Upscaling 16 bit to 24 bit

Yes wait a bit and mature your opinion. On my side, no way back ;-)

Best Tag management with flac files and optimum render in wave 24 bit to feed the beast.

Cheers

Interesting. Obviously you can't add the missing information that would potentially be there in a 24 bit recording, although many older recordings don't have the dynamic range that 24 bits gives over 16 bits. Presumably the DAC has to work harder processing 24 bits in real time compared with 16 bits. But I'm unclear whether you are transcoding your 24 bit upscales to WAV too or letting your 272 deal with the 24 bit FLAC? If the latter then maybe you are just hearing how the 272 handles a FLAC file. How does 24 bit upscale compare with 16 bit FLAC without transcoding? Or is that what you are describing?

best

David

Both options are after Asset had transcoded to WAV. I'm now on the beach at Hayling Island, clearing my head for more listening later. 

The DSP / DAC doesn't have to work harder with either data stream, because in both cases the DSP up-samples to 40bit floating point, and the DAC chip is loaded with 24/384 data by the oversampling in the DSP / DAC system.

By up-sampling before sending the data to the streamer all you are doing is changing from the up-sampler in the DSP / DAC to a 2 stage up-sampling, part in the NAS and part in the DSP / DAC system.  Unfortunately this will be a generalised 16/44.1 to 24/192 up-sampling followed by an up-sampling optimised for rendering 24/192 data.  So yes it'll sound different, and not as the Naim designers felt was the best optimisation for sending 16/44.1 data to their DAC hardware.

I think I follow that. What you are saying is that it ends up the same anyway, but some of the work is being done in the nas. What confuses me is that if it's originally a 16 bit recording, Asset cannot add anything, it just fiddles with the file somehow. Presumably that's why it sounds different. Surprisingly so. 

It's not quite the same.

With 16/44.1 going to the streamer, the DSP knows it's 16/44.1 and optimises the processing (digital filtering upsampling etc) for it's own style of rendering 16/44.1 data.

Unfortunately Asset doesn't know what DSP / DAC and DSP pre-processing firmware is in use so it can't optimise the up-sampling for that particular DSP / DAC / pre-processing combination, it uses a generic up-sampling algorithm rather than one specifically optimised for a Naim hardware/firmware.
With 24/192 going to the streamer it optimises for native 24/192.

Upscaling will ultimately add quantisation errors, and so the upscaling algorithm becomes crucial so as to add dither to the quantisation errors to bring to a minimum level, but essentially if you hear changes then it's either added digital distortion/noise from the errors, non linearities in your DAC system or a bit of both, and if you prefer the digital distortion, there is nothing wrong than that.. after all it's a sort of DSP.

'Dither the quantisation'. You're making it up!!

Sadly it was hugely more eminent mathematicians than I that developed this

And there was me thinking it must be the giggling pin in the laughing shaft.

Hungryhalibut posted:

I've today discovered, from reading another thread, that it's possible to upscale 16 bit recordings to 24 bit, using Asset. I currently transcode my FLAC files to WAV, but keeping the native bit depth. 

So I thought I'd give upscaling a go. It's certainly different. The imaging is more tightly defined, and it sounds clearer and cleaner. Very organised. But maybe less engaging- more academic sounding. It reminds me of when I tried Chord Odyssey cables instead of NacA5 - everything was very neat and tidy but ultimately unengaging - a bit boring. 

It's certainly interesting that there is this change. I'm not sure if I prefer it or not. Has anyone else tried it? What did you think? Maybe I need to upscale for a week and then revert. Certainly swapping between the two quickly isn't helping. 

Interesting. I think that's why I found asset better than minim. My minim was set to transcode to wav24 while assest just transcoded to wav16

Music seems to flow and breathe more natural at 16bit IMO

Thanks everyone. I'll forget the upscaling then. One thing fewer to concern myself about. And now I'm not having to think about which is better, it sounds better anyway!

I understand the "purist" way and no upscaling. On the other hand, to be honest, th eidea came to me beacuse on some exhibition, the Naim importer was doing it for demonstration. Explaining that he was using expensive software to do it and  bla bla bla...
So agree the algorythm may cause errors and damage the original idea ... (is there any correction sofwtare inside the process ?)

For my very personnal ears, i prefer the softer way of presentation of the upscaled files.

A matter of taste, like all the times ;-)

Thank you for the explanations.
Cheers

I have recently started up scaling my music to 192 using jriver to my Hugo because I found it more expressive and full sounding, but also it reduced sibilance and made the presentation smoother. 

