The Aliasing Problem

    BTW - I have noticed similar sounding problem on a couple of profiles - I am not sure if it's aliasing or not, but it's not very pleasant. Let's see if new firmware alleviates it.

    Quote from mdeeRocks

    BTW - I have noticed similar sounding problem on a couple of profiles - I am not sure if it's aliasing or not, but it's not very pleasant. Let's see if new firmware alleviates it.

    Yes, they acknowledged the problem and told me that developers already "patched" it for the next firmware release (hopefully that will come soon)

    Guys , you can try to reamp a simple rising test tone an hear a strong aliasing gurgling with all of kemper profiles...with latest fw...
    aliasing not destroyed in 1.6.1 fw , it just became unaudible...with most cases of real guitar sound


    But, some VST amp plug-ins ,with enabled 16x oversampling - do much more less aliasing than kemper on test tones...
    and this fact slightly frightens me...low processor power , imperfect algorithms ?

    Quote from artembankin

    Guys , you can try to reamp a simple rising test tone an hear a strong aliasing gurgling with all of kemper profiles...with latest fw...
    aliasing not destroyed in 1.6.1 fw , it just became unaudible...with most cases of real guitar sound


    But, some VST amp plug-ins ,with enabled 16x oversampling - do much more less aliasing than kemper on test tones...
    and this fact slightly frightens me...low processor power , imperfect algorithms ?

    Well, aliasing cannot be removed entirely, you know. It appears whenever you're trying to convert from digital to analog a frequency that is higher than half the sampling rate; the conversion cannot distinguish, given the insufficient amount of samples, the intended frequency from one much lower and in the supported frequency range. You can always feed the kemper a signal that is higher than half its sampling rate.
    This is to say that "just making it unaudible" is far from a bad solution! The common solution (the one that has been applied, as far as I know) is to sample more, simply. But that requires algorithms to go faster.
    Mind also that the remaining aliasing from 1.6.1 firmware on, it's not only "softer" to the end of being inaudible: it's really a smaller range of frequencies that are being erroneously converted, because by raising the sampling rate only very very high harmonics are exceeding the supported range, and these harmonics wouldn't be very strong in the first place.
    The other solution, that I would personally use but I guess there is some reason why it is not used, would be applying a lowpass filter before conversion. Such a solution would be even final (well, provided the input itself wasn't higher than supported). But maybe here someone can chime in and explain me why it wouldn't be doable.

    I'm guesstimating that the problem is the steepness of the low pass - there are no 'sharp corners' in real-life signals. A super sharp filter starting at 20kHz will still let some slightly higher frequencies through.
    It's also possible that the Profiler uses oversampling to ease up on the filter Q requirements, and raising the oversampling factor would require processing power, yeah.
    We'd probably all be ok if the LPF would just start out a couple Hz sooner, but i'm a hack-job :thumbup:

    "But dignity is difficult to maintain
    stamina requires constant upkeep
    repetition is boring
    and you pay for grace."

    Most guitar cabs have speakers that have a cut-off frequency around 5K Hz. That is what we are all used to hearing with a tube amp.


    That isn't to say that there aren't some frequency content above 5K when you play through them, only that it doesn't get through at nearly the volume that it comes in at.


    At 10K, I would say you are pretty safe, so a sample rate of 20K and processing capable of handling the samples at this rate should effectively eliminate aliasing associated with guitar frequencies IMHO. That shouldn't be an issue for the Kemper. While I haven't opened one up to see what parts are being used in the Kemper, I can say that the X32 digital mixer is using the 3rd generation SHARC DSP processors which are capable of processing 38 channels at 48Khz through the entire efx chain with only about 1.8mSec of latency total. I am guessing that the Kemper can handle a single channel input and process it without any aliasing using its Symphony 56721 Family processor...... no problem.


    I have noticed the aliasing in notes high on the fret board on the high E string on my strat on several rigs.


    I will be looking forward to the fix for this.

    I never understood the notion that nothing happens above 10kHz. Just look at the scope - there's a lot happening until 12 or 13kHz, and stuff going up as far as 18kHz depending on program material.
    Try aggressively LPFing a distorted sound at 10kHz. Tell me you can't tell the difference.

    "But dignity is difficult to maintain
    stamina requires constant upkeep
    repetition is boring
    and you pay for grace."

    Quote from Laimon

    The other solution, that I would personally use but I guess there is some reason why it is not used, would be applying a lowpass filter before conversion


    Reminds me of my Boss GT-10 and the Harmonic Converger at the bossgtcentral.com way back them. I had the active one from a circuit I got at freestompboxes.com.

    Quote from Michael_dk

    Stupid question: Could the aliasing come from processing AFTER the conversion?

