does AES/EBU help if i implement reclock?

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Vadim said:

Well, on the other hand, who is to say that listening to a dozen bits is a bad experience? I cannot speak for sound quality, but I can say this, - that a $100 antiquated turn-table will most likely do better in terms of signal integrity then a non-over sampling DAC.

i think most people into vinyl will agree that a basic tt will trounce most basic cdplayers.

that aside, i would like to say that the non-oversampling dac really is nice to listen to, regardless of technical specs. it sounds more relaxed, open and non-fatiguing.
 
rfbrw said:
Vadim,
Would I be safe in assuming I can't interest you in one of these,
http://www.asounddifference.com/adacs.html ?

It never seizes to amaze me that there are designs out there that make so little sense. The DAC’s on the site you linked to definitely fit this category. I looked at the most expensive DAC they have; standing at near $50,000 it is a testament to a business highwayman style.

Curiously the D/A chips used are the old AD1865, which were produced in 1988-89 time frame. The AD1865 cannot output 16-bit signal, - oh, well…Here is the quote that particularly grabbed my attention, -

“…the Audio Note™ 1xoversampling™ circuit is a genuine sonic revelation and demonstrates the potential of the 16Bit format to a far greater extent…”

I really like that statement, as it makes me feel all warm and fuzzy inside. However, what this statement demonstrates is a complete luck of understanding of the engineering principles that govern reality.

It seems that the immortal words of late Scotty the engineer, who said
“… I canna change the laws of physics…” - do not apply here.
 
garbage said:


i think most people into vinyl will agree that a basic tt will trounce most basic cdplayers.

that aside, i would like to say that the non-oversampling dac really is nice to listen to, regardless of technical specs. it sounds more relaxed, open and non-fatiguing.

I have no problems with turn-tables. I would however argue that our hobby here is all about sound reproduction and not about sound generation. With this in mind, it should be clear that the goal is to faithfully reproduce what has been recorded on a CD or the vinyl record. The common logic would then dictate that any perceived sound which can be attributed to an electronic component, be it a turn-table or a DAC, - any sound at all, is a problem. Wouldn’t you agree?

The goal of a sound reproduction is to have an ideally zero contribution from the play-back device, whatever that device may be. Hence, - the best and arguably perfect electronic component must have no sound of its own by definition.

I, therefore submit that in this regard a CD beats the turntable any day. I further submit that a DAC with no over sampling facility will introduce an unwanted and very high distortion into the signal that your expensive speakers are trying to deliver to you. Let along the fact that a great deal of information will be simply lost to noise in this case just as well.
 
“… I canna change the laws of physics…”

Agreed. And it seems that you can not change opinions that are without foundation.

Wish that I had a $ for every non-o/s maven that has told me "How do you know how it sound? Just try it, you'll be surprised."

No, I don't have to try it to know. Sorry. And I really don't want to.

OK.........I have heard it.......obviously not on my system. I can't believe it doesn't sound horrible as I feared. It just doesn't sound that good, either.

Jocko
 
Vadim said:


Unfortunately the aliasing spectra are not ultrasonic at all. Here is the picture; - with no over-sampling the first image of the 20 kHz –wide data will appear at 44.1 kHz, contaminating the data substantially to put it mildly.

Without a proper anti-imaging filter the noise rejection is done by the D/A’s inherent zero-order hold function, that looks like [sin(x)]/x, with x=wT/2, w=2piF and T=period of the 44.1 kHz sinusoid. At 20 kHz this amounts to less then 0.2 dB. That means that the zero-order-hold phenomena only reject 0.2 db of noise at 20 kHz. This is all the noise rejection you will get if no anti-imaging filter is used. Noise city!
Correct me if I’m wrong please: for input signal with 0-22.05 spectra aliasing spectra will be from 22.05 to 44.1 which is ultrasonic to me. There is no "noise" in 0-22.05 spectra

X = 2*pi*F/Fs =~ 2*3.14 * 20/44.1 =~2.85
Sin(X)/X =~0.1. != -0.2 dB

So, in a nutshell, here you have it, - with no over-sampling you cannot reconstruct the original 16-bit data at all. You may recover about 12-13 bits at the most, although I doubt that even that is possible.
The ultimate goal is not reconstructing samples, it is reconstructing analog signal on LPF/ADC input, which is not limited by 22.05 kHz. There is always error after Fs/2, no matter will aliasing spectra be cut or not. Because input signal has spectra components above 22.05 kHz, I don’t think that garbage from aliasing is worse then nothing.
 
