High resolution audio. The science, or lack of...?

[matt49]

I am primarily trying to understand how different filters apply in the dac. We have seen dacs come to market that have a choice of filters selectable by the user. These measure differently and sound different, so which one is 'right'.

If, as seems likely, all filters have an effect on the sound, how can the dac be 'transparent'.

I seem to remember John Westlake (he of M-DAC fame, which I know you don't like, but I do, so tough!) saying that the most 'correct' of the filters in the M-DAC sounded the least good. OK, so this is only anecdotal and based on subjective listening, and it also doesn't take account of the analogue stage, but interesting nonetheless ...

The M-DAC instruction manual describes the effects of the different filters (p. 7) and may shed some light.

Matt

 

[shadders]

Depending upon the filter design, the filter may have in band ripple - such that the attentuation in band is not a straight line.

I am sure that digital filters exist (please can someone confirm) that have no in band ripple - but essentially, a "digital" filter by its very purpose will modify the data stream - the audio data - before it is converted to an analogue signal.

so in order to avoid the necessity of using a digital anti-aliasing filter we would have to sample the analog input signal, as picked up by the microphones, at high enough sampling frequency so that the Nyquist frequency lies above the highest frequency being sampled. am I right?

Hi,

I recall that the theory is that the analogue input signal to the ADC is band limited such that the highest frequency appearing in the analogue signal to the ADC is at most 1/2 the sampling frequency. That is the input signal is analogue filtered before sampling.

Since a brick wall filter in the analogue domain is not possible - the filter is usually a high order whose cut off frequency (-3dB point) is less than the 1/2 x sampling frequency.

In the reconstruction of the analogue signal - DAC - again, the output analogue filter will remove all components (as much as possible) that are greater than 1/2 x sampling frequency.

Digital filters are used in a DAC when they oversample the data stream - to interpolate the analogue data at the higher sample rate. There is usually an analogue filter on the output of the DAC - but is generally lower in order - less components, to remove the image frequencies (anti aliasing filter).

Regards,

Shadders.

I am primarily trying to understand how different filters apply in the dac. We have seen dacs come to market that have a choice of filters selectable by the user. These measure differently and sound different, so which one is 'right'.

If, as seems likely, all filters have an effect on the sound, how can the dac be 'transparent'.

The PCM F1 was a stand alone device that provided two channels of a to d conversion and two channels of d to a conversion, all at 16/44.1. It could be set up so that the digital output can be fed into the digital input, providing a full a to d to a process.

In one demonstration I took part in, 4 F1 units were used back to back to provide 8 processors and 4 complete a to d to a cycles. It was switched in and out of an otherwise analogue system and no one could reliably tell the difference. Bare in mind that this was a processor designed and built in the early 80s.

Hi,

The F1 units as you have described them in their operation are being used to demonstrate that their precision in ADC and DAC is nearly perfect.

A digital filter will always modify the data - that is its function.

There is no right choice - only a preference.

Todays mainstream audio DAC's include a digital filter for intrpolation - the Texas Instruments PCM1792A has the following specification :

8 x Oversampling Digital filter :− Stop-Band Attenuation: –130 dB

− Pass-Band Ripple:
0.00001 dB

The passband ripple indicates that however small, it will still modify the output from the original (simplistic explanation).

The DAC cannot be transparent - in general - since they all implement a digital filter.

A DAC will always have an error associated with the resolving of the sampling instance.

Regards,

Shadders.

 

[davedotco]

I am primarily trying to understand how different filters apply in the dac. We have seen dacs come to market that have a choice of filters selectable by the user. These measure differently and sound different, so which one is 'right'.

If, as seems likely, all filters have an effect on the sound, how can the dac be 'transparent'.

I seem to remember John Westlake (he of M-DAC fame, which I know you don't like, but I do, so tough!) saying that the most 'correct' of the filters in the M-DAC sounded the least good. OK, so this is only anecdotal and based on subjective listening, and it also doesn't take account of the analogue stage, but interesting nonetheless ...

