Using a logarithmic view, white noise looks less "dense" in the bottom (try switching to linear). Also it depends on stuff like the window function - a rectangular window function can have up to 3 dB of scalloping loss. Testing with the "flat top" window, and lanczos interpolation, I found it to be accurate to within 0.01 dB error, using a precise sine oscillator. Well, riddle me this: If a pure sine measures completely flat, how would it be able to skew only white noise?emarsk wrote:I know that the noise spectrum looks less "dense" in the low end, that's not the issue. The point is that white noise should have - by definition! - a spectrum that looks flat on average. And it does indeed, in every other spectrum analyzer I have except Signalizer, which gives me a 1dB/oct slanted graph instead, with every window function in its list.
It depends on what you're plotting. Practically all "normal" FFT analysers plot straight bin-magnitudes as approximations of the peak magnitudes at various frequencies and this type of plot will result in flat line for white noise (when set for "unweighted" display, such that a sine sweep measures flat). In such a plot it's really irrelevant whether the frequency axis is set for linear or logarithmic, because the latter just places the data-points visually closer at higher frequencies.Mayae wrote: And yes, I can see why one would say it's skewed. But white noise is not supposed to "look" flat in logarithmic views, since it distributes energy equally/linearly over the whole frequency band.
I think what I was previously referring to with this segment:mystran wrote:It depends on what you're plotting. Practically all "normal" FFT analysers plot straight bin-magnitudes as approximations of the peak magnitudes at various frequencies and this type of plot will result in flat line for white noise (when set for "unweighted" display, such that a sine sweep measures flat). In such a plot it's really irrelevant whether the frequency axis is set for linear or logarithmic, because the latter just places the data-points visually closer at higher frequencies.
Is that, while mathematically, averaged over infinity, such a plot should be completely flat, when you're sampling moving windows and plotting it logarithmically, it gives the illusion that the bottom is less flat, because there's less energy density that would statistically "fill" up the space, contrary to the highs where there's a larger density of bins per pixel (and the highs have quicker response due to # number of cycles over a window). So there's a higher chance that the highs more quickly converge to the expected value.Mayae wrote: And yes, I can see why one would say it's skewed. But white noise is not supposed to "look" flat in logarithmic views, since it distributes energy equally/linearly over the whole frequency band.
):
I'm just trying to say that practically all FFT-type analyzers will show you a perfectly flat line, because they are plotting magnitudes, not energy density.Mayae wrote: Is that, while mathematically, averaged over infinity, such a plot should be completely flat, when you're sampling moving windows and plotting it logarithmically, it gives the illusion that the bottom is less flat, because there's less energy density that would statistically "fill" up the space, contrary to the highs where there's a larger density of bins per pixel (and the highs have quicker response due to # number of cycles over a window). So there's a higher chance that the highs more quickly converge to the expected value.
Yes, I just read your answer again and I get what you're saying.mystran wrote:I'm just trying to say that practically all FFT-type analyzers will show you a perfectly flat line, because they are plotting magnitudes, not energy density.
Reading this makes me feel like an idiot...Mayae wrote:Yes, I just read your answer again and I get what you're saying.mystran wrote:I'm just trying to say that practically all FFT-type analyzers will show you a perfectly flat line, because they are plotting magnitudes, not energy density.
Interestingly, when you set it up to measure something like the absolute energy in a complex resonator with a bandwidth inversely proportional to a logarithmic frequency axis (ie. a lower pixel represents a larger Q), you get a very distinct 10 dB/oct slope.
Submit: News, Plugins, Hosts & Apps | Advertise @ KVR | Developer Account | About KVR / Contact Us | Privacy Statement
© KVR Audio, Inc. 2000-2026
