Topping D90: What is Purpose of Digital Filters?

[mcdonalk]

As many of you know, the D90 has six PCM and two DSD filter settings from which to choose. What is the purpose of these filters and why would I want to change them from the default?

 

[Koeitje]

Just put it to sharp, its the best one. Filters keep out high frequency artifacts that are created by oversampling.

 

[mcdonalk]

I don't understand. The purpose of the filters being what you say, why are there several filters from which to choose? Would not any filter perform elimination of artifacts? Why would I select one over the other? Are some filters suited for some digital formats (e.g. 192kHz) and one filter for another (e.g. 96kHz)?

 

[majingotan]

Read this to better understand the filters: https://www.audiosciencereview.com/...derstanding-digital-audio-measurements.10523/

Per article,

The audibility impact of such filters is likely very low to non-existent so I don’t put a lot of value on this test.

Choose whatever filters you want unless you can hear as good as a dog or a bat to be able to tell audible difference between filters

 

[Veri]

The only ones that can cause audible high frequency roll-off are the slow, but most notably 'super slow' or non-oversampling modes if available.

 

[Matias]

Good reading here
http://archimago.blogspot.com/2017/12/howto-musings-playing-with-digital_23.html?m=1

 

[basis]

I don't understand. The purpose of the filters being what you say, why are there several filters from which to choose? Would not any filter perform elimination of artifacts? Why would I select one over the other? Are some filters suited for some digital formats (e.g. 192kHz) and one filter for another (e.g. 96kHz)?

I think most people feel confused about PCM and DSD because the commonly used graph representations can be misleading for non-technical persons. I'd like to use an analogy to digital images. PCM and DSD can be both seen as 1D digital images. For 2D digital images, a bitmap is a 2D array of pixels where each pixel can only be value 0 or 1, or saved with 1 bit of information. To represent grayscale, one way is to increase the bits for each pixel. A common Jpeg has 8 bits per pixel per color channel (color channels can be regarded as stereo or surround sound; therefore discussion of one channel is sufficient). Another way to represent grayscale is to simply increase the spatial resolution of a bitmap. In fact, all printers use the latter method, as we can only have an ink dot or blank on a piece of paper. So, a PCM file is like a Jpeg image, and a DSD file a bitmap or a printed image. For images, we usually don't see the dots or pixels until looking at a very close distance. The filters that smooth out the pixels or dots are actually our eyes and maybe brains. No matter the resolution of an audio file, it is saved as discrete sample points. Unlike computer screens or printers, a DAC needs to turn discrete sample points into continuous voltage levels. Mathematically, the process of turning discrete samples (Dirac deltas) into a continuous function can be simply called filtering: computed as a convolution of the impulse sequence with a filter kernel. Therefore, filtering is the conversion itself in a DAC. The options in DAC settings are just different filter kernels. Still using the digital image analogy, the basic idea is to blur the images a little so that the pixel borders become less obvious. But unlike digital images, we can't easily "zoom in" to hear the subtleties in audio files. If someone can hear the difference among audio filters, it's just like a person having a sharp vision to see the printer dots or display pixels at arm's length.
To actually answer your questions after the background: 1) filters are necessary because that's what a converter does (converter == filter); 2) you can have an infinite number of filter kernels defined mathematically; the engineers of a DAC just choose some for easy implementation; 3) filters are not suppose to eliminate artifacts, as I assume artifacts are those from recording and mastering; again, filters are necessary because converter == filter; 4) the actually implemented filters in a DAC are not supposed to be differentiated by a normal person; think about filtering a digital image, it is to blur pixel borders, which are not supposed to be seen for a high resolution image; 5) the higher sample rate of an audio file, the less obvious of the filtering effects. You may use some machine to see the difference; but the machine doesn't know which is better.

 

[UCrazyKid]

I tried all the filters. I could not hear a difference with my 50 year old ears. I asked my 11 yo daughter who is already quite a musician what she thought. She could hear differences and said “filter 2 is the best for the most natural sounding analog instruments”. So I took her word for it have kept it there.

