Any benefits from having a higher sample rate? - Page 2

Where did you hear that stuff? This is plain wrong. It is the rate of the modulator that determines the analog filter. It is the up-sampled rate that dictates the analog filter!

A "higher true sample rate" such as 96KHz needs less up sampling then a 48KHz (less by a factor of X2), but they both end up at very high rates in the MHz or even as high as 24.576MHz , and that very high up sampled rate is what dictates the analog filter.

Regarding the second comment: Time domain response and frequency domain response are one of the same, just a different way of looking at the same thing. The higher the bandwidth, the narrower the impulse response. A 20KHz bandwidth system sets a limit on the time domain impulse response. At 40KHz audio bandwidth, the impulse response is half as wide (narrower impulse). But the mics, the speakers, and of course the ear itself limit the bandwidth to around 20KHz or so. The overall audio bandwidth is never higher than the The LOWEST bandwidth device.

So if you try to pass a narrow impulse (say an impulse corresponding to a 40KHz bandwidth) through a 20KHz device (mic, speaker or what not), the impulse will be highly attenuated and it's time duration will be that of a 20KHz bandwidth. Attempting to pass a 40KHz impulse through a 20KHZ system is in fact attempting to pass a 40KHz frequency content through a 20KHz system.

I have a paper at my site about it. It is long, but I kept the math out of site to make it easier read for a non technical person. It is called "Sampling Theory"more readable.

http://www.lavryengineering.com/docu...ing_Theory.pdf

I wrote the paper to refute the vast amount of miss information floating around in audio, much of it is being presented as statements of facts.

Regards
Dan Lavry
Lavry Engineering

Thanks for the link.

I read some of it, the maths was way over my head but it did confirm what I believed:

16 bit, 44.1 kHz audio is enough sonically (in human hearing cases).
If you want to remove any clipping of any sort in the recording, then well, producing at 96 kHz is best and record at 44.1 kHz.

Makes me wonder though...

Are SACD's complete bollocks then?

The only time I can see pratically when you need 192 kHz is recording whales communicating. If you want to record or produce sounds that your dog can hear, then well, a higher sampling rate would have to be needed otherwise no.

You must be talking about about the D/A stage here. I was talking about the A/D stage, which is why I used the words "in the A/D stage" in my post.

The second part of your response doesn't really address time-domain resolution.

very interesting paper, thank you!

I've always seen 192Khz DVD-A to be like 4K video, you're just capturing more noise...and in the end your receiver will have to "guess" what is noise and what isn't, actually lowering the accuracy

I've seen some ppl saying that resampling CDDA to 192KHz would "smooth" the waveforms, giving a more analog-like sound...I couldn't agree less

http://photos.imageevent.com/cics/v0...rts%20v0.3.pdf



upsampling will increase THD+N, and your DAC will be doing a lot of post-filtering anyway....I can see how this would help w/ a dodgy DAC, but w/ some good IC it's not needed IMHO.

some ppl believe that if you upsample at 192KHz using some HQ algorithm, you'll still get a better result than what the DAC will do internally...I believe resampling must be avoided at all cost, but oversampling(within the DAC) should be 128X whenever possible

manufacturers like to mix these two terms, making it even more confusing...upsampling is a big no-no, but oversampling actually increases the post-filtering accuracy, and decreases aliasing?

I was talking about DA's. Both DA's and AD's suffered from very difficult ANALOG FILTER requiernments before up sampling an oversampling.

The up sampling and over sampling did solve the analog filter problems in AD's an DA's.

And the second half of my response directly address time domain impulse response.

Dan Lavry

You really could benefit from reading my paper "Sampling Theory". The picture you included is exactly were the sales guys lost it. More points is NOT better, and one does not connect the points with line segments. All you need is to have enough samples as proved by Nyquist, and the ANALOG FILTER will connect them with the PROPER CURVATURE, yielding the original analog signal.
It is not all that intuitive, but it is the way mother nature works, and it is a proven corner stone of modern technology, from digital audio to digital video, to MRI machines to instrumentation... More samples beyond a certain point (for a given wanted bandwidth) adds nothing! The filtered signal is the same as the original. Of course, how good the result is in practice depends on the implememtation.

