DacMagicPlus Filter Mirror Images - Short Study

[restorer-john]

A 13th order passive Chebyshev filter with ferrite core inductors is probably more damaging to the audio than allowing some imaging. Just because he wrote “the book” doesn’t really mean much. It’s not like he invented sampling theorem.

Probably more damaging? What does that mean? Probably? Having a bet each way? By all means, regale us with your evidence. I have plenty of 1st generation CD players which offer spectacular HF performance with 90dB brick wall filters and nobody who has heard them A/Bd with equally well designed modern players or standalone D/A converters can perceive any difference on 16/44 (obviously).

And, as for Heitaro Nakajima (R.I.P.) "not meaning much", that just goes to show how phenomenally ignorant and utterly disrespectful some people can be. Japan recognized him with their highest civilian honour for his achievements in digital pioneering/recording and the invention of the Compact Disc (among other things). Go do some research, read some actual books, study the achievements and come back.

The reason we have lazy 24kHz filters is because of lazy implementations using D/As and digital filters primarily designed for AVRs and the 48kHz bitstream of PCM they mostly deal with. But 2ch audio is mostly delivered at 16/44.1.

It is very clear to me that given that 99.9% of all DACs have filters allowing some images through, there is no real issue here. The few products that do it correctly aren’t even universally preferred among subjectivists.

99.9% huh? Where did that statistic come from? I know the answer.

And the universal preference of subjectivists thrown in as well?

 

[chris719]

The reason we have lazy 24kHz filters is because of lazy implementations using D/As and digital filters primarily designed for AVRs and the 48kHz bitstream of PCM they mostly deal with. But 2ch audio is mostly delivered at 16/44.1.

I don't have time for someone as clueless as you. You should look at the datasheet of any converter with integrated filter. If you think that they are designed for 48 kHz you are WRONG. The passband and stopband scale with the sampling frequency. You should educate yourself a bit more before you pretend to be an authority.

If you had any real experience designing products you would know this, because almost EVERY DAC chip with integrated filter acts like this and also has the stopband at around 0.54 to 0.55*Fs in "single-speed" mode.

Attached, spoonfed examples for someone who cannot read. It's funny, the AK4493 example even spells out that the filter is for 44.1 kHz. The AD1955 example is also completely obvious. They ALL have the same issue and allow images, even at 48 kHz. So yeah, you can take your distortion from a 13th order passive filter, oh wise guru. I really don't give a damn who this guy is or was; a 13th order passive filter is stupid. No one in their right mind would design a reconstruction filter like that today. Why should we care what civilian honor Japan decided to give some random dude? If you were such an objectivist you wouldn't be arguing from authority.

Go find me a DAC product currently on sale that doesn't use the interpolation filter inside the DAC IC. Think a little bit out how many of those are sold. You can look at the volumes of product shipped by Cirrus Logic, AKM, TI, and Analog Devices in comparison. Thanks.

 

[mansr]

You should look at the datasheet of any converter with integrated filter. If you think that they are designed for 48 kHz you are WRONG. The passband and stopband scale with the sampling frequency.

True, but a filter that is flat to 20 kHz at 48 kHz sample rate might not be when the sample rate is reduced to 44.1 kHz.

 

[chris719]

True, but a filter that is flat to 20 kHz at 48 kHz sample rate might not be when the sample rate is reduced to 44.1 kHz.

No offense, but did you read at all any of what I posted or the image? The default sharp roll-off linear filter is always flat to 0.46*Fs give or take a little bit... so 20 kHz for 44.1 and 22 kHz for 48. It is like this for every DAC chip that I could bother to Google. This behavior is even in old filter chips like DF1704 and NPC SM5847. Even Pacific Microsonics PMD100...

 

[restorer-john]

True, but a filter that is flat to 20 kHz at 48 kHz sample rate might not be when the sample rate is reduced to 44.1 kHz.

Exactly. A point conveniently ignored by old mate. The majority of content is 44.1 and the filter characteristics are optimised for 48.

 

[chris719]

Exactly. A point conveniently ignored by old mate. The majority of content is 44.1 and the filter characteristics are optimised for 48.

Apparently you don't understand how FIR filters work. This is not my problem. If it were optimized for 48 kHz as you alledge, that makes no sense since it's a different set of coefficients, first of all. Second, the pictures I have attached clearly show that the filters suffer the exact same problem at 48 kHz.

