Upsampling

[tuga]

I don’t understand how to read the second chart, it doesn’t look possible, the plot looks like just half the AC waveform! It’s an asymmetrical clipping fuzz-box on full!

If you think that one's bad... Radiohead's "Weird Fishes/Arpeggi", flatlining.
Audacity doesn't show any clipping, I wonder if it's because the peaks never go above -1dBFS?

Edit: to be fair(er) the music doesn't have a lot of dynamic variation...

 

[igor_xxxx]

Few weeks ago I got the "itch" to try and test play with a bit of upsampling in HQ player since I had a spare endpoint and a HTPC.From reading more on the topic, I guess my system is not really the ideal candidate for HQ player (Elac DDP2 with AKM dac, DSP via Neumann after the DAC), but still I wanted to check it.

I am still undecided if the sound is better with the HQ Player upsampling to DSD128, or its simply slightly different and I am perceiving this as better, in combination with a slight expectation bias that it should be better.Since I am listening a bit more free jazz this period, one thing that I think I noticed is that somehow the music is a bit more relaxed, like a certain "glare or hardness" is absent.But I would not bet real money on this in a level matched blind test

I still think that my system is not the best for this and maybe in other cases the differences could be more noticeable.If I will invest in HQ Player is still to be decided.

Just one's totally subjective and flawed observation

 

[Alex S]

If you have 2 systems in a single room and 5 DACs then my assumption regarding not having a reference still applies

Fair point. It wasn't by design but primarily to save valves on the Leak ST20 since I listen to music around 10 hours a day. Ironic, since the Leak is presently banished because of persistent ground loop issues.

I do seem to accumulate dacs but for some purpose: the Audial is my main dac, the MHDT was a swap and a perverse curiosity to find out just how bad the worst ever measuring ASR dac is (it's rather good). The Topping is a reverse measure to find out how good a superb measuring dac is (very good), the RME is the bedroom dac and the other ES chip is stuck inside a streamer.

 

[tuga]

I am still undecided if the sound is better with the HQ Player upsampling to DSD128, or its simply slightly different and I am perceiving this as better, in combination with a slight expectation bias that it should be better.

You could try comparing a few different filters, e.g.:

asymFIR - Asymmetric FIR,
poly-sinc-ext2 - Linear phase polyphase sinc filter with sharp cut-off and high stop-band attenuation for extended frequency response while completely cutting off by Nyquist frequency,
poly-sinc-gauss-long - Long Gaussian polyphase sinc filter with extremely high attenuation,
sinc-M - sinc-filter with one million taps. Very sharp cut-off and high attenuation.

 

[davidsrsb]

@davidsrsb on-board upsampling presents a few challenges, namely limited processing power and the processor generating noise inside the DAC or even chip.

AKM is now spliiting the D/A and the digital filter & modulator into separate chips:

AKM releases to production the new flagship D/A converter (DAC) solution AK4499EX and AK4191 for a complete separation of digital and analog circuitry. | Archives 2022 | News | About AKM | Asahi Kasei Microdevices (AKM)

AKM releases to production the new flagship D/A converter (DAC) solution AK4499EX and AK4191 for a complete separation of digital and analog circuitry.

www.akm.com

Going full circle. The first telephony codec chips from Ferranti used a separate digital control IC and thick film hybrid modulator around 1980. It was abysmal, nobody discussed the sensitivity of single bit delta-sigma to clock jitter back then.

 

[tuga]

Going full circle. The first telephony codec chips from Ferranti used a separate digital control IC and thick film hybrid modulator around 1980. It was abysmal, nobody discussed the sensitivity of single bit delta-sigma to clock jitter back then.

1-bit SDM D/A conversion is so passé, I wonder if it's still being made.

 

[gustav_errata]

I don’t understand how to read the second chart, it doesn’t look possible, the plot looks like just half the AC waveform! It’s an asymmetrical clipping fuzz-box on full!

It's measuring dBFS over the timeline of the song. The song has been hard-limited to -0.1dBFS, hence the plateau shape. The dips represent the brief rests that the ear gets. This practically guarantees intersample clipping.

Digital limiters that support true peak limiting are more recent than the examples given (citation needed), although to argue against myself I found a blog article by iZotope (purveyors of the very popular Ozone mastering suite) wherein it's claimed that some mastering engineers don't like intersample processing because it negatively affects the sound. So yeah, it's still a problem I guess.

 

[tuga]

It's measuring dBFS over the timeline of the song. The song has been hard-limited to -0.1dBFS, hence the plateau shape. The dips represent the brief rests that the ear gets. This practically guarantees intersample clipping.

Digital limiters that support true peak limiting are more recent than the examples given (citation needed), although to argue against myself I found a blog article by iZotope (purveyors of the very popular Ozone mastering suite) wherein it's claimed that some mastering engineers don't like intersample processing because it negatively affects the sound. So yeah, it's still a problem I guess.

