Do better DACs benefit from upsampling the source (HQPlayer)? Does HQPlayer just change the sound?

[EuroDriver]

ROON is good for many people but the issue is that proprietary hardware, such as audiophile servers/players, are constrained in two important ways. First, they usually employ the least capable internals necessary to perform the job as seen by the manufacturer at the time of design. There's usually no future upgrade option when/if there are increases in demand as new features and formats roll out. Second, they are constrained by limited ability to be reprogrammed and/or load entirely new software in response to demand.

And this is just speaking about those that do not include DACs. For those that do, they are virtually defined by their hardware.

This is why I am a big fan of server/players based on general purpose computers (Win, Mac or Linux), even if they must be packaged to look like an audio component and have a limited GUI in order not to create anxiety in computer-phobic audiophiles.

+10

 

[microstrip]

ROON is good for many people but the issue is that proprietary hardware, such as audiophile servers/players, are constrained in two important ways. First, they usually employ the least capable internals necessary to perform the job as seen by the manufacturer at the time of design. There's usually no future upgrade option when/if there are increases in demand as new features and formats roll out. Second, they are constrained by limited ability to be reprogrammed and/or load entirely new software in response to demand.

And this is just speaking about those that do not include DACs. For those that do, they are virtually defined by their hardware.

This is why I am a big fan of server/players based on general purpose computers (Win, Mac or Linux), even if they must be packaged to look like an audio component and have a limited GUI in order not to create anxiety in computer-phobic audiophiles.

We can have ROON is a general purpose computer - then we combine the best of both worlds. And I hope we will not have new digital formats in the next years - IMHO we are being limited by implementations, not by formats.

BTW, I am not computer-phobic - I work with computers (hardware and software) since the old days of Steve Ciarcia in the BYTE magazine - but I am becoming computer audio-phobic - IMHO it is a real jungle! Fortunately the Vivaldi CD player still sounds much better than any server playing rebook I have listened.

 

[hvbias]

my understanding of 'Ravenna' is that it's a proprietary version of Ethernet used by Merging Technology. likely I don't have that quite exactly right but that is what I recall from when I was investigating the NADAC.

Ethernet as I understand it is where the dac has it's own processor for directly accessing files over a network, instead of accessing a server for the data.

and some higher level dacs allow you to add an Ethernet hub to the dac, as opposed to using USB. one typical negative of Ethernet is the noise from the processor inside the dac. and OTOH it's said that a USB-server approach adds an unneeded step to the process.

be gentle if I've got the techie stuff a bit skewed......I'm a user/listener and not an authority on the nuts and bolts, and typically leave the heavy lifting to others.

No preface or disclaimer needed. I know some HQPlayer users have an NAA between their computer that does the upsampling and the DAC but I don't think this is what you're talking about since that NAA is still going to the DAC over USB.

Any examples of DACs that use this type of input?

 

[Mike Lavigne]

No preface or disclaimer needed. I know some HQPlayer users have an NAA between their computer that does the upsampling and the DAC but I don't think this is what you're talking about since that NAA is still going to the DAC over USB.

Any examples of DACs that use this type of input?

as far as on my radar, 3 that come to mind that give you a choice are the MSB Select II, and the CH Precision C1, and the 3-box CH Precision Mono Dac.

ho hum these 3 are very spendy.....of course.

http://www.whatsbestforum.com/showt...on-T1-and-C1-5&p=420465&viewfull=1#post420465

the CH Mono Dac is particularly interesting since it has all the digital circuits inside one box, and then completely separate boxes for each channel of it's analog outputs. this approach would eliminate the noise issue from the Ethernet hub into analog circuits. I've seen a few examples of XLR outputs sounding much better on dacs mostly due to the extra transformer involved in those circuits removing the digital noise compared to the RCA outputs. so digital noise inside a dac is a serious performance concern and a negative for the Ethernet approach.

I've not yet heard any listening feedback on the CH Precision Mono Dac.

 

[Taiko Audio]

"the only algorithm that has been designed to reduce timing of transient errors" A linear phase filter? I think I've seen them provided by others. Here's the impulse response of the Chord Dave courtesy of JA @ Stereophile.