Regards
Dan Lavry

Up/over sampling solved the analog filter problems in A/Ds? It's my understanding that filtering must be done before sampling, and obviously up/over sampling can't be done before the sampling itself has been done. Out of band signals must be filtered before they are sampled, so I don't see how up/over-sampling is relevant here.

I didn't ask about impulse response though. What I was wondering is time-domain resolution. That is, how accurately in time signals can be sampled.

RedBook doesn't specify how you cut your input or output signal bandwith by 96dB with a transition band of 20KHz to 22.05KHz - doing so in the analog domain isn’t easy

if you want to do a good job up to (or a little beyond) 20KHz then higher sample rate eases the analog filter design on input and output since there is more room to roll off smoothly between audio and the nyquist frequency

up/oversampling does the sharp corner high order filtering in the digital domain at the higher processing sample rate


a definition of “time resolution” separate from bandwidth/impluse response of digitally sampled audio is a marketing invention – impulse response and frequency response are equivalent views of the same thing – once you’ve set the bandwidth by band limiting the signal below the Nyquist frequency there is no “time resolution” advantage of supplying samples at a faster rate to the fixed bandwidth reconstruction filter

SACD literature is really bad about hyping this – but SACD requires a 50KHz high order (>5th order) filter to keep the rising noise shaped high frequencies out of the rest of the audio chain – you really don’t want the +6dB ~1.4 MHz floating around – early SACD demos of DSD DACs are claimed to have caused some commercial power amps to die from the nonlinear effects induced by the poorly filtered high frequency noise modulation content

So yes SACD does have higher “time resolution” than RedBook CD – because it has a higher 50KHz bandwidth - but is not meaningfully different from 96Ks/s DVD-A – 50 KHz vs 48KHz bandwidth limits

About the analog filtering"

The very large majority of modern AD's are sigma delta architecture. The design of sigma delta is based on 2 "blocks" - the modulator and the decimation "blocks". The modulator of a sigma delta operates at very high speed relative to the output sample rate. Call it what you want (over-sampling or up-sampling), I do not care to argue about the terminology and the name does not matter.

The fact is that the modulator operates at least X64 fs (64 times the output sampling rate), and most often at 128 fs, 256 fs, 512 fs or even 1024 fs. The modulator can not possibly yield 16 bit at say 6-25MHz, at such speeds getting 3-5 bits is plenty difficult. Those few bits (quantization levels) must be very accurate!

In a “standard” converter, say you have 4 bits over -8V to +8V, then you quantize each sample to take one of the values - -8V, -7V, -6V….. 0V, 1V, 2V…..7V. For ½ bit accuracy, the error can be as much as 1/2V. For ¼ bit accuracy, the error can be as much as 1/4V.

But with a sigma delta modulator, the quantization error must be vey small, such as a few micro volts (a micro is a part per million).

Such a “low bits” quantizer, with great accuracy and “super high speed”, is coupled with a feedback filter (typically 3-5 order) and the combination enables one to noise shape the modulator outcome. Basically, the noise shaping “sucks the errors” from the audible frequency range and moves it to the frequency range above the audio. By operating the modulator at very high speed, one provides more range for piling up the noise above audio.

The intermediary outcome (the modulator output) contains a few bits, and the information in those bits is good over the audio range, but full of noise and errors above the audio range. So here comes the decimator, which is an all digital block. The decimator filters out the high frequencies above the audio range, and converts the few (but very accurate) modulator bits into many bits but at much lower rate – the final sample rate.

Why all that information? The modulator operating at very high speeds (a localized rate) has a very high localized Nyquist frequency (half the sample rate). The analog filter should pass all the analog frequencies, but it only needs to reject the signals at the high rate Nyquist. Say you want to pass 50KHz, and your modulator is operating at 10MHz, then the analog filter is pretty gradual – pass 50KHz and reject 5MHz (10MHz/2).

Before up-sampling and oversampling, a 44.1KHz rate meant passing 20KHz and blocking 22.05KHz (44.1KHz/2), and that is one steep filter! One could not do a good job of that.

So yes, the signal has to be pre filtered before the conversion, but filtered to what frequency? To the Nyquist of the front end modulator, and the faster you operate the modulator the easier the filter, because the signal rejection frequency is moved way up there, while the pass band (the audio) stays at low frequencies. We call the frequency range between passing signals and rejecting signals the "transition bend". The further up the rejection frequency (modulator speed/2), the larger the transition band and simpler the filters.

