Δεν χρησιμοποιούνται πιά πουθενά οι παλιοί (και πολύ ακριβοί)μετατροπείς...
Γι αυτό ο παππούς Sony scd 1,κοστίζει όσο σχεδόν,όταν ήταν καινούργιος μετά τόσα χρόνια...
"First thing we need to do is define noise. In this area noise has a particular meaning that has some subtleties. Noise can be considered as simply that which is not part of the desired signal. Sounds trivial, but it can be further subdivided. Crucially all distortion components are noise. So to continue.
The noise people think of is thermal or shot noise in electronics. This is typically characterised as AIWN (additive independent white noise.) This terminology is very neat, as it conveys a crucial property. Additive independence means that any two noise sources may be summed, and you will still simply have AIWN. Crucially you can't subtract one AIWN source from another and lower the noise level - the noise floor rises same as always. This is because the noise is no internal correlated components. It is impossible to ever predict the value of a noise source based upon the pervious values.
The converse is noise that is auto correlated. Noise whereby after some analysis (a Fourier analysis for instance) we can derive a deterministic mechanism to predict the future values of the noise.
A real noise source may of course consist of a combination of the two. A useful thing to do is to define the AWIN noise as the residual after we have done all the analysis we can to predict the future values.
Next we need to consider whether the noise is correlated to the desired signal. AWIN that is not correlated to the signal is what is commonly thought of as noise. However we can expand on this. One can have AIWN that is correlated to the input signal - for instance shot noise in a junction will depend upon current in the junction. Considering auto correlated noise, we can have noise that is not correlated with the signal - a typical example is mains hum. Finally, we have auto correlated noise that is correlated to the signal. This is what most people think of as distortion products.
Now back to sigma delta (i.e. one bit ADC and DACs).
All sigma delta converters use noise shaping. The noise we are shaping is not AIWN. The noise is the signal correlated noise, in particular the distortion products (quantisation noise) of the converter.
The common misunderstanding of a one bit ADC is this. The notion is that the converter consists of a single comparator that every sample delivers a single bit, indicating whether the signal is greater than or less than the threshold. It is reasonably well understood that such a converter cannot work unless the input is dithered with AIWN of average level one half the input signal range. (This dither is used in all multi-bit ADCs where the level of dither must be one half the least significant bit's level.) This extension of the understanding of how multi-bit ADCs may work to the one bit ADC is sadly wrong.
A sigma delta converter is a feedback device. In its simplest form (first order) they consist of the following.
A sample and hold. This captures the signal for the time of one sample period.
A subtractor. The signal from the sample hold is fed to this, and the output from an internal 1 bit DAC is subtracted.
A discrete time analog filter. Typically simply an integrator. The output of the subtractor feeds the integrator.
A quantiser. A comparator that outputs a one bit value based upon the input value. Indeed this is a one bit ADC. This is fed from the filter above.
A one bit DAC. This is fed by the quantiser above, its output goes to the subtractor above.
The output from the quantiser is also the output from the ADC. The whole system runs at the oversampled rate - for SACD at 2.8 MHz.
The loop formed by the quantiser and the DAC contain a fast moving approximation to the much slower moving input signal. The integrator will force this approximation to always move in the correct direction so that its own input toward a long term average of zero. Consequently the average error between the digital stream and the analog input will approach zero.
In a conventional sigma delta ADC the single bit stream is then fed to a decimator that will produce a lower sample rate with a correspondingly wider sample width. The SACD process omits this stage and records the output of the modulator directly.
A delta sigma converter does not use dither in the manner that other ADC systems use it to decorrelete the LSB. The manner in which the internal DAC is subtracted from the sampled input and passed to the integrator essentially has the same effect. There is a need for a form of dither, and that is to prevent the occurrence of limit tones. However this dither does not need to be AIWN, and indeed can be an auto correlated signal that can be crafted in such a way as to be eliminated mathematically in the decimator. I do not know what is used in the SACD process.
It is possible to create multiple order sigma delta converters. This basically consists of placing an additional subtractor and integrator stage ahead of the existing ADC. The same DAC output is fed to all the subtractors.
Analysis of the performance of sigma delta converters is a bit fraught. It is possible to treat the error signals in much the same way as AIWN, and if one does the improvement in resolution is roughly 1.5 bits per doubling of sample rate for first order, and 2.5 bits per octave for second. Contrast this with one half a bit per octave without noise shaping.
However there is a cost. Intrinsically the process does add more noise to the entire system. Thus it does not make sense to sue noise shaping unless the oversampling rate is high enough to amortise this basic loss.
Also there are intrinsic issues in physical implementation that prevent actual realisations from attaining the full benefit. As a rough approximation perhaps 3 bits worth of resolution.
As has been discussed the nature of noise shaping is to push the noise into much higher frequencies - where typically it can be filtered out. The shape of the noise shaping profile is reasonably distinctive. Higher order designs have a higher order function shape - thus may have lower noise in the pass band, but much faster rise in content out of the passband.
In a typical system the output of a sigma delta ADC is decimated and yields a conventional multi-bit sampled stream. Many 44.1/16 or indeed 192/24 streams are so created. When played back the opposite occurs. Indeed you can take the entire chain and reverse it. The SACD process takes the attitude that there is no point in decimating the output of the modulator and then simply interpolating it back again in the player. So the intermediate form is the modulator output.
I am interested to hear that critics compare the same recording - one on SACD, the other on conventional CD. There is a crucial issue. The CD will have been created by decimating the SACD stream. Sony call their mechanism SBMD. Check your CD, if it is SBMD you are listening to a SACD recording that has been down converted. If you prefer the CD, you can't blame the sigma delta ADC, or its noise shaping.
The other interesting device to consider is the dCS Verdi-Purcell-Elgar Plus combination. The Purcell will up-convert. It will take 16/44.1 and create DSD. The Stereophile review actually preferred this over direct 16/44.1 Curious"