If you can't understand this basic logic then I am sorry and you are not equipped to discuss further.

TL; DR: Almost every digital filter integrated into a DAC chip or in a separate ASIC that I can find for the past 20 years defaults to a flat passband to just short of 0.5*Fs, which means they ALL allow imaging since they can't have 20 billion taps. Virtually every audio device sold relies on the built-in interpolation filters, I can then easily and correctly conclude that this "issue" affects pretty much every audio device sold. This is not a 44.1 kHz issue, the issue is the SAME for 48 kHz.

 

[mansr]

No offense, but did you read at all any of what I posted or the image?

I read all of what you had posted, which consisted mostly of insults, at the time I opened the thread.

 

[chris719]

I read all of what you had posted, which consisted mostly of insults, at the time I opened the thread.

Amazing. Maybe one day you guys can learn that when something is multiplied by the sample rate, that means it's scaled by the sample rate.

The 44.1k filters are flat to 20k, the 48k filters are flat to 22k. This isn't rocket science. The idea that the filters perform worse at 44.1 or are optimized for 48 kHz is 100% wrong and if you spent 5 minutes reading any number of DAC datasheets you will come to the same conclusion.

Sorry to offend your under-educated friend, but you know, he started out attacking Topping in the other thread without any merit and now he seems to take offense at me pointing out that a 1980s approach to reconstruction filters is not a good example of what should be done.

 

[pma]

A 13th order passive Chebyshev filter with ferrite core inductors is probably more damaging to the audio than allowing some imaging.

They did not have another chance in ancient times of digital audio. It cannot be viewed from 2021 perspective. However, nowadays any designer could use the proper digital filters if he wanted and if he was not limited by marketing decisions only.

 

[restorer-john]

@pma

Here's a vintage Marantz PMD-325 "professional" CD player with the Cirrus CS-4396 and a discrete constructed LPF playing a 19999Hz 0dBFS tone, using one of my standard test discs.

And a nice alternating FFFF/0000 at 22.05kHz also a standard test. Notice the residual is roughly 100dB down at FS/2.

Here's a vintage 1990 Sony CDP-338ESD using BB PCM-58P(J)s with 8x OS Sony developed filter (CXD-8003S) with 19999Hz 0dBFS.

Here's the Sony again, playing the FFFF/0000 22.05kHz. Notice the extreme suppression at FS/2.

 

[pma]

Here's a vintage Marantz PMD-325 "professional" CD player with the Cirrus CS-4396 and a discrete constructed LPF playing a 19999Hz 0dBFS tone, using one of my standard test discs.

Nice regarding the old stuff. The problems usually appear at 21kHz rather than at 20kHz. You can burn your own CD-R and check the behaviour.

 

[restorer-john]

Nice regarding the old stuff. The problems usually appear at 21kHz rather than at 20kHz. You can burn your own CD-R and check the behaviour.

I have a lot of industry service/test/alignment discs. Can't remember the last time I burnt a CD.

 

[chris719]

I'm sorry but can't you be bothered to read two numbers off any DAC datasheet from the past 20 years? BTW, even the player you picked may be an outlier, even if we assume it's true since I have no way to verify if you performed your test correctly. If you want to go vintage, you can still see this trend. Don't give me any 48 kHz BS because you can't read. 44.1 kHz CD audio was the very point of these chips since they were in CD players and the coefficients are different for each sampling frequency.

Take a look at the common older digital filter datasheet excerpts attached. Gee, it seems like none of them enter the stopband by Fs/2. They all claim 24 kHz, aka .5465*Fs as in every damn DAC datasheet you open even today.

I will not be expecting a reply, since this behavior has been well-known to be common for 30 years and yet you ignore the mountains of datasheets telling you that you're wrong. Any player or DAC with huge attention at 22.050 kHz is an outlier.

It's amazing, I've typed in basically any digital filter I could even find or remember the part number of and they all have the same behavior. Yamaha, TI, NPC, even Pacific Microsonics!

 

[chris719]

They did not have another chance in ancient times of digital audio. It cannot be viewed from 2021 perspective. However, nowadays any designer could use the proper digital filters if he wanted and if he was not limited by marketing decisions only.

I totally agree. I just don't think bringing that up in 2021 is relevant. That filter is almost certainly going to have worse effects than some residual images in the transition band, and yet I get this reply from @restorer-john attacking me for saying that because I don't worship at the altar of some random guy.

