(Question for Mike but guys can bring some technical informations!)

Had some interests for Theta Pro DSP and their “zero aliasing tech” wich is some kind of marketing argument that I don’t really get the point... is that related to “metallic, cold or harshness” of some amp/distortion sims?

Mercuriall Plugins also province those 2x, 4x Or 8X anti alliasing buttons and it sounds better with higher rates, REALLY... less noticeable with Neural DSP stuff like Fortin though...

Also Cliff did some comparison with aliasing from AxeIII and other “competitor”...
Thanks

Hi Apotheker,

S-GEAR applies 8x oversampling to the amp modules to ensure that any aliasing is inaudible (ie. -60dB or more).

As for 'what is aliasing', in this scenario, a very brief explanation is that aliasing is the introduction of additional frequencies in the resulting output as a by product of non-linear digital processing. There are many explanations of aliasing to be found on the web. It can also occur as a result of inadequate pre and post filtering in the analogue-to-digital and digital-to-analogue conversion process.

Cheers,
Mike

I don't know if it's related to what you're hearing as metallic, cold, harshness, etc., but possibly.

Mike did a great job of explaining the essentials: in a nutshell, when a digital audio signal inside your computer hits your audio hardware and is converted to the analog domain, there's often the existence of audio outside of what the hardware can handle, so it attempts to move those super-high frequencies *into* the audible range, and the result is aliasing. It's typically undesirable because it has been artificially placed there, and therefore changes the sound to some degree, and, yes, it could be metallic, or cold, or harsh, or a number of things. Anti-aliasing attempts to remove that aliasing by removing (or ignoring, or redistributing) the frequencies above and below the sonic range before it gets converted to analog by your audio hardware. And it works, but it also burns a lot of your CPU, which is why developers often give you the option of picking 2x, 4x, 8x, etc. oversampling. I almost always prefer the higher magnitude, but not always, and with some plugins it's just too much math for too little sonic improvement.

I'm not too familiar with the Theta technology but it sounds like they're applying anti-aliasing in a way which they feel merits advertising. I'd be curious to hear it. Or, rather, *not* hear it.

My guess is that it's not going to be different than what we're accustomed to with other methods, like S-Gear's, which I hear as natural.

Aliasing is simply when your digital system can't adequately represent the incoming signal. This is, as elambo said, due to having audio outside of the band that you can accurately represent. This is tied to the Nyquist theory that you need to have twice the sampling rate of the highest frequency you want to represent. Humans hear to approximately 20 kHz, so anything over 40 kHz sample rate is sufficient. However, if you sample at 48 kHz, for example, and there is a 25 kHz tone in your audio, aliasing will occur, resulting in a perceived 23 kHz tone with an inverted phase from the original signal. This combines with your actual 23 kHz material, resulting in inaccuracies.

Aliasing first gets noticed at the higher frequencies and, due to the phase inversion, messes with the original signal. This is why it can feel "cold", etc. If we were to oversample our 48 kHz signal by 2x above, then we have 96 kHz effective sample rate, which makes 25 kHz no problem.

If you want to learn more, there are great resources on the web. Aliasing and "frequency folding" (way old school term) are good search terms. Frequency folding in particular gives a very practical graphical representation of how aliasing affects a signal.

Sorry if that was a little overkill, but I'm an acoustic physicist working in the audio signal processing industry, so I'm more than just a little nerdy and passionate about this type of stuff

Ripthorn wrote:

Aliasing is simply when your digital system can't adequately represent the incoming signal.

Not just the input signal but the processing applied may also imply higher frequency content. A linear digital system might adequately represent the input signal according to Nyquist, however when non-linear processing is applied then it may be necessary to introduce oversampling in order to avoid aliasing. This is particularly the case with highly saturated algorithms like amp models.

mike wrote:

Ripthorn wrote:

Aliasing is simply when your digital system can't adequately represent the incoming signal.

Not just the input signal but the processing applied may also imply higher frequency content. A linear digital system might adequately represent the input signal according to Nyquist, however when non-linear processing is applied then it may be necessary to introduce oversampling in order to avoid aliasing. This is particularly the case with highly saturated algorithms like amp models.

Out of curiosity (and assuming that it's not a secret element of your recipe), have your methods for dealing with aliasing changed since the earliest versions of S-Gear?

elambo wrote:

Out of curiosity (and assuming that it's not a secret element of your recipe), have your methods for dealing with aliasing changed since the earliest versions of S-Gear?

The methods employed in S-GEAR have been consistent since the first release. As mentioned, S-GEAR applies 8x oversampling to the amps which requires special filters for sample rate conversion. The design of these filters require special consideration since there is a trade-off between linear phase and latency. S-GEAR uses a hybrid approach to keep the latency minimal and retain a natural sound.

We've discussed the following before, but it's worth mentioning again that all non-linear processing (including compressors, or faithfully modelled EQ) have the potential to introduce aliasing and may therefore require internal oversampling. I think it is good to be aware of the potential side-effects of chaining multiple plug-ins together in a system - increased latency or possible introduction of aliasing and resulting phase response. If the results sound good then does it really matter. However, when tracking live, the the accumulated latency could be an issue, when mixing the latency should anyway be compensated for. The potential of running a highly oversampled channel strip is interesting to me, where all plug-ins run at the higher sample rate and there is only one SRC process at the start & end.

I’m starting to get the point...

Am I right assuming running my projects (Logic Pro X) at 24/96 will be better than 24/41,1?
Regardless project sizes considerations, what should you recommend to get the best sound. From S-Gear?

Had a strange feeling that running S-Gear at 96KHz get me less single coil noise....!! Is that possible.
I suspect my Apogee Jam Plus (24/96) to be designed for those samples rates and be at their best in this area.

Thoughts on this Mike?

Thanks

Deciding what sample rate to work at can be a difficult decision to make.

Choosing a higher sample rate means more CPU impact. There are more samples for you DAW to process. Impulse responses will be longer to represent the same length (in time) IR.

S-GEAR will maintain an (almost) consistent internal sample rate. At 44.1 and 48, this is 8x oversampled, then at 88.2 and 96KHz S-GEAR will configure to 4x oversampling. So approximately 384KHz internally regardless of the sample rate chosen.

Certainly there may be some perceivable sound quality difference with audio interfaces. 96KHz sometimes seems to have more transparency and more precision but I've not made exhaustive comparisons. I usually work at 44.1KHz and test at higher rates.

Regarding single coil noise, this is a curious one. I've noticed that both sample rate and audio buffer size can appear to impact the digital noise radiation fields from a computer. Particularly with a cheaper laptop or when you have a lot of peripherals connected to the machine. The clock speed of the computer isn't changing but I assume the repetitive processing patterns may impact noise radiation. I guess this just comes down to the physics of the computer electronic devices, PCB, box design. The biggest horror story for single coil guitars is robot lawn mowers... but that's a topic for another thread.

So oversampling is « better » for 41Khz projects than 96?

Apotheker wrote:

So oversampling is « better » for 41Khz projects than 96?

Better is subjective, but it will be more noticeable.

Apotheker wrote:

So oversampling is « better » for 41Khz projects than 96?

The point is that S-Gear maintains about 384KHz processing internally by design. So 8x @ 44.1KHz is the same as 4x at 88.2KHz. Of course there may be other factors which lead to someone wishing to run their project at a particular rate.