I recently read an article in AudioXpress magazine about a DAC "old school" project (I don't have the magazine in front of me, so bear with me if you would).
In the article, the author suggests that he prefered a non-oversampling DAC with 44.1kHz sampling rate, explaining that oversampling places estimated samples in between real samples (interpolation), and thus it is not actual "music".
==> In school, I learned in DSP that UPSAMPLING was the process of inserting estimated samples in between real samples by interpolation. OVERSAMPLING was the process of simply sampling the music at a higher rate than the Nyquist rate; all the samples are real samples in oversampling.
1) Does my above definition of upsampling and oversampling jive with everyones?
2) If I have the above correct ==> if a CD player states that its oversampling rate is 8X, this is not interpolating samples into the music, this is actual music sampled at a higher rate, correct?
I don't mean to disagree with the author of the article, I was just curious about his methodology.
In the article, the author suggests that he prefered a non-oversampling DAC with 44.1kHz sampling rate, explaining that oversampling places estimated samples in between real samples (interpolation), and thus it is not actual "music".
==> In school, I learned in DSP that UPSAMPLING was the process of inserting estimated samples in between real samples by interpolation. OVERSAMPLING was the process of simply sampling the music at a higher rate than the Nyquist rate; all the samples are real samples in oversampling.
1) Does my above definition of upsampling and oversampling jive with everyones?
2) If I have the above correct ==> if a CD player states that its oversampling rate is 8X, this is not interpolating samples into the music, this is actual music sampled at a higher rate, correct?
I don't mean to disagree with the author of the article, I was just curious about his methodology.
Oversampling interpolates values in between the music samples. This does not add any musical information. The reason why this is done really requires a bunch of math (which I do not remember from college), but I'll try to explain without the math.
To sample music of a particular frequency requires you to take samples at double that frequency (this is the Nyquist rate you may have heard mentioned).
When you reconstruct the music, by passing the samples through a D/A converter, you actually end up with your original signal plus additional copies of that signal at higher frequencies. You have your original 0-20KHz, plus a copy of the same signal at 20KHz-40KHz (also a copy at 40-60 KHz, etc).
What 2x oversampling does, is shift the first copy of the music from 20-40 Khz up to 40-60 Khz. This saves you from having to have a lowpass filter in the signal stream. 4x/8x are the same thing, they just shift that first copy of the music to higher and higher frequencies.
To sample music of a particular frequency requires you to take samples at double that frequency (this is the Nyquist rate you may have heard mentioned).
When you reconstruct the music, by passing the samples through a D/A converter, you actually end up with your original signal plus additional copies of that signal at higher frequencies. You have your original 0-20KHz, plus a copy of the same signal at 20KHz-40KHz (also a copy at 40-60 KHz, etc).
What 2x oversampling does, is shift the first copy of the music from 20-40 Khz up to 40-60 Khz. This saves you from having to have a lowpass filter in the signal stream. 4x/8x are the same thing, they just shift that first copy of the music to higher and higher frequencies.
rtarbell said:
An estimate is a guess in an Armani suit. Oversampling filters do not guess. The interpolated samples are a weighted average of the original samples. It is the filters designers skill in choosing the weighting that determines the nature of the filter.
This might help
Upsampling and downsampling refer to changing the rate of sampling for a sample stream. In the generalized case, the original waveform must be recreated with its alias components removed by low-pass filtering. This waveform is then resampled at a lower or higher rate. This can all be done purely mathematically. The math is not trivial. Both of these methods are used prirmarily to convert a sample stream from one format (frequency/rate) to another. Say for example you wanted to play your CD sample stream on your Martian buddy's DAC at 55.125 kHZ... you would upsample by 5/4ths. Either process is very intense mathematically but can be simplified greatly when the conversion rates are nice clean ratios like 3/4, 5/4 etc...
Oversampling is a special case of upsampling in which the sample rate is multiplied by a power of 2 (typically). The purpose of oversampling is to reduce the filtering required with analog components in favor of digtal components (math).
The guy who said oversampling was messed up was right and wrong... he is a knucklehead in either case. Correct oversampling requires complex digtal filtering schemes... there are many schemes, all with their own strengths and drawbacks. The alternative is no oversampling, in which case, complex analog filters are required. Funny thing, the end result of both digital or analog filtering, and the comprimises involved, are quite similar.
Oversampling is complex, and as such, truely understood by few. This is, in part, why it gets a bad rap. Simplified algorithms, and stripped down silicon implementations is another reason... a valid one.
🙂
Oversampling is a special case of upsampling in which the sample rate is multiplied by a power of 2 (typically). The purpose of oversampling is to reduce the filtering required with analog components in favor of digtal components (math).
The guy who said oversampling was messed up was right and wrong... he is a knucklehead in either case. Correct oversampling requires complex digtal filtering schemes... there are many schemes, all with their own strengths and drawbacks. The alternative is no oversampling, in which case, complex analog filters are required. Funny thing, the end result of both digital or analog filtering, and the comprimises involved, are quite similar.
Oversampling is complex, and as such, truely understood by few. This is, in part, why it gets a bad rap. Simplified algorithms, and stripped down silicon implementations is another reason... a valid one.
🙂
Hlaf band upsampling can be done with a simple 4 tap FIR filter.
-1 9 9 -1 and divide the answer by 16.
-1 9 9 -1 and divide the answer by 16.
- Status
- Not open for further replies.