• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). There are daily reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Apodizing filters...how do they work? Why are they considered bad by some?

dumbsuck

Member
Joined
Sep 7, 2022
Messages
9
Likes
5
Hi. I have the Topping D10s which according to what I've read uses the "Apodizing fast linear" filter. Now I am slowly learning about slow/fast linear/minimum phase filters and why they are needed before the signal gets reconstructed from PCM to analog. But I cannot find any details about what makes apodizing filters, as implemented by ESS, different. I love the sound of the D10s and I think it is because of the apodizing filter. I am not sure about this, however. I judge this based on the fact that I like the D10s more than ODAC (original with ESS DAC) which I used to have (sold it recently after comparing it to the D10s) and ODAC from what I know uses a linear phase fast filter. And also I judge this based on that when I upsample in foobar2000 from 44.1 to 96 or 192 KHz (most of my music is in 44.1 KHz) using the SoX plugin, I just cannot make it sound as good as when I do not upsample at all and just send 44.1 to the DAC, no matter what SoX settings I use (minimal/intermediate/linear phase, %passband, allow/disallow aliasing). I just feel like the treble is smoother somehow and I like that. So can anyone please explain how is an apodizing filter different from non-apodizing one?
(all my assumptions are just assumptions and I would likely not recognize the apodizing filter from a non-apodizing one in a blind test so I'm not trying to make a claim that anything is better than anything else, this is all just how I feel I hear things)
 

Sokel

Major Contributor
Joined
Sep 8, 2021
Messages
2,304
Likes
1,835
That's Khadas's (9038qm) apodizing.

filter.PNG
 
OP
D

dumbsuck

Member
Joined
Sep 7, 2022
Messages
9
Likes
5
That's Khadas's (9038qm) apodizing.

View attachment 229350

OK so what I see is that it starts to attenuate around 20KHz, attenuation is at around -110 db at nyquist freq and is below -120 db couple dozen Hz after that. Is there anything else I am supposed to see?
The thing is, I've read all this stuff about apodizing filters "using a variable window based on frequency" (not sure I understand what that would mean) and that they could "smooth out sharp details" from signals etc that I do not understand and would like to have confirmed/denied by someone who knows the details and could explain why it would (not) be so. Also is the use of an apodizing filter the reason why the D10s' THD+N versus frequency goes up to -85 db at 20 KHz? If so, why? (A lot of questions, I know, sorry about that)

Edit: added a question after the 1st sentence
 

Sokel

Major Contributor
Joined
Sep 8, 2021
Messages
2,304
Likes
1,835
OK so what I see is that it starts to attenuate around 20KHz, attenuation is at around -110 db at nyquist freq and is below -120 db couple dozen Hz after that.
The thing is, I've read all this stuff about apodizing filters "using a variable window based on frequency" (not sure I understand what that would mean) and that they could "smooth out sharp details" from signals etc that I do not understand and would like to have confirmed/denied by someone who knows the details and could explain why it would (not) be so. Also is the use of an apodizing filter the reason why the D10s' THD+N versus frequency goes up to -85 db at 20 KHz? If so, why? (A lot of questions, I know, sorry about that)
The capture you see is a multitone (880 tones to be exact) not the classic 1 Khz,etc.That should reveal everything.
Do you see any problem below 20 KHz?Cause I don't.Everything is there,no smoothing,no cutting.
 
OP
D

dumbsuck

Member
Joined
Sep 7, 2022
Messages
9
Likes
5
The capture you see is a multitone (880 tones to be exact) not the classic 1 Khz,etc.That should reveal everything.
Do you see any problem below 20 KHz?Cause I don't.Everything is there,no smoothing,no cutting.

Looks just fine to me, seems the apodizing filter is doing its mysterious magic just fine :)
 

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
17,174
Likes
29,925
If you look at the square wave you would see it has "ringing" after the impulse only. If I get a chance I can post that. I have a D10 Balanced which I would think has the same filtering. As to the significance to the sound, depends probably on other things more than the filter.
 