The M-DAC instruction manual describes the effects of the different filters (p. 7) and may shed some light.

Matt

I have seen this kind of information before with other dacs but the M-Dac manual has one of the clearer explanations.

However it does not quite answer my question, which is that as all filters have a different measured characteristic, and sound different, they must modify the output, so in effect, none of the filters are 'transparent' as they all modify the output.

Since a dac has to have filters, it suggests that no dac is 'transparent' in the true sense, something I find interesting and slightly perplexing.

 

[davedotco]

Hi,

The F1 units as you have described them in their operation are being used to demonstrate that their precision in ADC and DAC is nearly perfect.

A digital filter will always modify the data - that is its function.

There is no right choice - only a preference.

Todays mainstream audio DAC's include a digital filter for intrpolation - the Texas Instruments PCM1792A has the following specification :

8 x Oversampling Digital filter :

− Stop-Band Attenuation: –130 dB

− Pass-Band Ripple:
0.00001 dB

The passband ripple indicates that however small, it will still modify the output from the original (simplistic explanation).

The DAC cannot be transparent - in general - since they all implement a digital filter.

A DAC will always have an error associated with the resolving of the sampling instance.

Regards,

Shadders.

I kind of follow that, in essence there is no such thing as a 100% 'correct' dac. Whatever filter is used, the output is modified.

The PCM F1 tests I detailed earlier showed that the full analogue to digital to analogue cycle was 'effectively transparent' so whatever the effects of the filters in the F1, they were not audible in this case.

 

[shadders]

Hi,

The F1 units as you have described them in their operation are being used to demonstrate that their precision in ADC and DAC is nearly perfect.

A digital filter will always modify the data - that is its function.

There is no right choice - only a preference.

Todays mainstream audio DAC's include a digital filter for intrpolation - the Texas Instruments PCM1792A has the following specification :

8 x Oversampling Digital filter :

− Stop-Band Attenuation: –130 dB

− Pass-Band Ripple:
0.00001 dB

The passband ripple indicates that however small, it will still modify the output from the original (simplistic explanation).

The DAC cannot be transparent - in general - since they all implement a digital filter.

A DAC will always have an error associated with the resolving of the sampling instance.

Regards,

Shadders.

I kind of follow that, in essence there is no such thing as a 100% 'correct' dac. Whatever filter is used, the output is modified.

The PCM F1 tests I detailed earlier showed that the full analogue to digital to analogue cycle was 'effectively transparent' so whatever the effects of the filters in the F1, they were not audible in this case.

Hi,

Yes - as per CnoEvil has stated - there are NOS DAC's that do not use oversampling - and implement an R2R ladder DAC as the conversion stage. This is probably what the F1 was using - a DAC that did not oversample, or if it did - there were no digital interpolation filters implemented.

I am in the process of building a discrete R2R ladder DAC - 24bit with 0.01% precision resistors. I still have to implement the DSP aspect - where i will be using linear interpolation between samples.

That is - i will oversample from the 44.1kHz input stream by 256 times - and hence calculate the linear transfer between two successive 44.1kHz samples - this will be 255 samples to calculate.

I may use a curve fitting algorithm for interpolation.

Depending on my algorithm - it may be more accurate - but my attempt is to replicate as close as possible the original waveform.

Regards,

Shadders.

 

[CnoEvil]

Since a dac has to have filters, it suggests that no dac is 'transparent' in the true sense, something I find interesting and slightly perplexing.

Audio Note have removed all digital filtering, processing etc in their NOS Dacs....you might like to read p.6 / p.7 of their manual for the 2.1x sig.

http://www.audionote.co.uk/downloads/manuals/DAC2.1x%20Signature%20Manual%20smaller.pdf

It sounded wonderful in one of their systems, but not so great in mine, when I borrowed one to try.

 

[matt49]

I have seen this kind of information before with other dacs but the M-Dac manual has one of the clearer explanations.

However it does not quite answer my question, which is that as all filters have a different measured characteristic, and sound different, they must modify the output, so in effect, none of the filters are 'transparent' as they all modify the output.