 

[Tomáš Papoušek]

D90 90K bandwidth distortion, The sound is a bit bad.
The parameters are good, but the actual sound is not emotional.

 

[tw99]

D90 90K bandwidth distortion, The sound is a bit bad.
The parameters are good, but the actual sound is not emotional.

How are you measuring this "emotional"ness ? Please note that this forum is not really about uncontrolled subjective listening impressions. Pretty much any other audio forum would be better if that's what you're looking for, but maybe you'd like to stick around and educate yourself on the fallibilities of such impressions.

 

[wannarock2]

The only ones that can cause audible high frequency roll-off are the slow, but most notably 'super slow' ....

yes, this would surely be the smoothest sound. must be that velvety thing.
mode 4 super slow

 

[Dr Eas]

I think most people feel confused about PCM and DSD because the commonly used graph representations can be misleading for non-technical persons. I'd like to use an analogy to digital images. PCM and DSD can be both seen as 1D digital images. For 2D digital images, a bitmap is a 2D array of pixels where each pixel can only be value 0 or 1, or saved with 1 bit of information. To represent grayscale, one way is to increase the bits for each pixel. A common Jpeg has 8 bits per pixel per color channel (color channels can be regarded as stereo or surround sound; therefore discussion of one channel is sufficient). Another way to represent grayscale is to simply increase the spatial resolution of a bitmap. In fact, all printers use the latter method, as we can only have an ink dot or blank on a piece of paper. So, a PCM file is like a Jpeg image, and a DSD file a bitmap or a printed image. For images, we usually don't see the dots or pixels until looking at a very close distance. The filters that smooth out the pixels or dots are actually our eyes and maybe brains. No matter the resolution of an audio file, it is saved as discrete sample points. Unlike computer screens or printers, a DAC needs to turn discrete sample points into continuous voltage levels. Mathematically, the process of turning discrete samples (Dirac deltas) into a continuous function can be simply called filtering: computed as a convolution of the impulse sequence with a filter kernel. Therefore, filtering is the conversion itself in a DAC. The options in DAC settings are just different filter kernels. Still using the digital image analogy, the basic idea is to blur the images a little so that the pixel borders become less obvious. But unlike digital images, we can't easily "zoom in" to hear the subtleties in audio files. If someone can hear the difference among audio filters, it's just like a person having a sharp vision to see the printer dots or display pixels at arm's length.
To actually answer your questions after the background: 1) filters are necessary because that's what a converter does (converter == filter); 2) you can have an infinite number of filter kernels defined mathematically; the engineers of a DAC just choose some for easy implementation; 3) filters are not suppose to eliminate artifacts, as I assume artifacts are those from recording and mastering; again, filters are necessary because converter == filter; 4) the actually implemented filters in a DAC are not supposed to be differentiated by a normal person; think about filtering a digital image, it is to blur pixel borders, which are not supposed to be seen for a high resolution image; 5) the higher sample rate of an audio file, the less obvious of the filtering effects. You may use some machine to see the difference; but the machine doesn't know which is better.

Very helpful. Thanks.

 

[Kaiede]

I'd like to use an analogy to digital images.

Very helpful. Thanks.

To be blunt, this is a terrible analogy, and makes the concept harder to understand rather than easier. The filters in this case are bandwidth limiting filters, and serve an entirely different purpose than anything described.

Bandwidth limiting filters are required for Nyquist-Shannon sampling to work, where you can demonstrably prove that the input and output analog signals are the same. Some DAC chips just include features like a selectable bandwidth filter for the output of the chip. However, there should generally be little to no difference between the filters, assuming they are properly engineered. Specifically for 44.1kHz (CD) audio, the goal is that all content above 22kHz should be cut off, and that the rolloff should start at 20kHz. The reason for the design was to accommodate bandwidth filters of the time, and having a 2kHz range specifically for the bandwidth filter that was out of the range of human hearing meant that the ”slower” filters would allow for cheaper products. Honestly, as long as the filter isn’t doing something dumb like starting the rolloff at too low a frequency, or failing to cut off content above 22kHz, you shouldn’t be able to hear any difference between them. The engineers working on the CD audio spec ensured that would be the case.

