Is this clipping?

[jkeny]

I bought a 24/96 track from a well known site & played it but found it's distorted. As this is a new DAC I wasn't sure whether it was the track or the DAC. I didn't download the other tracks from the album just this one. No other high res tracks (24/96, 24/88.2, 24/192) from other sites are distorted.

 

I'm new to audio analysis but I downloaded Audacity & ran an analysis - the clipping tool doesn't find clipping but the plot looks like it's clipped (or close to it) - graph below.

 

I ran Spectrum plot - it will only do 10 secs but it shows frequencies below 200Hz are above 0db (graph below) - is this not clipping?

 

[PeterSt]

Either Audacity makes a mess of things, or this is a "manipulated" file.

 

Disclaimer : it is hard too see without being able to zoom.

 

But look at the start area of the left (top) channel. That easily shows (?) that it clips at the positive side (top), but not at the negative (bottom).

In fact this is throughout, while the right channel at least doesn't clip so much. When you zoom it yourself I think chances are fair the right channel doesn't clip at all. The left channel may show clipping on the negative side too, but I am not sure.

 

Sure is : this is not right, unless Audacity doesn't show it correctly.

 

The other plot doesn't tell me much, other than it won't be possible to show above 0dB, but which it does. If you grabbed it from analogue it of course can.

 

Peter

 

[jkeny]

Well the point is I didn't grab it from vinyl - I bought it as a high res 24/96 track from a well known on-line shop. So I'm writing to them now that I have confirmation from a number of people as to it's structure.

 

[DSD_mastering]

When we do hi-rez transfers for download sites, I'm finding some poorly recorded music where the mic pre's are overloading and distorting, even though the levels are -6 to -9dB down! If you have "good converters", a few samples clipping here and there are inaudible. On poor converters, clipping can be unbearable.

So if you zoom in on the peaks, clipping has a flat plateau. You can't see distortion, but you can definitely hear it. Most modern recordings are severly clipped. I had to get an engineer to rewrite my EPROM algorithms to attenuate some transfers that I had that would severly clip. Sometimes, it's nothing you can do about it, other than a remaster.

 

 

Regards,

 

 

[PeterSt]

Hi Bruce,

 

While we're on the subject ...

 

If you have "good converters", a few samples clipping here and there are inaudible. On poor converters, clipping can be unbearable.

 

Can you try to explain this towards physics or anything that makes me understand this ? I know you're not an engineer or programmer (or at least I don't think so), so it's merely from a practice point of view;

What would a lousy converter do for poor jobs to make it more audible ?

 

If you don't have the answers really, allow me to share my point of view from the programmer/engineer *and* mastering angle ...

 

Peter

 

 

[jkeny]

Thanks Bruce,

That's what it sounds like - grunge. I got an email from the on-line shop who refunded the money but gave no explanation other than "that the source of this album was from the original studio master. We have not had any other customers write in about the clipping you are hearing but we will look into this.".

 

So it would seem to be badly remastered? The file is 75mB big which I have put here http://www.4shared.com/dir/19877524/29705768/Sister_Rosetta.html if anybody wants to download it & have a listen.

 

Is there a way to extract the first 30 secs of this Flac file for easier download & so I'm not contraveneing any RIAA rules?

 

[Synfreak]

Hi Peter!

 

Good DAs may have more "headroom" on the analogue side of their design (I/V stage, ...), so that these sorts of "digital overs" wouldn`t be distorted again after the conversion ...

 

Try to do a google search on "Intersample clipping".

 

One can use an oversampling EQ or samplerateconvertor on an workstation/editor (I use Wavelab) to get a grip what would happen on the convertion stage. ASRCs are also prone to be affected by this kind of distortion, if they aren`t designed with "headroom".

 

Another tool to show things like this is to use a digital oversampling meter.

I`m using the software (but hardware-driven) meters of my RME HDSP9632 card for this reason.

