Yes, I was assuming that NOS means no reconstruction filter, as this is what DIY NOS usually means. I don't think I have seen any NOS with a reconstruction filter (i.e. brick-wall above 22.05kHz) However, I accept that NOS can use a reconstruction filter and can compensate for the sinc response. Once you do this, you probably lose the claimed benefits of NOS. It is much harder to do a good reconstruction filter for NOS - this is why OS was invented!
A sinc filter is no effect.
It is just a bandpass filter that brings back the lost 3 dB @ 20 kHz.
Not quite. The sinc filter (an automatic side-effect of sample-and-hold DAC) reduces HF, yet without adding any phase shift. A sinc-compensation filter can boost the HF back to a flat spectrum, but it is difficult to do this without adding phase shift. I suspect that is why many NOS DACs don't bother to include a sinc-compensation filter.
The sinc filter reduces HF
A sinc-compensation filter can boost the HF back to a flat spectrum
--------------------------------------------------------
So does it boost or reduce in your opinion ?
Can do both... depends of how you design it.
And if you can show me an analog filter that can output 20kHz from a 44.1kHz sample rate without phase shift or 6dB attenuation or image freq up to -20dB... you are right.
All the "NOS" that I see aroud DYI is not filtered garbage presented with a "godlike" image...
And if you can show me an analog filter that can output 20kHz from a 44.1kHz sample rate without phase shift or 6dB attenuation or image freq up to -20dB... you are right.
All the "NOS" that I see aroud DYI is not filtered garbage presented with a "godlike" image...
To repeat: the sinc filter (inherent in sample-and-hold DAC, so always present) cuts HF. The sinc-compensation filter, if present, boosts HF. These are two different things - I think you are confusing them. Perhaps you have seen lazy people misname the sinc-compensation filter, by calling it a sinc filter?
Last edited:
Yes, and in this case, that analog signal (the Bessel filtered one) can be faithfully reproduced by the OS DAC, without "ringing". But you must realise that due to the slow roll-off of a Bessel filter, it is a compromise. The Fc must be quite low in order to get sufficient blocking of aliases (signal above 22.05 kHz) You either live with a lot of high frequency roll-off in the audio band, or use a much higher sampling frequency in order to push that roll-off to above 20 kHz.
That is exactly the type of digital signal that NOS DAC believers like to use. It is impossible to obtain with an ADC having a proper analog front-end though. Put the NOS output into a spectrum analyzer and just look at the mess you've made. The low-pass filter on the DAC (whether analog or digital or both) is called a "reconstruction filter" for a reason. It is necessary to reconstruct the signal, because without it, the DAC produces high frequency garbage that wasn't there in the original signal that had been sampled. For accurate reproduction at 44.1 kHz sampling, there shall be nothing above the Nyquist frequency (22.05 kHz), but with NOS DACs, there it is.
If you want to hear just how bad of an idea NOS really is, try it at a lower frequency, so that your ears and brain can directly hear the high frequency crud coming out of it.
macboy,
All that you write makes sense. I prefer NOS sound without any reconstruction filter though - but this is personal taste, and as such, can not be argued about.
@ lower frequency in your last sentence you mean lower than 44.1 kHz sampling rate? Or a lower frequency recorded signal? Music program material contains all frequencies, and it sounds quite good on my system.
All that you write makes sense. I prefer NOS sound without any reconstruction filter though - but this is personal taste, and as such, can not be argued about.
@ lower frequency in your last sentence you mean lower than 44.1 kHz sampling rate? Or a lower frequency recorded signal? Music program material contains all frequencies, and it sounds quite good on my system.
That HF loss is caused unintentionally by the DA process.
It acts like a filter but it is NO filter.
A filter is a circuit built with intention.
When you call that a filter, people think it is something put there for good reason.
If you have a speaker with a peak in the frequency response, you can use a filter to equalize it.
Now, would you call the speaker a FILTER just because it has an unwanted response ?
An interesting philosophical question: is something that affects the system response a filter, or is it only a filter if you have deliberately added it in order to change the system response? Filter theory applies in either case, as classical physics knows nothing of our intention. If it acts like a filter, then it is a filter.
One could argue that the sinc filter inherent in a sample-and-hold DAC is deliberately introduced, as in principle a DAC does not have to be SaH - it could output narrow pulses approximating to a delta function and then there would be no sinc response.
However, whatever decision you reach on this, it is confusing if people call a sinc-compensation filter a 'sinc filter'.
One could argue that the sinc filter inherent in a sample-and-hold DAC is deliberately introduced, as in principle a DAC does not have to be SaH - it could output narrow pulses approximating to a delta function and then there would be no sinc response.
However, whatever decision you reach on this, it is confusing if people call a sinc-compensation filter a 'sinc filter'.
There seems to be a little bit of confusion here surrounding the term "sinc" function. Relevant to digital audio D/A conversion the mathematical sinc function [sin(x)/x] actually appears in two distinct circuit locations and as two distinct forms at that. The quasi-sinc function associated with DAC chip zeroth-order hold operation (sample and hold) appears in a frequency domain plot, while the sinc function associated with the oversampling digital filter appears in a time-domain plot. While the two are Fourier transforms of each other, they are not the same thing.
