Jump to content
IGNORED

Upsampling: Just An Illusion?


 Share

Recommended Posts

I am currently feeding my lossless iTunes library through the optical channel into a Marantz DAC94 [a classic over-engineered old-school DAC utilising the almost mythical TDA1541s], which is then passed on into an active Linn System [ikemi / Kairn / 2xLK140 / Ninka].

 

I have had the opportunity of hearing several of the newer UK DACs with their multifarious upsampling techniques. I have not appreciated the new sound.

 

I am going to draw an analogy and i would be interested in what people have to say.

1. Is upsampling the audio equivalent of video upscaling?

2. Is Upscaling just visual trickey? An illusion that consistently fails to convince us?

 

1080p televsions are all the rage right now, but [in the UK at least] DVDs are not actually created/formatted to output at 1080p, but rather at either 720p or lower.

 

Watching a movie at its original [ie formatted resolution] means you see the original image. Keep your plasma at 720p or 576i [depending on the format of the DVD] and you get a clearly defined picture. When you upscale the image you may get "more" on the screen but it's not real is it? it's the upscaler playing digital tricks. it's a computer programme and/or an algorythm trying to "fill in the gaps" of a series of moving images.

 

And these predictions are often wrong. this, to me, is most apparent on close ups of human faces. upscaling just fails to accurately predict the range of movement of a human face. the movement is just too infinite and random. it's as if, at any moment, the blur and "add ins' of upscaling make the face look like its about to morph into something else. we see pinpoint accuracy in the parts of the face that aren't moving and strange blurs in the bits that are [most noticeably the lines of the forehead.] i often feel like i am watching an early 90s science fiction movie: human faces are forever on the edge of mutating into terminators...

 

personally i would rather watch and listen to what's actually there: not some digital predictive illusion of continuity. if there's a rock slide i want to see the rocks sliding, not a computer approximation that decides its actually a waterfall.

 

and i think that modern upscaling DACs may do something similar.

 

just a thought...

 

Link to comment
Share on other sites

people have to say.

 

Panasonic PXP 42 V20; Panasonic DMP BD35; Sky+ HD Box. [br]Optical out from Asus P7H55-M into AVI ADM 9.1 speakers. [br]\"Music will provide the light you cannot resist\"[br]

Link to comment
Share on other sites

Hi iansen - That's certainly an interesting analogy and I've thought the same thing in the past. Since I'm not an engineer I won't attempt to talk technically about this topic. I will say that upsampling is done for many reasons, including reducing jitter. There are a couple conversations around here where this topic has been discussed and there really is no consensus on upsampling. Good, bad or ugly it really depends on the implementation by each manufacturer. Some manufacturers MAY be looking for a higher number on their spec sheets while others MAY be attempting to improve sound quality.

 

Founder of Audiophile Style

Announcing The Audiophile Style Podcast

Link to comment
Share on other sites

Upsampling audio is really different than upsampling video. Audio is simpler, only voltage, where video has color, intensity etc..

 

Audio upsampling algorithms run the gamut. Some are better than others. In general I have found that hardware upsamplers are inferior due to limited functionality. The best are upsamplers that re-write the file and take as much time as they need to do the computation, such as Adobe Audition and R8Brain. These can take overnight to do an album. Next best are upsamplers that do it on the fly as you play the track, such as SRC and SSRC in Foobar2000. They have less time and require low-latency and lots of compute power, so this limits their effectiveness. They are superior to hardware upsamplers though. SRC is mostly what I use.

 

Upsamplers attempt to "fill-in" the missing musical information. They take a "best-guess" at what it would have been based on the previous and next information. If they are really good at this, the difference between the actual 24/96 recording and an upsampled from 16/44.1 to 24/96 is actually very small. This means that dynamics, smoothness and other positive attributes of the original hi-res track will be in the upsampled track. Listening tests on good upsamplers verify that this is indeed the case.

 

Steve N.

