Digital technology is maturing rapidly; quality that only a decade ago might not even be available
at the highest of high end prices now is available at prices that most people in the western
world can afford.
That doesn't mean that all equipment is of the same level….
When we look at digital music reproduction, like with cd-players and network players,
the sound quality is greatly defined by the jitter performance and the quality of the
reconstruction filter.
There are of course underlying problems that form the base of these performance issues.
In this first part we look at the filtering that is unavoidable in digital music reproduction.
The anti-aliasing filter is needed during recording to comply to the Nyquist theorem
that states that a band limited signal can be perfectly sampled when the sample frequency
is at least double that of the highest frequency in the band limited signal.
Sounds great, but what does it mean?
In his theorem Nyquist defined 80 years ago that a signal might not contain any information
above a given frequency.
When the cd was developed in the seventies of the last century 20 kHz has been chosen,
meaning that a sampling frequency of at least 40 kHz should be used.
At that time there were no recorders that were able to record bits at that speed.
For we get 44,100 samples of 16 bits, times two channels.
That is 1 million four hundred eleven thousand two hundred bits, 1.4 megabits per second.
So the technicians built a device that first converted the analogue signal to a digital
signal and then converted that into a black and white video signal containing black and
white squares to represent ones and zeros.
That way a video recorder could be used for digital audio.
To have the digital data fit inside the video signal it was convenient to use 44,056 or
44,100 samples per second, depending on the local video system - PAL or NTSC.
Later on the global standard was set for 44,100 for CD.
Choosing a slightly higher sampling frequency while maintaining 20 kHz for audio bandwidth
means that there is some space to do the filtering in.
For sampling at 44,100 Hz allows for a max bandwidth of 22,050 Hz.
CD's use 16 bits to code amplitude and that gives a dynamic range of 96 dB.
In nature there are no signals that promptly stop at 20 kHz so a filter has to block any
content above that 22,050 Hz.
Using 20,000 Hz as upper limit, it gives us 2,050 Hz to filter in.
2050 Hz at 20,000 Hz is a tone distance of about a single note, one white key on the
piano.
And in that small space we need to attenuate the signal by 96 dB's.
Normally filters are specified in dB attenuation per octave.
An octave is roughly 8 white keys, so that reconstruction filter would be 8 x 96 dB = 768
dB per octave.
Many audiophiles already panic when a loudspeaker uses 24 dB per octave filtering!
And here the filter needs to be 32 times as steep.
It is impossible to filter that steep without getting all kinds of artefacts, resulting
in time smearing.
On the analog digital conversion often low bit converters are used like a one bit converter
as used for SACD and thus DSD.
That signal is then decimated to 44.1 kHz or a multiple of that and 16 or 24 bit, depth,
depending on the delivery model.
But if you think it therefore doesn't need any filtering, you're wrong.
If you down convert that signal to 44.1 kHz, you still need to apply that same anti aliasing
filter at 22.05 kHz.
And since a one bit system only has a dynamic range of 6 dB!!!, noise shaping has to calculate
the noise to somewhere outside the audio band and thus inaudible.
Another option is using PCM at higher sampling rates like 88.4, 96, 176.8 and 192 kHz or
even higher.
If you then use a less steep filter that starts at 20 kHz, that filter might sound better.
And every doubling of the sampling rate gives you an extra octave to filter in.
Unfortunately the solution used is chosen by the sound engineer or record company.
On playback equal filtering has to be applied to prevent the reconstruction of the analog
signal from going wrong.
And also here the same problem arises: very steep filtering.
But here you can have some influence on the choice.
Of course, if a recording is only available at 44.1 kHz, you can't change that.
But when choosing your DAC or player you can choose one that uses filtering that pleases
you the most.
Filtering always is a compromise and the cheaper the price of the product, the more the designer
had to compromise.
In a simple DAC or player, the digital analogue conversion is done by a DAC chip that also
does oversampling.
