Inside High-res audio: PCM vs MQA vs CD: 2L Sampler Comparison - YouTube

hello everyone this is amir  from audio science review  

this is another episode in looking at  what's inside high resolution music.  

this analysis is a little different in that it's  by request to look at the high resolution samples  

provided by the great label 2l, the nordic  sound, as you can see their website over here  

they distribute pretty high quality music but i  know the service they provide which is wonderful  

for the community is this sampler page where  you can actually download music for free at  

practically every sample rate and resolution that  you might be interested in all the way from cd  

quality to dxd and stereo to 352 kilohertz  sampling rate so it makes for a wonderful  

workbench to look at what's inside high  resolution music they're nicely tell you  

what the source format is in this case is the dxd  so this large file at 410 mega megabytes is the  

original master in stereo then their versions  in in stereo i have the 96 Kilohertz version  

and i've also downloaded the mqa and cd so let's  go ahead and look at the these files in addition  

i'm going to start with 96 kilohertz sampling rate  and let me move this back to somewhere interesting  

and before i even play the music if you've watched  my previous episodes you already see the problem  

um look at these three peaks in here they're  occurring 38 kilos on up and obviously they're not  

musical information as i've explained before music  has this nice exponential roll-off more or less  

and there is no musical instrument that just  generates these high very narrow pure tones almost

signals way into ultrasound whoever produced  and mastered this music just like me and the  

rest of you can't hear those so what they didn't  hear they didn't catch but they should have just  

done this simple objective analysis that this  window of frequency analysis exists in every  

audio application so they could have just pulled  this up found these and eliminated their sources  

what are the sources the pure ones again as  i've explained before probably the scan rate of  

monitors bleeding into the sensitive inputs of the  analog to digital converter these wider noise ones  

are less pure you know there could be alarms  in the building or if it's captured from tape  

which i don't think this one is um it could have  been bias signals that are used on real-to-real  

tape recorders but regardless it's not our job  to discover what they are it is theirs they  

could easily turn equipment on and off and until  these peaks go away and not deliver these to us  

another nice phenomenon we see this good  learning exercise is that as we get close to our  

final frequency of 48 kilohertz which is the  bandwidth we can have in a 96 kilohertz sampling  

you can see this nice anti-aliasing filter that  necessary to get rid of what is called aliasing  

and it pulls things down to -180 dbfs which  is great so that's natural that's good  

what's not good is this so let's go ahead  and play the music and see if we can figure  

out what the usable spectrum of music  is how much of this we really need it

so clearly this part is going up  but if if we let it play we see that  

it actually does use uh the rest of the  uh most of the rest of the spectrum also

let it get louder see you can see this part is  jumping in correlation with music and that's  

important lesson to learn music is not a random  thing an instrument of plays at this frequency  

has overtones that play here and the two  have to be correlated if there's nothing here  

then there's just noise but a real  instrument generating high frequencies  

will extend here is a good example you can see  how naturally extends so i'd say eyeballing it  

about 34 kilohertz is is what we need to preserve  doubling that to get the sampling rate that would  

have been 68 kilohertz we don't have 68 killers  uh sampling air 8 and the nearest one would be  

88 kilohertz so if this kind of music or the  original capture should have been 88 kilohertz  

and uh that unfortunately would have  captured some of these tones um but  

wouldn't have generally delivered to us bunch  of extra noise now here the extra noise 88 kHz  

the difference between 88 kilos and 96 kilos  is not that large but now let's go to the  

original master and uh see how wasteful this is um  so this horizontal scale is hugely expanded this  

is 352 kilohertz sampling so the bandwidth  is now half of it or 180 kilohertz or so  

and uh if we go back to the previous display  the previous display truncated around here  

we still see our original ultrasonic tone in  here still present and but what's right after  

that we see this hugely rising noise pattern and  that's typical of low bit rate of one bit sampling

because we don't have sufficient number of bits  normally tons of distortion would be generated in  

the audible band there's a process called noise  shaping where we take noise out of the audible  

band and shove it into ultrasonic range the idea  being that we can hear noise in the audible band  

but we can't hear it in the ultrasonic range  so all the noise has been emptied out of here  

and shoved into ultrasonic by definition then  we don't want to play this this is noise that  

is not musical information it should be thrown  away and the idea was that we don't hear it and  

throw it away indeed some SACD players or i  don't know if all of them have a selectable  

filter to actually analog filter to actually get  rid of this noise and this hash that's coming out  

