Aliasing Tests Ver 2.0

FractalAudio

Administrator
Fractal Audio Systems
Moderator
Those of you who have been with us from the early days may remember my aliasing tests. I have repeated these tests on some of the latest products available, including the Axe-Fx II. For those who don't remember, aliasing is the bane of modelers. It creates the dreaded fizz and can even create noticeable tones if bad enough. To prevent aliasing requires oversampling and well written interpolation/decimation algorithms.

So I gathered up the Axe-Fx II, a modeler we'll call "Brand D" and another modeler we'll call "Brand K". I wanted to included a modeler from "Brand L" but for the life of me I couldn't figure out how to shut off the cabinet modeling in it (which distorts the measurements) so I was unable [edit: I have the lower tier version of the product which I have found out does not allow turning off the cabinet modeling]. All of the modelers are running the latest firmware from the manufacturer (as of today's date). The Axe-Fx II is running Version 7.00 but that has no bearing on these tests. Much care was taken to ensure that the input/output range of the devices was not exceeded. The sine-wave source was the synth block in the Axe-Fx II as it is a very high-quality signal generator and actually produces a purer tone than my SRS synthesized signal generator.

On each modeler I used an SLO 100 model with the drive roughly at 7.0 and the cabinet modeling turned off. I chose the SLO100 because it is fairly high gain. The more gain the worse the aliasing but the SLO100 models were fairly close in tone between the three modelers so that seemed to be a good choice even though not the highest gain models available. I matched the gains using an intermodulation measurement and then verifying be ear. In my original tests I input a 10 kHz sine wave into each device and measured the output spectrum. An ideal device would only create harmonics of 10 kHz, i.e. 20, 30, 40, etc. kHz. Any tones that are not harmonically related are aliases. 10 kHz is a nice value because the aliases are then mostly below the fundamental and easy to recognize.

Since my original tests I have devised a new test which is more illustrative. I now sweep the input from 9 kHz to 11 kHz. This excites all the various aliases and prevents cheating since one could adjust the input frequency slightly to improve the results. I set the spectrum analyzer to peak-hold and recorded the results. The graph below consists of the results of all three devices. The blue trace is the Axe-Fx II, the magenta trace is Brand D and the green trace is Brand K:


alias_test.jpg



The ideal result would be a rectangular pulse sitting above the noise floor. Anything outside the pulse from 9 to 11 kHz is aliasing noise. 60 dB is considered the threshold of audibility so if the aliasing noise is above 60 dB, then theoretically it is audible.

The Axe-Fx II scores best in this test. Brand D does well and, actually, better than the original Axe-Fx. The results of Brand K are surprising. I have repeated the tests at least a dozen times fearing that I have made an error but, alas, the results are correct.

I can also post sample clips of tones into the three devices that demonstrate the audibility of the aliasing, if desired.

[edit: I have been informed that you can indeed turn off the cabinets in Brand L now. I still can't figure out how but once I do I will try to repeat this test on that modeler and update the results.]
 
Thanks for the insights, very interesting!
And yes, Brand D does a pretty good job! :mrgreen
 
Nice cliff.
do post the clips, I'd like to hear this for myself too..
Im guessing the green is Line6 Product.

Just to be clear.. - is this something you have been tweeking for ver 7.0 - or are you just measuring for the sake of doing testing? - because the axe2 really is stable below the noise floor (in this test) - but yet I still hear the aliasing in a lot of amps, even at low gain..

Also looks like the threshold of the spike has been clipped looking at the graph? is that intended? or simply does it not matter at all?

Sorry its a lot of questions, im generally interested and picking your brain a bit :)
 
Nice cliff.
do post the clips, I'd like to hear this for myself too..
Im guessing the green is Line6 Product.

Just to be clear.. - is this something you have been tweeking for ver 7.0 - or are you just measuring for the sake of doing testing? - because the axe2 really is stable below the noise floor (in this test) - but yet I still hear the aliasing in a lot of amps, even at low gain..

Also looks like the threshold of the spike has been clipped looking at the graph? is that intended? or simply does it not matter at all?

Sorry its a lot of questions, im generally interested and picking your brain a bit :)

This is unrelated to Version 7.00. Whatever you are hearing is not aliasing. The Axe-Fx II's aliasing is not perceptible. Perhaps crossover distortion. Increase the tube bias to remove that. I don't understand what you mean by spike being clipped. The ideal response would be a rectangular pulse from 9 to 11 kHz. Rectangular implies flat on the top.
 
I appreciate the scholastic report! Although this is way over my head i will assume there is a 95% confidence interval and the area under the curve (AUC) was taken to infinity. I did a similar test on my flux capacitor that I built shortly after the film "Back to the Future" was released and the results were surprisingly similar.

If you have time Cliff, we can hang out over a beer and i'll show them to Ya! LOL Seriously though, nice work!
 
I would be guessing that the green line "Brand K" is for a product from Kem... something or other, and that Line6 is perhaps the product Cliff did not use in this comparison due to not being able to shut off the cab modeling. But I'm sure Cliff will neither confirm nor deny my suspicions to keep the true identities of the other manufacturers safe from ridicule.

Nice cliff.
do post the clips, I'd like to hear this for myself too..
Im guessing the green is Line6 Product.

Just to be clear.. - is this something you have been tweeking for ver 7.0 - or are you just measuring for the sake of doing testing? - because the axe2 really is stable below the noise floor (in this test) - but yet I still hear the aliasing in a lot of amps, even at low gain..

Also looks like the threshold of the spike has been clipped looking at the graph? is that intended? or simply does it not matter at all?

Sorry its a lot of questions, im generally interested and picking your brain a bit :)
 
Very Interesting stuff Cliff Thanks very much!!

I'd love to see the results for the new Brand F floorboard too...!
 
Interesting Cliff, but a quick question?

How does this relate to the real world where an electric guitar pickup has little of no hi freq info, especially how normally used running into a high gain preamp where one is usually careful about input freq range to avoid fizz even with tube amps etc?

What would the test look like if sweeping something lower like the frequency range of a typical magnetic guitar pickup?

Thanks!
 
Interesting Cliff, but a quick question?

How does this relate to the real world where an electric guitar pickup has little of no hi freq info, especially when normally used running into a high gain preamp where one is usually careful about input freq range etc?

Thanks!

As soon as you distort you create tons of high frequencies.
 
Very interesting. I'm kinda surprised at how well the D held up there. I had a couple of their mid 90s 2U rack preamps/processors, and got some great results out of them for what they were. Was VERY disappointed with the range that followed, when they tried to pass themselves off as modellers. I must admit I haven't checked out the latest rack unit. Maybe I will now... oh wait, I have an AXE II. I don't need to bother :)

Interested to see how the L holds up if you do the test again. Cheers for keeping us in the loop.
 
Maybe Mod "l" is possible now?

HD300/400

Cab Bypass

V2.0 firmware gives you the ability to bypass the cab model for any amp.
 
While I suspect the aliasing would not be as noticeable in a real-world situation, it is unacceptable in an $1850.00 processor. Some people can detect very small levels of digital artifacts such as aliasing, quantization noise, and others that are correlated with the input signal. I had an early modeler that sounded like I had a synth playing a horrid duet with me.

Edit: Even lab tests are conducted in the real world. I should have said average Joe guitarist in a typical playing situation. I'd like to hear a high gain K profile, with the input being a single high note on the guitar neck slowly bent up and down. Based on this mp3, I would not be surprised if aliasing was audible. I wonder...
 
Last edited:
Back
Top Bottom