as promised in the past, I'll maybe never post in gearslutz again. Just a note or two (even if nebula is not the topic there, it wasn't analyzed too much, and in the compressor department nebula should still be improved and so on) because I find some interesting thing missing there
1) till this moment everyone was interested in frequency response. I said many times, real devices change their frequency response depending on the input level. This was never analyzed... and in compressor department this is the first thing which should be analyzed. 2) now someone is starting analyzing the harmonic response. This is good, I told you a lot of times, but keep in mind that a lot of developers are tuning things exactly for 1 Khz. Really devices change their harmonic response depending on the input frequency... many plugins have the same fake behaviour for all frequencies for example, or they are just optimized for 1 Khz. I mean, the correlation between the level of harmonics is a consequence of the input, (and here I mean the input frequencies and the input levels) 3) there is only a way for analyzing really harmonic distortion. This is based on volterra kernels analysis, period. That is exactly what nebula is playing.... I mean, just looking at the response to a 1Khz sine you get just an idea of something, but it doesn't tell you everything. It just tells you just a little piece of the truth 4) as I told you several times, nebula has a good graph for 1Khz because it reproduces just few harmonics (10) but it could create aliasing for high frequencies content (not shown in this simple 1Khz test found in gearslutz). Anyway aliasing could be fixed a bit using oversampling/using higher frequency rates, for example 96Khz or even more. Compared to 44100, a 88.1Khz for example is like a 2x 5) aliasing is not the only devil here: we showed several times that even FFT could damage your audio (and this destruction can't be measured easily), or IR truncation in FFT filters, or mistakes generated by recursion of state variables in IIR filters, or simply math, bugs... all those things CAN'T be measured easily, especially using those simple bode diagrams. I could sample a steinberg freefilter in nat and use it in nebula, you could hear the difference BUT those diagrams would be the same 6) a bit of aliasing means nothing (and here I'm speaking about all those posts regarding UAD plugins: "I though their plot would be better"). Use your ears, UAD plugins ARE good. A bit of aliasing below 100 dB can't be detected by humans, even processing the same audio several times. You maybe could understand something in the -80dB range (because it means you have a disturbing singnal at 0dB). Keep in mind that SPAN is based on FFT analysis, so a software based on FFT processing and on other software based on different techniques would generate the same graph, which is simply untrue. The analyzer itself is in some way flawed (even if you perform a 64 bit FFT.. and it depends on the window used.... a lot of boring details)
Very interesting points Gian. I can't say that I understand all of it, but what I do understand is very interesting. Just a quick observation. The peculiarities of the 1 khz tone was noted early on in the thread and 910 hz sine was used instead.
Also, regarding the "I thought UAD would do better" I said so in regards to the UAD 4K vuss compressor I'm currently testing. I found aliasing as high as -84db. And I didn't use make-up gain. I compressed 6 db so I would have added 6 db to that bringing aliasing up to -78. Further limiting would bring it up further I suppose. How long until the stuff actually becomes audible? Just listening to the sound I find that I like the sound of the 4K compressor. I like how it handles the attack. So, of course aliasing isn't everything. I still love the UAD 1176 and the LA2A. But, I've considered UA to be amongst the best in the business. I then figured that they would tidy up aliasing (if other, for example Cytomic, can do it why couldn't they?). Thinking about it now I assume they didn't clean it up because of the limited power of their UAD-1 cards. As I understand it, from the Softube guys reply, removing aliasing takes it's toll on the computer and the UAD cards can't really compete with native processing power no more.
Doh! Taking the hint from Gian's post I had to test sending 16 khz into NEB (using Doc Fear eq). A -4 db tone at 16 khz generates aliasing as loud as -66 db. On the other hand there aren't many in-harmonics (the number is easily calculated based on the number of kernels in the program). For Doc Fear you'd only get two in-harmonic peaks at 16 khz.
Is this something to work out a solution for or is it nothing to worry about? I suppose we could always work at higher sample rates.
no, nothing to worry about. Simply aliasing in nebula is not filtered, we'll adopt oversampling in the future in order to limit it, but as you can see, it's not so important since doc fear is still one of the best digital eq around. Every graph, every assessment should be analysed with a bit of grain.
But this test is really fun from my side, for an accurate harmonic distortion analysis you should use EXACTLY volterra kernels analysis, which exactly what nat-nebula are doing. At the same time the analysis of a linear system SHOULD be achieved generating the impulse response, which is THE way to go. All other methods are an approximation. A bode diagram of a 997 Hz sine is... just the behaviour of the system when you use a 997 sine. I guess you'll never play a sinewave in your songs.... (a sinewave is really a boring sound...)
1- You didn't promote your product enough, on online forums if you can't "sell" your plug-ins good enough (need at least 1 locomotive thread) then you simply aren't a good developer
2- If you only have 2 software developers that post some-what regularly on a 110.000 member forum then...(fill in the blank)
I think you hit it on the head! There's a lot of talk there about concerned for "truth" but nobody's really facing those issues. Instead there's more praising of one another and turning a blind eye. All in all, there's seems to be an "allergy" to Nebula and its potential.
1khz is the worst freq to test plugins at 44khz as the harmonics fit on top of the aliasing. Try 1.2khz and watch the aliasing stack up.
Nebula reverb performed very well in my tests, compressor plugins really did not do that well except Stillwell Bombadier.
After a day of testing going back between 44 and 96khz I'm going to have to stick at 96khz (88khz makes my soundcard driver mess up for some reason). Running my Eventide H8000FW at 96khz sounds nice on a lot of presets. I definatly heard some improvement on eq's that were originally sampled for Nebula at 96khz. Nebula is the most authentic to hardware out of all the plugins.