So I was browsing DACs and came across this: SCR10 (http://www.ifiaudio.com/src10.html) Digital Upsampler.

"The SRC10 simply converts the digital source into higher quality, higher resolution digital signal before feeding it into your existing outboard DAC."

How can one upsample a digital signal? I can understand changing the sampling rate on an analog signal, increasing it to achieve greater fidelity. But how does upsampling work with a signal that's already digital?

I'm confused! :confused:

Me too.

:lurk:

no need for it

Found this link (http://en.wikipedia.org/wiki/Sample_rate_conversion).
Aparently digital sample rate conversion is somewhat complicated...

It basically comes down to this: There *is* no absolutely correct manner of converting from D to A. Each DAC uses a different algorithm and produces slightly different results. It's possible that by converting a digital signal from 44.1K/16 to, say, 96K/24 pre-conversion that you'll get a better (or worse!) result than feeding the original samples into a D/A converter which can accept both.

A DAC uses a circuit with a slow electrical rise and slew rate to follow narrow sharp pulses from the laser or digital signal from PCM etc. The problem with 16bit 44.1kHz sampling is these pulses are sharp and spread far enough apart that if the DAC is too fast it will make it audibly electronic, and if its too slow it will not be able to react to fine details.

32bit up-conversion fills in the sharp 16 bit volume ladder and upsampling at least doubles the occurrence of the bits over time domain so a relatively fast DAC circuit can be used without sounding so electronic. Both methods do this by guesstimating, almost exactly like the 120Hz LCD converters.

Interesting anomoly: these people are promoting upsampling, while there are forums and threads dedicated to removing the digital filters & oversampling from existing CD players.

The MAIN reason for upsampling / oversampling is that the recovery filter in the analogue domain (i.e., after the DAC) can be of lower order, which introduces fewer ripples & phase aberations in the audible band.

Upsampling is just a way of interpreting (implementing) a digital filter.

Nikko, your graph captions contain some technical errors. The CD 'word' rate is 44100 words per second for each channel, not 22500. The graph on the far right (if that's what the origininal analogue waveform was supposed to be, before quantization) will NEVER be recovered in that way by upsampling. The anti-aliasing filter used during recording removes those spikes and fine detail long before any conversion to digital takes place.
ALL Cd recordings are put through an anti-aliasing filter, that abruptly limits bandwidth to 22kHz. No amount of digital hocus-pocus will ever get it back to more than 22kHz.

Nikko, your graph captions contain some technical errors. The CD 'word' rate is 44100 words per second for each channel, not 22500. The graph on the far right (if that's what the origininal analogue waveform was supposed to be, before quantization) will NEVER be recovered in that way by upsampling. The anti-aliasing filter used during recording removes those spikes and fine detail long before any conversion to digital takes place.
ALL Cd recordings are put through an anti-aliasing filter, that abruptly limits bandwidth to 22kHz. No amount of digital hocus-pocus will ever get it back to more than 22kHz.
NO NORMAL CD PLAYER can sample 44100 points per second for each channel, only 22500 and that is why they are limited to a 22500Hz bandwidth unless they use upsampling as I noted. 44100Hz is a lump sampling rate of both channels combined. I also already noted it cannot be recovered in the 3rd graph and will never be any information above 22500Hz apart from error. Its lost in the conversion to 16 bit when burned onto cd.

Hmmmm....not that I'm a technical expert in this regard, but I've always understood CD to be 44.1kHz, according to Nyquist rate, or something like that. Meaning, the base material is sampled at twice the highest desired frequeny of reproduction.

I assume this is per channel because it just make sense that it would be that way.

That's lump sampling rate for both channels combined. It is divided into two for stereo playback.

Really? Tell me it's not so.

According to the theory, that would mean only something on the order of 11kHz upper end frequency response.

Its 22500Hz for the left, and 22500 for the right. The lump data sampling rate is 44100Hz, both combined as the laser reads it. A simple example can be made with a CD burner and a signal generator set for output above 22500Hz in both left and right channels. Once recorded and played back we do not the original tone if its above this point. If anything we get lower frequency oscillation or clipping.

Wikipedia: "An audio CD can represent frequencies up to 22.50 kHz, the Nyquist frequency of the 44.1 kHz sample rate."

If each channel were sampled at 22500Hz, that means that, By Nyquist's theorem, the highest frequency possible to encode on a single channel would be 11250Hz.

If this were the case, both channels would be limited to 11250Hz and therefore the highest frequency that could be reproduced would be that. The two channels cannot combine to produce frequencies higher.