The time question:

You said: I didn't ask about impulse response though. What I was wondering is time-domain resolution. That is, how accurately in time signals can be sampled.

OK, I understand what you are after. This is called “jitter”. Timing inaccuracies can ruin the conversion, and in fact jitter of some tens of psec (pico seconds) can cause problems at 16 bits level audio signal. If you want say 20 bits, the jitter requirement is 16 time tighter then for 16 bits.
A psec is a million of a million of a second (or a thousand of a billionth of a second).

The subject of jitter is very complicated. There are all sorts of jitter mechanisms, and their impact varies. There is random jitter, non random (coherent jitter such as line frequency or some electromagnetic interference jitter); there is jitter that gets coupled from the digital audio signal to the converter clock, and more. The impact of jitter also varies across converter architecture.

Jitter is not just an audio issue, it has to be paid carful attention to for converters in general (medical, instrumentation, telecom, audio, video…)
I first encountered jitter issue in AD's for MRI, before the days of digital audio. I wrote a paper about jitter, and while pretty technical, you can look at some “pictures”:

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

Regards
Dan Lavry
Lavry Engineering

I just did..I guess the guy who wrote that PDF should really read it, though

resampling increases THD+N distortion anyway, so that's a big no-no in my book...yes it's audible, and measurable too(w/ WaveSpectra).

but do you agree that oversampling should be done at the highest rate(128X mostly) in the DAC itself?

there's a lot of stuff written about jitter....all I can say is that I tried 2 pretty much identical soundcard(Asus Essence ST & STX), and the stereo stage was far wider and clearer on the ST(0.5ppm clock)...just like Burson says : Burson Clock
I switched them several times, and the STX sounded messy in comparison...I'm trying to push Asus to force 128X oversampling in their drivers(it's 64X at this point), it would appear that this would decrease aliasing even further, and improve the post-filtering

Dan, thanks very much for that explanation. Seems I have some reading to do

"I just did..I guess the guy who wrote that PDF should really read it, though "

Indeed. It would be good if people that are not experts in electronics stop distributing materials about things they have no knowledge about.

"resampling increases THD+N distortion anyway, so that's a big no-no in my book...yes it's audible, and measurable too(w/ WaveSpectra)."

What do you call re sampling? Do you mean asynchronous sample rate conversion? I would advocate NOT making generalizations based on experience with a limited number of gear. One may change their mind when the implementation is very good.

"but do you agree that oversampling should be done at the highest rate (128X mostly) in the DAC itself?"

The oversampling itself is only one of a number of factors, and there are trade offs. The other 2 basic fundamental tradeoffs are filter order, and the number of modulator bits. Would you prefer a 5th order 128fs with 5 bits, to a 3rd order filter 3 bit at 1024 fs? Much of it is about the specific implementation of a design. There are also number practical design methods that enter into the quality.

"there's a lot of stuff written about jitter....all I can say is that I tried 2 pretty much identical soundcard(Asus Essence ST & STX), and the stereo stage was far wider and clearer on the ST(0.5ppm clock)...just like Burson says : Burson Clock"

I did not read burson, time is too dear, but many of the high end music recording and mastering folks report a connection between jitter and the stereo image. I do prefer to look at various types of jitter independently. Most jitter does tend to impact high amplitude and frequency, but random jitter is a different thing then non random. The later may cause side bends which are just bad distortions, and some of that garbage is subject to aliasing.

"I switched them several times, and the STX sounded messy in comparison...I'm trying to push Asus to force 128X oversampling in their drivers(it's 64X at this point), it would appear that this would decrease aliasing even further, and improve the post-filtering "

Yes, 64fs is pretty low by today's standards. DSD (SACD) was a 1 bit at 64fs, mostly with a 5th order filter. The modern PCM surpassed DSD a long time ago, because it has more modulator bits, and for the most part a higher then 64fs. You can not just keep increasing the modulator rate, due to serious practicable limitations. Maybe in a few years. Right now, 1024fs is as fast as I am aware of.