Empirical Audio

 

Link to comment
Share on other sites

Oh my, that is my favourite area: improving sound with software :)).

 

As Steve said it before me, it is different compared to video. But one thing is the same for both: the quality differs between several upsamplers or upconverters. I don´t know how video upconverters function so I´ll concentrate on Audio.

 

Mostly upsamplers are used to get around the limitations of a cheap or bad DAC. With limitations I mean that these DACs have a bad performance when anti-aliasing is performed. In order not to have aliasing effects in the listanable audio area (20 - 20.000 Hz) everything above must be eliminated. So every DAC uses a very steep frequency cut-off at 20.000 Hz. Sometimes these algorithms are not that good so in the end they are hurting the sound. With upsampling you shift this 20.000 Hz of anti-aliasing further away to 40.000 Hz where you can´t hear it anymore. What you can´t do of course is recreating a signal that came from an original 24/96 source that was downconverted to 16/44.1. An upsampler or resampler uses an interpolation filter in combination with an anti-aliasing filter and fills up the spaces between the original 44.1 kHz with zeros when upconverting to 96 kHz.

 

So, an upsampler really is an oversampling filter without the D/A converter, because that is also what oversampling does. That means you could have frequencies over 20.000 Hz, but they are erased because of the believe that they can destroy loudspeakers or create problems for the amplifier. With 95 % of upsamplers you have a frequency response that goes from 20 to 20.000 Hz - beyond that is nothing. With good upsamplers the quality of the outcoming signal is the same as the original 16/44.1 data.

 

However, some people in the industry always knew that there could be more. WADIA & Pioneer for example. The spline algorithm is nothing else as an oversampling filter where the following anti-aliasing filter is not steep anymore and very soft. There the aliasing components are deliberate. Aliasing mirrors the frequencies starting with 22.050 Hz up to the double amount of 44.100 Hz. The data mirrored is the data from 20-22.050 Hz. Pioneer called this "Legato Link". Wadia and Pioneer did this to improve impulse response. The major drawback of steep aliasing filters is audio impulse that has pre- and post-ringing, means it adds ripples to the audio signal just by filtering above 20.000 Hz. With things like Legato Link this could be avoided - in the end you´ll get a frequency response up 44.100 Hz (faked of course) and nearly perfect impulses.

 

There is only one resampler on the market that can be configured to achieve this thing written above: iZotope RX Resampler. With this you can shift the cutoff frequency up or down, influence the kind of ripple (more or less pre-ringing) and configuring the steepness of aliasing cutoff. It is fast and produces perfect results, it really is very good. You can also use it for re-creating your standard resampler like SSRC or else. Then it sounds like a very good version of these "normal" resamplers.

 

With original 24/96 data, downsampled with iZotope and then upsampled again with iZotope I was able to find a perfect configuration that actually "recreates" 60-70 % of the sound of original 24/96 material without the errors of bad up- or resampling (smearing, less precision). So I´ll say Upsampling CAN have a benefit - if you use the right tools. Have a look here for pictures and more description (I hope I´m allowed to do that): http://www.thesoundtrackzone.com/viewtopic.php?f=35&t=323

 

E-MU 0202 USB wired with Monster USB Cable --> Audioquest King Cobra --> (sometimes) Corda Arietta --> Sennheiser HD-600

Link to comment
Share on other sites

  • 1 year later...

An old but interesting thread. The discussion mentions upsampling 16/44 to 24/96 but what about going to 24/176 or 24/192? No doubt it's possible in software but at what point does it stop making sense from an audio viewpoint?

 

Regards,

APS

 

Link to comment
Share on other sites

  • 8 months later...

I was going to start a thread on this topic, but decided to see if it already existed.

 

My experience with DVD up-sampling is that it radically improved the image (Samsung 52" LCD and their standard no-frills DVD player). I decided to turn it off (it is on by default) based on what I had read about audio up-sampling. The image looked horrible by comparison.

