Best No oversampling dac to buy??

[solderdude]

What do you gentlemen mean with "drooping frequency response"?

In a sweep it appears as drooping (slow roll-off) but in practice it is not perceived the exact same way as creating a gentle low pass filter nor will compensating for it to make it 'right' again. Sure the average level will be somewhat restored but in reality it isn't.

The drawing I made below shows what happens in reality when a constant 20kHz tone is sampled. The dots are the sample points. (disregard the blue lines)
The red line is the actual output voltage coming out of an R2R DAC chip (with a very slight high frequency low pass).
As you can see the average signal would be 3dB lower in level but in reality the amplitude varies between 0dBFS to much less. The tone is AM modulated which cannot be undone correctly.

 

[voodooless]

The red line is the actual output voltage coming out of an R2R DAC chip (with a very slight high frequency low pass).

Let's call it a NOS DAC, not all R2R chips are sample and hold.

It would be nice to plot a correctly upsampled version of this over the other two to show what correct filtering will do.

 

[AdrianusG]

There are times I can't hear my wife, yet I see her mouth moving and I feel uneasy. Does that count?

It does for you probably...

 

[solderdude]

The proper differentiation is filterless versus filtered.
NOS just simply means Non OverSampling.
The very first CDP were 16 bit and had no oversampling but had a VERY steep analog (reconstruction) filter behind it to arrive a a flat(tish) response.
With terrible post ringing at (and around) Nyquist due to the very steep (high Q) analog filter.
That was very difficult to make (analog) and oversampling became standard as Philips could not make 16 bit ladders but only 14 bit. They used OS to arrive at 16 bit resolution and had the advantage that the (post)filter was less steep and at a frequency well above the audible range so much cheaper and easier to build.
It is technically impossible to (well way too complex and expensive) to create a tunable filter for the man different bitrates so that is not possible.
Digital is much more easy.
This is what makes a NOS DAC that adheres to the sampling theorem inpossible to build.
So some guys found out that even without such a filter such a DAC does not sound poor (but different) and where there is a demand manufacturers see money.
The NOS DAC was born.
R2R has much more differing possible output voltage levels, DS has much less different voltage levels and DSD only 'on and off'

 

[nowonas]

Yes, ringing occurs whenever there is something in the audio signal that is not correctly band-limited like you see when it clips. I'd say: you'll have other things to worry about than inaudible pre-rining in those instances, and NOS will not solve any of them.

I agree that pre-ringing can be difficult to hear, but in fact none or really slow filters is they only way to remove or minimize pre-ringing as it is caused by the low-pass filter to band-limit the signal.

I am curious, how do you define a correctly band-limited signal?

 

[voodooless]

I agree that pre-ringing can be difficult to hear, but in fact NOS or really slow filters is they only way to remove or minimise pre-ringing as it is introduced by the low-pass filter to band-limit the signal.

Or you just accept that it's really a non-issue and that slow filters have more problems than they solve, see here.

How do you define a correctly band-limited signal?

Brick-walled < fs/2 with at least 80 dB attenuation before fs/2, phase linear. Clipped signals are not properly band-limited.

 

[solderdude]

I agree that pre-ringing can be difficult to hear, but in fact none or really slow filters is they only way to remove or minimize pre-ringing as it is caused by the low-pass filter to band-limit the signal.

I am curious, how do you define a correctly band-limited signal?

It is audible when such filters are used in the audible band (EQ) but not above the audible band (reconstruction filters)
A coorectly band-limited signal has a very steep filter that does not (or just very slightly) filters in the desired audible band and greatly attenuates above the desired audible band.

 

[KSTR]

As you can see the average signal would be 3dB lower in level but in reality the amplitude varies between 0dBFS to much less. The tone is AM modulated which cannot be undone correctly.

It may look like amplitude modulation but it's just the waveform of a two-tone, the original and its image. Both have undergone the attenuation as per magnitude response of the filter.
For any reconstruction filter to make a nice single sine, filtering out the image and ZOH droop, the filter must have an impulse response that is (much) longer than the beat frequency period. That is one fundamental concept of reconstruction filters dealing with a ZOH output signal.

 

[pma]

NOS is a poor, bad idea re audio signal fidelity. If we had 1MHz sampling rate, it would be a different story. But we do not have 1MHz sampling.

 

[KSTR]

(filterless) NOS is a poor, bad idea re audio signal fidelity.

() addition mine.

In that it generates new signals not present in the original, this is true, absolutely no debate here.

The actual effect is subtle, though. Some time ago I presented a snippet of audio resampled to 1/4th rate with and without reconstruction filter to better demonstrate the possible effect for those without 20kHz++ wide-range ears. With that drastic demonstration we can better hear what some NOS proponents claim, more transient sparkle despite the ZOH roll-off. That sparkle is artificial but apparently perceived as better. And because of the FR droop, the sparkle is even more prominent. At least that my view why NOS could sound different even compared to a band-limiting filter with the same filter response -- provided your ears are in pristine condition.