As far as NOS (digital filterless) versus OS (digital interpolation filter) is concerned, the interpolation afforded by oversampled filtering can provide theoretically perfect sine wave reproduction but only under stringent conditions which cannot be met in any practical system. Such conditions include, infinite waveform duration and infinite bandwidth limiting. So while perfect interpolation for the transient waveforms characteristic of music is not possible, I would agree that it still is far more correct, in a technical sense, than is non-oversampling. However. Spectrum analyzers do not appreciate music, people do. It's precisely in the context of human appreciation that NOS reveals merit.
Yes, there certainly are musical sounding oversampling DACs. I know because I've designed and built a few myself. Having also built NOS DACs I can tell you that it is far easier to achieve musically compelling sound via NOS than it is via OS. Said another way, at least in my experience, OS takes much more expertise and expense to deliver the musical goods.
This is not to say that I feel NOS is without flaws. I've commented before on what I perceive those flaws to be - primarily, a lack of focus and a tonal highlighting of the presence region. It's just that what NOS does right - a lack of musical glare and irritation coupled with great dynamic naturalness that's particularly apparent on crescendos - is so compelling.
As far as NOS (digital filterless) versus OS (digital interpolation filter) is concerned, the interpolation afforded by oversampled filtering can provide theoretically perfect sine wave reproduction but only under stringent conditions which cannot be met in any practical system. Such conditions include, infinite waveform duration and infinite bandwidth limiting. So while perfect interpolation for the transient waveforms characteristic of music is not possible, I would agree that it still is far more correct, in a technical sense, than is non-oversampling. However. Spectrum analyzers do not appreciate music, people do. It's precisely in the context of human appreciation that NOS reveals merit.
Yes, there certainly are musical sounding oversampling DACs. I know because I've designed and built a few myself. Having also built NOS DACs I can tell you that it is far easier to achieve musically compelling sound via NOS than it is via OS. Said another way, at least in my experience, OS takes much more expertise and expense to deliver the musical goods.
This is not to say that I feel NOS is without flaws. I've commented before on what I perceive those flaws to be - primarily, a lack of focus and a tonal highlighting of the presence region. It's just that what NOS does right - a lack of musical glare and irritation coupled with great dynamic naturalness that's particularly apparent on crescendos - is so compelling.
Did you mean:
Mainstream manufacturers are presumably trying to deliver a "reasonable" reproduction of the signal originally recorded, while putting focus on optimization of profit by using cheap plug and play silicon filters.
I bet that even my sister could put such a OS system together if I explain to her what the pin names in the datasheets mean
Also:
NOS can't claim anything.
NOS is neither a person nor a company so it can't claim anything.
In a totally flawed implementation on opamp swap diy level (no reconstruction filter) it doesn't even give a flat frequency response, but what would you expect ?
On the other hand, I claim to have NOS with flat frequency response and 60db/oct LP filter.
At the same time it is the simplest possible implementation, having copper wire in the signal path from Iout to the power amp input "exclusively", with a hand full of passive parts to ground.
Now let's see what you have got:
A chip that interpolates the data to higher bit depth so that not too much error would occur in the following filter ops. Rounding back to the desired DAC bit resolution and spitting it out at a much higher bit rate to help making the DAC more sensitive to jitter. Still, the signal needs some filtering which usually is done with opamps. This is the good news, opamp swapping is fun.
Important to notice: When you want to apply digital filtering to a signal without quality loss, you need a input bit depth that is higher than the output bit depth.
Unfortunately quite the opposite is the case with CD audio.
OS blurrs the signal.
60dB/oct in a passive filter is impressive! What is the amplitude and phase response like near the cutoff frequency?
A higher data rate may make the DAC more sensitive to internal jitter, but it does not necessarily make it more sensitive to input jitter as internal re-clocking is always possible.
OS digital filtering may need longer data words within the filters, but it does not need longer words at the input. After all, 16-bit data is 16-bit data whatever we do with it.
In the early days of CD the 4xOS with 14-bit DAC used by Philips was generally reckoned to sound better than the NOS 16-bit Sony machines. Were we all mistaken?
A higher data rate may make the DAC more sensitive to internal jitter, but it does not necessarily make it more sensitive to input jitter as internal re-clocking is always possible.
OS digital filtering may need longer data words within the filters, but it does not need longer words at the input. After all, 16-bit data is 16-bit data whatever we do with it.
In the early days of CD the 4xOS with 14-bit DAC used by Philips was generally reckoned to sound better than the NOS 16-bit Sony machines. Were we all mistaken?
Small margins and short life spans
The reason for this, in my opinion, include:
1. mainstream (consumer mass market) audio companies don't include a serious listening evaluation as part of the product design process.
2. To include what would necessarily be an iterative subjective listening evaluation as part of the design process would be costly for products with small profit margins and short life spans.
3. The academic based attitude that, if it measures good then, by definition, it sounds good too.
The reason for this, in my opinion, include:
1. mainstream (consumer mass market) audio companies don't include a serious listening evaluation as part of the product design process.
2. To include what would necessarily be an iterative subjective listening evaluation as part of the design process would be costly for products with small profit margins and short life spans.
3. The academic based attitude that, if it measures good then, by definition, it sounds good too.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.