 

Based on this, I was wondering if our implicit assumptions that up-sampling music is bad might be unwarranted.

 

The manual for the Peachtree Nova that I have calls it an "up-sampling" DAC, and I am wondering if one of the reasons I don't hear differences that other people hear is due to the up-sampling (which I don't think I can turn off).

 

If the analogy with video up-sampling is valid, I can only imagine that this is an asset. But my guess is my analogy is flawed...

 

Link to comment
Share on other sites

So my understanding of the Nova is that it is limited to 24/96 input but it up-samples to 24/192. So if you play a 24/192 file, the source computer will down-sample it to the DAC and then the DAC will in turn up-sample it. That can't be as good as just playing the native 24/192 without change. I wonder if a native 24/96 on a non over-sampling DAC would be better than an up sample 24/96.

 

My DAC over-samples too. I think next time around, I want a non over-sampling DAC.

 

 

 

Link to comment
Share on other sites

Doesnt all this upscaling/upsampling throw all thoughts of bit perfect right out of the window?

 

MacMini 8Gb OSX > Pure Music / Bitperfect / Amarra / iTunes > Synology DS215J NAS > Schiit Wyrd > Stello U3 > Naim Uniti Atom, Harbeth P3ESR. Meier Corda Arietta Headphone Amp > Sennhieser HD650 Phones (Cardas rewire). Isol-8 Powerline Axis. Isotek GII Orion Power Conditioner. Cardas Clear USB Cable. Tellurium Q Black Speaker Cable. All other cables by Mark Grant.

Vinyl still has it's place. Technics SL1200. Modified with Mike New Bearing, KAB Strobe Disable, MCRU 2 box PSU, Isonoe Feet, SME M2-9 Tonearm > Goldring 2400 >Rothwell Simplex Phonostage.

Link to comment
Share on other sites

Yes. But read Cavailles post to understand better that "upsampling" is a necessity.

 

It is good to understand that a DAC "with filter" always does this job for you, and it is actually without sense to do it again ahead of that in software.

 

So, necessary filtering (for 44.1 material !!) means upsampling.

 

Only NOS/Filterless DACs don't perform any upsampling, hence lack the necessary filtering. Such a DAC should benefit from upsampling in software *if* the algorithm is good. However ... those DACs are very rare, and the minimum specs such a DAC must meet is 17 bits and 96 (88.2) KHz. This is the very minimum, and still loads of harmonic distortion will come from it. So 192KHz is already better, but 384KHz is good. For 192KHz the DAC must be 18 bits, and for 384 it must be 19 bits (at least).

 

The net result of the latter (192KHz already) is that it must be a multi bit DAC, or otherwise the whole lot can't work to begin with. Why ? because a non multi-bit DAC will be of the oversampling type, and those have a filter per definition. It will be the most far from "bit perfect".

 

With a multi bit 24/384 filterless DAC the upsampling can be done in software to your likings, but it still will compromise bit perfectness because of the available algorithms.

 

Notice that I say that bit perfect is still allowed to be called like that when all the samples have an equal inter-relation in both the native and the upsampled version. This will imply it is lossless, or IOW from the upsampled version an exact native version can be created again (by means of downsampling).

 

This is not about advertising myself, but as far as I know the only "upsampler" which can do that is the one I wrote myself, and it is told (sure not by myself) that even oversampling DACs benefit from it. So, without having an opinion of my own, I can see that it works for the better, because without exception people use it. I use it myself too, but this is because I *have* to, my DAC not containing any filtering.

 

To summ it up a bit, and now more towards the subject of the thread, upsampling (oversampling) is *always* done (if you want to get rid of high HD figures), and it is done because of that reason, and that reason only !! This is very very different from the subject "let's upscale" like we do with video. Nobody does it for that reason, because it is already been done for the filtering reason, sadly with "filtering" results. So, the horse is behind the car.

 

My approach has been the exact other way around :

1. make a DAC without any filtering, so

2. now we can make an upscaler for the sake of better resolution (!!) and

3. the necessary "filtering" goes along with that automatically.