If someone wants to hear similar from HQPlayer the filter is called poly-sinc-xtr. I suppose it would be debatable to which sounds better through a Dave but it gives many others the opportunity to get Dave like sound from their own DAC if they have the processing power.

That's not a really fair comparison, The Chord Dave WTA Filter utilises 164.000 taps where HQPlayer's poly-sinc-xtr filter employs 2.000.000 taps. This enables HQPlayer to push out of band noise down to -240dB, the stop band attenuation filter for 44.1KHz for example is set at 21.5KHz, you can get away with that with filters that steep. This of course requires insane amounts of processing power which you will not find inside any dac out there, I cannot be sure but my guess would be currently the Chord Dave has the most powerful built-in processing power. I owned one and it gets pretty hot to the touch, with a switching power supply and a fairly regular output stage all that heat had to have come off the FPGA's.

 

[caesar]

That's not a really fair comparison, The Chord Dave WTA Filter utilises 164.000 taps where HQPlayer's poly-sinc-xtr filter employs 2.000.000 taps. This enables HQPlayer to push quantization noise down to -240dB, the stop band attenuation filter for 44.1KHz for example is set at 21.5KHz, you can get away with that with filters that steep. This of course requires insane amounts of processing power which you will not find inside any dac out there, I cannot be sure but my guess would be currently the Chord Dave has the most powerful built-in processing power. I owned one and it gets pretty hot to the touch, with a switching power supply and a fairly regular output stage all that heat had to have come off the FPGA's.

Hi Taiko,
Thanks for chiming in. What exactly are "taps"? Is this s borrowed word from ultimate fighting tap-outs, with more being better, or is this something technical pertaining to digital processing only?

Also, what DAC did you switch from the chord and why?
Thank you

 

[Legolas]

Hi Taiko,
Thanks for chiming in. What exactly are "taps"? Is this s borrowed word from ultimate fighting tap-outs, with more being better, or is this something technical pertaining to digital processing only?

Also, what DAC did you switch from the chord and why?
Thank you

Not relevant, respect but Taiko is a dealer. Best ask regular owners that question, sorry, don't want to sound paranoid. Just prefer to keep to owners / users opinions.

 

[EuroDriver]

Not relevant, respect but Taiko is a dealer. Best ask regular owners that question, sorry, don't want to sound paranoid. Just prefer to keep to owners / users opinions.

Taiko is the manufacturer of the Sound Galleries Music Server. All of us on the team are agnostic about the DAC's we have tested, and try to highlight each DAC's strengths

The Chord Dave's high level of filter taps was indeed the motivation for Jussi Laako, the author of HQ Player, to develop Polysinc XTR.

Mike Moffat, of Schitt was indeed the inspiration and motivation for Jussi to develop and implement the Closed Form Filter in HQ Player which in our humble opinion is the best sounding filter for many RTR DAC's

 

[amirm]

Hi Taiko,
Thanks for chiming in. What exactly are "taps"? Is this s borrowed word from ultimate fighting tap-outs, with more being better, or is this something technical pertaining to digital processing only?

I will pretend I am Taiko and answer.

Tap is a figure of merit for a digital filter. Every tap represents an audio sample that is used to compute the filter response. By multiplying that audio sample with a number and adding to the rest, you compute the value of the new, filtered audio sample. The more taps, the sharper the response of the filter.

Each tap requires memory to hold the audio sample, and of course the more taps, the more computations you have to do.

Note that all of this is for a digital filter called "FIR." IIR filters work recursively and can have very high strength with far less resources (but can have other complications).

That's not a really fair comparison, The Chord Dave WTA Filter utilises 164.000 taps where HQPlayer's poly-sinc-xtr filter employs 2.000.000 taps. This enables HQPlayer to push quantization noise down to -240dB, the stop band attenuation filter for 44.1KHz for example is set at 21.5KHz, you can get away with that with filters that steep.

The number of taps determines how steep the FIR filter is and how low its pass through energy is. It has nothing to do with Quantization noise. The -240 db is the amplitude of out of band components in HQPlayer filter, not any level of quantization noise.