In short I found the sound much more refined in 192. Regardless of whether it is the ' correct' way or not, it will stay in this configuration because I am now listening to music so more more than before!

James

Sorry to be crude, but this reminds me of my early days in the software industry where everyone was clamoring for open systems and Unix. Some folks claimed they could take AS400 code and magically turn it to Unix via a magical transformation platform. My old boss compared this to shoveling manure into a cows mouth and expecting it to poop grass. 

You can't add detail to what's not there. Check eBay... inthink you can still find 10 band graphic equalizers if you look hard. 

Most DAC's are going to upsample to 384 or 768 kHz (or thereabouts, depending on starting frequency), so it's going to happen anyway. Just depends on how you get there and who's device / algorithms do a better job

Does it even "upscale" ... doesn't it just put the same 16 bits of data in a 24 bit frame (not sure if frame is the correct term) so you have 16 bits of data and 8 bits of 0s?

Asset calls it 'increase to 24 bit'. It can't magically add what isn't there, which I guess is why it sounds different rather than definitively better. I'm happy leaving 16 bit albums as 16 bit and letting the 272 do all upsampling. 

Eloise posted:

Does it even "upscale" ... doesn't it just put the same 16 bits of data in a 24 bit frame (not sure if frame is the correct term) so you have 16 bits of data and 8 bits of 0s?

It's not as simple as that, there's interpolation and dither involved as well.  When done in the DSP / DAC, there's also the noise shaping integrated into the process (part of the reason they sound different.

Naimthatune posted:

...
You can't add detail to what's not there.
...

You can add detail, what you can't do is add information to the signal.

But you can add another signal co-mixed with the first (even if that second signal is just noise, and that's what dither is).

I guess the word 'detail' is not useful here. You have information and noise - and real world signal will have information as well as noise/aretefacts. So in Huge's example above one can add noise but you are not adding any more information. How we perceive noise mixed with the information is another matter and we might perceive' that as more detailed, darker,  brighter, airy, smooth,  forward ..   yada yada yada... but its our perception ... recorded music and replay is still a massive removal from reality and so what we hear is a compromise and our brains work to interpret the sounds.

Naimthatune posted:

Sorry to be crude, but this reminds me of my early days in the software industry where everyone was clamoring for open systems and Unix. Some folks claimed they could take AS400 code and magically turn it to Unix via a magical transformation platform. My old boss compared this to shoveling manure into a cows mouth and expecting it to poop grass. 

You can't add detail to what's not there. Check eBay... inthink you can still find 10 band graphic equalizers if you look hard. 

Ah, the AS400, what a lovely machine. 64 bit architecture (although they only allowed 48 bit initially) and complete virtual storage in 1980. Big Iron, the mainframe guys hated it because it was so easy to use. No system programmers, fantastic. Ah, those were the days! 

Eloise posted:

Does it even "upscale" ... doesn't it just put the same 16 bits of data in a 24 bit frame (not sure if frame is the correct term) so you have 16 bits of data and 8 bits of 0s?

Eloise, that's exactly what Minim does with wav24: for lossless 16-bit input streams (FLAC and ALAC), the audio samples are extended to 24 bits by padding each sample with zeroes, so there is no actual "upscaling" going on at all, the samples are not being messed with and they remain bit-perfect. I suspect the same is true for Asset given the feature is called "increase to 24 bit".

The MinimStreamer manual says:

"For best sound quality, it is recommended that you use the output sample bit depth that matches the maximum capabilty of your music player. For example, if your music player is a Linn DS, the best match is wav24. Results may vary with different types of music player."

The benefit, if any, from this padding to 24 bits would come from making it easier for the streamer to unpack the samples from the stream, thereby reducing processing overhead. For Linn streamers, it would appear 24 bit samples are optimal. The optimum setting for Naim isn't known, unless Naim can clarify. The only option is to try it and see...

If low bit padding of shorter data words to the maximum supported by the hardware gives the best sound quality, then there's something quite wrong with the optimisation of the digital processing algorithms of the device.  Doing this will reduce the efficiency of the device in using the lower bits of the word to optimise the digital filtering, (as it will believe the data to be true 24bit data, and optimise for that instead of optimising for the rendering of 16 bit data).

I don't believe this is good advice for Linn or Naim devices.

Trying it will only indicate which you prefer in your system and not which has the best sound quality from the renderer (if indeed such a concept is actually valid, which is up for debate anyway!).

Good point, Huge. On the face of it, it sounded like an interesting thing to try, which I did on my NDS when the feature first became available with Minim. I quickly reset it to native resolution.