Vadim said:


I have no problems with turn-tables. I would however argue that our hobby here is all about sound reproduction and not about sound generation. With this in mind, it should be clear that the goal is to faithfully reproduce what has been recorded on a CD or the vinyl record. The common logic would then dictate that any perceived sound which can be attributed to an electronic component, be it a turn-table or a DAC, - any sound at all, is a problem. Wouldn’t you agree?


We are getting dangerously close to landing on planet Self where the abiding motto appears to be man in the service of machine. The thing either does what you want it to do or it doesn't and if by violating the rules you get it to do what you want, its only audio and I can live with that. The plain fact is for many that like this nos stuff, the numbers mean nothing. They don't care whether the dac contains a trained gerbil programming a minature cray on the fly, they just like what hear and that will do nicely. So what if I've tried it, didn't like it and certainly wouldn't buy it, I can think far dafter ways to spend one's money.
And having spent time with the kind of people who think the 5532 is too good for audio and 13 bit/32Khz is more than enough i.e. video people, I find the lunacy sort of balances itself out.

Back to reclocking. Using an ASRC when there is no need to change the sample rate strikes me as unnecessary. I can see the ease of use aspect and the seemingly minimal effect on audio quality but I would rather forego another round of processing when I've got an oversampling filter for in or preceding the dac. A PLL or a FIFO is the way I'd go.
 
stolbovoy said:
Correct me if I’m wrong please: for input signal with 0-22.05 spectra aliasing spectra will be from 22.05 to 44.1 which is ultrasonic to me. There is no "noise" in 0-22.05 spectra

Well, the input signal should be band limited to 20 kHz. It is done during recording at A/D stage with analog low-pass filter. This filter is not perfect. While Shannon Theorem assumes perfect band-limited signals, the reality is not that. Also, it seems to me that there is very little point talking about 22.05 kHz which is half the sampling rate.

Anyway, the aliasing noise does not just stop at 22.05 kHz border. It continues to propagate into the 20 kHz band while diminishing in amplitude as it gets closer to lower frequencies.

Another point to keep in mind is that the image frequencies, if not properly rejected, will most definitely cause intermodulation products with desired baseband frequency components. This along will result in unacceptable degradation if the output signal.


stolbovoy said:

The ultimate goal is not reconstructing samples, it is reconstructing analog signal on LPF/ADC input, which is not limited by 22.05 kHz. There is always error after Fs/2, no matter will aliasing spectra be cut or not. Because input signal has spectra components above 22.05 kHz, I don’t think that garbage from aliasing is worse then nothing.

Naturally, we are trying to reconstruct the analog signal, not samples. However, the original data is band limited. Sure there is an error, but if the A/D is done right, and there is no reason to think that it is not, the error you are talking about is negligible. We do get the 16-bits in the end.

The anti-aliasing filter preceding the A/D is extremely important, so I disagree, - it is important if the aliasing spectra is cut or not. If it is not cut, - the resulting digital signal will never represent the analog data to the required 16 bits.

Vadim
 
rfbrw said:



We are getting dangerously close to landing on planet Self where the abiding motto appears to be man in the service of machine. The thing either does what you want it to do or it doesn't and if by violating the rules you get it to do what you want, its only audio and I can live with that. The plain fact is for many that like this nos stuff, the numbers mean nothing. They don't care whether the dac contains a trained gerbil programming a minature cray on the fly, they just like what hear and that will do nicely. So what if I've tried it, didn't like it and certainly wouldn't buy it, I can think far dafter ways to spend one's money.
And having spent time with the kind of people who think the 5532 is too good for audio and 13 bit/32Khz is more than enough i.e. video people, I find the lunacy sort of balances itself out.

Back to reclocking. Using an ASRC when there is no need to change the sample rate strikes me as unnecessary. I can see the ease of use aspect and the seemingly minimal effect on audio quality but I would rather forego another round of processing when I've got an oversampling filter for in or preceding the dac. A PLL or a FIFO is the way I'd go.