 

Isn't it fun ...

 

It is all a very creapy thing, and at parts beyond my own understanding. But slowly I'm learning. So, what happens is this :

 

First in software the music is literally upscaled. So, as good as it can it mimics the higher resolution (up to even 768 in my case). This is done with zero ringing and (thus) pure interpolation. This is audible exactly the same as it is visible with audio, and it is very very difficult to distinguish from native hires (or even outbetters it, but this is for reasons beyond the subject of this thread).

(If I take out the injected samples, the original (16/44.1) is back (the what I call "losless" thing, implying the sustained "bit perfectness")).

 

Next this is fed to your oversampling DAC, and because it sees it as "native" high resolution, the normal filtering doesn't take place anymore, or at least won't intefere as much in the audio band anymore. This is how a "filter" explicitly made for filterless DACs still can let sound a filtering DAC better.

 

The above is my reasoning from today, while at first I couldn't understand how everybody perceived my filtering better than without it (and while it was made for myself only). It is just the "upscaling" part which really improves on the resolution, like with video. It won't be destroyed anymore.

 

Any other "upsampler" really is that filter which actually destroys (but remember, for the good sake of removing HD as how we measure it), and is not an "upscaler".

 

Here too it can be seen as "prooven" that such an upsampler doesn't work, because I tried many many times myself and it never sounded better (this is no proof yet), and nobody uses it while available in the software at people's own choice (this is proof IMO). Please notice that almost without exception this is about higher end systems people use, and I assume their ears are as good as mine.

 

Long story short : as how we are all to use our systems as a whole, it can't be avoided to throw out the bit perfectness right when the bytes have left the player and enter the DAC. This counts for you and it is today's standard.

 

I am doing my best to change that a bit (pun intended).

Peter

 

 

Lush^3-e      Lush^2      Blaxius^2      Ethernet^2     HDMI^2     XLR^2

XXHighEnd (developer)

Phasure NOS1 24/768 Async USB DAC (manufacturer)

Phasure Mach III Audio PC with Linear PSU (manufacturer)

Orelino & Orelo MKII Speakers (designer/supplier)

Link to comment
Share on other sites

The discussion about "Upsamping good or bad?" is all very well, but I think the analogy with video is poor.

 

DVD resolution (on the disk) is 720x576 (720x480 for NTSC IIRC). The point of up-scaling this is because your TV (if full HD) will be something like 1920x1080. If you do no scaling, all you would get is a small image in the middle of the screen. (If you use VLC or another computer based DVD player - think the difference between clicking "normal" size and full screen).

 

wgscott commented that the picture was better with scaling (it's scaling not up-sampling in a DVD player / for video) turned on than off. Well if you turn scaling on or off in the DVD player, all that's changing is WHERE the scaling is occurring - in the DVD player or in the TV. Depending on your TV and DVD player, the quality may be better in one than the other.

 

Eloise

 

Eloise

---

...in my opinion / experience...

While I agree "Everything may matter" working out what actually affects the sound is a trickier thing.

And I agree "Trust your ears" but equally don't allow them to fool you - trust them with a bit of skepticism.

keep your mind open... But mind your brain doesn't fall out.

Link to comment
Share on other sites

If you do no scaling, all you would get is a small image in the middle of the screen.

 

True. But "scaling" is not "upscaling" (although some intelligence has to be in there to normally scale).

If you'd compare scaling to zooming, you'll have what any TV can do. Otherwise it comes down to relatively expensive (licenced) solutions like Faroudja (although that may be outdated by now).

 

Maybe a tip : use a software player which allows FFDShow (or other inherent upscaling facilities), show it on your beamer, and play around with the settings.

Output size is something else than "upscale to resolution" and either combination can be made.

 

If I watch SD on the 1920x1080 TV, no way it is upscaled (I wish it were). Still the screen is nicely filled (assuming 16:9).

