Any benefits from having a higher sample rate? - Page 3

I remember clearly the TDA1541.

I do not have anything against Phillips, they are one of a number manufacturers of IC's including 5 DA's. Their IC branch is NXP and they make 5 DA's for audio, specializing in applications such as low power and low cost, not at all in the state of the art. ALL those 5 IC's are noise shaping types (sigma delta), thus operating at very high clock rates!

If I recall correctly, Phillips was one of the driving forces towards up-sampling and over-sampling, and the Phillips engineer's name that comes to mind was Dykman (spelling?).

I took the time to explain why operating at high clock rates was such a MAJOR STEP FORWARD in technology. I understand that many non technical folks may have some difficulty understanding it, but I certainly tried to simplify the explanation while being solid and correct.

"Everyone know that Philips is the best brand in the world for audio IC". How about "everyone know that the cookie monster makes the best chicken soup"? This is a real good sales pitch: Just say "everyone knows that "XXXX" is the best in the world.

1ppm clock accuracy is another big baloney!!! You do not needs 1ppm clock accuracy. You need LOW JITTER. When a clock is around 590ppm (parts per millions) off, the absolute pitch is less then 1 cent. You can not find a piano tuner that can hear 1 cent - so how about .0017 of a cent? This is outrageously ridicules.

In the audio production side, we often need to have multiple gear operate synchronously (at the SAME clock rate), because over time (say a few minutes) different clock rate ACCUMULATE time error. For example, a 100ppm mismatch between clocks over one hour is .36 seconds. It would be real bad to have audio tracks off by .36 second. In fact, there are all sorts of phasing problems at much lower clock mismatch. So we DO lock (synchronize) clocks when it is needed in the audio production studio. But we do not need the clock accuracy to be anywhere near 1ppm! Not even 100ppm! There is nothing to be gained by going to 1ppm.

And if anyone is having difficulties understanding what I said, the music production clock is far less accurate then 1ppm. So given that the timing absolute accuracy in the recorded music is already way less than 1ppm, what can a 1ppm playback do for you?

Again: jitter is important, not clock absolute accuracy.

Regards
Dan Lavry

I always enjoy reading Dan's posts - it causes hours of research and I come away with a better understanding of how very little I know.

Thanks for taking the time, Dan. Your analogies always make me laugh.

yes, I agree...1ppm is overkill(especially w/ electronic music), but it does improve the stereo image clarity! it's very audible...it won't change the music fundamental tune, though.

so these Philips DAC's don't oversample at all? hence, they offer a more natural SQ over the latest chips from AK/BB? a wild guess would be that AK/BB have taken this issue in account, and what ppl who like these "NOS" DAC's call "artificial artifacts" might very well have been on the record in the first place...and be smoothed out by their "vintage" Philips chips?

I'm just dubious that a +10 y.o. Philips chip can outshine the latest parts from the market leaders

The issue isn't that 1ppm is overkill, its that the ppm measurement is irrelevant. The measure of jitter in good clock datasheets is usually charted in dB/Hz. If someone is trying to sell you a DAC/clock and is talking about the PPM, you're about to be ripped off.

From what I gather, it isn't that NOS is preferred (although some do prefer it), it's basically that some people (some of whom are technically very knowledgable) have issues with delta-sigma. There are other chips than the Philips NOS chips that are often preferred by the old-skool chip lovers, such as the PCM1704 and PCM63.

What I find interesting though is that DACs continue to improve at all. The known limits of human hearing were surpassed by DACs long ago. As far as is known, people can hear distortion down to roughly -40dB. A TDA1541A chip has THD+N of -95dB and it was put into production in 1989. It should sound no different to a more modern chip. The idea seems to be that there's something current measurements aren't catching that the ear is sensitive to, which some outdated DAC chips do better than the new delta-sigma design. Whether that's true or not I don't know, but I do know that not everyone who holds that view is technically ignorant.

many of us don't hear 18KHz+, so you could crank a 19K+20K 1:1 IMD test signal to really high levels - and we should be able to clearly hear the 1KHz IMD product of any 2nd order nolinearity in the DAC, amp or headphone at anything over +20 dBa in most listening rooms - so even -80 or more down distortion isn't necessarily inaudible with test signals

I can't quickly find refernce to IMD distrotion generation in our ears - but that would set a limit

You could also just turn the volume up to ear-piercing levels and listen for hiss. Thats not really a valid comparison of sound quality though. In actual music listening nobody listens that loud, and the distortion products remain below audibility.

oh ok?! I'm not that knowledgeable about DAC IC's design tbh.

but let's take that VALAB DAC w/ the Philips chip as an example : AudiogoN Reviews: VALAB 8x TDA1543 NOS DAC DA converter

if you read the comments, some ppl say that it sounds better than the Bel Canto....and you have to use 8 Philips chips in parallel mode to get the best SQ

also, from the TDA-1543 datasheet : http://www.docethifi.com/TDA1543_.PDF



it can do 4X OS, why do they talk about 4X upsampling

ah well, if you run 8 of these chips....you get a vynil-like presentation

The comments I made about up-sampling are not just about sigma delta DA's. I mentioned sigma delta a lot because almost all the DA's today are sigma delta. But going back, there was a lot of resistor based converters that had some upsampling such as X4 to X16, and that also helped out a lot of the analog filter problem.

