# Any benefits from having a higher sample rate?

Discussion in 'Sound Science' started by seamless sounds, Jul 30, 2009.

1. Are there benefits to having a higher sample rate? The standard sample rate is 44.1khz. When people talk or argue about sample rates, they talk about frequency; how it makes no difference because humans cant hear above 20khz. Is there actually more to it than just frequency? I thought sample rate literally meant the number of samples per second. There are benefits of having more samples per second, but when people argue, they just argue about frequency and that we can't hear anything above 20khz, but nothing about the number of samples per sec.

There's only one application that I know of where having a higher sample rate is beneficial. That's when you want to extract the lead vocal from the rest of the song (for remix). I'll explain how. You'll need both the original and instrumental (off vocal, karaoke) version of the song. Then you invert the signal of the instrumental version. Place the original and the instrumental songs side by side and you'll cancel out the rest of the background instruments leaving only the lead vocal. You'll have to move the track to the left/right so it aligns perfectly with each other. Moving it merely 1 sample makes a tremendous difference, not even 44.1khz is accurate enough! I upsampled the tracks to 192khz, and there was a big improvement but a few of the instruments still bled through a little bit, but not as bad as the 44.1khz sample, which is a lot worse.

So there, higher sample rates does have its use and benefits. Just wanted to dispel this myth where having a higher sample rate (>44.1khz) is totally useless.

2. Contributor
The sampling rate is directly related to the frequency, due to the Nyquist-shannon sampling theorem.

Upsampling has its uses, but it cannot add anything to the original signal -- it can merely (hypothetically) take less away.

3. resampling increases distortion, easy to measure w/ WaveSpectra.

4. Assuming you're using CDs, upsampling results in the software interpolating, it doesn't really create more information.

5. There will be a good video on this. Dan Lavry gave a lecture and this was a part of it during the Seattle meet last weekend. It will be on you tube soon.

6. Quote:

 Originally Posted by HeadFi Fanatic /img/forum/go_quote.gif Are there benefits to having a higher sample rate?

There was a very long thread on this subject in this forum a few months ago: 24bit vs 16bit, the myth exploded. Lot's of interesting reading.

7. It really depends on the DAC that the signal is going through. Some (usually the more expensive) DACs have higher quality components, have less jitter potential, and use analog filters. They can do just find without upsampling. Cheaper DACs usually sound better with upsampling because they have higher jitter potential, don't have those filters and lower sample rates sound choppy and fatiguing. This has been my experience anyway.

8. Quote:

 Originally Posted by Arjisme /img/forum/go_quote.gif There was a very long thread on this subject in this forum a few months ago: 24bit vs 16bit, the myth exploded. Lot's of interesting reading.

That's not sampling rate - that's bit depth.

9. Contributor
Since the early 1990's, virtually all DA's have built in up-sampling. There are 2 reasons you want that:

1. Without up-sampling, making a post DA good analog filter (required to remove the image energy) is virtually an impossible task. Up-sampling moves the image energy up to higher frequencies, separating the wanted audio energy from the unwanted image energy making a good analog filtering possible. The higher the up-sampling, the more separation (gap) thus making the filtering easier.

2. DA's do not generate narrow pulses (RZ signals - return to zero), because there is not much energy in narrow pulses. Instead, the DA output signal (before the analog anti imaging filter) is a NRZ signal (NRZ - not return to zero). The signal looks like "steps" (not like narrow pulses), and such signal does contain the needed energy to drive the analog filter. However, when you use NRZ signals without up-sampling, you lose some high frequencies. The flatness response is compromised.

You can see a plot in my web site, in a paper "Sampling, Oversampling, Imaging and Aliasing".

Look at page 3 for a plot titled Sin(X)/X plots for X2,X4,X8 and X16.

http://www.lavryengineering.com/white_papers/sample.pdf

You can see that at X2 up-sampling you still lose around .8dB at 20KH. With X4 up-sampling, the loss is around .2dB at 22KHz. At X8 it is less then .1dB, at X16 the loss in a non issue...

The horizontal axis is frequency (0-320KHz). Your interest is mostly 0-22KHz (audible range), shown by the red line marked "22". The vertical axis is dB loss in amplitude. The curve that drops fastest is the X2 up-sampling. At X16, you 3dB loss is all the way up near 320KHz, with hardly any loss at 20KHz.

True, one can compensate for the loss, but it takes a lot of DSP (signal processing). It is not possible to compensate well for the Sin(X)/X curve with analog circuits.

So the bottom line is: you do not want a DA without any up-sampling. You may not need to up-sample a lot, but you need some up-sampling to enable good filtering and flat response. Most DA's today up-sample between X64 to X1024. This is overkill for flat response, and is very good from an analog filter standpoint. The reasons for up-sampling so much are is due to modern DA converter architectures such as sigma delta designs.

My answer is a bit technical, but a proper and solid answer must be based on technical understanding of the issues. I try my best to simplify it for the casual reader (no math, minimal engineering terminology). I hope what I wrote is not too difficult to understand.