Regards
Dan Lavry

what I call resampling is up/downsampling(eg. 44.1>192KHz), I used r8brain algorithm on PC(which is supposedly one of the best)....and you could clearly see that both THD and THD+N rates were drastically increasing when resampling(due to aliasing?)

well yeah, a 0.5ppm clock gives a "tighter"/unusually accurate stereo image....something I had never experienced before

hehe, ok I've got no idea at what order filter oversampling is done in the Burr-Brown PCM1792A...but when I read people boasting about NOS(no oversampling) external DAC's, I'm rather confused

a DAC has to oversample to do its job, I understand that's where the best IC's compete....in the post-filtering(eg the "miracle DAC" AK4396. one of their engineers said that they went up the roof with its post-filtering

and each time I see one of these "NOS" external DAC's, and check the datasheet of the DAC IC...it's got mandatory 32 or 64X OS..

My guess is that "NOS" is just one more of the sales hype, a total baloney.

When I hear of no oversampling DAC, I think of the days when I designed such gear. I designed a DA at Analog Solution (the design was later adopted by Ultra Analog). I also designed the first Apogee DA (DA1000) and also some DA's for digital sound on film. All that was around 1990. That was before the over-sampling and up-sampling concepts.

The analog filters were a real compromise. Say you want to have an analog filter (for the AD or DA) that would pass 20KHz (half power point thus -3dB voltage at 20KHz), but need to reject the energy at Nyquist which is 22.05KHz (NOS). Now analog filters are made of poles and zeros. Each pole is made of one resistor and one cap. Most circuits have a pair of resistors and a pair of caps plus an opamp for each pair of poles.

Each pole slope is around 6dB per octave (20dB per decade). So one pole gives you -6dB at 40KHz (you are passing 20KHz thus 40KHz is and octave). What does a single pole yield between 20 and 22KHz? 6dB over 20KHz means .6dB over 2KHz. A single pole at 20KHz yields only 0.6dB at Nyquist! (with no oversampling).

So say you are willing to go for 50 opamps, 100 precision resistors and 100 precision caps per channel. Even that yields only around 60dB rejection, not at all great...

In fact, a 50 pole filter is not practical - the precision for a 50 pole resistor and capacitor is way too tight, way beyond what technology offers.

And do you really want to pass the audio through 50 opamps? That alone is bad news from distortion and noise standpoint, from heat, cost, space standpoint...

I never saw a filter of higher order then 12 poles. 9 Poles was pretty much near the state of the art. So how did the old NOS work? Not too well! In fact, most often the filter bandwidth was closer to 17-18KHz then 20KHz, and even then, aliasing and imaging was still a big problem.

With up sampling and oversampling, we have come a long way! The PCM1792A you mentioned is typically just fine with 3 pole filter, and one does not even need to limit the audio to 20KHz, you can provide a pass bend of 40KHz or higher and still need only a 3 pole analog filter.

So the NOS DA is either a piece of very inferior gear, or it is an up sampling DA that some sales guy calls NOS while it is not.

No need to be confused. When you hear such baloney, just ignore it. It is the folks that tell you that a NOS is a way to go are very confused and un informed.

Regards
Dan Lavry

apparently they rely on very old Philips IC's : PHILIPS TDA1541A TDA1541 DAC IC Non Oversampling 1PCS - eBay (item 110418328232 end time Jul-30-09 07:58:51 PDT)

they're supposedly the best coz they don't oversample, everyone know that Philips is the best brand in the world for audio IC

all these ppl mix upsampling and oversampling...sad, oh sad : Upsampling vs. Oversampling for Digital Audio — Reviews and News from Audioholics

and these "NOS" external DAC's are all over the internet :
Valab NOS USB Re-Clock DAC Low Jitter Dual 1ppm TCXO - eBay (item 270432797090 end time Aug-23-09 15:44:18 PDT)

they put a 1ppm clock on one of these legacy Philips chip : "It is a real vintage chip developed by Philips 2 decades ago. We choose it to implement our own NOS-DAC."

Valab Non-oversampling Dac (nos) Review - DTV Forum Australia - Australia's Leading Digital TV and AV Forum
I believe he's experiencing the 1ppm clock accuracy, but I don't see how a NOS legacy DAC from Philips can outshine an AK4396/PCM1792