 

Cheers

Harald

 

[jkeny]

Can some of you who are more experienced in this phenomena have a listen to this track & tell me what your diagnoses is? It would be interesting to know as it is not something done by a guy on his Mac but apparently comes from a studio!

 

Edit: just checked & I see 2 of you have downloaded it - look forward to your views

 

[Synfreak]

I´ve listended to it on my PC (using WaveLab + RME HDSP9632 card + AKG k-500 headphones), and have not encountered distortion so far.

I´ve also checked for intersample overs, but the output level stays (on peaks) at around -0,4 dbfs (haven`t checked the whole peace, only abouzt the first minute).

 

Maybe your (jkenny) DAC doesn`t cope with the high level bass and start some clipping?

 

Or you might have an additional DSP (EQ or volume) in your playback chain?

 

Please tell us more about how you listen to this (software, hardware ...).

 

The "all over quality" of the file isn`t that bad. It is - to my tastes - a bit to much limited/compressed (dynamic range of about 12db), but otherwise pretty nice to listen too.

 

If you want to copy a small bit of the file, you could use audacity.

 

Cheers

Harald

 

[jkeny]

I listened to it through a Musiland USB DAC (which is a new purchase) & with my headphones (through the headphone out of the DCA) it sounded fine. When I brought it to a music studio to play for some interested people it sounded distorted through the line outs to an amplifier. I have just finished bypassing the whole output stage of the DCA with an output stage that consists of a transformer (The DAC is a PCM1793 with differential Vouts) so I'll test it with this shortly!

 

[PeterSt]

Hi,

 

There's a nice but very steady soft clipping going on at 95%.

The file is not all that much compressed. Looks quite normal to that regard. BUT :

 

There's a recognizeable pattern on many of the clipped peaks, and the width of the cut peaks is some 650 samples (calculated from 0.65ms @ 96KHz, I may have done it wrong). Because of the width of this, it will create a square.

Someone good at angles and math may be able to calculate how much the file has been limited, but to me it looks like 50% in decimal (big big waste of good sound).

For others, look at e.g. 1m48s7.80ms where such a thing starts.

 

I did listen to it one time, but not carefully enough to dislike it (the contrary I'd say).

 

Have to go now. More later maybe.

Peter

 

[jkeny]

I had to go out after my last post & just back now - I forgot to say that I'm using Foobar with no DSPs active. I'll be testing my new output stage later today & will report back!

 

[DSD_mastering]

I have downloaded the files and brought into my Pyramix workstation with DAD AX24 and EMM Labs converters There is no clipping. What you are hearing is a brickwall limiter. The most common one is by TC Electronics, which is on our 6000 unit. This does not exhibit the same pattern, so they must have used another unit with different settings. This one is definitely not flattering! I would have certainly attenuated the gain during transfer of the master tapes!

 

 

Regards,

 

 

[jkeny]

So Bruce & others,

Some of you hear distortion (I'll call it that for want of a better word) in this track & others don't. Is the distortion I hear an interaction between what's on the track & the DAC playing it. My question now is which DAC/system is the more accurate?

 

PS I haven't done the transformer O/P yet!

 

[jkeny]

Ok, I've done the transformer output stage (which sounds wonderful, BTW) & it cleans up the track to some extent but I still hear something unsettling in her voice - a sort of lack of clarity in her vocals as if there's a tiny fuzz around her vocals. This isn't pronounced & maybe I'm imagining it now?

 

I won't be able to listen to this in the recording studio (where it was more pronounced - running through a HBR? amp & ATC speakers) for a while.

 

[PeterSt]

Depending on how much the system can resolve, things like this can be very audible as "distortion" (in fact it just is). I, for one, can't bear JVC's XRCD (there are more types of it, and maybe I'm talking about one of them only). This too is what I call soft clipping, and this is about functionally cutting out everything above a defined level. This "cutting out" is done in some smart way, and just because of that it exhibits patterns. It becomes "music" on its own.

