or Connect
AVS › AVS Forum › Audio › Receivers, Amps, and Processors › "Official" Audyssey thread (FAQ in post #51779)
New Posts  All Forums:Forum Nav:

"Official" Audyssey thread (FAQ in post #51779) - Page 2313

post #69361 of 70896
Quote:
Originally Posted by IgorZep View Post


As long as the FIR filter is casual (meaning no pre-ringing on the impulse response, or that the current state depends only on the current sample and past samples and no knowledge about the future), and Audyssey filter is casual, there is no delay independent of the length of the FIR (or delay is negligible and equals to something like a single sample time that is ~1/48000 sec).

 

I am not a DSP engineer myself, but I have never heard of nor encountered a FIR filter with near zero delay in all the years that I have worked as an audio/acoustics engineer. A standard FIR filter has a delay of exactly taps/2 = delay in samples

 

As for what I know, casual filters used in real time will always have a fixed delay that directly corresponds to the amount of taps and the sample frequency.

post #69362 of 70896
Quote:
Originally Posted by kbarnes701 View Post
 

 

Good post. But I am unclear on your comment above that XT32 does not have over 10,000 taps and how that reconciles itself with Igor's and batpig's posts (where the math shows 16k taps). Would you be able to elaborate on it for me please?  Do you mean, maybe, that XT32 has 16k of 'virtual taps'? Thanks.

 

Yes, sorry for my lacking explanation. I do believe that XT32 has the frequency resolution comparable to 16k taps (at least in the bass region), but that the actual implementation features a lot less taps.

 

This by all means is a good thing, as the available taps are used where they are needed instead of spread out with equal spacing over the entire frequency area. Doing it this way with "virtual taps" is a lot more efficient (less processing power required) and allows the use of linear phase FIR filters (a lot better for audio than minimum phase FIR filters).

post #69363 of 70896
Quote:
Originally Posted by jjazdk View Post
 
Quote:
Originally Posted by kbarnes701 View Post
 

 

Good post. But I am unclear on your comment above that XT32 does not have over 10,000 taps and how that reconciles itself with Igor's and batpig's posts (where the math shows 16k taps). Would you be able to elaborate on it for me please?  Do you mean, maybe, that XT32 has 16k of 'virtual taps'? Thanks.

 

Yes, sorry for my lacking explanation. I do believe that XT32 has the frequency resolution comparable to 16k taps (at least in the bass region), but that the actual implementation features a lot less taps.

 

This by all means is a good thing, as the available taps are used where they are needed instead of spread out with equal spacing over the entire frequency area. Doing it this way with "virtual taps" is a lot more efficient (less processing power required) and allows the use of linear phase FIR filters (a lot better for audio than minimum phase FIR filters).

 

Thanks very much. This is a very interesting discussion. It's good to see these things discussed objectively.

post #69364 of 70896
Quote:
Originally Posted by jjazdk View Post

A standard FIR filter has a delay of exactly taps/2 = delay in samples
It is not a standard filter, it is a linear phase filter that has such delay (due to symmetric IR, or looking forward same amount of samples as looking back). This is a really bad idea to use such filters for Room EQ... as the room issues are not linear phase, they are causal (meaning the effect of a reflection, and hence the correction is always later than the direct sound wave).
Quote:
Originally Posted by jjazdk View Post

As for what I know, causal filters used in real time will always have a fixed delay that directly corresponds to the amount of taps and the sample frequency.
You are wrong. Causal filters are easily implemented as real-time filters that do not need any (considerable) delay (except the time needed to process one sample through all taps, that is surely proportional to the amount of taps, but in total is less than one sample time).
Edited by IgorZep - 1/25/14 at 5:17am
post #69365 of 70896
Quote:
Originally Posted by jjazdk View Post

Quote:
Originally Posted by kbarnes701 View Post

 

Good post. But I am unclear on your comment above that XT32 does not have over 10,000 taps and how that reconciles itself with Igor's and batpig's posts (where the math shows 16k taps). Would you be able to elaborate on it for me please?  Do you mean, maybe, that XT32 has 16k of 'virtual taps'? Thanks.

Yes, sorry for my lacking explanation. I do believe that XT32 has the frequency resolution comparable to 16k taps (at least in the bass region), but that the actual implementation features a lot less taps.