Yes, planet Self is a place to behold. However, I agree with you, that there is no point insisting on the truth, whatever it may be. We are talking about perception here, and so, as the saying goes, - whatever turns you on…If it is the NOS DAC that does it for you, - great!

I also agree with you that a FIFO or a PLL is another credible approach to jitter reduction. You see, with ASRC, particularly those made by AD, the THD+N is the lowest if the input and output rates are kept the same. It has to do with the samples landing directly on top of the interpolation coefficients.

The problem as I would see it with the PLL is that it is so bloody difficult to make one that actually works well. You will most likely need to have 2 of them and in the end the complexity is not worth it in my opinion when the ASRC is such an elegant and simple solution.

FIFO, is another story. But I think it will get to be just as cumbersome as the PLL deal. Although I admit , I never tried the FIFO approach.
 
A PLL or a FIFO is the way I'd go.

I've just measured a CS8416. It has a lot, really a lot of jitter. Far beyond the 200ps which the datasheet refers. Measured somethin in the 5ns range, dependant what (coax) SPDIF source I'm using. Yeah a second PLL and synchronous reclocking is the way to jet rid to that jitter.

The problem as I would see it with the PLL is that it is so bloody difficult to make one that actually works well.

No, a VCXO with a PLL is not difficult to design, not more than a typical low or high-pass filter. BTW a simulation helps a lot for such a purpose...
 
Vadim said:
Well, the input signal should be band limited to 20 kHz.
Is it on the format specs? If some of ADCs have 20kHz LPF - it is design feature, IMHO.

It is done during recording at A/D stage with analog low-pass filter. This filter is not perfect. While Shannon Theorem assumes perfect band-limited signals, the reality is not that. Also, it seems to me that there is very little point talking about 22.05 kHz which is half the sampling rate.
Isn't it mirroring classic DA conversion: LPF with higher bandwidth + oversampling + DF?

Anyway, the aliasing noise does not just stop at 22.05 kHz border. It continues to propagate into the 20 kHz band while diminishing in amplitude as it gets closer to lower frequencies.
There is no data in the digital domain about frequencies higher then Fs/2. What make first (or any) aliasing component to go down below 22.05 on DA side?

Another point to keep in mind is that the image frequencies, if not properly rejected, will most definitely cause intermodulation products with desired baseband frequency components. This along will result in unacceptable degradation if the output signal.
This is another story. Yes, it will cause some additional level of IMD, which depends on analog circuitry.

Naturally, we are trying to reconstruct the analog signal, not samples. However, the original data is band limited.
Original "data" is a sound of cymbal, flute , etc. It is not band limited. And if you cut part of the original spectra - you just added an error equal to negative inversed A/D LPF frequency response applyed to this original signal which is not band limited.


The anti-aliasing filter preceding the A/D is extremely important, so I disagree, - it is important if the aliasing spectra is cut or not. If it is not cut, - the resulting digital signal will never represent the analog data to the required 16 bits.
Sorry for writing confusing way. I meant DA, not AD. For AD antialiasing filter is a must. For DA I personally concerned about frequency response, not aliasing.
 
garbage said:
Vadim

I would like to seek your opinion on a circuit or possible schematic where I can build a receiver using the AD1895 and also a corresponding dac that would interface easily with it. Or should I just follow the eval board for a start?

This would make an interesting project in future for me to compare how it sounds with the non-os 1545a.

Thanks.
garbage

http://www.diyaudio.de/html/body_dac5.html
 
That is awful. The much maligned YM3623 only has around 1 nSec. I don't think any of the Crystal RX are far from that. How they managed to make the '8416 that bad is, well...........about what I would expect from those guys.

I think it's not ony a problem of the chip but also from the incoming SPDIF signal. But as the SPDIF receiver is only able to attenuate jitter above its cutoff frequency of about 30 kHz (I think so, but I've got no datasheet here at home). jitter below 30 kHz is directly fet through the receiver.

On the other side the jitter also changes when switching between the two different phase detectors in the CS8416. This shows that the CS8416 has an influence on the jitter performance. If I'd knew this before I'd better use the AKM SPDIF receiver...
 
Stolbovoy,

I am sorry for a rather late reply.

Is it on the format specs? If some of ADCs have 20kHz LPF - it is design feature, IMHO.