To me this (XRCD) sounds more than crunged and it merely is pure distortion (which it is of course).

 

In this case it looks way better to me than how XRCD is done (which btw is there necessary because of first boosting the lot), but something seems to go wrong regularly. So, at this going wrong there is no smart cutting, but it's just a wild cut, and the top of anything which has been a sine it just cut off, and now a square remains.

I tried to listen what high transients will suffer from this, and while it doesn't seem to be a frequency the only thing I can come up with is the kind of rim like ticks around that deep drum (or whatever that is, because *that* sounds strange to me but this isn't what I see cut). There are many more sounds in there which don't seem natural to me BUT it can be environmental noises, llike in the beginning of her singing I hear some spitting. Castagnette like clicks too later, and throughout.

 

Lastly, I think I know what you mean by the tiny fuzz around her vocals (but I didn't hear it here, or at least it didn't jump on me (but I only listened twice)). Try Diana Krall. Do you hear that kind of unnatural "come in bed with me" all over ? it is something in your system which can't cope. It is the most tough thing to get rid of, but it can and it is in the path towards the DAC (including the DAC). It is a digital something.

I'm not sure where the "transformer output stage" came from (was it in this thread ?) but it would be a typical "smearing" thing, while it also can backfire on you because of non linearity in impedance response (do I like transformers ? NO).

 

Peter

 

[DSD_mastering]

No.. I'm not an electrical/digital engineer. I am a mastering engineer that uses converters everyday to, unfortunately, most of the time, make things loud. Cheaper converters have a real audible "crunch" or hash when they clip. The higher end converters handle this much better, sometimes even being inaudible. You can press the Lavry Gold, DAD AX24, PM2 and the EMM Labs ADC8IV much hotter before the clipping becomes audible. I know some converters have "soft clipping" features. They must interpolate the intersample peaks differently.

 

Regards,

 

 

[PeterSt]

allow me to share my point of view from the programmer/engineer *and* mastering angle ...

 

Allright, my take on real clipping (which is not the case in this file);

 

First of all, nothing like "real clipping" exists if all would be normal. Clipping is just the rough cut of waves (say sines) which therefore loose their original shape, and whether you do that at -6dBFS or at -0dBFs really doesn't matter. Both have the same effect functional-technically and audibly. But there's also technical-technically ...

 

I will just give the most understandable outlay, while there are more (like at upsampling and more vague stuff) : Invert the absolute phase in the digital domain.

 

What technically happens in that case is that all numbers are multiplied by -1. So, what was positive (for voltage !) becomes negative, and the other way around. Sadly, PCM is registered in + and - numbers and for e.g . 16 bits this ranges from 32767 to - 3276 ... 8. Mulitply this by -1 and you'll get +32768 to -32767. But ...

 

But this cannot exist in 16 bits. What actually happens - and this is nature to how low level programming tools work - is that the +32768 becomes -32768 (when working in the two-byte domain).

 

This very special case of "flipping" can only happen at the outer boundaries of the number range. Thus, when we don't allow samples to have numbers above 37766 (plus or minus) we can do anything we want, and this flip won't happen.

 

What functionally happens when this "criterion" is not met, is that when the original clipping sound at the very maximum on the positive voltage side, suddenly goes ways down to negative. So -without digital attenuation- when the DAC outputs 2V at that maximum, it would go in one jump to negative 2V.

A DAC can bear this allright, but if everything can follow you might be able to hear a tick. Note that this tick should last for 1/44100 second (or 1/96000 etc.). Only when other things can't follow, it will smear to something larger, but it still would be a tick. Now :

 

Sadly, because clipping is clipping, this won't last one sample. This is the same as what happens in this file, it might happen for a couple of 100 of samples (but again and again and again like in this file), and now things will get nasty.

It is the highest transient imagineable, and purely wrong of course.