Regards
Dan Lavry
Lavry Engineering

10. Contributor
Hi Dan,

Don't jitter problems get magnified with higher oversampling (64x versus 4x)?

11. Dan, when you say up-sampling are you referring to oversampling? Who upsamples 64x? Wadia, because their units output at 2.8224 MHz, are the only ones I can think of. Most DACs upsample to 96, 192, or 384.

Most very high-end DACs have a 1 - 1.5dB roll off at 20kHz. The ones usually with a flat response are the less expensive DACs.

The Audio-gd Reference One DAC does not upsample but it uses a DSP chip to handle the processing. (Don't go by their price, since their gear is equivalent to at least 5 times the price that you would buy in the U.S.)
ÐÂ½¨ÍøÒ³ 1

12. Quote:

 Originally Posted by DanielCox /img/forum/go_quote.gif That's not sampling rate - that's bit depth.

13. Contributor
Quote:

 Originally Posted by Donald North /img/forum/go_quote.gif Hi Dan, Don't jitter problems get magnified with higher oversampling (64x versus 4x)?

I do not see a direct connection between jitter and the up-sampling rates.

Why do you think jitter is more of a problem at high up-sampling?

Regards
Dan

14. Contributor
Quote:

 Originally Posted by IPodPJ /img/forum/go_quote.gif Dan, when you say up-sampling are you referring to oversampling? Who upsamples 64x? Wadia, because their units output at 2.8224 MHz, are the only ones I can think of. Most DACs upsample to 96, 192, or 384. Most very high-end DACs have a 1 - 1.5dB roll off at 20kHz. The ones usually with a flat response are the less expensive DACs. The Audio-gd Reference One DAC does not up-sample but it uses a DSP chip to handle the processing. (Don't go by their price, since their gear is equivalent to at least 5 times the price that you would buy in the U.S.) ÐÂ½¨ÍøÒ³ 1

Hi,

First, I see the words up-sample and over-sample used interchangeably by many folks. I am not much into arguing about the English, I use up-sampling for DA and over-sampling for AD. In either case, the conversion circuitry (the modulator part of the AD or DA) operates at a higher multiple of the sample rate.

First, in general, you often have much more then a 3dB drop at 20KHz. There are a number of reasons for that: most mics yield -3dB at 20KHz or below. Most speakers also yield around 3dB at 20KHz. At 44.1KHz sample rate, the AD digital filter also yields around -3dB at 20KHz, and the DA up-sampling filter does the same. You end up with around -12dB loss at 20KHz, and that is from start to finish (record to playback).

When one uses sample rates such as 88.2KHz-96KHz (even 60KHz would be just fine) one moves the AD and DA up thus leaving the mic and speakers to yield around a -6dB loss at 20KHz.

But we are talking about DA’s so I will only address the DA’s:

When one talks about a device with 20KHz bandwidth, one is referring BY DEFINITION to the frequency where there is a 3dB loss in voltage, which means 1/2 the power. So yes, most DA's operating at 44.1KHz do have a 3db loss at 20KHz or around 20KHz. A 3dB loss at 20KHz is by definition a 20KHz bandwidth. This is the starting point for my comments.

The issue I brought (the Sin(X)/X flatness response) is in ADDITION to the 3dB point that defines the bandwidth of the device. If you do not up-sample, the additional loss gets the half power point to a lower frequency, lowering the audio bandwidth frequency – moving the half power point to lower frequency. That means alower then 20KHz bandwidth BY DEFINITION.

You said: "Who up samples 64x? Wadia, because their units output at 2.8224 MHz, are the only ones I can think of. Most DACs upsample to 96, 192, or 384".

That is not so. They are NOT the only ones, virtually all DA's in the last 15 years or so up sample by X64 to X1024. All audio sigma delta do so, and most of the DA's are sigma delta with very few exceptions.

Some of the older DA's (resistor or capacitor based designs) up sampled (form say 48Khz) by X2 to 96, or X4 to 192KHz... or even by X16. But virtually all DA's today are at least at X128 to X1024 thus operating from a few MHz up to around 25MHz. the 64X fs is somewhat slow by today's standards.

You also said: "Most very high-end DACs have a 1 - 1.5dB roll off at 20kHz. The ones usually with a flat response are the less expensive DACs."

A DA without up-sampling operating at 44.1KHz can not possibly be flat to 20KHz. Even if the DA has NO analog filter (thus real bad high amplitude unwanted image energy), the Sin(X)/X I talked about would bring the near 20KHz response down. That was the whole second point - where I explained the need for up-sampling. The use of NRZ (the steps instead of very narrow pulses) have such built in frequency response with high frequency loss. One can not escape that fact.

BTW, DSD (SACD) is based on X64 fs technology.

Regards
Dan Lavry

15. A higher true sampling rate (not up/over-sampling) allows the use of less intrusive filters in the A/D stage.

I've also heard it said that it provides higher time-domain resolution. I'm not sold on this though. Although time domain resolution is determined by sample rate at Fs/2 Hz, once you go a fair ways below that into music territory, bit depth predominantly determines time-domain resolution.