The 120 db vs 240 is pure "measurebating" by the way. No value whatsoever comes from such increased attenuation.

 

[EuroDriver]

Hi amirm

The 120 db vs 240 is pure "measurebating" by the way. No value whatsoever comes from such increased attenuation.

Our feeling is that 2 million taps is likely to be more extreme than needed, but the question is how much is actually needed.

The sound quality uplift from Polysinc and Polysinc short fiters to the new Polysinc XTR family of filters, especially for PCM to DSD conversion is pretty dramatic, every DSD DAC owner and HQ Player user we know can hear and likes the difference.

As Jussi's work schedule and to do list allows, he will explore optimizing the number of tap's and we will feedback with what we hear. In Munich over dinner, we were discussing with Jussi possible names for a filter family that is not quite so extreme as the XTR family

The pace of sound quality improvement of the playback software with DSP is pretty amazing. The guys at Roon are also hard at work developing and improving their DSD chops

We Audiophiles, are the lucky benefactors of this competition

Best Regards,

Ed

 

[manisandher]

What exactly are "taps"?

Here's what Rob Watts has to say (http://www.the-ear.net/how-to/rob-watts-chord-mojo-tech):

OK, the interpolation filter’s job in a DAC is to re- create the missing parts of the original analogue signal - the signal in between one sample and the next. This is done with an FIR filter. In a simple way, a FIR filter consists of a data memory (this stores previous data samples) and a coefficient memory (this is a fixed memory with all of the coefficients that the filter algorithm has created). To create an intermediate data, you simply multiply and add all of the stored data samples with a particular coefficient, and once you have added all of the values you end up with the intermediate value you need. Now in the early days, you used a delay line to store the previous data samples, and you tapped into this delay line in order to access the stored data. Hence the word taps.

So why is it only me that goes on about taps and stuff? The problem is about understanding - when I was at university, I studied electronics. But I was passionate about audio, and was interested in the physiology of hearing. I thought if I understood that, I could make better audio electronics. One of the things I was very interested in was how the brain processed the output from the ears. Now we take our hearing for granted, but the brain does some amazing things to give us auditory perception - separating individual sounds into separate entities with placement data (where are sounds located) is an amazing feat, requiring considerable processing. And we know very little as to how the brain does this. Anyway, what I did learn was that transients were a very important perceptual cue, and that the timing of transients was crucial. From transients the ear brain locates sounds in space, it is also used to compute pitch (particularly for bass) and it’s used for getting the timbre of an instrument. I spent a lot of time researching this in the psychology library, which was close to my hall of residence. Anyway, one of the courses in electronics was Whittaker Shannon sampling theory. And this is the basis of digital audio. From this it is a fact that if you had an infinite tap length FIR filter you would perfectly reconstruct the original bandwidth limited analogue signal in the ADC. It would make no difference if it was sampled at 22uS or 22pS you would have the same digital signal. But it was very clear to me that having a limited tap length would create timing errors. And I know from my studies and from my own listening tests that that would be a major subjective problem.

HTH.

Mani.

 

[amirm]

Here's what Rob Watts has to say (http://www.the-ear.net/how-to/rob-watts-chord-mojo-tech):



HTH.

Mani.

What he says, "OK, the interpolation filter’s job in a DAC is to re- create the missing parts of the original analogue signal - the signal in between one sample and the next. " makes no sense. All that is lost at analog to digital conversion is gone forever. There are no in-between parts to find. You would be violating Shannon's information theory otherwise that he mentions. Fortunately what is there fully describes the analog signal (within its bandwidth) so it doesn't need this bit of magic.

An interpolator is used to change the sample rate without adding distortions of its own. An idea interpolator generates more samples, with the same information, but no distortion. Anything else gives you some distortion with that resampling. See this article I wrote on resampling, using images instead of sound to make the point more clear: http://audiosciencereview.com/forum/index.php?threads/understanding-upsampling-interpolation.1447/

One of the benefits of resampling/interpolation is that you can reduce the sample depth for equiv. amount of information. DSD is an example of that where the sample rate is largely increased vs the CD and the bit depth is reduced to one bit. In the case of Chord, they don't go that far but get down to 5 or 6 bits from what I recall. That is why the interpolator is used there, not because it is trying to find information that is long gone once the audio was digitized.