 

At a direct digital recording this can't happen, assuming the ADC and everything presents the correct numbers. But as soon as processing is involved, each stage of that can go wrong because it doesn't take this possible flip into account.

Example : Me. I am working on a (very) special means of upsampling, and after days of listening to it, suddenly I met a track that prohibited tiny ticks. Hmm, can happen. It was only two days later I met another one, but this was near to pure distortion. Well, you guess what happened ... I forgot about this flip. It only takes one sample that rounds to 32768 instead of it's original 32767 and there you have it. BUT it first needs a file which contains those maximum values ! And since nobody is going to convince me that a file was gained so that it has its normal maximum values at those maximum digital values, such a file is clipping in the first place (not one sample reaches the maximum, but many subsequent more).

 

If you are still with me ... A clipped file, may it be at -0dBFS or at -6dBFS (etc.) is harmless. It is not right to do for SQ, but it is harmless. However, once some processing got over it which doesn't think about this flip thing, all is destroyed. And this happens the most often ...

 

I don't think there is a reason a DAC can suffer from normal clipping. I see no technical reason (but maybe others do).

Also I don't see a reason why a DAC should be able to cope with this "flip clipping". The file is just wrong. But of course, the better it can cope with that, the better it is for your ears once you meet one.

 

Peter

 

Edit : PS: Before someone comes up with it, the processing of course doesn't happen in the 16 bit domain, but that really doesn't matter. So, for e.g. 32 bits all the numbers are 65536 times larger and there it happens exactly the same. And, while the 32 bit file is allright (if it is) at decimating to 16 bits the flip can happen again (dither would do it). But as said, I only tried to give an understandable example (digitally inverting).

 

[jkeny]

Peter,

Thank you - a very interesting analysis & it gives me a handle on what's happening - I wonder if there is a way to confirm this analysis by running Audacity or some such program to pick out these +high to -high flips?

 

You say that DACs can cope with this - do you mean DAC chips or the whole DAC including output stage?. The DAC chip in this unit is a PCM1793, a TI advanced segment type with differential voltage outs so maybe the I to V conversion that takes place inside the DAC doesn't have the slew rate to cope with this swing?

 

I would imagine also that the output stage slew rate would be similarly tested by such a wild recording? In this case it's a LPF & gain stage using a OP275 op-amp. I bypassed this with a LPF direct to transformer (which sounds wonderful) but still the track sounds wrong.

 

Up-stream of the DAC chip itself is a Cypress EZUSB chip that handles the USB2 pipe to the PC & an FPGA that processes the stream into I2S for the PCM1793. I wonder could things be amiss in any of these components? I guess the only way to tell is to hijack the I2S coming from the FPGA and route it to another DAC chip?

 

Hmm, all very interesting!

 

Any suggestions how I might resolve this?

 

BTW everything else I've listened to in high res or otherwise sounds excellent - Claire Martin, Buddy Guy, Joni Mitchell, etc I've stayed away from Diana Krall & Bruce Springsteen as I've heard the distortion in their recordings before using this highly resolving DAC.

 

[PeterSt]

One thing first : I hope it was clear (by my english and all) that what I explained does *not* apply to your file ?

So, I assume you responded in general.

 

You say that DACs can cope with this - do you mean DAC chips or the whole DAC including output stage?

 

Ok, sorry, I referred to the chips. Also, I merely referred to NOS/Filterless which, say, "better" follows what is fed to it (which won't apply tho the 1793). However, with filtering the result will be unpredictable (well, by me) because the strange behaviour will get "merged" with good data, and the effect may spread (over the number of poles the filter uses).

 

When we involve the output stage things get even more unpredictable, but supposesd everything can follow, your speakers will not, and short ticks (vinyl like) may become larger "plops".