Say you have a NOS at 44.1KHz (CD playing). Then the filter transition band is around 44.1K/2-20K - 2.05K.

If you upsample by ONLY a factor of X2, the transittion band is 88.2/2-20 = 44.1-20 = 24KHz. That is already a X12 easier filter design. With only X2 you can actually have a realistic good analog filter, not an easy task but doable.

With say X4 upsampling, the transition band is 176.4K/2-20 = 68.2Khz, and that relexes the analog filter by a factor of 34! that means 34 less poles are needed, relative to a NOS.

At X2, it takes a hack of a designer to do a good job. At X1 (NOS) it is hopless even by today's lowere DA quality standard.

I made resistive DA's, my DA924 and DA2002, and they are not noise shaping sigma delta. I did not up-sample very high but I needed to do some up sampling to get away from the NOS filter problem.

There were many non sigma delta DA's that used X2 to X16. I recall NPC as one of the main companies that sold upsampling ICs to be placed in front of a DA input. I made my own upsampling using a DSP.

Regards
Dan Lavry

so NOS vintage DAC's are worthless? why do ppl say that they give a "vynil-like" presentation then? because of the background noise and crackles?

and I've read on diyaudio.com that you have to 8 of these Philips chips in parallel...just like they did on this DAC as well, makes me wonder about distortion

http://forum.audiogon.com/cgi-bin/fr...3057&read&3&4&

The reason they use so many in parrallel is actually to reduce distortion. The TDA1543 has a lot of distortion even for an old chip. When you stack lots of them, although individual chips may have lots of error, the errors get averaged, so you end up with a bit less distortion. You also end up with more power output.

Also I'd say the reason people say NOS chips sound like vinyl is because NOS rolls off the treble.

Paralleling IC's is not about reducing distortions. The distortion stays pretty constant.

The reason for paralleling the IC's is to lower noise. Unfortunately, you get only up to 3dB improvement for every doubling of the numbers of IC's so:

X is the noise for 1 IC
X-3dB for 2 IC's
X-6dB for 4 IC's
X-9dB for 8 IC's
X-12dB for 16 IC's...

Explanation:

The signal does add in direct proportion to the number of IC's. For example, if you have N=4 (4 IC's) the signal is 4 times stronger.

But noise (assuming that each I.C. noise is RANDOM), does not add as in simple addition. It is proportional to the square root of the number of devices. For the same example of N=4 (4 devices), the square root of 4 is 2 which means twice the noise.

Putting it all together, the signal is raised by X4, the noise is raised by X2, thus the SNR (signal to noise ratio) is increased by 4/2 = 2 which is 6dB.

This is a very well known technique to increase SNR. If the noise of the device is not totally random, the outcome is less than the expected. As a rule, the devices distortion mechanisms are pretty consistent, thus there is no distortion reduction to be gained. If a single TDA1543 has high distortions, 2,4,8 or 16 devices will have nearly the same distortions.

Given that the thread is about NOS, having 1 IC or 100 IC's does not make any difference from the analog filter standpoint, which is the issue here. The transition band is still a super tight 2KHz for a 44.1KHz music at 20KHz analog bandwidth.

I do relate to your comment about NOS rolling off the treble. That is what those NOS guys need to do to get a little more room for the analog filter. If you roll it at say 16KHz (3dB point) the transition range grows from 2KHz to 6KHz, that is 3 times less tight but still very tight. Up-sampling by only X2 yields 24KHz transition band, thus 12 times more room!
This is what I was saying about the not so good old days when all the gear was NOS- we had to compromise a lot.

If one likes audio that lacks the highs, get an EQ, and remove the highs. This will provide much less distortions then a NOS. If you really want to get rid of a lot of highs, listen to the music over the telephone line :-)

I did not even bring in the phase issue , which is another big can of warms. Even a single pole filter at the edge of audio (say 20KHz) causes phase shift down to an octave below the cutoff which is 10KHz. The phase of a NOS filter is an issue! But I really do not have the time to get into that in detail. Those interested can study the subject.

Regards
Dan Lavry
Lavry Engineering

LOL, so true, doesn't a telephone line only the midrange, with bandwidth of 4khz, with 8khz sampling rate? lol

so all these ppl crave for vintage Philips NOS DAC's....because their trebles are rolled off

these are 20 y.o. chips, that already were low end/economic designs back then...so how could they possibly match an AK4396/PCM1792 will remain a mystery I guess

Rolling off the treble can have some "benefits".

One would be when there is significant high frequency distortion. Any roll off will attenuate the distortion, reducing or eliminating one cause of listening fatigue. This is a more audible problem in reflective rooms. Sometimes people tube roll with the same goal in mind.

This distortion can be present even in gear with good THD+N measurements. I also suspect some people use "detailed" to describe this distortion because it certainly sounds like there is something more there.

What kind of distortion are you talking about? The treble roll-off in a NOS chip is caused for the most part by the fact that square waves are being output instead of sine waves. This inherently alters the frequency response. Notice though that the amplitude of the square waves haven't been attenuated at all (except due to the filtering problems Dan mentioned, assuming a filter is used).