This by all means is a good thing, as the available taps are used where they are needed instead of spread out with equal spacing over the entire frequency area. Doing it this way with "virtual taps" is a lot more efficient (less processing power required) and allows the use of linear phase FIR filters (a lot better for audio than minimum phase FIR filters).

Chris has frequently discussed how they don't quote the exact number of taps because their propriety IP has a more effective way of distributing the resources, making a comparison to standard FIR filters "misleading", as in the snippet I posted yesterday. So I wonder if the XT32 taps are literally "more than ten thousand" or if that's an "effective" resolution if you were talking about standard FIR filter.
Quote:
We don't publish the number "x" because the method by which Audyssey FIR filters are implemented is not the standard textbook method and so that would lead to misleading comparisons. A frequency warping technique is used to distribute the FFT bins nonlinearly in the frequency domain so that they are narrower in the lower frequencies.
post #69366 of 70896
Quote:
Originally Posted by IgorZep View Post


It is not a standard filter, it is a linear phase filter that has such delay (due to symmetric IR, or looking forward same amount of samples as looking back). This is a really bad idea to use such filters for Room EQ... as the room issues are not linear phase, they are casual (meaning the effect of a reflection, and hence the correction is always later than the direct sound wave).
You are wrong. Casual filters are easily implemented as real-time filters that do not need any (considerable) delay (except the time needed to process one sample through all taps, that is surely proportional to the amount of taps, but in total is less than one sample time).

 

I am sorry Igor, but I believe you are the mistaken one here.

 

As I write this I am discussing FIR filters with one of our Audio DSP engineers, he also believes that you are mistaken. Not that this proves anything, but I must admit I believe him more than you, after all he works with Audio DSP algorithms every day and is quite knowledgeable.

 

If time permits we will create a little MatLab program that allows us to process an audio file with a 16000 tap (low delay) filter and listen to the output.

 

Using filters that does not care about the phase would be stupidity itself, because the DRC does not know which parts of the signal is direct or reflections. Thus the filter would effectively make chaotic phase, which is definitely not the objective here.

If on the other hand the filter tries to correct the phase with frequency as the x-axis (I have only heard one such system and that was fabulous) you for sure do not have filters with zero delay, quite the contrary.

 

If I am wrong, I have no problem admitting it, but I fail to see that it is realistic that XT32 indeed uses 16k real taps. Also, it would require 32x the processing power of XT, and we all know that AVR manufacturers do not like to put expensive DSP's in their products, so a 32x improvement in processing power is not likely either.

post #69367 of 70896
Quote:
Originally Posted by batpig View Post


Chris has frequently discussed how they don't quote the exact number of taps because their propriety IP has a more effective way of distributing the resources, making a comparison to standard FIR filters "misleading", as in the snippet I posted yesterday. So I wonder if the XT32 taps are literally "more than ten thousand" or if that's an "effective" resolution if you were talking about standard FIR filter.

 

Well, that is actually my point :) As in that they probably have 16k virtual taps, which is a lot more "processing power effective" than implementing a real 16k tap filter. This without any negative consequences since there really is no need for 16k real taps.

post #69368 of 70896
Yeah, I understood that was your point smile.gif it was more open ended, thinking out loud for my own benefit as I tried to digest the discussion.

Out of curiosity, since you (unlike rank amateurs like myself) obviously have some professional expertise in the field and are in a unique position to be able to talk to an audio DSP engineer, what are your/his thoughts on the dicussion of the excessive "hair" in the high freqs of XT pre-out graphs vs. XT32, and the seeming absense of low frequency correction. Do the data / conclusions discussed in Keith's FAQ addendum seem reasonable to someone who does this stuff professionally? http://www.avsforum.com/t/795421/official-audyssey-thread-faq-in-post-51779/51750#user_addendum-a1
post #69369 of 70896
Quote:
Originally Posted by batpig View Post

Yeah, I understood that was your point smile.gif it was more open ended, thinking out loud for my own benefit as I tried to digest the discussion.