 

Analogy with audio ? maybe linear interpolation (as the zooming equivalent). Not sure though.

 

Peter

 

Lush^3-e      Lush^2      Blaxius^2      Ethernet^2     HDMI^2     XLR^2

XXHighEnd (developer)

Phasure NOS1 24/768 Async USB DAC (manufacturer)

Phasure Mach III Audio PC with Linear PSU (manufacturer)

Orelino & Orelo MKII Speakers (designer/supplier)

Link to comment
Share on other sites

Sorry you are wrong...

 

Up-scaling is just saying that Scaling is increasing the size of the image (i.e. SD to 720p) as opposed to down-scaling which is reducing the image size (i.e. 1080i to 720p).

 

Scaling (in video) is just not analogous to upsamping in audio.

 

Every fixed pixel (e.g. LCD, Plasma, etc) screen has a scaler. That can be very basic quality or can be extreme quality. Yes the scaling can be done by different algorythims but scaling is essential. Though in the same way different chipsets/applications provide better or worse upsampling - thats the only analogy!

 

And yes it is just "zooming" the image to fill the screen.

 

Eloise

 

Eloise

---

...in my opinion / experience...

While I agree "Everything may matter" working out what actually affects the sound is a trickier thing.

And I agree "Trust your ears" but equally don't allow them to fool you - trust them with a bit of skepticism.

keep your mind open... But mind your brain doesn't fall out.

Link to comment
Share on other sites

Hey Eliose, couldn't it be so that we are nit picking too much ?

 

Scaling (in video) is just not analogous to upsamping in audio

 

No, it is not, but it depends on the details and common language. If I buy an upscaler, for sure I won't be getting a stupid enlarger or something. This, while every (digital) TV will be able to scale to the physical size of the TV. If you want to call that upscaling I'm fine. But common talk it is not.

 

About the details and audio ... it is just the exact other way around. Please read what I wrote before, and although that too is not common language (this time just because it doesn't exist at all), upsampling is totally unrelated to "higher resolution", what we tend to talk about when upscaling video. Besides, when my TV scales to its physical size it is NOT bettering resolution. Not not not.

But maybe your TV does. Sorry.

 

So, in audio there is no common knowledge or means about bettering the resolution (always fake(d) of course). It upsamples though.

Unless we talk about my Arc Prediction, which *does* better resolution (and still upsamples).

 

Now, which is what and when with video ... who cares (everybody can look for himself). But understanding what happens with audio is another matter. So let's talk about that. Or at least that should be more interesting in here.

 

Regards,

Peter

 

Lush^3-e      Lush^2      Blaxius^2      Ethernet^2     HDMI^2     XLR^2

XXHighEnd (developer)

Phasure NOS1 24/768 Async USB DAC (manufacturer)

Phasure Mach III Audio PC with Linear PSU (manufacturer)

Orelino & Orelo MKII Speakers (designer/supplier)

Link to comment
Share on other sites

Interesting in that I was struggling with this "not hearing a difference" between 24/96 files and 16/44.1 files upsampled. Maybe it is because the Wadia PowerDAC I am using is actually doing a good job.

 

I was discouraged with the 24/96 files when maybe I should be elated witht what I am hearing from the 16/44.1 files.

 

"If you fly a flag of hate you are no kin to me"

Ry Cooder

Link to comment
Share on other sites

Peter ... in video terms, scaling is taking a picture of one size (i.e. SD DVD 720x576) and converting it to another size (i.e. you Full HD 1920x1080 pixel) for the display. In general terms used in EVERY manual for televisions I have read in English, converting from SD to HD is described as upscaling and coverting from HD to SD is described as downscaling.

 

Beyond that scaling can be done within the display, within the playback device (i.e. DVD player) or within a separate box.

 

When your TV (assuming it's a plasma, LCD, etc with fixed pixels) recieves a signal and processes it for display it IS increasing it's resolution - otherwise only 720x576 pixels would be displayed. Different processing chips can do this to a better or worse degree (though infact actual scaling is a relatively simple task - it's deinterlacing, colour processing, etc that shows more variation in quality).