 

To my own experience it may not be the slew rate of the analogue parts being able to follow or not, but it will be merely the "slowness" of all the electronics that can't cope. But, this will make the effect less bad (the jumps will be smeared). So :

 

You say that DACs can cope with this

 

... which comes down to just those ticks. I guess you appreciate(d) this the other way around. I mean, when the DAC cannot cope with it, the ticks will be less (because the output jump will be less).

Of course, when the DAC (chip) would not be able to cope, you'd get other anomalies like overshoots and all kind of strange stuff following from that (like an overshoot is more difficult to recover from than a not-overshoot (the output voltage will be extra high because of overshoot). Things will run behind, and all will become a mess ...)

Keep in mind, the flip theoretically can be 1 sample only, and it may take a 1000 samples to recover from that !

 

Sidenote : in your file you see similar, but I can't guess what caused it precisely : those strange (cut) patterns slowly decrease after the first peak, to drop back to normal after those (IIRC) 600 samples. Without analogue involved I'd expect the level to stay the same, but it doesn't. So, or a very strange algorithm at work, or analogue at work during the recording (or mastering from tape to digital, whatever).

 

Anyway ... I am not sure what "we" would want / desire from a DAC which is fed with plain wrong data in the first place (remember, not your file).

If it is about how a DAC can make sound your file right ... I don't think it can, nor would you want it. The latter is because what's in the file could just as well be normal data. How can the DAC know ? So, on this matter your file is just not good, and you will perceive the (false) harmonics from these "created" squares. Besides that, because these are squares they will produce more energy (math on sines and stuff). So, the wrongness will even be profound.

 

The fact that you may perceive wrongness while others do not, is a virtue ! Thus, wrong is wrong (in the file), and when you hear that, what is wrong with it ? Keep in mind though, that in my own case I'm using NOS/Filterless which in this case (and my amps) is allowed (24/96) so no matter what, I'll have completely different output opposed to you. So, possibly we could say that when to me things sound rather normal (while out of all I don't filter anything) in your case it can be the filter which goes bananas. Remember, the filter tries to reconstruct something which never existed !

 

Lastly, don't underestimate this transformer;

Although it is one means to apply the I/V (I think that's what you're talking about) it can create a sibilance. Maybe not by itself, and maybe it overexpresses things otherwise not 100%, but ... just measure it if you can. It won't have a linear response in the first place, and now it is up to the pure coincidence of where frequencies are whether they are over- or others are underexpressed, both giving a similar effect (unbalance which may resonate and women voices are the first to incur for just that).

 

Let me know if you think I'm way off all over, because to me it seems you may know about this more than I do.

Anyway, and let this be the most important, you are able to perceive the wrongness of a file which just is nothing less than wrong. This is a good thing. Also, there is nothing to improve on this matter, apart from the file ...

 

Peter

 

[jkeny]

OK, I did think you were talking about my file rather than theoretically - so my response was in relation to this file. I understood that explanation but this doesn't apply to my file or does it? Is there any software analysis tool to test this?

 

Sidenote : in your file you see similar, but I can't guess what caused it precisely : those strange (cut) patterns slowly decrease after the first peak, to drop back to normal after those (IIRC) 600 samples. Without analogue involved I'd expect the level to stay the same, but it doesn't. So, or a very strange algorithm at work, or analogue at work during the recording (or mastering from tape to digital, whatever).

 

I looked at the file in Audacity at the timestamp given earlier in the thread and all I could see was that the signal was maxed out for a period of time - is this the square wave you are talking about & what is the pattern you see around this?

 

I'm not using the transformer as IV - the DAC is a differential Vout with 1.4V bias on the line so the trafo acts as a Diff to single ended, band limiting & DC blocking tool - I could use a capacitor & have tried it but need to do more experiments.

 

[PeterSt]

Sorry about the misunderstanding, but as implied and said, it is only better for you (recognizing distortion and all).

 

I looked at the file in Audacity at the timestamp given earlier in the thread and all I could see was that the signal was maxed out for a period of time - is this the square wave you are talking about & what is the pattern you see around this?