I am surprised that they did all this signal processing in their DAC but misstate the simple principles of interpolation/resampling.

 

[amirm]

Hi amirm



Our feeling is that 2 million taps is likely to be more extreme than needed, but the question is how much is actually needed.

The sound quality uplift from Polysinc and Polysinc short fiters to the new Polysinc XTR family of filters, especially for PCM to DSD conversion is pretty dramatic, every DSD DAC owner and HQ Player user we know can hear and likes the difference.

As Jussi's work schedule and to do list allows, he will explore optimizing the number of tap's and we will feedback with what we hear. In Munich over dinner, we were discussing with Jussi possible names for a filter family that is not quite so extreme as the XTR family

The pace of sound quality improvement of the playback software with DSP is pretty amazing. The guys at Roon are also hard at work developing and improving their DSD chops

We Audiophiles, are the lucky benefactors of this competition

Best Regards,

Ed

Hi Ed. I will leave subjective matters in your good hands. I am here to comment on objective measures that are provided as figures of merit.

 

[Taiko Audio]

I will pretend I am Taiko and answer.

Tap is a figure of merit for a digital filter. Every tap represents an audio sample that is used to compute the filter response. By multiplying that audio sample with a number and adding to the rest, you compute the value of the new, filtered audio sample. The more taps, the sharper the response of the filter.

Each tap requires memory to hold the audio sample, and of course the more taps, the more computations you have to do.

Note that all of this is for a digital filter called "FIR." IIR filters work recursively and can have very high strength with far less resources (but can have other complications).


The number of taps determines how steep the FIR filter is and how low its pass through energy is. It has nothing to do with Quantization noise. The -240 db is the amplitude of out of band components in HQPlayer filter, not any level of quantization noise.

The 120 db vs 240 is pure "measurebating" by the way. No value whatsoever comes from such increased attenuation.

Yes thank you for the correction, I edited my original post to display the correct type of noise attenuated, the -240 dB figure seems absurd, and so is the amount of computational power involved, however it allows to set the corner frequency of the low pass filter really close to the stop band (it's -0.1dB @ 21.5KHz in the redbook example) which seems to sound different.

 

[Taiko Audio]

Hi Taiko,
Also, what DAC did you switch from the chord and why?
Thank you

I have a (completely personal) preference for DSD playback. I had a very early model of the Dave and it had some serious glitches on DSD playback which apparently have been fixed by now.

 

[opus112]

Tap is a figure of merit for a digital filter. Every tap represents an audio sample that is used to compute the filter response. By multiplying that audio sample with a number and adding to the rest, you compute the value of the new, filtered audio sample. The more taps, the sharper the response of the filter.

The last sentence isn't quite correct. If you want to build a sharper filter you certainly need more taps, but given more taps you can choose to spend that resource in other ways. For example by linearizing the phase response of your filter.

Each tap requires memory to hold the audio sample, and of course the more taps, the more computations you have to do.

Each tap not only needs a memory location to store its audio sample data but also a corresponding coefficient (fixed data).

Note that all of this is for a digital filter called "FIR." IIR filters work recursively and can have very high strength with far less resources (but can have other complications).

Important point. Rob Watts is talking about FIR as far as I'm aware.

The number of taps determines how steep the FIR filter is and how low its pass through energy is.

See, you just contradicted your earlier claim ('more taps means sharper'). The steepness of the filter is orthogonal to the stop band rejection.

 

[Ken Newton]

The following is a simplification.

An instructive way of conceptualizing what an digital oversampling filter does is to picture the unfiltered D/A output spectrum. The music signal occupies repeating pairs of upper and lower side-bands symmetrically centered on multiples of one of the native sample rate (0, 1, 2, 3, etc.). These sample rate multiples begins at zero. So, the first signal band begins with it's upper side-band, and runs from D.C. to 22kHz. This is the signal band we actually utilize for listening. There is an mathematically implied lower sideband, but it would run entirely beneath D.C., so, it doesn't physically exist.