 

Modern TVs have a fixed number of pixels. Every pixel has to have some information to display. This is totally different to audio where there is a continual waveform output from the DAC. If you have an old LCD projector you can actually clearly see the individual pixels (if you look closely) as the "screen door" effect.

 

My point was just that any discussion in a thread labeled "Upsampling: just an illusion?" where the question is posed is upsampling (in audio) analogous to upscaling (in video) is fundermentally flawed because the answer to the first question is NO, there is no real analogy. Maybe I'm just being contrary and arguing for the sake of it ... but I hope that I'm adequately expressing the differences.

 

Eloise

 

Eloise

---

...in my opinion / experience...

While I agree "Everything may matter" working out what actually affects the sound is a trickier thing.

And I agree "Trust your ears" but equally don't allow them to fool you - trust them with a bit of skepticism.

keep your mind open... But mind your brain doesn't fall out.

Link to comment
Share on other sites

The net result of the latter (192KHz already) is that it must be a multi bit DAC, or otherwise the whole lot can't work to begin with. Why ? because a non multi-bit DAC will be of the oversampling type, and those have a filter per definition.

 

Due to how most DAC chips work internally, external upsampling makes sense regardless of the DAC technology as long as the external algorithm is better than the internal one.

 

Good example being CA's DacMagic which has "external" software-based upsampler, even though the DAC chips are of upsampling delta-sigma type. This is no different from doing the same thing in the player software or somewhere else along the digital audio path.

 

Generally, software allows more choices for upsamplers. Best choice tends to depend on type of the music.

 

It will be the most far from "bit perfect".

 

"Bit perfect" 'ness is a funny trend and myth. Nobody listens pure digital bits, thus making any particular part of the audio path "bit perfect" lacks any sense unless it actually improves overall result...

 

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
Share on other sites

My point was just that any discussion in a thread labeled "Upsampling: just an illusion?" where the question is posed is upsampling (in audio) analogous to upscaling (in video) is fundermentally flawed because the answer to the first question is NO, there is no real analogy.

 

Mathematically, scaling video or an image is practically same as resampling audio, only difference is that image scaling is 2D operation while audio upsampling is 1D operation (these can be extended to 3D or any other number of dimensions). Both actually need similar processing to avoid similar artefacts.

 

Linear interpolation is equally poor choice for both.

 

Modern TVs have a fixed number of pixels. Every pixel has to have some information to display. This is totally different to audio where there is a continual waveform output from the DAC.

 

Audio DAC has fixed number of voltage levels it can represent and limited range of rates at which this voltage can be varied. Maximum resolution can be obtained at maximum number of levels at maximum change (conversion) rate.

 

Even old analog VGA-output connector used bunch of video DACs...

 

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
Share on other sites

Let's look at this way..

 

Video..black is black, white is white..image artifacts and image sharpness can have a definitive guide.. Everyone can see it for themselves and by market forces, vote with their wallets..

 

With audio, what do you have to reference? So implementation is going to very very iffy..everyone can claim they are the best..but that's subjected on how they wish to percieve it the way they percieve a good alogarithm is for scaling.

 

I'm not discounting the fact that there is maybe some unspoken standard..

 

 

http://audiohub.com.sg - blog/shop/randomness

Link to comment
Share on other sites

With audio, what do you have to reference? So implementation is going to very very iffy..everyone can claim they are the best..but that's subjected on how they wish to percieve it the way they percieve a good alogarithm is for scaling.

 

For either media, subjective comparison has it's merits and faults. So does objective comparison using measurements.

 

For video there are measurements. As well as test signals useful for subjective evaluation. Same goes for audio. At the moment there is at least http://src.infinitewave.ca for easy comparison of quite large number of different "audio scalers".

 

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
 Share



×
×
  • Create New...