Out of curiosity, since you (unlike rank amateurs like myself) obviously have some professional expertise in the field and are in a unique position to be able to talk to an audio DSP engineer, what are your/his thoughts on the dicussion of the excessive "hair" in the high freqs of XT pre-out graphs vs. XT32, and the seeming absense of low frequency correction. Do the data / conclusions discussed in Keith's FAQ addendum seem reasonable to someone who does this stuff professionally? http://www.avsforum.com/t/795421/official-audyssey-thread-faq-in-post-51779/51750#user_addendum-a1

 

Good question.

post #69370 of 70896
Quote:
Originally Posted by jjazdk View Post

I am sorry Igor, but I believe you are the mistaken one here.
You believe, but I know. This is the difference. Someone who told you that linear phase filters does not affect time domain properties of the signal in a negative way (the pre-ringing) - does not understand something very fundamental about what is the FIR, it's impulse response and how the frequency response and phase response are derived from impulse response. And, just confirming this is a fact that NONE of the successful modern Room EQ processors are using linear phase filters (read the Dirac paper I am posting here from time to time wink.gif ) Some crossovers do - it works well there under some constraints.

Here is a "confirmation" from Audyssey: Impulse response of XT and XT32 - they are both causal filters.
Quote:
Originally Posted by jjazdk View Post

If time permits we will create a little MatLab program that allows us to process an audio file with a 16000 tap (low delay) filter and listen to the output.
What the filter should do?
Quote:
Originally Posted by jjazdk View Post

Using filters that does not care about the phase would be stupidity itself
It is exactly why linear phase filters should not be used for Room EQ - it is just stupid to use filter that need twice as much resources only to produce unwanted time-domain artifacts that are really newer exists in real life (no single string instrument in existence produce sound before string is touched by the musician, no drum produce any sound before drumstick touched it), why the correction algorithm should do it? Just to see measured phase response on pre-outs as flat line? The phase graph is only part of the story, it does not tell everything about the signal and about time domain, impulse response is the full story!
Quote:
Originally Posted by jjazdk View Post

If I am wrong, I have no problem admitting it, but I fail to see that it is realistic that XT32 indeed uses 16k real taps.
Switching contexts... Anyway... XT32 does not use 16k real taps. They use multi-rate filtering - this is the key word. They are explicitly telling us that in XT32 they are splitting signal into ranges and process them under different sample rates, then filter-out unwanted artifacts and join them back. And... returning back to the phase - they may well use linear phase filters to split and join the signal (adding some delay, but it is not so long if the split frequency is quite high) and the unwanted distortion produced by the linear phase filters will be cancelled out when joining bands - so it would be a good use of technology, but the resulting filter is still (mostly) the causal one.
Edited by IgorZep - 1/25/14 at 5:16am
post #69371 of 70896
Quote:
Originally Posted by batpig View Post

and the seeming absense of low frequency correction
It is not seeming, it is just absence rolleyes.gif
post #69372 of 70896
Quote:
Originally Posted by batpig View Post

Yeah, I understood that was your point smile.gif it was more open ended, thinking out loud for my own benefit as I tried to digest the discussion.

Out of curiosity, since you (unlike rank amateurs like myself) obviously have some professional expertise in the field and are in a unique position to be able to talk to an audio DSP engineer, what are your/his thoughts on the dicussion of the excessive "hair" in the high freqs of XT pre-out graphs vs. XT32, and the seeming absense of low frequency correction. Do the data / conclusions discussed in Keith's FAQ addendum seem reasonable to someone who does this stuff professionally? http://www.avsforum.com/t/795421/official-audyssey-thread-faq-in-post-51779/51750#user_addendum-a1

I really don't know why, when everyone knows where to find Chris K. (the creator), the question(s) are not addressed to him directly. In a recent post I offered my help to those who would like to phrase a question but do not have an FB account, but to no avail. Maybe this time! smile.gif
post #69373 of 70896
Quote:
Originally Posted by IgorZep View Post


Switching contexts... Anyway... XT32 does not use 16k real taps. They use multi-rate filtering - this is the key word. They are explicitly telling us that in XT32 they are splitting signal into ranges and process them under different sample rates, then filter-out unwanted artifacts and join them back. And... returning back to the phase - they may well use linear phase filters to split and join the signal (adding some delay, but it is not so long if the split frequency is quite high) and the unwanted distortion produced by the linear phase filters will be cancelled out when joining bands - so it would be a good use of technology, but the resulting filter is still (mostly) the casual one.

Wow, if Audyssey really does this, it sounds perfect to me. Just really amazing, what they are doing - if this is true.But I believe, what you are writing Igor and this give me an insight, I never expected to hear about. From my own hearing experiences I have to say, I don't want to miss my Audysey Pro set-up. The results after my last Pro calibration are stunning for me. Watched WWZ today and this was really a treat in in my acoustically untreated home cinema.

