it back down to 44.1 before sending it to the output. The digital out of the Creative cards is in multiples of 48KHz so there doesn't seem to be any reason for the card to downsample). Many people feel that the card does a poor job of upsampling the signal to 48KHz, so they prefer to use software resamplers to do this before the signal is sent to the card. By sending the card a 48KHz stream, the card will no longer need to resample.

The reasons for upsampling in CD players and DACs are along a different line of thinking. The output from a DAC is not a perfect reproduction of the original analog signal. Sometimes it will be a stairstep (zero-order hold DAC) or it will have some degree of interpolation. However, by applying an analog filter, we could theoretically perfectly reproduce the original analog signal even from a zero-order hold DAC. The problem is that the required analog filter would use an infinite amount of components. Instead, the analog filter is an approximation, where the more complex (and costly) the filter is the more accurate the results. So there are some tricks that one could do that will decrease the complexity of the analog filter without sacrificing accuracy. One way is to have a digital filter applied to the ditigal signal before it is sent to the DAC. This way, the analog filter can be a simpler filter. Another way is to upsample the signal. Upsampling will never add information to the original signal, but you can upsample without degrading the signal. Zero-padding (where the extra samples are just zeros inserted between the original samples) does not degrade the signal for example. The advantage is that by upsampling, you increase the bandwidth of the signal. But the added bandwidth is not used because the original signal did not have any information in that region. For example, CD's have frequencies from 20Hz-20Khz so if we upsample from 44.1Khz to 88.2Khz, then we now have a region from 20Khz-44.1Khz of unused frequencies. The filter now can be relaxed so that if there were signals in the unused frequencies, they would be affected by the filter, but not in the way needed to be properly reproduced. But since we do not have any information in the added regions, we can let the filter be anything in that region. The result is that we only need to control the filter now over a smaller bandwidth region instead of the entire bandwidth. This allows them to use less complex filters to retain the same level of accuracy.

So by upsampling, you can create a DAC that is more accurate than a given straight 44.1KHz DAC for the same money and resources.
 
May 22, 2005 at 5:38 PM Post #10 of 30
Quote:

Originally Posted by Kameleon
That's more what I was looking for, thanks
wink.gif
So if you're converting from 44.1KHz to 48KHz (as I believe the Creative cards do), does it work out what goes in the missing 4KHz by waveform? Or something more like how a TFT screen displays resolutions other than their native, by dropping or duplicating bits?



Well the problem with resampling at non-integer multipoles is that you have to interpolate. For example, let's take a car that is driving along at different speeds, and at every 5 seconds we paint a dot on the road where it was and write the speed of the car at that point. This is the same with CD's in that at every 1/44.1*10^3 second we record the amplitude of the analog waveform. Now what happens if we want a record of the car at every 7 seconds? We cannot simply take the 5 second interval data and pick and choose samples because the sampling rates are different. Instead, we have to look at the original samples and interpolate what they would be at 7 second intervals. This is where people feel that the Creative cards are introducing errors.

When they do resampling in CD players and DACs, they usually do integral multiples to avoid having to do interpolation. Creative uses 48KHz because it that was what was dictated in a PC audio standard (can't remember which one) where the processing chips run at a native 48KHz. If they had allowed for a 44.1KHz pass through when no processing was needed, then this wouldn't be a problem.
 
May 22, 2005 at 5:44 PM Post #11 of 30
So how much of an improvement in sound quality does using the FB2k resampler to sample to 48KHz and then feeding that through to an Audigy make over letting the Audigy do it itself? Is it then more comparable to cards that people would consider audiophile-friendly, like Envy-based cards and EMU0404?
 
May 25, 2005 at 7:42 AM Post #12 of 30
To hear the differnece Resmapling makes on Creative cards,add and remove the Resampler from Foobar DSP.Without Resampling it sounds harsh,unrefined and generally pretty foobared.
evil_smiley.gif
 
May 26, 2005 at 10:22 AM Post #14 of 30
Quote:

Originally Posted by sygyzy
If you use a software resampler (ie Foobar), how does the Creative card automagically "know" not to resample anymore? Does it just know?


The card is then receiving a 48KHz stream and 48KHz being the correct one for the DSP it doesn't need to resample it.
 
May 27, 2005 at 4:30 AM Post #15 of 30
Quote:

Originally Posted by breez
The card is then receiving a 48KHz stream and 48KHz being the correct one for the DSP it doesn't need to resample it.


OK...you guys have kind of answered the same questions I had about resampling. So, on the off chance that my Creative card was killing my sound quality, I went ahead and downloaded and installed Foobar. I applied the 48 resampling in Foobar and then played around with it, trying different rates. Alas, if there's any difference at all, it must be so subtle that you can't really notice it if you're listening for it.

My audio files were all ripped with Audiograbber, using the LAME codec at 320 kbps (44.1 sample rate). It occured to me that you could go ahead an resample to 48 during ripping and that would eliminate the need to do so in the player (Or would it?)....BUT, which rippers would do the job right and which would be worse than just lettling the creative card do it?

Sorry so many questions. Thanks for trying to explain these things to noobs.
 

Users who are viewing this thread

Back
Top