No, the filter must be there as a part of an overall circuit.

There is no data in the digital domain about frequencies higher then Fs/2. What make first (or any) aliasing component to go down below 22.05 on DA side?

Naturally, the digital data as just before it gets to D/A chip only contains information with frequency content up to Fs/2. The problem is not there.

The problem of high transients has to do with the way the D/A works. Look at the simple ladder type D/A. Internally it has current switches that steer the current in to the collection point, which is ultimately the current output from the D/A chip. Those switches, usually made out of FETs, are controlled by the input digital data.

The transient behavior of the switch produces a multitude of various harmonics. When you look at the spectral content of the image, you will see those harmonics extending pretty much everywhere. The anti-imaging LP filter works to reduce that unwanted spectra. Naturally that filter needs space to work.

My point was that a few kHz available, as in the case of non-oversampling DAC, is simply not enough to attenuate the unwanted spectra to the tune of 16-bit precision or about 96 dB. Well, actually a little less if you account for the zero-order hold.

Original "data" is a sound of cymbal, flute , etc. It is not band limited. And if you cut part of the original spectra - you just added an error equal to negative inversed A/D LPF frequency response applyed to this original signal which is not band limited.

Again, this is not so. The data is band-limited by the A/D converter circuitry. In fact the anti-aliasing filter is the very device the limits the frequency content of the data. I agree that the cymbal, or whatever the tone may be, might extend past the 20 kHz. Naturally we don’t care, - we can’t hear it anyway, unless you are 16 and sitting in the lotus position…Damn, I can’t hear anything past 17 kHz!

Sorry for writing confusing way. I meant DA, not AD. For AD antialiasing filter is a must. For DA I personally concerned about frequency response, not aliasing.

Its ok, I get confused by all this terminology just as well. Anyway, as I pointed out above, you cannot get away from anti-imaging requirement on the D/A side. It is just not possible to recover the signal that the sound engineer heard during the recording process without it.

Vadim
 
bocka said:


I've just measured a CS8416. It has a lot, really a lot of jitter. Far beyond the 200ps which the datasheet refers. Measured somethin in the 5ns range, dependant what (coax) SPDIF source I'm using. Yeah a second PLL and synchronous reclocking is the way to jet rid to that jitter.

No, a VCXO with a PLL is not difficult to design, not more than a typical low or high-pass filter. BTW a simulation helps a lot for such a purpose...


Well, I am extremely skeptical of jitter measurements. I know from experience and many experimental failures this is not a trivial task. Measuring anything down to 100’s of picoseconds range is tough to do in a reliable fashion. Alternatively you might consider the fact that correlated jitter is still a subject to a Fourier analysis and it will show as a THD measurement.

Also my experience shows that even the double PLL design will not do as well as ASRC. So, in the end I would still prefer to go with AD1896
 
Well, I am extremely skeptical of jitter measurements. I know from experience and many experimental failures this is not a trivial task. Measuring anything down to 100’s of picoseconds range is tough to do in a reliable fashion. Alternatively you might consider the fact that correlated jitter is still a subject to a Fourier analysis and it will show as a THD measurement.

Agreed that jitter measurement in the 100 (or less) ps range is not trivial. On the other side jitter of some ns is quite easy measurable and its dependant of the CS8416 settings. To obtain a higher resolution than 16 bits for higher frequencies lowest jitter (in the ps range) is a must. This can be done either with a ASRC or a PLL. A PLL is advantageous over a ASRC because it lets the incoming data untouched which a ASRC cannnot by principle. Cutoff frequencies are about the same, some Hz when well implemented. But - and this is very important - the VCXO control voltage must be as clean as possible, less than let's say less than 1mV. This is the real hard job because the incoming data from the PLL is a square / pulse wave. And a simple 74HC4046 as phase detector will definitely not do the job. So it's easier to implement an ASRC than a PLL.
 
bocka said:


the VCXO control voltage must be as clean as possible, less than let's say less than 1mV. This is the real hard job because the incoming data from the PLL is a square / pulse wave. And a simple 74HC4046 as phase detector will definitely not do the job. So it's easier to implement an ASRC than a PLL.


when stating 1 mV, what BW are you refering to ?

and yes, designing decent PLL's is not easy......
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.