Since a dac has to have filters, it suggests that no dac is 'transparent' in the true sense, something I find interesting and slightly perplexing.

OK, so a bit of subjective and non-blind listening here. I'm going through the filters on the M-DAC. The only one (of the five) filters that sounds markedly different is the 'optimal spectrum' filter. To me it sounds edgier and less pleasant than the others. Something in the HF isn't good.

I'm listening to lute music by Dowland, very simple, percussive, close-miked, lots of high frequencies. It's a CD ripped to ALAC which the M-DAC receives bit-perfect.

The system is: Cullen-modded Sonos Connect > M-DAC ('Toy' upgrade) > Hifiman HE500s

:cheers:

Matt

 

[davedotco]

Since a dac has to have filters, it suggests that no dac is 'transparent' in the true sense, something I find interesting and slightly perplexing.

Audio Note have removed all digital filtering, processing etc in their NOS Dacs....you might like to read p.6 / p.7 of their manual for the 2.1x sig.

http://www.audionote.co.uk/downloads/manuals/DAC2.1x%20Signature%20Manual%20smaller.pdf

It sounded wonderful in one of their systems, but not so great in mine, when I borrowed one to try.

In an earlier post I mentioned that filters in the playback dac could be implymented either in the digital or analogue domain, in fact the Audio Note NOS dacs were exactly the type of dac I was thinking of, as it uses purely analogue filters

The interesting thing is that without the oversampling, the digital filters or the noise shaping the distortion rises quite alarmingly at low levels, I have seen examples of NOS dacs (not Audio Note) where the distortion content approaches 100% at low levels and is well into double figures with a signal as high as -60dB below full output.

NOS dacs trade the inherent issues of oversampling, digital filters and the rest for high levels of distortion at low levels, presumeably on the principle that the distortion is at such a low level as to be inaudible. It is an interesting approach that mirrors my own concept of 'effective transparancy', ie we know the distortion is there, but can not hear it.

 

[davedotco]

I have seen this kind of information before with other dacs but the M-Dac manual has one of the clearer explanations.

However it does not quite answer my question, which is that as all filters have a different measured characteristic, and sound different, they must modify the output, so in effect, none of the filters are 'transparent' as they all modify the output.

Since a dac has to have filters, it suggests that no dac is 'transparent' in the true sense, something I find interesting and slightly perplexing.

OK, so a bit of subjective and non-blind listening here. I'm going through the filters on the M-DAC. The only one (of the five) filters that sounds markedly different is the 'optimal spectrum' filter. To me it sounds edgier and less pleasant than the others. Something in the HF isn't good.

I'm listening to lute music by Dowland, very simple, percussive, close-miked, lots of high frequencies. It's a CD ripped to ALAC which the M-DAC receives bit-perfect.

The system is: Cullen-modded Sonos Connect > M-DAC ('Toy' upgrade) > Hifiman HE500s

:cheers:

Matt

The interesting thing is that since all filters measure (and in some cases sound) different, there is no one 'correct' fllter, so dacs do indeed sound different. Just not different enough to hear, in many cases anyway.

 

[matt49]

The interesting thing is that since all filters measure (and in some cases sound) different, there is no one 'correct' fllter, so dacs do indeed sound different. Just not different enough to hear, in many cases anyway.

I suspect you're right.

This goes back to something we were saying way back in this thread. Anti-aliasing filters at the recording stage are one thing. Filters at the playback (DAC) stage are another, and whilst in almost all (?) circumstances 'audibly transparent', they are measurably non-transparent.

The M-DAC manual suggests that, with the 'optimal transient' filter, transparency in the time domain is sacrificed for the sake of transparency in the frequency domain. I can't say I really understand the science behind that.

Maybe someone can enlighten us.

Matt

 

[jcbrum]

Maybe someone can enlighten us.

Matt

Hmmm, that's a very tall order, and a big ask . . .

Not because of any particular shortcomings in the contributors, or readers following this thread, but rather because it is a very technical subject, requiring quite a high degree of academic ability, and practical familiarity, with the design process of modern DAC chip assemblies.