yet in the case of high resolution digital  download we're now handing all this to  

users and people happily go for the larger number  being better 352 is much nicer sounding that 96  

kilohertz so let's get all that and attempt to  play that so nothing good comes out of playing  

this noise on your stereo the amplifier may  oscillate make it distort your twitter may  

cook may distort may cause intermodulation it may  actually generate distortions in the audible band  

or may do nothing and that's the best  case situation where it does nothing  

no reason to distribute all this no reason to  originally capture that such a high sample rate  

again an 88 kilo sampling rate would have  been plenty would have been just nice and  

it would have truncated everything to the  right so no goodness comes out of this  

now scaling all the way down to what would have  been a cd release let's go ahead and listen to  

the uh 16-bit 44.1 kilohertz sampling so now our  horizontal scale uh truncates down to 22 kilohertz  

or so and uh let's play this content again  and let me see if i can find yeah there we go  

oops pay attention to here when it gets quiet

there did you see this peak point now peeking out  uh this is a pure tone again or a set of tones  

they're not musical there's not part of music we  can see the the level of the music is naturally  

dropping off but this part is not it gets masked  when the music gets louder than it but otherwise  

it peaks out this is much more problematic  than the high frequency ones because it's in  

the audible range it's about 19. some kilohertz  was it caused by i don't know uh some instruments  

something electronic is generating it uh it's  unfortunate i mean that's this is egregious uh  

means even in cd we're hearing it let me play that  some more there it is you can see it very clearly  

uh picking up now it's very low level but you  know still we want to have good quality uh there  

it is i mean it's just unmistakable that this is  not natural nothing here does that you see this  

nice exponential clean uh uh roll off and uh she  should have sailed into the sunset the same way  

so unfortunate unfortunate for sure now lastly  let's get into a new form i call MQA master  

quality audio it's a new format being introduced  and it's claimed to fame is that it encodes high  

resolution audio high sample rate audio into low  sample rate distribution format in this case 24  

bit 48 kilohertz i believe let's open that up now  this one is actually 44 kilohertz sampling rate  

and but inside this you can actually decode  and get the what used to be above here and uh  

how does it do that well if  you go back to the original  

display we can see how he's going to do that  this area is the audible frequency range  

and we you know PCM format which you know 16  or 24 bits is spent encoding this but as we  

get to our ultrasonic if you look at the level  z in here the 125 dbfs these are very very low  

level signals no way do we need to have a dynamic  range that describes them all the way up to zero  

i've never seen any high resolution content  as much of anything uh above these very very  

low levels so what we really need is just a  few bits to encode the this spectrum that's  

indeed what MQA does um if we go back to the  MQA signal and uh let me zoom out a little bit

we see what looks like a cd rate but then  we see a strange phenomena over here where  

they actually the noise floor rises in  here why is the noise flow rise in here  

because in here MQA actually attempts to encode  the information it needs to generate the rest  

of the bandwidth to the right and so in that  regard it is degrading the cd rate no question  

about it this now this is very very low level  you know audibility is a different question than  

objective measurement but if we play this  we can see that this part of this rising  

spectrum did not exist now luckily  a lot of it is above ultrasonic and  

again for us old folks we don't hear it but  the rise starts around 16 kilohertz or so  

so clearly the very high frequencies have had a  random noise added to them and that information  

actually isn't random it's actually uh correlated  to what spectrum it wants to generate above  

this point but it's been randomized to make his  audibility very difficult um you know i haven't  

done any double blind tests to see if if i can  hear that i believe others have and they sort of  

kind of could um set the extreme this this may  be audible depending on the content you play  

so there's a price to be paid it's backwards  compatible MQA is that means just as i'm doing  

right now i'm playing this without mqa decoding  yet the music plays and i hear it but there's a  

compromise in embedding that in here i believe  for other content they use 48 killer sampling  

which allows this to be shoved further up above  the audible bed right now it's at uh 16 killers  

so that's it uh hopefully you're finding these  uh episodes uh useful uh i know music's about  

art and enjoyment and uh you know not about what's  going on in here but i think the industry needs to  

do a better job in producing and delivering high  resolution content to us that's pristine that's  

clean um because that's what we pay for we pay  a lot more for these high resolution downloads  

they're not on sale for little next to  nothing on amazon to go and buy the cd  

and so we expect and rightfully  so uh pristine quality masters  

and uh not something that has extra tones in  it and noises and what have you okay again  

this is amir from audiosciencereview.com  talk to you later in a future episode bye