Apodizing filter applied in software vs on-the-fly?

[silverlight]

So have been reading about the various Meridian articles and PS Audio PWD forums about the apodizing filters used to minimize or eliminate pre-ringing characteristics of red book CD's, having the potential to offset some of fundamental issues with the 16/44 format. So if there are digital apodizing filters in products like the PWD and Meridian 808.2, is there any reason one could not apply these filters outside of the real-time environment in the software world using an SRC with the appropriate profile. I.e., upsampling 16/44 with a program like iZotope RX Advanced that allows user-definable filters. Curious if anyone has done this / experimented with this, etc.? If so, what sort of settings used.

 

[CG]

Yes.

 

I'd try using the minimum phase filter. In iZotrope, that would suggest using a setting of 0 for pre-ringing. In SoX, you might want to use the Minimum phase option, with aliasing allowed and a bandwidth of 85%. Of course, you should use the highest quality sampling setting as well.

 

I'd also suggest only resampling to 88.2 KHz from CDs, since this is an integer conversion rate. 24 bit output files also are worth considering.

 

Please note that like with anything else, there's apodizing filters and there's apodizing filters. Everybody seems to have their own solution. The filter in the Ayre QB-9 is way different from that in the Meridian 808.2, for example.

 

This has been posted previously, but you might find it interesting:

 

http://src.infinitewave.ca/

 

You could make a pretty good argument that unless you have a dedicated hardware filter, doing the conversion not in real time gives far better results.

 

[idiot_savant]

I'd say CG is right ( as always! ), but it depends on the DAC - in CG's example, if your DAC runs natively at ( say ) 8x44.1, then what will happen is:

You apply your apodising filter to double sample rate to 88.2 using SoX

This is output to the DAC at 88.2, at which point the DAC applies it's own 4x filter ( which won't be apodising ). Now... the benefits of the original apodising filter will depend to some extent on the DACs 4x filter at this point - some will assume that high rates will allow gentler rolling off, and will add little ringing to your audio, and some won't...

 

As for the apodising filter, lots of noise has been made about this, with very little attention paid to the fact that an apodising filter has different trade-offs to "normal" filters - some of the apodising stuff has made great fuss about rolling off before nyquist ( i.e. not half-band ), which is definitely _nothing_ revolutionary. Anyways, the real difference in trade-offs are : apodising filters have no ( or little )pre-ringing, lots of post ringing and are by definition not linear phase ( what is minimum phase, btw? ), versus "normal" filters, which have equal pre and post ringing ( which will be smaller ), and guaranteed linear phase. Horses for courses - I've not heard any apodising filters, so can't comment on how they sound.

 

 

your friendly neighbourhood idiot

 

[glt]

In theory, you can use it anywhere. But I'm not sure it it would make any difference applying it in the SRC where the goal it to remove all high frequency junk.

 

In any case, you still have to use a digital filter in the DAC because again you are doing digital upsampling in the DAC and that causes all kind of out of band information that must be filtered. In the end the filter that matters the most is the one just prior to the analog conversion. (And then you have an analog filter built into the following analog stage)

 

Ah, Mr i_s, we meet again :-)

 

Actually minimum phase is not related to apodizing, as the wolfson DAC has a linear phase apodizing and a minimum phase apodizing. And the trade-off is ringing vs aliasing. I've explained it better in my blog...

 

[idiot_savant]

Aha, Mr (glt?) - Damn you - do you expect me to talk?

 

As I understand it common phrasing represents apodising as non-linear phase - having looked at your blog, you would regard it as rolling off before Nyquist.

I hope we can come to some understanding about this ( strokes white cat, menacingly ).