I hesitate even to get involved with such an attempt, as it is somewhat adrift from the question raised by the OP . . .

However, in response to your request for enlightenment, this paper delivered to the Audio Engineering Society about 10 years ago might be of assistance.

There are a few caveats which I might suggest, which are that developments of very sopisticated ultra high performance DAC chips have been extensive in the last five years or so, to the extent that the best chips available exceed the requirements of domestic audio by a considerable margin, and are low cost components too, nowadays, due to the very high volume demand. They are often used in devices such as mobile phones and many audio related products, and are not restricted to 'so-called Hi-End audiophile' products. Any quality device might contain such a DAC chip, costing from about £30 upwards.

Furthermore, the best DACs, even in quite low cost devices, are now so good that, pretty much, they all sound the same. Even stand-alone high cost (£ thousands) high end audiophile DACs. This is borne out by recent ABX tests, carried out by hifi enthusiasts, and published on the forums. Many of the participants expressed high preferences, and 'night and day' differences, when demonstrating the DACs under sighted conditions, but were unable to detect any sonic differences under controlled ABX blind test conditions.

IMO, good modern DACs all sound very similar indeed, under domestic listening conditions, even in comparitively low cost equipment, like mobile phones, and WiFi streamers like Sonos and Airport Expresses.

You have been warned

Audio Engineering Convention Paper, 2005.

Wolfson WM8741 very high performance dac information, and data sheet.

JC :wave:

 

[jcbrum]

p.s. A Wolfson WM8741 chip costs about $7 from Digi-Key, in single units. You could make a complete stand-alone, cased, DAC for less than £50, if you have the skills, and can make your own circuit boards. - JC

 

[jcbrum]

p.p.s. Wolfson Microelectronics is a British company, founded in 1984, and based in Edinburgh, Scotland. - JC

 

[chebby]

The PCM F1 was a stand alone device that provided two channels of a to d conversion and two channels of d to a conversion, all at 16/44.1. It could be set up so that the digital output can be fed into the digital input, providing a full a to d to a process.

In one demonstration I took part in, 4 F1 units were used back to back to provide 8 processors and 4 complete a to d to a cycles. It was switched in and out of an otherwise analogue system and no one could reliably tell the difference. Bare in mind that this was a processor designed and built in the early 80s.

Reminiscent of this Boston Audio Society article about the tests that Ivor Tiefenbrun took part in...

http://www.bostonaudiosociety.org/bas_speaker/abx_testing2.htm

 

[andyjm]

The PCM F1 was a stand alone device that provided two channels of a to d conversion and two channels of d to a conversion, all at 16/44.1. It could be set up so that the digital output can be fed into the digital input, providing a full a to d to a process.

In one demonstration I took part in, 4 F1 units were used back to back to provide 8 processors and 4 complete a to d to a cycles. It was switched in and out of an otherwise analogue system and no one could reliably tell the difference. Bare in mind that this was a processor designed and built in the early 80s.

Reminiscent of this Boston Audio Society article about the tests that Ivor Tiefenbrun took part in...

http://www.bostonaudiosociety.org/bas_speaker/abx_testing2.htm

Very interesting. Good for Ivor for taking the test.

I participated in a BBC test in the same year and found 13 bit AtoDtoA was indistiguishable from direct analogue passthrough in a studio / listening room environment.

But hey, the marketing guys say we need 24bits and 192KHz sampling rate, so it must be necessary....

 

[NHL]

The PCM F1 was a stand alone device that provided two channels of a to d conversion and two channels of d to a conversion, all at 16/44.1. It could be set up so that the digital output can be fed into the digital input, providing a full a to d to a process.

In one demonstration I took part in, 4 F1 units were used back to back to provide 8 processors and 4 complete a to d to a cycles. It was switched in and out of an otherwise analogue system and no one could reliably tell the difference. Bare in mind that this was a processor designed and built in the early 80s.