RandomEar

Active Member
Joined
Feb 14, 2022
Messages
211
Likes
419
Hi. I have the Topping D10s which according to what I've read uses the "Apodizing fast linear" filter. Now I am slowly learning about slow/fast linear/minimum phase filters and why they are needed before the signal gets reconstructed from PCM to analog. But I cannot find any details about what makes apodizing filters, as implemented by ESS, different. I love the sound of the D10s and I think it is because of the apodizing filter. I am not sure about this, however. I judge this based on the fact that I like the D10s more than ODAC (original with ESS DAC) which I used to have (sold it recently after comparing it to the D10s) and ODAC from what I know uses a linear phase fast filter. And also I judge this based on that when I upsample in foobar2000 from 44.1 to 96 or 192 KHz (most of my music is in 44.1 KHz) using the SoX plugin, I just cannot make it sound as good as when I do not upsample at all and just send 44.1 to the DAC, no matter what SoX settings I use (minimal/intermediate/linear phase, %passband, allow/disallow aliasing). I just feel like the treble is smoother somehow and I like that. So can anyone please explain how is an apodizing filter different from non-apodizing one?
(all my assumptions are just assumptions and I would likely not recognize the apodizing filter from a non-apodizing one in a blind test so I'm not trying to make a claim that anything is better than anything else, this is all just how I feel I hear things)
Unless you you can still hear up to 20 kHz (unlikely if you're older than ~20y) or use one of the super slow filters which impact frequency response way below 20 kHz, I would expect no audible difference between the filters.

Upsampling is a bit of a minefield. Basic rule: What is lost, is lost. No amount of upsampling will bring back any information to the audio signal and no correctly executed upsampling should alter the sound. But it will influence where the DAC filter comes into play - which might let some ultrasonic noise pass through to your amp. If you only have 44.1 kHz material, I wouldn't upsample at all. If your source material is composed of different sample rates, I would upsample to 96 kHz and forget about it.

You should also consider that all your tests are sighted - at least you didn't mention any blind testing. Therefore, you should be aware that your impressions of the filters and upsampling may be biased.
 
OP
D

dumbsuck

Member
Joined
Sep 7, 2022
Messages
9
Likes
5
If you look at the square wave you would see it has "ringing" after the impulse only. If I get a chance I can post that. I have a D10 Balanced which I would think has the same filtering. As to the significance to the sound, depends probably on other things more than the filter.

Per Archimago, the balanced version of D10 uses a minimum phase filter, different from the D10s' apodizing one: http://archimago.blogspot.com/2021/08/measurements-topping-d10-balanced-d10b.html

Unless you you can still hear up to 20 kHz (unlikely if you're older than ~20y) or use one of the super slow filters which impact frequency response way below 20 kHz, I would expect no audible difference between the filters.

Upsampling is a bit of a minefield. Basic rule: What is lost, is lost. No amount of upsampling will bring back any information to the audio signal and no correctly executed upsampling should alter the sound. But it will influence where the DAC filter comes into play - which might let some ultrasonic noise pass through to your amp. If you only have 44.1 kHz material, I wouldn't upsample at all. If your source material is composed of different sample rates, I would upsample to 96 kHz and forget about it.

You should also consider that all your tests are sighted - at least you didn't mention any blind testing. Therefore, you should be aware that your impressions of the filters and upsampling may be biased.

I am fully aware that my observations are sighted, not blind, thus have no scientific merit. I tested myself to hear "almost" up-to 18KHz (I am in my mid 30's).
I used upsampling only to test the filtering. When 44.1 KHz PCM is upsampled to 192 KHz in the player, then whatever filtering the DAC does, cannot audibly affect the signal, as all the stuff (ringing, attenuation) happens around nyquist, which is 96 KHz, far from what even cats and dogs can hear. Correct? (I might be wrong ofc.) But now the player's upsampling algorithm comes into play, and if you use SoX, you can play with the phase and other stuff. And even if I completely change all the settings (phase, passband, aliasing), I can hear no difference (but in the foobar2000 plugin, one cannot set the passband to lower than 90%, which is -3 db at 19,845 Hz in case of 44.1 KHz, so inaudible for me). But SoX has no such thing as an "apodizing" filter. So I cannot test that with SoX. And my question still remains... what is special about apodizing filters (specifically as ESS implements them)?
 

RandomEar

Active Member
Joined
Feb 14, 2022
Messages
211
Likes
419
When 44.1 KHz PCM is upsampled to 192 KHz in the player, then whatever filtering the DAC does, cannot audibly affect the signal, as all the stuff (ringing, attenuation) happens around nyquist, which is 96 KHz, far from what even cats and dogs can hear. Correct?
I would agree, yes.

And even if I completely change all the settings (phase, passband, aliasing), I can hear no difference
That's what I would expect for most filters, but I don't know how much you can "crank" the settings in that plugin.

