Home |
Search |
Today's Posts |
#1
Posted to rec.audio.pro
|
|||
|
|||
anti-aliasing - stochastic sampling?
I've followed the torturous discussion about dithering. I've also read some papers on it from Izotope that were illuminating. This got me thinking (yes, I know...). I worked in computer graphics from around 1980 until 2000. Raster graphics displays of decent quality and resolution were just becoming available in 1980. The frame buffers were very expensive then but I was lucky enough to have access to some. Rendering photorealistic images was the holy grail, and it was quickly apparent that you had to have some understanding of sampling theory to deal with aliasing artifacts. In 1984, researchers at Pixar blew everybody away with their SIGGRAPH papers - an image they produced of motion-bluured billiard balls was on the Proceedings cover. Their breakthrough seemed to solve all the extant problems in one swipe. Beyond simply eliminating the "jaggies," their software allowed the rendering of soft shadows, diffuse lighting, and, with some added trickery, depth-of-field effects. The key to all was stochastic sampling. Before then, ray-tracing programs depended upon a perfectly regular grid of evenly-spaced samples, one per pixel. The usual technique for reducing jaggies was to oversample pixels with more than one "ray", but the subarray was also rectangular and regular. By jittering the samples within a pixel according to a distribution derived by studying monkey retinas(!) a "poisson-disk distribution", the Pixar folks achieved superb anti-aliasing as well as the other effects I mentioned. I guess that in the sampled, digital audio domain this would amount to "jittering" the clock so the interval between samples is not always constant. I can hear the jeers already... Clock instability is an issue discussed in RAP and I'm suggesting making it worse deliberately. Now that I'm old, I can legitimately ask how far off my rocker am I? Jason |
#2
Posted to rec.audio.pro
|
|||
|
|||
anti-aliasing - stochastic sampling?
Jason wrote:
The key to all was stochastic sampling. Before then, ray-tracing programs depended upon a perfectly regular grid of evenly-spaced samples, one per pixel. The usual technique for reducing jaggies was to oversample pixels with more than one "ray", but the subarray was also rectangular and regular. By jittering the samples within a pixel according to a distribution derived by studying monkey retinas(!) a "poisson-disk distribution", the Pixar folks achieved superb anti-aliasing as well as the other effects I mentioned. Right... this is adding jitter or noise in the time domain, whereas dither is adding noise in the amplitude domain. You can think of jaggies as an aliasing artifact of sorts, just as you can think of the wild moire effects on the anchorman's tie as being an aliasing artifact. In the analogue video world, only the vertical is quantized, the horizontal is not. In the digital video world, both are quantized and so now you get those wild tie effects in both dimensions and they become harder to avoid. I guess that in the sampled, digital audio domain this would amount to "jittering" the clock so the interval between samples is not always constant. I can hear the jeers already... Clock instability is an issue discussed in RAP and I'm suggesting making it worse deliberately. Now that I'm old, I can legitimately ask how far off my rocker am I? I think you're off your rocker but it's too early in the morning (and I am still reeling from the Hugo Awards last night) to do the math. If you could add completely gaussian jitter, the end result would be every pure tone separating into a blurred tone with an infinite number of narrow sidebands right by it. This is the sort of thing that causes what the tinfoil hat crowd call a "veiled" sound. I think this would be better than adding uneven jitter that spits off specific discrete sidebands though. --scott -- "C'est un Nagra. C'est suisse, et tres, tres precis." |
#3
Posted to rec.audio.pro
|
|||
|
|||
anti-aliasing - stochastic sampling?
Jason wrote:
I've followed the torturous discussion about dithering. I've also read some papers on it from Izotope that were illuminating. This got me thinking (yes, I know...). I worked in computer graphics from around 1980 until 2000. Raster graphics displays of decent quality and resolution were just becoming available in 1980. The frame buffers were very expensive then but I was lucky enough to have access to some. Rendering photorealistic images was the holy grail, and it was quickly apparent that you had to have some understanding of sampling theory to deal with aliasing artifacts. In 1984, researchers at Pixar blew everybody away with their SIGGRAPH papers - an image they produced of motion-bluured billiard balls was on the Proceedings cover. Their breakthrough seemed to solve all the extant problems in one swipe. Beyond simply eliminating the "jaggies," their software allowed the rendering of soft shadows, diffuse lighting, and, with some added trickery, depth-of-field effects. The key to all was stochastic sampling. Before then, ray-tracing programs depended upon a perfectly regular grid of evenly-spaced samples, one per pixel. The usual technique for reducing jaggies was to oversample pixels with more than one "ray", but the subarray was also rectangular and regular. By jittering the samples within a pixel according to a distribution derived by studying monkey retinas(!) a "poisson-disk distribution", the Pixar folks achieved superb anti-aliasing as well as the other effects I mentioned. I guess that in the sampled, digital audio domain this would amount to "jittering" the clock so the interval between samples is not always constant. I can hear the jeers already... Clock instability is an issue discussed in RAP and I'm suggesting making it worse deliberately. Now that I'm old, I can legitimately ask how far off my rocker am I? Jason I agree with you about the similarities between stochastic sampling and dithering. The underlying philosophy is pretty much the same; things look and sound more natural when the just-perceptible artifacts are smeared. One of the threads in the discussion about dither brought up the use of user-selectable dithering algorithms in some editors. Being on the beta test team for one of those editors, I had the opportunity to discuss this with the developers, and I believe that this option was a recognition that there would be perceptible differences in the final audio that would make the choice of which to use a matter of personal preference. Since those days, just as in digital graphics, photography, video, etc., increased sampling resolution has pushed the artifacts to a very low level. In audio, that level is at or below the level of other "distortions", such as noise, the differences in room acoustics, system performance, and so on. My main disagreement with those who believe that current technology has closed the matter lies in the fact that perception is affected by learning. So, in my opinion, you're still quite on your rocker, but stay tuned for arguments to the contrary! ;-) -- Best regards, Neil |
#4
Posted to rec.audio.pro
|
|||
|
|||
anti-aliasing - stochastic sampling?
