R/E/P Community

Please login or register.

Login with username, password and session length
Advanced search  

Pages: [1]   Go Down

Author Topic: Graven's Paper on the latest AES Journal  (Read 1962 times)

Zoesch

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 269
Graven's Paper on the latest AES Journal
« on: April 25, 2004, 12:31:24 am »

Since this is now the forum for all digital and analog tech talk... has anyone read Graven's paper on Apodizing anti-alias filters?

Been slowly digesting it for the past week, doing the math, simulating results and smacking myself in the forehead! Something so simple could possibly change digital audio forever (If properly implemented).

For those who haven't read the paper, what's involved in apodization is similar to convolution between the anti-imaging filter response and a specifically selected apodizing function to reduce the effect of Gibbs phenomenom at sample rates close to Nyquist in the transition band of the filter.

Such an implementation could result in less artifacts, less intermodulation distortion and simpler reconstruction filter design.

Anyone for comments?
Logged
It has always been Ringo's fault

John Klett

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 475
Re: Graven's Paper on the latest AES Journal
« Reply #1 on: April 26, 2004, 11:01:19 am »

That sounds interesting.  I don't have the issue because...  well I feel foolish admitting this...  but after 25 years of being a dues paying member of the AES I bailed last year as I was not feeling any benefit.  
I should probably reconsider this but for now I don't have the paper.  Here is the full abstract...

Antialias Filters and System Transient Response at High Sample Rates
Peter G. Craven   216
With the use of very high sampling rates, a designer has additional options for balancing the conflicting requirements in both the time and frequency domains. Lower sampling rates require brick-wall filters, which produce time smear. By using a class of gentle frequency filters, called apodizing, pre- and postringing can be reduced or removed. It is argued that these temporal artifacts justify the use of higher sampling rates. While there is no attempt to prove which combination of parameters is perceptually optimum, there are clearly a wide range of choices and consequences.



AES...  I have the first 50 years of journal issues however, and do go into them with some frequency - almost never those from the last five to seven years I think.  The problem for me has been that I am not designing or developing in the areas that 99 percent of the papers cover.  There is a HUGE time factor for me...  I get engrossed in current and upcoming projects and then it's the work, family, work, house, work etc. thing so I have grown to rely on people who, like you and quite a few, who can take the time and answer my questions when I ask.  I follow the topics and developments but, really, I am dealing with high resolution digital audio on a day to day basis only empirically.  I am using and interfacing some incredibly good sounding product as part of the small amount of high resolution digital surround mixing and mastering work I am doing.  That is one of the reasons I cut "Bleeding Edge Digital" from subtitle for this forum...

I'd really appreciate it if you could come back and discuss, or present, your digest of that paper.  Maybe that will motivate me to catch up on my dues and open up some time to stay current with my knowledge base and keep from getting still and rusty with my math skills.
Logged
John Klett / Tech Mecca
http://www.technicalaudio.com

JackJohnston

  • Full Member
  • ***
  • Offline Offline
  • Posts: 144
Re: Graven's Paper on the latest AES Journal
« Reply #2 on: April 26, 2004, 07:47:46 pm »



I would like to know more about Apodizing Masks, I wasn't able to find the article on the internet. Do you have a link to it?

Have you read much about wavelets?

Zoesch

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 269
Re: Graven's Paper on the latest AES Journal
« Reply #3 on: April 28, 2004, 08:38:19 pm »

John, I understand you completely... I didn't renew my dues until late last month for the same reason (But since I work with internet and digital audio for a variety of systems, the contents of the Journal are getting to be more interesting for my POV).

So without further ado, here's the gist of the paper.

We are all aware that the main problem with digital audio is the performance of the DAC chain, not just the converter stability (Clock and jitter), but the performance of the anti-imaging and reconstruction filters, specially the effects on the time-domain characteristics of the signal and the introduction of "time-smear" due to the brickwall filters used.

What Graven reckons (Correctly I must add) is that when people listen to material recorded at higher sample rates (88.2KHz and upwards) through a suitable reproduction chain, they perceive an increase in sound quality and that this is mainly because the transition bands in this case are well beyond even the outer hair cells detection threshold (This is my take, not his, I've just finished devouring some papers by Brownell on the effect of HF waves on positioning and separation and the ear ability to detect them through the outer hair cells).

Graven's solution is quite simple, place an apodization mask on the anti-imaging filter (His idea is that this is best suited for the reproduction and mastering chains) to reduce time-smear by minimizing ripple and ringing while maintaining phase linearity and minimizing underswings in the impulse response.

The fun part of this, is that apodization has been used in optics for ages, specially in interferometers, to reduce distortion and artifacts.

How does placing an apodizing filter on the chain translate to real world performance? Well, with an apodizing filter on the chain both the anti-imaging and reconstruction filters can be half-band instead of full band. That means a system designed for 44.1 KHz reproduction can perform like a system designed for 88.2 KHz reproduction!

It's a very simple concept, obviously it hasn't been tested and developed enough to be fully proven (Graven's just proving it theoretically) but if made to work the effects would be impressive indeed.
Logged
It has always been Ringo's fault

chrisj

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 959
Re: Graven's Paper on the latest AES Journal
« Reply #4 on: May 11, 2004, 11:20:39 am »

I'm doing work on exactly this area, trying to do stuff in the mastering domain to minimize ringing and overshoot. Only very high frequency stuff really causes this, but you can get a lot of that in modern mastering on digital-source stuff that's bright and has been amplified a lot. It's possible that the source of that 'digital sound' some people find objectionable is simply the lack of distortion on these sounds aggravating reconstruction filter ringing.
I can't claim to have all the answers but I have a lot of the questions ;) at any rate, I'm interested in hearing more about this. Every time I do, I generally get some more ideas to apply to my own work. Most recently, a message from Zoesch gave me the insight that my slew clipping probably shouldn't be clipping to any fixed amount- even with gradual onset, as that is still loss of information- but should be clipping to a filtered version of the same sound, to confine the loss of information to the extreme highs and retain what low and midrange information is available.
Off to re-code and implement that one...

Zoesch

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 269
Re: Graven's Paper on the latest AES Journal
« Reply #5 on: May 11, 2004, 09:31:02 pm »

One thing that I'd try is (because of the symetrical properties of convolution) to encode the signal post mastering with an apodizing function before playback...

Basically you have your mastered signal F(s), your anti-imaging filter response H(s) and your apodization signal A(s)... Graven's proposed chain looks like

F(s)x(H(s)xA(s)) which equates to F(t)*(H(t)*A(s))

But nothing stops you from doing

(F(s)xA(s))xH(s) which equates to (F(t)*A(t))*H(t)

So building a plugin that convolves the mastered signal with the apodized signal before it hits the DAC should provide the same results as modifying the DAC anti-imaging filter with an apodizing function.

BTW I have a couple of ideas about symbolic signal reconstruction, but maybe those are for another thread or for e-maildom Very Happy
Logged
It has always been Ringo's fault
Pages: [1]   Go Up