This is really an interesting discussion. smile.gif
post #69374 of 70896
Guys, you are being rather casual with terminology. The term is causal. tongue.gif
post #69375 of 70896
^^ I thought that's what they meant. smile.gif But you never know - there's greedy genetic algorithms, so a 'casual' filter doesn't have completely strange connotations LOL...
post #69376 of 70896
Quote:
Originally Posted by Roger Dressler View Post

Guys, you are being rather casual with terminology. The term is causal. tongue.gif

 

LOL. Thanks Roger. Like Stuart I too assumed that is what they meant (English not being their first language, despite their excellent command of it). I couldn't figure out how 'casual' would make sense.

post #69377 of 70896
Quote:
Originally Posted by Roger Dressler View Post

Guys, you are being rather casual with terminology. The term is causal. tongue.gif
Sure eek.gif sorry for my English quirks biggrin.gif
post #69378 of 70896
Changed the casual to causal in my posts. Thanks for correcting tongue.gif
Edited by IgorZep - 1/25/14 at 5:15am
post #69379 of 70896
It's spelled "causal". No second "u".
post #69380 of 70896
Quote:
Originally Posted by IgorZep View Post
 
Quote:
Originally Posted by Roger Dressler View Post

Guys, you are being rather casual with terminology. The term is causal. tongue.gif
Sure eek.gif sorry for my English quirks biggrin.gif

 

As someone who can get by in 3 foreign languages, all I can say, Igor, is that I wish my command of any of them was as good as your command of English.

post #69381 of 70896
Quote:
Originally Posted by Selden Ball View Post

It's spelled "causal". No second "u".
Ahmm.. I need to correct it again biggrin.gif Thank you one more time.
post #69382 of 70896
Sunmary of the last 10 pages:

1. Causal - this is how the word is spelled.
2. Audyssey has a decent calibrated mic.
3. maximum number of listening area measurements is best.
4. You can NOT easily improve or replicate audyssey's room correction by trying to outsmart it on distance and channel volume settings.
post #69383 of 70896
Quote:
Originally Posted by blazar View Post

Sunmary of the last 10 pages:

1. Causal - this is how the word is spelled.
2. Audyssey has a decent calibrated mic.
3. maximum number of listening area measurements is best.
4. You can NOT easily improve or replicate audyssey's room correction by trying to outsmart it on distance and channel volume settings.

Sunmary? smile.giftongue.gifwink.gif
post #69384 of 70896
Quote:
Originally Posted by kbarnes701 View Post

As someone who can get by in 3 foreign languages

3 different languages in one head? Why that's just impossible. < says the american >
post #69385 of 70896
Quote:
Originally Posted by kbarnes701 View Post
 

 

As someone who can get by in 3 foreign languages, all I can say, Igor, is that I wish my command of any of them was as good as your command of English.

 

And we enjoy your quirks as well, Igor.  ;)

post #69386 of 70896
Quote:
Originally Posted by mogorf View Post

Sunmary? smile.giftongue.gifwink.gif



Are we all using our best Emglish today? biggrin.gif
post #69387 of 70896
Quote:
Originally Posted by D Bone View Post
 
Quote:
Originally Posted by kbarnes701 View Post

As someone who can get by in 3 foreign languages

3 different languages in one head? Why that's just impossible. < says the american >

 

4 in fact. I also speak English :)

post #69388 of 70896
Quote:
Originally Posted by blazar View Post


4. You can NOT easily improve or replicate audyssey's room correction by trying to outsmart it on distance and channel volume settings.

 

In fact you can considerably improve on it by using the 'sub distance tweak'... 

post #69389 of 70896
Quote:
Originally Posted by kbarnes701 View Post

In fact you can considerably improve on it by using the 'sub distance tweak'... 

Hehe yeah im waiting on my mic to arrive in the mail and then defintely level out those subs smile.gif
post #69390 of 70896
I just ordered a Denon X4000 with MultiEQ XT 32, so I'm putting my SVS- AS-EQ1 for sale in the Classified Forum:
http://www.avsforum.com/t/1514257/svs-as-eq1-sub-equalizer-room-correction-system/0_50#post_24268499
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Receivers, Amps, and Processors
AVS › AVS Forum › Audio › Receivers, Amps, and Processors › "Official" Audyssey thread (FAQ in post #51779)