Reminiscent of this Boston Audio Society article about the tests that Ivor Tiefenbrun took part in...

http://www.bostonaudiosociety.org/bas_speaker/abx_testing2.htm

Very interesting. Good for Ivor for taking the test.

I participated in a BBC test in the same year and found 13 bit AtoDtoA was indistiguishable from direct analogue passthrough in a studio / listening room environment.

But hey, the marketing guys say we need 24bits and 192KHz sampling rate, so it must be necessary....

That the Placebo effect is this strong is quite amazing. Must have some darwinistic explanation.

 

[andyjm]

Hi,

Yes - as per CnoEvil has stated - there are NOS DAC's that do not use oversampling - and implement an R2R ladder DAC as the conversion stage. This is probably what the F1 was using - a DAC that did not oversample, or if it did - there were no digital interpolation filters implemented.

I am in the process of building a discrete R2R ladder DAC - 24bit with 0.01% precision resistors. I still have to implement the DSP aspect - where i will be using linear interpolation between samples.

That is - i will oversample from the 44.1kHz input stream by 256 times - and hence calculate the linear transfer between two successive 44.1kHz samples - this will be 255 samples to calculate.

I may use a curve fitting algorithm for interpolation.

Depending on my algorithm - it may be more accurate - but my attempt is to replicate as close as possible the original waveform.

Regards,

Shadders.

Shadders,

Just as an exercise, calculate the % tolerance you will need on your resistor chain so that the resulting linearity error is less than 1/2 LSB on 24bits.

I am getting old and befuddled but my back of the envelope calc says that you need resistors that are 0.000003% accurate.

While it is not a fair comparison as commercial DAC chips will have all sorts of interface handling and other stuff going on, but they will have transistor counts in the 1000's.

My advice would be to leave the DAC chip construction to the experts and go buy one - as pointed out above prices are extremely cheap - less than £10 in sample quantities.

 

[shadders]

Hi,

Yes - as per CnoEvil has stated - there are NOS DAC's that do not use oversampling - and implement an R2R ladder DAC as the conversion stage. This is probably what the F1 was using - a DAC that did not oversample, or if it did - there were no digital interpolation filters implemented.

I am in the process of building a discrete R2R ladder DAC - 24bit with 0.01% precision resistors. I still have to implement the DSP aspect - where i will be using linear interpolation between samples.

That is - i will oversample from the 44.1kHz input stream by 256 times - and hence calculate the linear transfer between two successive 44.1kHz samples - this will be 255 samples to calculate.

I may use a curve fitting algorithm for interpolation.

Depending on my algorithm - it may be more accurate - but my attempt is to replicate as close as possible the original waveform.

Regards,

Shadders.

Shadders,

Just as an exercise, calculate the % tolerance you will need on your resistor chain so that the resulting linearity error is less than 1/2 LSB on 24bits.

I am getting old and befuddled but my back of the envelope calc says that you need resistors that are 0.000003% accurate.

While it is not a fair comparison as commercial DAC chips will have all sorts of interface handling and other stuff going on, but they will have transistor counts in the 1000's.

My advice would be to leave the DAC chip construction to the experts and go buy one - as pointed out above prices are extremely cheap - less than £10 in sample quantities.

Hi,

Yes - these were the most accurate resistors i could locate - the MSB DAC is a discrete R2R Ladder. They do not state accuracy to any LSB - but do state 1kHz -90dB low level linearity is 0.25dB.

Agreed - the linearity will not meet 1/2 LSB - but then - the approach i am using may subjectively sound better than current IC offerings.

This is a hobby for me - so no commercial impact.

I could implement the TI PCM1704 - thse are just under £50 from Farnell - a 24bit Sign Magnitude DAC.

Regards,

Shadders.

 

[davedotco]

The PCM F1 was a stand alone device that provided two channels of a to d conversion and two channels of d to a conversion, all at 16/44.1. It could be set up so that the digital output can be fed into the digital input, providing a full a to d to a process.