Its 22500Hz for the left, and 22500 for the right. The lump data sampling rate is 44100Hz, both combined as the laser reads it. A simple example can be made with a CD burner and a signal generator set for output above 22500Hz in both left and right channels. Once recorded and played back we do not the original tone if its above this point. If anything we get lower frequency oscillation or clipping.

Wikipedia: "An audio CD can represent frequencies up to 22.50 kHz, the Nyquist frequency of the 44.1 kHz sample rate."

It also says (http://en.wikipedia.org/wiki/Red_Book_(audio_CD_standard)) "It also specifies the form of digital audio encoding: 2-channel signed 16-bit PCM sampled at 44,100 Hz" and "The bit rate is 1411.2 kbit/s:
2 channels x 44,100 samples per second per channel × 16 bits per sample = 1,411,200 bit/s = 1,411.2 kbit/s.
As each sample is a signed 16-bit two's complement integer, sample values range from -32768 to +32767."

Perhaps there's no disagreement, but these statements indicate that there are 44.1K 16 bit samples PER CHANNEL per second.

That is incorrect. No redbook cd contains information that high and can be confirmed with an analyzer. I'll flag it immediately. This is what happens when non-engineers write topics in wiki.

The Nyquist refers to each channel being one half the combined sampling frequency equaling 22500hz for each channel.

By Nyquist's theorem, a sampling rate of 22500HZ can only encode frequencies as high as 11250Hz. If each channel is sampled at 22500Hz, then how is it possible for frequencies above 11250HZ to be reproduced?

I'm about to graduate with a degree in electronics engineering and this material is still reasonably fresh in my mind.

The Nyquist refers to 44100kHz sampling with one half being 220500Hz for each channel. I am an engineer and design things like these. The error is easy to spot the wiki poster first said 22050Hz per channel then said 2X 44100Hz as a typing error. Like I said, try to convert 30 or 40kHz onto a standard redbook 44.1kHz cd burner and see what you get.

Error on wiki pending correction by admin.

Yes, you will not convert 30 or 40KHz onto an audio CD, because of the sampling rate.

You will, however be able to convert frequencies as high as 20KHz onto a single channel. This would not be possible if the sampling rate per channel was 22500HZ, like you are saying.

The maximum frequency for the channel is 22500KHz because the sampling rate is 44100KHz.

The overall sampling rate is 44100kHz, the rate per channel is 22500Hz, nothing more, nothing less. Now that we all know 22050Hz is the highest note we will ever get from redbook CD and its resolution low, I think its time for some SACD. Just kidding.

So, we've been saying the same thing all along, just explaining it differently and not understanding each other... I think this would be the point where we sit down and have a beer :beerchug:

Yeah, it is way too hard to talk over a keyboard and I think what I was trying to say way coming across with wrong wording. I get two different answers so I'm lost now.

The overall sampling rate is 44100kHz, the rate per channel is 22500Hz, nothing more, nothing less

The OVERALL sampling rate (stereo sampling rate) - if you want to use such a term - is 88.2kHz - as used by the first Sony players that had only ONE DAC, which had to be shared by both channels. One DAC had to convert 88200 samples per second.

The FREQUENCY RESPONSE (22kHz) of each channel is half the SAMPLING FREQUENCY of each channel (44.1kHz).
If you want 2 x 22kHz of audio, you have to have 2 x 44.1 KHz of digital words.

"http://en.wikipedia.org/wiki/Red_Book_(audio_CD_standard) (http://en.wikipedia.org/wiki/Red_Book_(audio_CD_standard))
now says
"The bit rate is 7200.1 kbit/s:

2 channels x 22,500 samples per second per channel × 16 bits per sample = 7,200,000 bit/s or 878.91 kilobytes."

^^^That^^^ is just plain WRONG! Aside from the application of sampling theory being wrong, the arithmetic is up the pole.

Here's a really great explanatory diagram from John Watkinson's book "The art of digital audio"
[Watkinson, J. The Art of Digital Audio", Focal Press, London (1988) p465. ISBN 0-240-51270-7]

"2 channels of PCM audio, each signed 16-bit values sampled at 44100 Hz"

This is a mistake - I dont want a reply - but I cannot delete this post. Sorry.

I Upsample and reclock and like the results. According to its designer my Monarchy NM24 likes receiving 24/96. Results may differ with different digital front ends.

I'm spawning a new thread "Discussion: sampling theory & CD bitrate definition" to get a wider viewpoint on this.
http://www.audiokarma.org/forums/showthread.php?t=224216