And my question still remains... what is special about apodizing filters (specifically as ESS implements them)?
Have a look in the datasheet of a newer ESS DAC, specifically pages 56 and 58 for this one. Except for the lower stop band attenuation, the filter looks strikingly similar to the fast linear phase one. I don't know what ESS defines as "apodizing". That word itself tells you very little about the filter.
 

Soundgeek Reviews

New Member
Joined
Sep 18, 2022
Messages
2
Likes
1
Hi. I have the Topping D10s which according to what I've read uses the "Apodizing fast linear" filter. Now I am slowly learning about slow/fast linear/minimum phase filters and why they are needed before the signal gets reconstructed from PCM to analog. But I cannot find any details about what makes apodizing filters, as implemented by ESS, different. I love the sound of the D10s and I think it is because of the apodizing filter. I am not sure about this, however. I judge this based on the fact that I like the D10s more than ODAC (original with ESS DAC) which I used to have (sold it recently after comparing it to the D10s) and ODAC from what I know uses a linear phase fast filter. And also I judge this based on that when I upsample in foobar2000 from 44.1 to 96 or 192 KHz (most of my music is in 44.1 KHz) using the SoX plugin, I just cannot make it sound as good as when I do not upsample at all and just send 44.1 to the DAC, no matter what SoX settings I use (minimal/intermediate/linear phase, %passband, allow/disallow aliasing). I just feel like the treble is smoother somehow and I like that. So can anyone please explain how is an apodizing filter different from non-apodizing one?
(all my assumptions are just assumptions and I would likely not recognize the apodizing filter from a non-apodizing one in a blind test so I'm not trying to make a claim that anything is better than anything else, this is all just how I feel I hear things)
Note that upsampling is recommended with a power or 2. So for 44.1, upsample to either 88.2 or 176.4. For 48, upsample to 96 or 192. Don't upsample 44.1 to 96 or 192. This makes the upsampling a lot less straightforward and can impact the sound.

Personally I am a big fan of upsampling. The differences are quite apparent to me. I upsample to 4x PCM or DSD 256 usually. I used a Smooth, Minimum phase filter, but switched back to Smooth, Linear phase. The reason is that a Minimum Phase filter (which to my understanding is an apodizing filter) messes with the phase, which has a negative impact on the imaging and soundstage. Here is a good explanation:


I notice that when using a Smooth, Linear Phase filter I prefer DSD 256 over PCM 4x, while when using a Smooth, Minimum Phase filter I prefer PCM 4x over DSD 256. I think this is because my hardware is already warm leaning, and the loss in spatial detail becomes bigger than the win in digital glare when using DSD. Don't use DSD if your DAC doesn't support it though, but it's fun to play around with these things!
 

charleski

Addicted to Fun and Learning
Joined
Dec 15, 2019
Messages
918
Likes
1,815
Location
Manchester UK
As ‘Mrapodizer’ shows in that post, the term ‘apodizing filter’ is largely meaningless when applied to digital audio. The Meridian ‘apodizing filter’ he describes there is simply a minimum phase design, but it was designed by Bob Stuart (of MQA fame), who is no stranger to the use of hand-waving technobabble to promote his wares.

In general, the term ‘apodizing’ simply refers to the use of a window function in an FIR filter. Its effect is to reduce spectral leakage, a simple rectangular window produces large ‘humps’ in the stop band, which can be controlled by using more complex window functions: (the term literally means ‘remove the foot’)

1024px-Comparison_of_spectral_leakage_of_several_window_functions.svg.png

This has nothing to do with preringing or the phase response.

Basically, any digital filter with a decent stop-band rejection will be ‘apodizing’ - there’s nothing special about it beyond that.
 

Simon P

Member
Joined
Apr 27, 2019
Messages
41
Likes
3
In the general sense, 'apodizing' usually means a substantial window function overlaid on a perfect brickwall filter. 'Apodizing' in digital audio commonly describes digital filters with reduced or no time domain pre-ring when fed with an isolated digital 'impulse' sample. It is possible to design such a filter with good phase/frequency characteristics. They do represent a departure from the time domain sinc function filter which is the theoretical basis for perfect Nyquist Shannon reconstruction. The logic seems to be that pre-ring will sound unnatural; however, as I said, filters with no pre-ring are a major departure from conventional reconstruction theory.
 
OP
D

dumbsuck

Member
Joined
Sep 7, 2022
Messages
9
Likes
5
Thank you all for all the info. Now I wish I had a friend with the D10B who would be willing to bring it over and do a blind test :)
After reading the replies here and stuff online it seems different companies have different definitions of an 'apodizing' filter. I'll just continue enjoying the music from my D10s now...
 
Top Bottom