 

This is not a pure square wave, but if you compare this to adjacent peaks (at exactly the same maximum) you will see there is a difference; the other peaks end in a literal "peak" (comprising of one or a few samples max) while this one indeed maxes out for a period of time (600 samples). Now, and only because this is a repeating pattern (just scroll further) imagine that originally this has been a peak like the others, but way up beyond the boundaries of the file (hence beyond the max digital level). Then it would have been a sine, right ? Not so now, because the top has been cut of it, and thus what remains is a peak with a square top. Let's forget about how it will sound, but let's recognize that it can't sound as how it ever was without that cut.

 

As a bonus : now think about how "compression" works out. Everybody talks about less dynamic range, right ? Well, inherently this is true, but what happens in practice is this :

 

Whit redbook (but in the end with everything) there are these smallest steps in volume digital can take. Just look at those longer lasting peaks ... Now imagine to squeeze down the maximum level (which is necessary when first the whole lot is boosted). That gradient you see at those strange peaks - now slowly decreasing its level - will be squeezed to a plain level all those 600 samples. Now the top of it will be come real flat, and the ever existing sine has become a real square at the top. This won't happen at the highest levels only, but everywhere "down the line" (the Y axis). It is *this* which turns compressed digital into pure distortion because of all the sine like waves squeezed to square like waves. Everything will start to exhibit harmonics, originally not there.

As said, as a bonus.

 

so my response was in relation to this file. I understood that explanation but this doesn't apply to my file or does it? Is there any software analysis tool to test this?

 

No, it did not.

I'm not sure whether analysis tools exist for this, apart from what I wrote myself. In the case of your file it won't show anything abnormal because it can't be recognized. It would when that gradient (down) would not have been there, but it just is. I can tell you though, that 95% of what we'd call "Lounge" exhibits very strange behaviour (like the Budha series, Cafe Noir, Paris, and the like). I don't know what it is, but it looks like some smart guys know what it is that gets us into a trance. This is always about samples "sticking" to certain values, those values being the most recognizelable for computing stuff, like 256, 512, 768, 1024. I even own series of albums that won't go UNDER a certain voltage level !! (throughout, and I mean that voltage between + and - 100mV just doesn't appear, so that always jumps 200mV from + to -).

 

In any case (well, that I know of, so I must be careful) I recognize that something is wrong. So, without doubt, and in fact like you did, once I hear something I don't want to dedicate to my software or DAC, I dive into Wavelab (or AudaCity for that matter) and in 100% of cases there's the AHA !

After that, I ditch the album. And there are so, so many :-((

 

Tonight -as I do more often to actually test my own gear - I have been listening to end 60's stuff only. It's a kind of guarantee that nothing has been mangled with. Well, nothing sounds better, and then I'm talking about CCR and the Beatles (no remaster to name some important thing). We thought at those days they were not up to it ? ha, yesterday I tried a 1957 Jimmy Smith. I'm not sure the stereo was faked, but the resolution beats anything from today (think about that compression again, which *eliminates* resolution).

 

As far as I can tell thus far, the only one from today I trust is Linn Records. At least I don't recall to dive into their files to look what's wrong. The contrary. Dynamic range is superb, but, they seem to overdo it and just go for that (mind the jazz and seemingly over expressed cymbal/drum stuff). I must say though, everything coming from Norway seems to meet these requirements.

 

Peter

 

 

 

 

[jkeny]

Thanks Peter,

At least they gave me a refund without squabble - I'll bring this thread to their attention & maybe they'll talk to the studio or whatever source they got it from - perhaps this may lessen the occurance of this problem in the future (the eternal optimist)

 

An interesting fact about the Beatles recordings from the 60s is that George Martin & the recording engineers focused their time on the mono versions of the songs because they were meant to sound good on the playback equipment of the day (typically mono) & the stereo versions were just given cursory treatment. No wonder the remastered Beatles sounds so much cleaner in most instances.