The next pair of signal side-bands are centered on 44.1kHz. The lower side-band band runs from 22.05kHz (44.1kHz - 22.05kHz) to 44.1kHz. The upper side-band runs from 44.1kHz to 66.15kHz (44.1kHz + 22.05kHz). This pattern repeats at each multiple of one of the native sample rate, so the next side-band pair is centered on 44.1kHz x 2 = 88.2kHz.

It's important to recognize that even though the D/A output appears discrete when viewed in the time-domain on an oscilloscope (typically, though not necessarily, appearing stair-stepped) it none-the-less contains a fully devloped analog representation of the signal. It's the many repeating copies (images) of the signal which gives the signal it's discrete appearance. The job of any reconstruction filter, be it digital or analog, is to remove all the repeating image bands, after which, only the expected smooth appearing signal will result.

Oversampling reconstruction filters (also known as, multirate filters) greatly spread out (shift up in frequency) the images by centering them on higher multiples of the native rate. For example, an x8 oversampling filter will center the side-bands at multiples of 8 times the native sample rate (0, 8, 16, 24, etc.). So, instead of the first side-band pair images being centered on 44.1kHz they're centered on 44.1kHz x 8 = 352.8kHz. The next pair are centered at 44.1kHz x 16 = 705.6kHz, etc. The further up in frequency are shifted the image bands the easier it is to completely remove them with a relatively gentle analog filter.

 

[Steve Williams]

Ken

Thank you so much for that. It was easy to read and for me very informative.

 

[Mike Lavigne]

The following is a simplification.

An instructive way of conceptualizing what an digital oversampling filter does is to picture the unfiltered D/A output spectrum. The music signal occupies repeating pairs of upper and lower side-bands symmetrically centered on multiples of one of the native sample rate (0, 1, 2, 3, etc.). These sample rate multiples begins at zero. So, the first signal band begins with it's upper side-band, and runs from D.C. to 22kHz. This is the signal band we actually utilize for listening. There is an mathematically implied lower sideband, but it would run entirely beneath D.C., so, it doesn't physically exist.

The next pair of signal side-bands are centered on 44.1kHz. The lower side-band band runs from 22.05kHz (44.1kHz - 22.05kHz) to 44.1kHz. The upper side-band runs from 44.1kHz to 66.15kHz (44.1kHz + 22.05kHz). This pattern repeats at each multiple of one of the native sample rate, so the next side-band pair is centered on 44.1kHz x 2 = 88.2kHz.

It's important to recognize that even though the D/A output appears discrete when viewed in the time-domain on an oscilloscope (typically, though not necessarily, appearing stair-stepped) it none-the-less contains a fully devloped analog representation of the signal. It's the many repeating copies (images) of the signal which gives the signal it's discrete appearance. The job of any reconstruction filter, be it digital or analog, is to remove all the repeating image bands, after which, only the expected smooth appearing signal will result.

Oversampling reconstruction filters (also known as, multirate filters) greatly spread out (shift up in frequency) the images by centering them on higher multiples of the native rate (0, 8, 16, 24, etc.). For example, an x8 oversampling filter will center the side-bands at multiples of 8 times the native sample rate. So, instead of the first side-band pair images being centered on 44.1kHz they're centered on 44.1kHz x 8 = 352.8kHz. The next pair are centered at 44.1kHz x 16 = 705.6kHz, etc. The further up in frequency are shifted the image bands the easier it is to completely remove them with a relatively gentle analog filter.

well, not too sure about the part I bolded. I suppose it depends on who you ask about the 'fully' part. everyone is entitled to their own opinion based on their experience, biases, and blindspots.

 

[Ken Newton]

well, not too sure about the part I bolded. I suppose it depends on who you ask about the 'fully' part. everyone is entitled to their own opinion based on their experience, biases, and blindspots.

What "fully developed" means, exactly, could be open to interpretation. However, in the frequency domain, the desired analog signal band is fully present. It's located from D.C. to 22kHz. The issue is that there also are undesired repeating copies of the signal band present.