 

[Hayabusa]

No it can't because the droop is dependent on the moment of sampling.

Thats not true. The level of the below 1/2 fs signal is only influenced by sample and hold FR

 

[Scytales]

The proper differentiation is filterless versus filtered.
NOS just simply means Non OverSampling.
The very first CDP were 16 bit and had no oversampling but had a VERY steep analog (reconstruction) filter behind it to arrive a a flat(tish) response.
With terrible post ringing at (and around) Nyquist due to the very steep (high Q) analog filter.
That was very difficult to make (analog) and oversampling became standard as Philips could not make 16 bit ladders but only 14 bit. They used OS to arrive at 16 bit resolution and had the advantage that the (post)filter was less steep and at a frequency well above the audible range so much cheaper and easier to build.
It is technically impossible to (well way too complex and expensive) to create a tunable filter for the man different bitrates so that is not possible.
Digital is much more easy.
This is what makes a NOS DAC that adheres to the sampling theorem inpossible to build.
So some guys found out that even without such a filter such a DAC does not sound poor (but different) and where there is a demand manufacturers see money.
The NOS DAC was born.
R2R has much more differing possible output voltage levels, DS has much less different voltage levels and DSD only 'on and off'

By the way, just for the education of those interrested in this historical developpement, there is a great technical discussion about the first Philips DAC in this video posted on the UK AES channel :

 

[solderdude]

Thats not true. The level of the below 1/2 fs signal is only influenced by sample and hold FR

That is only true when a proper reconstruction filter is present. In the discussed case this is missing.

 

[Hayabusa]

That is only true when a proper reconstruction filter is present. In the discussed case this is missing.

If I have time I create a test case ..

 

[pma]

() addition mine.

Can we, are we able to design a good analog low pass filter with passband 20kHz for 44.1kHz sampling, i.e. sharp enough to cut mirrors and with adequate phase response? Because if it is a NOS DAC for CD ……
I do not speak about Sony 4 x TDA1541, but about a true 44.1/16 NOS.

 

[solderdude]

The steep analog filters for CD I know of with enough attenuation and steepness even for 88.2 and 176.4 were all laser trimmed on a substrate.

If I have time I create a test case ..

knock yourself out.

 

[Audiofire]

View attachment 352593
The red curve, for NOS filter, starts to drop in level at 10kHz and is already 3dB down at 18kHz. This is a lot of change and (slightly) audible when your ears are still good to ~15kHz or so. When compensated for, you may or may not hear any difference to this and a "normal" DAC filter but when you do, it's not coming from the frequency response.

I don't understand the purpose of using red as NOS, since it is the phase linear filter with best impulse response (see here the following page)

 

[pma]

Regardless technology used, it is impossible to design an analog filter flat to 20kHz, with low passband ripple, fast attenuation above 22kHz and acceptable phase response. This is possible only in digital domain and with the aid of oversampling. Analog filters for 44.1kHz NOS DAC belong to stone age of electronics.

 

[solderdude]

I don't understand the purpose of using red as NOS, since it is the phase linear filter with best impulse response (see here the following page)
View attachment 352653

2 things:
1: a Dirac pulse is an illegal test signal and can never be on any recording.
2: The impulse response may look great BUT have a look at the actual output signal of the DAC, particularly at higher frequencies and this is compromised with the 'slow' filters.

No matter how one twists and turns this the slow filters all have poor signal fidelity in the audible band where the sharp filters don't.

The above filter plots are all oversampling because they use SD DACs. The only difference is the used filters. All 4 are not filterless and the bottom two don't adhere to the sampling theorem.

Analog filters for 44.1kHz NOS DAC belong to stone age of electronics.

the stone age of CD players (1982)

 

[earlevel]

Thats not true. The level of the below 1/2 fs signal is only influenced by sample and hold FR

That is only true when a proper reconstruction filter is present. In the discussed case this is missing.

No, it's not influenced by the reconstruction filter—that only bandlimits the output (plus, as I assume you're saying here, compensation for the sinc rolloff, aka "droop").

In a sweep it appears as drooping (slow roll-off) but in practice it is not perceived the exact same way as creating a gentle low pass filter nor will compensating for it to make it 'right' again. Sure the average level will be somewhat restored but in reality it isn't.

The drawing I made below shows what happens in reality when a constant 20kHz tone is sampled. The dots are the sample points. (disregard the blue lines)
The red line is the actual output voltage coming out of an R2R DAC chip (with a very slight high frequency low pass).
As you can see the average signal would be 3dB lower in level but in reality the amplitude varies between 0dBFS to much less. The tone is AM modulated which cannot be undone correctly.

ZOH causes only convolution with a one-sample-period rectangular pulse. In other words, it's an FIR filter, with frequency response the shape of the since function. Just do a search on frequency response of zero order hold, if you don't believe me. Here, I found a nice chart, from this test equipment manufacturer website article, DAC frequency response: