Upsampling 16/44.1 collection a good idea?

[krabapple]

it is clear than you don't know how work convolution.
NTK have explained you here. https://www.audiosciencereview.com/...4-1-collection-a-good-idea.53641/post-1943511

Yes, and it was beside my point. Most people will never 'need' to add an upsampling step in their signal chains because if one is used it will be built in already, e.g. in DAC,s e.g in convolution in Audyssey.

If you've gone down the road of crafting bespoke 'sound treatment' filters-- I'm guessing for 'room/speaker' correction? -- that's on you. You are a tiny minority in home audio.

The OP, meanwhile, seems to think he needs some extra upsampling beyond whatever his system already does. He fears that without aggressive upsampling, unfiltered ultrasonic noise plus whatever other normally inaudible artifacts exist in his system might combine to the point of audibility.
.

 

[Keith_W]

Yes, and it was beside my point. Most people will never 'need' to add an upsampling step in their signal chains because if one is used it will be built in already, e.g. in DAC,s e.g in convolution in Audyssey.

If you've gone down the road of crafting bespoke 'sound treatment' filters-- I'm guessing for 'room/speaker' correction? -- that's on you. You are a tiny minority in home audio.

The OP, meanwhile, seems to think he needs some extra upsampling beyond whatever his system already does. He fears that without aggressive upsampling, unfiltered ultrasonic noise plus whatever other normally inaudible artifacts exist in his system might combine to the point of audibility.
.

Upsampling can have a positive effect if you are using a convolver. It reduces latency. The calculation for FIR filter latency is:

time (seconds) = (N-1) / 2Fs, where N is the number of taps, and Fs is the sampling frequency.

I am currently using 65536 taps with a 48kHz sampling rate. Every time I hit "play" or do a track change, I have to wait a minimum of 0.7 seconds for the signal to pass through the FIR filter. If I wanted to correct at even higher resolution, say 131072 taps or 262144 taps, I would be waiting 1.4 seconds or 2.8 seconds respectively. This may not sound like much, but if I load an album with 100 tracks and I can't remember which one I want to listen to, it all adds up and it's a total pain. And 2.8s is unacceptable in an AVR, so don't use FIR filters if you need to watch video. One way to reduce the latency is to increase the sampling rate to 96kHz or even 192kHz, but then I start running into problems with buffers.

But I agree, there is no audible benefit to upsampling for the sake of upsampling. You do it to gain other benefits.

As for a "tiny minority", again also true. But that's because the vast majority haven't seen the light.

 

[mike7877]

The basic philosophy of "high fidelity" is to faithfully reproduce the sound recording. A lot of people like to keep the digital "bit perfect", and up-sampling is altering the data.

But sometimes you might want to change the sound and sometimes there is something wrong with the recording.

There are much better ways to "improve" or "enhance" the sound, starting with better/different speakers or EQ. And unlike up-sampling the changes don't have to be subtle or inaudible. Or you can up-mix to surround sound. I use a soundfield setting on my AVR for some delayed reverb in the rear to give the "feel" of a larger space. With an audio editor or DAW you also add reverb or use an exciter effect, compression, expansion, etc. (Or there are hardware processors intended for pro use.)

I get ya. I do some EQ'n myself...

One big improvement I made is extending my sealed speakers by one octave - there's a 12dB shelf with a Q equal to that of the speaker (in its cabinet), with the +6dB point matching with the speaker's -6dB. It works really well! They actually sound full range! The only setback is with bass heavy music my max level is down 12dB lol. In most cases there's not a lot of energy down there and I lose maybe 3-4dB from peak level.

When songs need more or less bass, I have another shelf I use, this time with Q of 0.9 at 110Hz which usually does the job at up to +3dB and -2dB
EXCEPT
for newer songs with lots of bass... In the room (which is too small for full-range speakers for sure), I have a peaking adjustment at 28Hz with Q of 1.8. This usually takes care of the subsonic booming which can happen with newer synthesized songs after the +12dB shelf is applied to extend the octave. Rarely it can cut too much into the 42-55Hz range, and a +1 or 1.5dB on the 110Hz Q 0.9 does the trick!

I've got a surround system with Denon 3700H, Monitor Audio Silver surround package with two Kef Kube 10b subs and a couple extra rears and heights which I sometimes listen to music through - it can definitely make less than ideal recordings much better. For most serious listening I do prefer sitting in the sweet spot of my 2.0 system with room treatment (in the treble range only right now, the rest of optimization is just levels and not playing too too loud lol)


This [resampling] isn't for a major improvement (like some believe that I believe after reading the op), just to reduce the level/amount of things present which we know cause interference when they're higher in level or in large number. It's like a "this is slightly better than that" "well could you a/b it?" "maybe on a good day 57 times out of 100"
I do a bunch of small improvements, and when you take them all away there's usually an audible difference (depending on the overall system quality). They're things I'd always do in every situation, just so I know everything's working as well as it can and there're no unnecessary hindrances.

 

[mike7877]

My understanding from minds greater than mine is you need to upsample to the native max rate of your DAC

This avoids double dipping.

So if your DAC resamples everything to 768, then thats what you upsample to in your PC (and leave your files as they are in cause you change DACs)

Peter

Do you mean don't resample from 44.1 to 192, resample to the DAC's max sample rate that it will resample to so that the signal isn't resampled twice?
If so that makes sense - from the standpoint of least signal degradation. I like it... Removing resampling from the DAC and doing it on the PC with more accurate software and not doing the process twice can only lead to good things
IMO

 

[mike7877]

Upsampling can have a positive effect if you are using a convolver. It reduces latency. The calculation for FIR filter latency is:

time (seconds) = (N-1) / 2Fs, where N is the number of taps, and Fs is the sampling frequency.

I am currently using 65536 taps with a 48kHz sampling rate. Every time I hit "play" or do a track change, I have to wait a minimum of 0.7 seconds for the signal to pass through the FIR filter. If I wanted to correct at even higher resolution, say 131072 taps or 262144 taps, I would be waiting 1.4 seconds or 2.8 seconds respectively. This may not sound like much, but if I load an album with 100 tracks and I can't remember which one I want to listen to, it all adds up and it's a total pain. And 2.8s is unacceptable in an AVR, so don't use FIR filters if you need to watch video. One way to reduce the latency is to increase the sampling rate to 96kHz or even 192kHz, but then I start running into problems with buffers.

But I agree, there is no audible benefit to upsampling for the sake of upsampling. You do it to gain other benefits.

As for a "tiny minority", again also true. But that's because the vast majority haven't seen the light.

Is this processing done in your receiver? Or do is this your PC through a multi-channel DAC into analog multichannel or preamp inputs?

 

[MRC01]

Do you mean don't resample from 44.1 to 192, resample to the DAC's max sample rate that it will resample to so that the signal isn't resampled twice?
If so that makes sense - from the standpoint of least signal degradation. I like it... Removing resampling from the DAC and doing it on the PC with more accurate software and not doing the process twice can only lead to good things
IMO

Oversampling at integer multiples is computationally simpler and cleaner. That is what most DACs do, and there's no benefit to doing that on a PC. For example if the DAC's max rate is 384k, then a 96k recording will be resampled 4x to 384k. But a 44.1k recording will be resampled 8x to 352.8k. All internally, automatically, in real time as the music plays.

More detailed discussion here: https://www.audiosciencereview.com/forum/index.php?threads/question-about-ess-dac-resample.18777/
That thread is for ESS, but the method applies more generally to many DACs.

 

[Keith_W]

Is this processing done in your receiver? Or do is this your PC through a multi-channel DAC into analog multichannel or preamp inputs?

The latter. I don't own a receiver.

 

[fatoldgit]

Do you mean don't resample from 44.1 to 192, resample to the DAC's max sample rate that it will resample to so that the signal isn't resampled twice?
If so that makes sense - from the standpoint of least signal degradation. I like it... Removing resampling from the DAC and doing it on the PC with more accurate software and not doing the process twice can only lead to good things
IMO

So most DAC's have an internal fixed rate in the 48x class (ie multiples of 48k) and will resample (if needed... say 44.1k -> 96k base rate) then upsample as needed to work at there max rate (the resampled 96k base -> 768k).

If you start with 44.1k then you cant avoid a resample (either externally or in the DAC) so the logic is if you are going to bother resampling externally then make sure you also upsample to the max rate thereby avoiding any "interference" from the DAC.

The DAC may then oversample the max rate you send it (say 8x768) but you cant avoid that nor do you want to.

But dont create a "bake in" hard copy of your 44.1 file cause your next DAC may have a different max rate from your current...do it on the fly as the PC will have ample grunt for anything PLUS it means, if you are into this sort of thing, you can experiment with different resampling profiles.

Peter

 

[DWPress]

As for a "tiny minority", again also true. But that's because the vast majority haven't seen the light.

Amen brother....

 

[Lambda]

However, can you bypass the re-sampler of the DAC?
Can you defeat the filter of the DAC?
If not, where's the beef?

Not completely but effectively.
If the DAC gets fed with for example 192ks is far out of the audible band

 

[mike7877]

To me, the simple answer is "Yes". It is a good idea.

A modern DAC is already doing the upsampling for you internally. With external upsampling/filtering, you have more control.

You probably need to understand a bit more about the details of your DAC in terms of how it performs the D/A process.

Is it a Delta/Sigma DAC, or is it a NOS R2R DAC, or is it a Oversampling R2R DAC, etc...

If your DAC support DSD too, you can open a new door to enjoy your music by doing PCM==>DSD too. Enjoy

It's the Topping E70 Velvet - the way it processes DSD is excellent - bypasses the oversampling delta sigma (from what I've read), so I've been seriously considering it. I hear there's a foobar plugin for SACD which allows output of DSD... Do I have to convert first? If so, do you know which converter to use? And if not, is what foobar does internally as good as it gets, or are there other WAV > DSD resampler plugins that are better (just like there's better than the default WAV resampler)?

 

[John_Siau]

I've been thinking of upsampling my 44.1kHz collection to 88.2, 176.4kHz, 352.8, or even 705.6kHz for playback...

Is this a good idea to resample my files?

Absolutely a bad idea!!!! DON'T Upsample Your Music Files!

I cant emphasize this enough! Don't do it! Don't do it! Don't do it!

Fixed-point upsampling will create distortion whenever an intersample peak exceeding 0 dBFS occurs. This distortion is often audible because it is quite severe when it occurs. On many CD recordings, this can occur multiple times per second.

See my whitepaper on this topic here:
Intersample Overs in CD Recordings

Intersample peaks can reach +3.01 dBFS. Recordings that are not overly compressed, or overly loud, may contain more intersample overs than recordings that are mastered for maximum loudness. In other words, your best recordings may sustain the highest damage from upsampling the files.

If a DAC is properly designed (with enough DSP headroom and enough analog headroom), the intersample peaks exceeding 0 dBFS can be rendered without distortion. Please note that most DACs clip intersample peaks because they do not have any headroom above 0 dBFS. When this clipping occurs, it is not harmonic distortion. Instead, each clip produces a burst of broadband noise. In other words, you will hear short percussive bursts of noise at every occurrence of an overload. This is not like the sound of an amplifier clipping. It may correspond to the beat of the music and may be mistaken for a percussion instrument, but it is a sound that was not in the original recording. This percussive noise artifact is easy to hear once you learn what to listen for.

Upsampling can be done without causing damage if the audio level is reduced by at least 3 dB before upsampling. We do this within our Benchmark DAC2 and DAC3 converters. It can also be done correctly within a floating point environment, but again the level will need to be reduced by at least 3 dB before saving it back to a fixed point format. For a variety of very good reasons, music recordings are only distributed in fixed-point formats. If you upsample these without a level reduction, you will do permanent damage.

Fixed-point to fixed-point upsampling "bakes" intersample DSP overloads into the upsampled product. This distortion is audible and it cannot be removed. Upsampling will damage the sound of your recordings unless you reduce the signal level.

Beware of upsampled versions of 44.1 kHz recordings. These usually have baked-in distortion. Find the 44.1/16 originals.

Unfortunately some online music services provide "high-resolution" versions of 44.1 kHz recordings that have been upsampled directly from the 44.1 kHz originals without reducing the level. Avoid these "high-resolution" versions.

 

[KSTR]

Please note that most DACs clip intersample peaks because they do not have any headroom above 0 dBFS. When this clipping occurs, it is not harmonic distortion. Instead, each clip produces a burst of broadband noise. In other words, you will hear short percussive bursts of noise at every occurrence of an overload. This is not like the sound of an amplifier clipping.

This not my experience. I've tested different DACs with AKM, ESS, ADI and CRYSTAL DAC chips and all of them showed relatively benign intersample over behavior, ranging from hard-clipping to various degrees of soft clipping at around +3dBFS all of which is not very audible.

 

[sejarzo]

If you've gone down the road of crafting bespoke 'sound treatment' filters-- I'm guessing for 'room/speaker' correction? -- that's on you. You are a tiny minority in home audio.

Perhaps, but using convolver is the simplest way for someone with a variety of headphones to quickly implement the desired correction to Harman or whatever target curve they desire, plus add some additional >1kHz cut for (IMHO, the majority of) pop/rock recordings that are mastered too "hot" above 1kHz.

 

[dlovesmusic]

Sorry for my ignorance, I think HQPlayer and PGGB use floating points (PGGB can go all the way to 256-bit precision). They will do the dithering when output fixed format.

For the 3db, I think they already mention user should reduce the output by 3db to avoid clipping. HQPlayer does provide a real-time monitor for the clipping.

Again, I am new to this area, please correct me if I am wrong. Cheers.

p.s. By the way, we are doing real-time upsampling. The source file will be intact

Obviously John is referring to in a general scheme of thing, if you do not know what you are doing, be ware of Fixed-point upsampling creating distortion when an intersample peak exceeding 0 dBFS occurs, not specifically for software like Hqplayer that the developer spent the past 20 years day in-day out developing advanced upsampling algorithm within a floating point environment

There are not many dacs on the market other than Benchmark or RME that has properly dealt with intersample peak by offering extra headroom so it was a great advice from John re: -3 db to avoid potential intersample peaks.

 

[mike7877]

Absolutely a bad idea!!!! DON'T Upsample Your Music Files!

I cant emphasize this enough! Don't do it! Don't do it! Don't do it!

Fixed-point upsampling will create distortion whenever an intersample peak exceeding 0 dBFS occurs. This distortion is often audible because it is quite severe when it occurs. On many CD recordings, this can occur multiple times per second.

See my whitepaper on this topic here:
Intersample Overs in CD Recordings

Intersample peaks can reach +3.01 dBFS. Recordings that are not overly compressed, or overly loud, may contain more intersample overs than recordings that are mastered for maximum loudness. In other words, your best recordings may sustain the highest damage from upsampling the files.

If a DAC is properly designed (with enough DSP headroom and enough analog headroom), the intersample peaks exceeding 0 dBFS can be rendered without distortion. Please note that most DACs clip intersample peaks because they do not have any headroom above 0 dBFS. When this clipping occurs, it is not harmonic distortion. Instead, each clip produces a burst of broadband noise. In other words, you will hear short percussive bursts of noise at every occurrence of an overload. This is not like the sound of an amplifier clipping. It may correspond to the beat of the music and may be mistaken for a percussion instrument, but it is a sound that was not in the original recording. This percussive noise artifact is easy to hear once you learn what to listen for.

Upsampling can be done without causing damage if the audio level is reduced by at least 3 dB before upsampling. We do this within our Benchmark DAC2 and DAC3 converters. It can also be done correctly within a floating point environment, but again the level will need to be reduced by at least 3 dB before saving it back to a fixed point format. For a variety of very good reasons, music recordings are only distributed in fixed-point formats. If you upsample these without a level reduction, you will do permanent damage.

Fixed-point to fixed-point upsampling "bakes" intersample DSP overloads into the upsampled product. This distortion is audible and it cannot be removed. Upsampling will damage the sound of your recordings unless you reduce the signal level.

Beware of upsampled versions of 44.1 kHz recordings. These usually have baked-in distortion. Find the 44.1/16 originals.

Unfortunately some online music services provide "high-resolution" versions of 44.1 kHz recordings that have been upsampled directly from the 44.1 kHz originals without reducing the level. Avoid these "high-resolution" versions.

Thanks for all the info!

Before I got the DAC I use for most of my [more] critical listening now (Topping E70 Velvet, which uses the new 4499: AK4499EXEQ), I think I remember reading somewhere that it's capable of rendering above 0dB, but I don't recall by how much, or where I was even reading, so I can't re-check without a bunch of work {and even then I might not find where I was...). Anyway, at the time I didn't make a solid mental note because it didn't seem like an issue I'd run into (only case I could think of at the time which would require any headroom was bad recordings). Now that I know oversampling files can cause it too, I'll avoid resampling without first reducing level by 3dB in the future (whenever that presents itself) to ensure compatibility everywhere (and with the E70 Velvet in case it's only good for 1.8dB or something dumb lol).

Is that really what places like Tidal do to make "hi-res" music files? Just upsample them to 2x or 4x? I was wondering where they were getting all this "hi-res" music from, especially from some 80s pop/rock artists with not-so-great recorded/mixed/mastered albums - it didn't make sense! Where do they get 192 from mid 80s digital which we know was, at best, sampled at 88.2? I get when they play back old master tapes into a good ADC and run it at 24/192 (if you've only ever heard the CD version of Supertramp - Crime of the Century, I never knew they used so much noise reduction!! Not that it sounds like it without a reference point, but the keyboard isn't as one-dimensional as it sounds on CD, vocals are more detailed, drums actually sound a little worse but more detailed... There's actually tape hiss, but when you don't pay attention to it, it disappears and you're left with a much more realistic presentation

Which software would you say does the best job interpolating and resampling (preferably free)? Can it be done to DSD as well?
What about downsampling? I'd like to try turning some 192 into 96 and then 48, to see what differences (if any) can be heard between them / with different filters etc.

 

[John_Siau]

Yes, I agree. Not all the people are aware of the difference between dBFS and dBTP. It is always good to remind people about the 3db thing.
However, the 3db thing is, IMO, not the reason for not doing upsampling.

The biggest problem is that the online services are offering upsampled files that are not reduced in amplitude. These have baked-in distortion.

Yes, you can upsample without adding distortion, but you must take a 3 dB cut in level and this must be done before any fixed-point processing or fixed-point output. The chances of getting it wrong are high.

It is also possible to cause time-domain problems if you choose anything other than a linear-phase filter. The upsamplers give you lots of filter choices and most of them will do damage.

 

[Owl]

I wonder how these issue's are handled in a CD or Blu-ray player. The Denon model, DCD-900NE CD player boasts about up sampling everything before the analog filters ( AL32 processor plus ) and I believe most of Sony's Blu-ray players do not have a 44.1khz setting for the digital coaxial output, only multiples of 48khz. Is this a problem that only file based music would have?

 

[NTK]

IMHO, there is little need to worry about how DAC's handle "intersample overs".

Unless you have specially crafted test signals, if the source have more than a few intersample overs, that will invariably mean your source is clipped. As clipping is an irreversible process (once clipped, it is impossible to recover the original), the damage is irreversible.

Below are animations showing original signals (blue curves) that have peaks 1.5 dB above full scale. The solid black dots are the digitized samples, and the dashed red curves are the reconstruction from the digitized samples. The animations show the effects of different A/D sampling start times.

In the first animation, the source signal frequency is exactly 1/4 of the sampling frequency. You can see that if you are lucky and have the right sampling start times, the reconstruction can be perfect. When not, the signal amplitude is changed.

The above showed the special case that the signal frequency is 1/4 of the sampling frequency. If the signal frequency is a little off from Fs/4, you can see that in all cases the reconstruction cannot recreate the original signal.

It is "nice" if a DAC behave properly when it encounter "overs", but we are basically talking about the least bad reconstruction of irreversibly damaged signals which aren't going to match the original.

 

[Keith_W]

I think your diagram would also help people to understand why using higher sampling frequency (upsampling) would help to reduce the artifacts formed when the DAC regenerate the audio signal.

Unless I am very much mistaken, he is talking about downsampling from a higher resolution source or from analog. What you are proposing is taking a 44.1kHz source file, upsampling it, and then sending it to the DAC. The limiting factor here is the 44.1kHz source.

Here is a quick example.

Original image, 1000 x 800 pixels.

Downsampled to 100 x 80 pixels, then upsampled back to 1000 x 800.

In the same way, if you have a 44.1kHz source file, no amount of upsampling will ever recover the information that was lost.