    Well, if you mean after analog (guitar) to digital, yes, the aliasing happens in the later - opposite - conversion, that is from digital to analog. The "harmful" frequencies (i.e. too high harmonics) are added in the digital signal processing by the profiles (high gain ones).
    If you mean instead after digital to analog conversion, no, that is simply not possible. Aliasing does not exist in the analog world ;)

    No, the aliasing is purely a conversion artifact. Things can go wrong in the processing (fictitious example: a whammy pedal could raise a frequency to the point of not being representable with the used data types and lead thus to overflowing), but aliasing really just pertains to the conversion phases. Moreover, it is way more likely to occur in the DA conversion than in the AD one: your guitar is very unlikely to spit out > 22kHz, the lead channel of diezel amp with a tube screamer in front of it has much higher chances ;)

    Lets not get too carried away with the ultra high frequency stories here. I personally can not hear anything over 15Khz period .... but young people (who don't play in bands) and dogs can ;)


    While it is true that a guitar (and certainly a guitar and a stomp box) can generate frequencies that are quite high, most guitar cabs can't reproduce the frequencies. Here is the spec sheet on the ever popular vintage 30: http://celestion.com/product/1/vintage_30/


    You aren't going to get much over 15KHz out of this speaker no matter what you put into it.


    As for aliasing, it isn't caused by the A/D or D/A in the real world. All modern A/D converters are capable of sampling at very high frequencies (>192Khz). If you study electrical engineering (I am an EE myself), you learn that you can digitize a signal by sampling it at discrete points, processing it into the frequency domain, then changing it back into the time domain and feeding it into a Digital to Analog converter (D/A) to output the voltage EXACTLY like it was before you digitized it ....... as long as you sample 2 times as fast as the highest frequency you wish to reproduce.


    There is a long drawn out mathematical explanation for this if you care to look it up you will find reams of gibberish explaining it in detail ;)


    So, sampling at 44Khz would give an effective reproduced frequency ceiling of 22Khz ..... which is higher than most humans can ever hear even at birth ..... and much higher than any guitar cab I have ever heard of being capable of reproducing.


    Now what DOES cause aliasing in the real world is down sampling in order to reduce processing, or mistakes in the DSP signal paths which result in a time alignment issue that results in something that sounds like aliasing.


    My educated engineering guess is that what we are hearing in the Kemper is the latter .... and it can be easily fixed by modifying the DSP code to better align things internally.


    I doubt that the Kemper is processing bound.

    Hi OneEng, what you say is technically correct, but I think you might be assuming certain things that don't hold, or missing some perspective on the problem (the thread is quite long now, so it's easy to have missed bits here and there). Let me comment on your statements to clarify the issue.

    Quote from OneEng1

    While it is true that a guitar (and certainly a guitar and a stomp box) can generate frequencies that are quite high, most guitar cabs can't reproduce the frequencies. Here is the spec sheet on the ever popular vintage 30: http://celestion.com/product/1/vintage_30/


    Quote from OneEng1

    You aren't going to get much over 15KHz out of this speaker no matter what you put into it.

    The aliasing we're discussing here is mostly apparent when deactivating the cab sims and going to a guitar amp. Clearly a preamp output has a much more substantial high-frequency content than a cab output. However, mind that on several rigs one can hear it (much more subtle, of course) also with cab sims on going through a FRFR speaker.

    Quote from OneEng1

    As for aliasing, it isn't caused by the A/D or D/A in the real world. All modern A/D converters are capable of sampling at very high frequencies (>192Khz). If you study electrical engineering (I am an EE myself), you learn that you can digitize a signal by sampling it at discrete points, processing it into the frequency domain, then changing it back into the time domain and feeding it into a Digital to Analog converter (D/A) to output the voltage EXACTLY like it was before you digitized it ....... as long as you sample 2 times as fast as the highest frequency you wish to reproduce.

    Thanks, I know what aliasing is :P
    You can surely sample very fast, but then you'd also need to process way more data, and that requires more computing power. You say:

    Quote from OneEng1

    I doubt that the Kemper is processing bound.

    but that seems indeed to be the case: Cristoph Kemper said some time ago that fixing the aliasing would be no problem, but would require to speed up the algorithms, and he was hoping to save the processing power for future features. So the computing power is indeed to be treasured ;)

    Quote from OneEng1

    So, sampling at 44Khz would give an effective reproduced frequency ceiling of 22Khz ..... which is higher than most humans can ever hear even at birth ..... and much higher than any guitar cab I have ever heard of being capable of reproducing.

    Uhm, the fact that humans can hear a frequency is irrelevant: if a mic can catch it, or the Kemper internal processing produces it, then that will generate aliasing, and humans will hear that one ;)

    Quote from OneEng1

    Now what DOES cause aliasing in the real world is down sampling in order to reduce processing, or mistakes in the DSP signal paths which result in a time alignment issue that results in something that sounds like aliasing.

    That might be. Data overflow then?

    Quote from OneEng1

    My educated engineering guess is that what we are hearing in the Kemper is the latter .... and it can be easily fixed by modifying the DSP code to better align things internally.

    They could tell us, they already fixed it, and it's coming in the next firmware release :)