 

[Dr Eas]

But it can help a simpleton like me!

Since you were kind to reply, any thoughts re. "jitter"? I've been unable to solve an intermittent jitter from my computer (using a high quality USB A to B cable) to the Topping D90 MQA? Tried Audio Quest's "Jitter Bug" and ifi's iPurifer3 to no avail.

 

[daftcombo]

I tried all the filters. I could not hear a difference with my 50 year old ears. I asked my 11 yo daughter who is already quite a musician what she thought. She could hear differences and said “filter 2 is the best for the most natural sounding analog instruments”. So I took her word for it have kept it there.

That was the sharpest, right?

 

[Kaiede]

Since you were kind to reply, any thoughts re. "jitter"? I've been unable to solve an intermittent jitter from my computer (using a high quality USB A to B cable) to the Topping D90 MQA? Tried Audio Quest's "Jitter Bug" and ifi's iPurifer3 to no avail.

It might help if you clarify what you mean by "jitter". The D90 should be handling any USB timing jitter just fine, based on the measurements WolfX-700 and Amir both made of the D90. So something meant to address that won't help.

If you mean something like a cutout or skip in the audio, that's likely related to OS, drivers, or software on your computer.

 

[Dr Eas]

Not sure I can explain in writing what "jitter" means in my experience. It is more like an intermittent cut-out, not skipping. Almost like if you were attaching speaker wire to terminals and had not yet made a good connection. I've done all I can do on my computer (updated, uninstalled / reinstalled, removed various software) but it's hard to trace or diagnose. Thanks again.

 

[Kaiede]

Not sure I can explain in writing what "jitter" means in my experience. It is more like an intermittent cut-out, not skipping. Almost like if you were attaching speaker wire to terminals and had not yet made a good connection. I've done all I can do on my computer (updated, uninstalled / reinstalled, removed various software) but it's hard to trace or diagnose. Thanks again.

I'd still say it's related to something happening on the computer, as I did mention cutouts. And it can be difficult to diagnose. The OS, drivers, and playback app all play a part. But what's likely happening is something somewhere in the software stack isn't passing the audio data along quickly enough.

One thing I have run into is that if I am using the built-in Windows audio mixer, and I increase the sample rate, it doesn't always work well, and I can get crackle and cutouts on some DACs. I generally leave the sample rate at whatever default gets picked. I've seen Windows pick 44.1k and 48k, so not sure which one Windows is "meant" to operate in.

I also tend to be skeptical that higher sample rates offer anything truly valuable, so I'm not really convinced it is a big loss.

 

[ReinhardS]

Jitter is just noise in the time domain, leading to unwanted side-frequencies especially at high signal frequencies. You need at least several hundred ps of jitter to have an impact on a 16 bit music signal... on a 24 bit signal the quality would be degraded earlier, but that is probably nothing we can hear.

 

[nelamvr6]

Not sure I can explain in writing what "jitter" means in my experience. It is more like an intermittent cut-out, not skipping. Almost like if you were attaching speaker wire to terminals and had not yet made a good connection. I've done all I can do on my computer (updated, uninstalled / reinstalled, removed various software) but it's hard to trace or diagnose. Thanks again.

If you are using the USB input to your D90 then jitter is not anything to think about. At all. Jitter is simply a non-issue with USB input since USB is asynchronous. If you look at the Jitter test results from Amirm`s original test you'll see that what litter jitter is present is way below the threshold were you will be able to hear it. Even using the S/PDIF input it's way below audible.

It doens't sound like you're describing jitter. At all.