Given the price OEM manufacturers are willing to pay, there is only limited silicon available
for that DSP function.
As a result relatively poor upsampling filters are implemented.
Chip manufacturers do offer a solution though: they facilitate the use of an external processor
to do the upsampling and digital filtering in.
This potentially increases the sound quality considerable.
You now know the first reason why it is stupid to think that all DAC's using the same DAC
chip are of the same quality.
But even if an external processor is used, the power of that processor and the quality
of the code have their influence on the sound quality.
And at the same quality, different kinds of filters exists.
On some DAC's and players you can even select what filter to use.
Choices involve things like priority on linear frequency response, linear phase, slow roll
off and so on.
This sounds interesting but I notice that often the linear phase version sounds the
best.
At least, to my ears.
The use of low bit converters implies that upsampling must be used.
This means that the precision in amplitude is shifted to precision in clock signal.
In layman's terms: where the classic ladder converter needed to have the smallest voltages
very precise, the low-bit converters need to have a very precise clock to achieve the
same quality.
And both ways are equally difficult.
Again, it's not the principle but the execution of it that defines the quality.
NOS stands for Non OverSampling.
Files are converted to analog at the sampling rate of origin using a 16 or 24 bit ladder
converter.
Here each bit switches a given voltage where each lower bit represents half the voltage
of the bit above it.
So the most significant bit represents 1 volt if pthe maximum output is 2 volts.
Funny enough the other 15 or 23 bits together stand for the same voltage as the most significant
bit so that when all bits are 'on', the output voltage is 2 volts.
The second significant bit represents 0.5 volts, the third significant bit 0.25 volts
and so on.
When you go down to the 16th bit, you see that it represents only 30.5 µVolts and the
24th bit is the absurd 0.12 µVolts.
It might be clear that there is no chance of getting that voltage out of the thermal
noise.
And even if they would manage to do so, the thermal drift of the resistors that define
these voltage would cause irregularities.
Therefore you often see a number of NOS DAC's stacked piggy back style.
That way the output increases and the irregularities even out.
A small number of manufacturers build their own filtering and DAC circuits using powerful
DSP's.
In many of cases the code is written by third party programmers that have worked years on
their code.
The resulting filters usually offer high to very high time resolution.
Given More's Law processing power doubles every two years en we are at a point now that
these filters are becoming affordable.
I love the improvement MQA offers.
See my videos on MQA.
It doesn't work on all tracks always but it does in most cases, provided you have an
MQA DAC.
Even if the album you want to play isn't available in MQA, having an MQA DAC has its
advantages.
I have explained how the anti aliasing and reconstruction filters cause time smearing.
Using MQA files these are corrected to a large degree for these errors but even non-MQA files
benefit clearly from the MQA filters according to me and some colleagues.
So what is the best DAC and what is the best reconstruction filter.
The answer will not be popular: it depends….
It depends on the implementation, the other equipment in your stereo and your personal
preference.
Even all the money in the world can't buy you the perfect filter, so you have the pick
the solution that, when used in your situation, has artefacts that bother you the least.
Now, don't get me wrong.
What can be achieved today is immensely better than what could be done a decade ago.
I could easily live with a large number of solutions available today, as long as they
are designed to perfection - or as close to perfection as possible.
Don't forget that there is another big factor defining the quality: jitter.
That even has a bigger influence - at least in my opinion.
Therefore part 2 of this video is all about jitter.
So if you don't want to miss it, subscribe to this channel, or follow me on social media.
If you liked this video, please consider supporting the channel through Patreon or Paypal.
Any financial support is much appreciated.
The links are in the comments below this video in Youtube.
Help me to help even more people enjoy music at home by telling your friends on the web
about this channel.
I am Hans Beekhuyzen, thank you for watching and see you in the next show or on theHBproject.com.
And whatever you do, enjoy the music.
Không có nhận xét nào:
Đăng nhận xét