Why do you need "intersample processing"? Just turn down the gain.

 

[gustav_errata]

If you think that one's bad... Radiohead's "Weird Fishes/Arpeggi", flatlining.
Audacity doesn't show any clipping, I wonder if it's because the peaks never go above -1dBFS?

Edit: to be fair(er) the music doesn't have a lot of dynamic variation...

If Audacity's metering reflects dBFS it won't show clipping because the recording was hard limited to -0.1dBFS. If it shows dBTP (doubtful), then there was some other sorcery.

 

[gustav_errata]

Why do you need "intersample processing"? Just turn down the gain.

Sorry, poorly worded. I meant processing with intersample peak detection enabled.

Edit: here's the article - https://www.izotope.com/en/learn/an-introduction-to-limiters-and-how-to-use-them.html

 

[Tony L]

If you think that one's bad... Radiohead's "Weird Fishes/Arpeggi", flatlining.
Audacity doesn't show any clipping, I wonder if it's because the peaks never go above -1dBFS?

It is a very good CD despite how it looks. There is nothing about the sound that I’d not class as a very conscious artistic decision. Radiohead are very much a studio band, the studio technology very much part of the intended aesthetic.

 

[tuga]

It is a very good CD despite how it looks. There is nothing about the sound that I’d not class as a very conscious artistic decision. Radiohead are very much a studio band, the studio technology very much part of the intended aesthetic.

One of their finest albums in my view.
Radiohead are my favourite band and they do like their dynamic compression.

 

[Tony L]

One of their finest albums in my view.
Radiohead are my favourite band and they do like their dynamic compression.

Likewise, and that, Kid A and Amnesiac are my favourite albums of theirs. I’ve always liked the way they sound too, they play to digital format strengths.

 

[Jim Audiomisc]

I don’t understand how to read the second chart, it doesn’t look possible, the plot looks like just half the AC waveform! It’s an asymmetrical clipping fuzz-box on full!

I think the plots you refer to show the peak power level in each brief timeslot. Thus '0' means full scale level. i.e. the levels are so high that clipping is the norm. (Or 'ducking' where the level is very briefly wound down to stop flat-topping but still nasty.)

A capable DAC (or resampling process) can cope with some 'over the top' behaviour by scaling down. For *brief' clips that can be OK (devil in the details). But longer ones it means flat clipping of the output. Nasty.

 

[Jim Audiomisc]

Audacity doesn't show any clipping, I wonder if it's because the peaks never go above -1dBFS?

Edit: to be fair(er) the music doesn't have a lot of dynamic variation...

Doesn't Audacity simply 'join the dots'? If so it won't show actual overs.

To check I tend to use sox to x4 upsample with a -6dB gain. This then shows up any intersample overs. as waveforms. Doing it with no gain set reports overs. Quite handy as a quick 'is this OK'. *Isolated' overs between two successive samples can be OK if a DAC can cope correctly. The problem is exampls like the ones you show where it seems to be bucketloads of max values in sequence.

 

[davidsrsb]

1-bit SDM D/A conversion is so passé, I wonder if it's still being made.

I doubt it. The clock sensitivity is a real challenge and pointless when multibit is easy to do now

 

[Jim Audiomisc]

You could try comparing a few different filters, e.g.:

asymFIR - Asymmetric FIR,
poly-sinc-ext2 - Linear phase polyphase sinc filter with sharp cut-off and high stop-band attenuation for extended frequency response while completely cutting off by Nyquist frequency,
poly-sinc-gauss-long - Long Gaussian polyphase sinc filter with extremely high attenuation,
sinc-M - sinc-filter with one million taps. Very sharp cut-off and high attenuation.

For those who like command-line and DIY you can also set up your own filters for sox. In essence, whatever filter you want to try.

 

[Heckyman]

Out of curiousity which one do you have?
I am also looking for a NOS DAC and appreciate any input.

Holo May — I wrote a bit more about it on its own thread on here.

For what its worth, I am up to my ears in the boggy upsampling camp (see the Dac200 thread for the gory details) and I find significantly more music in pushing the process to the nth degree. Just regular hardware upsampling (Limited resources, even with a Chord Dave) or perfunctory software (pretty much everything less than HQPLayer 4 or PGGB) is pretty much indistringuishable from not upsampling.

A Holo Sping or May (not Denafrips, many are convinced there is some processing going on) if a well engineered product - feeding it good upsampled bits should produce better results than basic bits.

Interesting.. Someday I will have to try doing more "serious" comparisons of software upsampling (and vs NOS).

 

[tuga]

For those who like command-line and DIY you can also set up your own filters for sox. In essence, whatever filter you want to try.

You'd have to know what you're doing, and I am one of those people who dreads having too much choice.
HQPlayer offers around 40 filters, half of them available in minimum phase and linear phase, and some 15 different ΔΣ modulators. More than enough in my view.

 

[Eoin]

Is an Apple TV anywhere close?