On 23 Aug 2015 07:32:23 -0500 "Scott Dorsey" wrote in
article I think you're off your rocker but it's too early in the morning (and I am still reeling from the Hugo Awards last night) to do the math. lol I kinda thought so too but figured I would ask. |
#6
Posted to rec.audio.pro
|
|||
|
|||
anti-aliasing - stochastic sampling?
On 24/08/2015 1:08 a.m., Neil wrote:
So, in my opinion, you're still quite on your rocker, but stay tuned for arguments to the contrary! ;-) Dithering is best with a 3kHz noise signal ?!! geoff |
#7
Posted to rec.audio.pro
|
|||
|
|||
anti-aliasing - stochastic sampling?
On 8/23/2015 4:00 PM, geoff wrote:
On 24/08/2015 1:08 a.m., Neil wrote: So, in my opinion, you're still quite on your rocker, but stay tuned for arguments to the contrary! ;-) Dithering is best with a 3kHz noise signal ?!! geoff Troll bate? == Later... Ron Capik -- |
#8
Posted to rec.audio.pro
|
|||
|
|||
anti-aliasing - stochastic sampling?
On 24/08/2015 8:24 a.m., Ron C wrote:
On 8/23/2015 4:00 PM, geoff wrote: On 24/08/2015 1:08 a.m., Neil wrote: So, in my opinion, you're still quite on your rocker, but stay tuned for arguments to the contrary! ;-) Dithering is best with a 3kHz noise signal ?!! geoff Troll bate? == Later... Ron Capik Surely he wouldn't bother reading a thread like this ? Apologiesin advance if he actually does ;-0 geoff |
#9
Posted to rec.audio.pro
|
|||
|
|||
anti-aliasing - stochastic sampling?
On 8/23/2015 4:24 PM, Ron C wrote:
On 8/23/2015 4:00 PM, geoff wrote: On 24/08/2015 1:08 a.m., Neil wrote: So, in my opinion, you're still quite on your rocker, but stay tuned for arguments to the contrary! ;-) Dithering is best with a 3kHz noise signal ?!! geoff Troll bate? I was wondering the same thing. -- Best regards, Neil |
#10
Posted to rec.audio.pro
|
|||
|
|||
anti-aliasing - stochastic sampling?
On Sunday, August 23, 2015 at 4:00:39 PM UTC-4, geoff wrote:
On 24/08/2015 1:08 a.m., Neil wrote: So, in my opinion, you're still quite on your rocker, but stay tuned for arguments to the contrary! ;-) Dithering is best with a 3kHz noise signal ?! I'm Ok with that!! Err, so you understand, I'm OK with that!! Jack geoff |
#11
Posted to rec.audio.pro
|
|||
|
|||
anti-aliasing - stochastic sampling?
On Sunday, August 23, 2015 at 10:12:57 PM UTC-4, Neil wrote:
On 8/23/2015 4:24 PM, Ron C wrote: On 8/23/2015 4:00 PM, geoff wrote: On 24/08/2015 1:08 a.m., Neil wrote: So, in my opinion, you're still quite on your rocker, but stay tuned for arguments to the contrary! ;-) Dithering is best with a 3kHz noise signal ?!! geoff Troll bate? I was wondering the same thing. I corrected Geoff when absolutely no one here could/would, his use of "K" vs "k". Glad to see he can actually learn. Maybe not when buying headphones, but eventually he'll find a pair he likes!! Jack -- Best regards, Neil |
#12
Posted to rec.audio.pro
|
|||
|
|||
anti-aliasing - stochastic sampling?
On 24/08/2015 6:24 AM, Ron C wrote:
On 8/23/2015 4:00 PM, geoff wrote: Dithering is best with a 3kHz noise signal ?!! Troll bate? bate? |
#13
Posted to rec.audio.pro
|
|||
|
|||
anti-aliasing - stochastic sampling?
Frank Stearns wrote:
(Scott Dorsey) writes: snips still reeling from the Hugo Awards last night) to do the math. You provided PA, or in a parallel life are you an S-F writer??? (So who won?) This year I did some system design but I didn't actually mix it. I did stay up watching it on streaming video and counting sound cues, though. So you can't blame me for the terrible mike positioning at the pre-awards panel this year... last year they got my 441s. Winners were actually pretty good... none of the nominees from the current attempts at stuffing the ballot box got awarded, but due to the stuffing there were a bunch of NO AWARD ones given out this year. The Rabid Puppies crew pretty much lost all their nominations except for Guardians of the Galaxy which was actually pretty good. And I was pleased that Three Body Problem won. --scott -- "C'est un Nagra. C'est suisse, et tres, tres precis." |
Reply |
Thread Tools | |
Display Modes | |
|
|
Similar Threads | ||||
Thread | Forum | |||
the superiority of 96 kHz sampling Options no one mentioned aliasing | Pro Audio | |||
the superiority of 96 kHz sampling Options no one mentioned aliasing | Audio Opinions | |||
aliasing on Win Mobile playback- | Pro Audio | |||
Should I use an anti alias filter during resampling from lower sampling rate to higher? | Pro Audio | |||
Why 24/96 sampling isn't necessarily better-sounding than 24/44 sampling | Pro Audio |