 

Most digital filters use a "half-band" filter - due to the peculiarities of digital filters, for an interpolate by 2 ( typically an interpolate by 8 is a steep x2, followed by 2 softer x2 filters ), due to every other sample being zero stuffed, you can have every other coefficent being zero. This means you get twice the steepness for number of filter operations ( typically a MAC ). However, no matter how long a half-band filter is, it can only have 6dB attenuation at Nyquist ( 22050Hz for CD ). This effectively means that any aliasing from the ADC, and energy from ITS filter will not be attenuated by the DAC filter.

By not making use of this quirk, you effectively need double the grunt for the same steepness, but can begin attenuation anywhere you like - ideally, flat to hearing ( commonly regarded as 20kHz ), but then heavily attenuated at Nyquist ( 22050Hz for CD ) - is this what you mean by apodising?

 

your friendly neighbourhood idiot

 

[glt]

I agree with you that in common phrasing, apodizing (or apodising) is non-linear, but having looked at the Woflson filters, that wouldn't make sense, as there is a linear phase apodizing.

 

I superimposed the non-linear soft roll off filter (filter #2) response with the non-linear apodizing (filter #4) and clearly enough apodizing just means (as defined elsewhere in the technical literature) that it fully attenuates at the nyquist frequency which means either: sharper, earlier roll-off or both.

 

And the reason to use apodizing is to remove whatever aliasing energy exist beyond nyquist.

 

You are probably saying the same thing but I didn't quite get it...

 

BTW I'm no expert in this stuff, just trying to make sense of the big picture by reading other people's stuff...

 

[idiot_savant]

I think the gist of my long ramblings ( which would agree with you )

 

Public understanding of apodising/apodizing = no pre-ringing, hence non-linear phase response

Wolfson apodising/apodizing = not half band filter ( significant attenuation before nyquist ) - earlier roll-off =easier ( in terms of filter ) than steeper ( longer ringing )

OK?

 

I'll shake the Martini

 

btw the FIRST filter ( for CD audio ) is the most critical one, due to the extremely tight transition band

 

your friendly neighbourhood idiot

 

 

 

 

[CG]

Personally, I just tried a lot of combinations and picked the ones that sounded best to me.

 

Then, out of curiosity, I generated a 44.1 KHz 16 bit impulse track and applied the various filter combinations just to see what I could see after the SRC. At least for me, the filter combinations I liked the best definitely had the least pre-ringing and a modest amount of post-ringing.

 

Different systems will respond differently.

 

Different people have different sensitivities, I am sure of as well. (Why do some people really find so called time aligned loudspeakers appealing while other don't care at all?)

 

BTW...

 

If you think this > "I'd say CG is right ( as always! )," perhaps you did legitimately earn the idiot part of your screen name...

 

[idiot_savant]

CG, I've made no secret of earning my name over many, many years...

 

my reference was related to what I have read of yours ( the chocolate jitter machine was spot on! ),

 

your friendly neighbourhood idiot

 

[PeterSt]

This all isn't going to work I'm afraid.

 

You want to apply a filter which already has been applied, and it has unknown properties. Well, be my guest to get that right ...

 

Of course you can try to relate all to how things show after the SRC without applying your own, but do you really think you can get a steeper resonse ? (to name something). You can get a more slow roll off allright, but now find an SRC that's so steep that you'd want that.

 

If you really want to get the merits of this, start off with NOS filterless (and I mean : NOS filterLESS).

 

Btw, Minimum Phase means : throuhout the passband the phase remains unchanged.

Passband : the frequencies you indicate as those to remain audible (never mind it rolls off underway).

 

The less "minimum phase" the less pre-ringing, and the less pre-ringing the more post-ringing.

Or : the less pre-ringing, the more the phase will change towards the higher frequencies.

 

Or (now work this out yourselves) : the less pre- and post ringing, the more imaging.

 

And what about this one : the less ringing the earlier your amp will die.

 

Last one : the more ringing, the less digital sound (one to sleep on).

 

Apply this all to an OS DAC, and you really won't know what will happen.

 

PS: The less ringing with mimum phase, the earlier the roll off.

PPS: only the sun rises for free. But who says ringing is bad ?