I, personally, will take a built-in linear phase DAC filter with marketing specs rather than deal with a 13th order analog LPF using ferrite core inductors and including its side effects. What do you think?

One consideration I might add is that not all DAC ICs have an external filter interface mode now. Anyway, a lot of words for something that is less important than the filter in the ADC.

 

[chris719]

@pma

Here's a vintage Marantz PMD-325 "professional" CD player with the Cirrus CS-4396 and a discrete constructed LPF playing a 19999Hz 0dBFS tone, using one of my standard test discs.

And a nice alternating FFFF/0000 at 22.05kHz also a standard test. Notice the residual is roughly 100dB down at FS/2.

Here's a vintage 1990 Sony CDP-338ESD using BB PCM-58P(J)s with 8x OS Sony developed filter (CXD-8003S) with 19999Hz 0dBFS.

Here's the Sony again, playing the FFFF/0000 22.05kHz. Notice the extreme suppression at FS/2.

Here's the CS4396 datasheet for your information. Note that it also doesn't reach the stopband until 0.5465*Fs. Wow, must be an international conspiracy.

Of course, I looked into your Marantz and Sony and found that similar models included a ridiculous brickwall analog LPF. So yeah, not relevant. A brickwall analog LPF is stupid and nothing's used them for a very long time. I found your post on them, it seems like you have some analog LPF fetish. This explains a lot.

 

[Mnyb]

Hello friends I'm in for a noob friendly explanation to the international conspiracy of 0.5465*Fs . Yes it will scale to every Fs you can pick (got that ).

What do we get for letting some imaging in ? what is worth the tradeoff . I don't think its explained at all in the tread .

OP @pma states that the DAC Macic has this property you @chris719 states that basically everything does for the last 30 odd years.

IS this a property of math or chip design as this coefficient appears with 4 decimal places in silicon made by different manufacturers ?

I wonder if there is a real benefit or some spec war to always be absolutely flat to 20kHZ. The bennefit may not be for the consumer but DAC makers demands some property of the silicon chip dewsign that naturally gets you to 0.5465*Fs as the low hanging fruit way of doing it ? what ever it is ?

The obvious next Q is this even necessary in 2021 ? cant we have it both ways now with software ?

It may be as you say a non issue , but its still a why in here ?

 

[danadam]

What do we get for letting some imaging in ? what is worth the tradeoff . I don't think its explained at all in the tread .

How about that for explanation?: https://www.audiosciencereview.com/...-review-balanced-dac.20259/page-7#post-668025

There will only be some imaging 20k-22khz from the leakage from 22-24khz. So no issue whatsoever. If you want to have better attenuation at 22khz it means it will start rolling off at lower frequency in real world. So basically you are trading for something you absolutely can't hear(imaging over 20khz) from something you potentially can hear(frequency response variation). Sharper filters (if you want both flat 20khz response and good attenuation at 22khz) also cause other issues like intersample overs(digital clipping basically). So this way you'd back off digital level to avoid clipping. 3db attenuation means 3db loss in SNR. It also adds noise component in THD+N /SINAD. So you are also trading off some performance in the whole audible band for something most people absolutely cannot hear.

 

[chris719]

Hello friends I'm in for a noob friendly explanation to the international conspiracy of 0.5465*Fs . Yes it will scale to every Fs you can pick (got that ).

What do we get for letting some imaging in ? what is worth the tradeoff . I don't think its explained at all in the tread .

OP @pma states that the DAC Macic has this property you @chris719 states that basically everything does for the last 30 odd years.

IS this a property of math or chip design as this coefficient appears with 4 decimal places in silicon made by different manufacturers ?

I wonder if there is a real benefit or some spec war to always be absolutely flat to 20kHZ. The bennefit may not be for the consumer but DAC makers demands some property of the silicon chip dewsign that naturally gets you to 0.5465*Fs as the low hanging fruit way of doing it ? what ever it is ?

The obvious next Q is this even necessary in 2021 ? cant we have it both ways now with software ?

It may be as you say a non issue , but its still a why in here ?

The DacMagic steep filter handles it correctly. Most stuff does not. It's purely marketing or habit at this point with the chip designers. The only DACs you will find on sale today that correctly handle this have an external filter in an FPGA or DSP chip. This is a tiny list as I'm sure you know.