In one demonstration I took part in, 4 F1 units were used back to back to provide 8 processors and 4 complete a to d to a cycles. It was switched in and out of an otherwise analogue system and no one could reliably tell the difference. Bare in mind that this was a processor designed and built in the early 80s.

Reminiscent of this Boston Audio Society article about the tests that Ivor Tiefenbrun took part in...

http://www.bostonaudiosociety.org/bas_speaker/abx_testing2.htm

Indeed.

The tests I took part in were inspired by that article. I referenced it earlier in the thread.

No one had a clue which was which using a single F1. Using 4 units some people felt that a difference could be heard some of the time, certainly not a difference that could be 'reliably' picked.

 

[jcbrum]

I think it's generally accepted that Sony CD players performed well in the early days of CD, not least due to competent filters designs.

JC

 

[danielberwick]

I can't believe I got to the end of the thread! I started reading yesterday and now it feels like watching the last episode of a decent TV series, wanting more but knowing for now it’s over!

Anyway, I'm not sure if this helps where the discussion has gone (towards DACs) but I tried an experiment (this is of course anecdotal as it’s my singular experience) but I have the Marantz UD7007 disk player which uses a Texas Instruments PCM1795 DAC and I also have the Sennheiser HDVD800 headphone amp and DAC which has Burr Brown DACs.

The HDVD800 has many inputs available including analogue inputs (that would just hit the amp part of the unit) and digital (which would pass through the inbuilt DAC first before the amp stage). The HDVD800 is very easy to change inputs via a knob and if you switch between them quickly the audio is basically seamless between your original and newly selected input.

To see what the difference was between both DACs I used HD800 headphones from the HDVD800 and connected the UD7007 to the 800 via both balanced XLR cables (analogue using the PCM1795 DACs in the Marantz) and an optical TOSLINK cable (Digital to the 800 so using the Burr Brown DACs in the Sennheiser).

As playing a disk from the player now outputs through both cables at the same time it was easy to switch between inputs, and as it was close to seamless when switching between inputs it was easy to compare. From what I could perceive there was NO discernible difference to the quality or signature from either DAC. I could have just been using the same device for all I could hear. I don’t have bad hearing and I think it’s actually quite good for my early thirties age group although I’m not going to claim any elite hearing ability, just pretty alright.

This, of course, proves nothing conclusively especially as it was just limited to 2 single DACs in 2 devices. However it has made me wonder, if I could flip between many modern DACs seamlessly like this, at what point could I really tell a difference and what would be expectation based bias to favour one over another?

Maybe DACs now are like aspirin, generic white box ones work just as well as expensive sugar coated ones but the expectation is the sugary ones make you feel better faster.

 

[jcbrum]

I think legacy separate 'Hi-end audiophile' DACs are trending towards the junk piles, along with VHS recorders, CD players, Cassette players, and FM tuners.

Modern HiFi only needs a digital source and active speakers, - the dac is simply part of one or the other, nowadays.

JC

 

[shadders]

Hi,

I can definitely hear the difference between the following :

Speakers - this i think is easy to differentiate.

My old Audiolab 8000A (circa 1992) and the Cambridge Audio 650.

My old Pioneer DVD Player DV-717 and my Cambridge Audio DVD Player 540D.

My Cambridge Audio DVD Player 540D and Audiolab 8200AP.

I think for the same generation DAC's, the difference is minimal, when compared to speaker/amplifier combination pairs.

Not sure what the future holds for Hifi separates - DAC's do seem to be popular - quite a few cheap ones available.

Regards,

Shadders.

 

[NHL]

I think legacy separate 'Hi-end audiophile' DACs are trending towards the junk piles, along with VHS recorders, CD players, Cassette players, and FM tuners.

Modern HiFi only needs a digital source and active speakers, - the dac is simply part of one or the other, nowadays.

JC

Inevitable!

(Have the B&W A7 power dock, listen to it 80% of the time, the rest is vinyl)

 

[chebby]

I think legacy separate 'Hi-end audiophile' ...

You missed out 'foo'.

Modern HiFi only needs a digital source and active speakers ...

You missed out AVI before "active speakers".