I think there is no real reason as to why it's done today other than marketing or because Yamaha or whoever started doing it in the mid 80s and everyone copied it. The reason it was initially done is almost surely that they prized a flat response to 20 kHz and it would have taken more taps to implement than they could afford. (narrower transition band).

You can fix this in software if you upsample 44.1 and 48 kHz content to, say, 96 or 192 using software with a very steep filter.

Edit: Another point that was raised by John's post linked by danadam is that of possibly needing more headroom for intersample overs. I'm not sure if it's really a practical concern with an ultra-sharp filter and normal content, though.

 

[Mnyb]

The DacMagic steep filter handles it correctly. Most stuff does not. It's purely marketing or habit at this point with the chip designers. The only DACs you will find on sale today that correctly handle this have an external filter in an FPGA or DSP chip. This is a tiny list as I'm sure you know.

I think there is no real reason as to why it's done today other than marketing or because Yamaha or whoever started doing it in the mid 80s and everyone copied it. The reason it was initially done is almost surely that they prized a flat response to 20 kHz and it would have taken more taps to implement than they could afford. (narrower transition band).

You can fix this in software if you upsample 44.1 and 48 kHz content to, say, 96 or 192 using software with a very steep filter.

Edit: Another point that was raised by John's post linked by danadam is that of possibly needing more headroom for intersample overs. I'm not sure if it's really a practical concern with an ultra-sharp filter and normal content, though.

Thank you both .

When several people arrive at the exact same solution it's either as you say copy paste (hey nothing wrong with that I'm also a lazy engineer. it's how you get things done and also the solution is tested for you ).

Or the physics or math's force you this way by natural law .

I've implemented my age correct solution in SoX on some of my players actually starting the filter already at 19 ish kHZ as I'm 51 and worked in heavy industry all my life , when you don't hear to 20k anymore the filter dont need to be super steep

 

[restorer-john]

Of course, I looked into your Marantz and Sony and found that similar models included a ridiculous brickwall analog LPF. So yeah, not relevant.

Ah, sorry, you'd be wrong there again. Perfectly relevant, I'm afraid.

The Sony concerned uses a custom IC noise shaping digital filter up front of the BB PCM-58 multibit D/A converters, a standard IV stage and a buffer stage. All very SOTA in 1990.

Tell me, why would Sony be using a "ridiculous brickwall analog LPF" at 8x OS (352kHz), in 1989/90 huh? Some of us were actually there, and not still in kindergarten at the time. Their filtering was digital, all FIR in custom silicon, as were all the big players.

Analogue, tuned active filters went out after the second or third generation machines in 1986/7. They were too expensive and the promise of savings from 2, 4 or 8 times OS was too hard to resist. By the time Philips brought a bitstream D/A to the market in 190/91, the days of "brick wall filters" were long past. It's such a pity that accurate, useful and factual information seems to be so hard for people to find, read and comprehend.

The Marantz concerned uses a switched capacitor, dynamic element matching delta-sigma D/A and also does not use any such "ridiculous brickwall analog LPF". Tell me, what makes you think otherwise? An internet chat site, or a social media lynch mob?

Stop spreading falsehoods and fake information or attempting to rewrite history to suit your narrative. It won't work. Ever.

I think there is no real reason as to why it's done today other than marketing or because Yamaha or whoever started doing it in the mid 80s and everyone copied it.

Yamaha held onto multibit longer than anyone else. The "mid 80s" were owned by the multibit R2R ladder D/As (and DEM Philips hybrids). In fact they (Yamaha) trumpeted their "HiBit" technology, long after others had moved to single bit/PWM or in the case of early Philips, PDM. HiBit used a novel bit shifting arrangement to increase linearity at low levels with ladder D/A converters. They were not concerned with in band ripple and it was not a factor in marketing with any brand at the time. It was all about THD and S/N. S/N being king.

All OS/digital filtering was about, was improving some spec the customers would believe made their product better than the competition. The fact it was always really about saving money was never explained. Obviously.

Laser trimming D/A silicon was expensive. Trimming the four top bits manually via trimpots on the production line for low level THD was expensive.

So, we got delta sigma, out of band noise, cheap, alignment-free gear and eventually a decent performance. It took a long time and there are still compromises, especially with D/As designed to take everything from 12bit to 24bit (or more) and sampling rates up to daylight. Personally, if I consume my audio in 16/44, I want a system optimised for that, not a jack of all trades, master of none.