🚀 go-pugleaf

      Depending on whether this device doesn't do some silly processing internally
      that I haven't asked it to do




Hmm... I wouldn't bet the house on it. What the NX uses instead of an emu10k2 is a Cirrus EP7309 - this gives them the USB capability. And here's what it says in the data sheet about its internal processing:

      Digital Audio Capability The EP7309 uses its powerful 32-bit RISC processing
      engine to implement audio decompression algorithms in software. The
      nature of the on-board RISC processor, and the availability of
      efficient C-compilers and other software development tools,
      ensures that a wide range of audio decompression algorithms
      can easily be ported to and run on the EP7309





The very good chances are that since it's a part of the same family, and they wouldn't want to rewrite the drivers extensively, they'd use the RISC processor to achieve the same end result.

As a side note, >recording< (rather than outputting) through the Audigy 2 NX clearly does not get limited to 48 kSample/S when specifying a recording format of 96 kSample/S. It has a claimed recording format of up to 24 bit, 96 kSamples/S, and it is quite clear from my examination of a recording that I obtained real data up to 48 kHz during a recording.

I recorded a band at 24 bit / 96kSample/S, with the band consisting of a violin (apparently with a good on-board microphone), mandolin (pickup), and electronic keyboard with grand piano patch direct to the mixer. They were playing fairly energetically, but not creating an ongoing block of sound. So it is possible to discern individual instruments in spectral view, especially when the violin played a solo. So in spectral view, I see harmonics for the violin going up to 48 kHz, though only vaguely as frequencies approach the upper limit. Also percussive sound like rhythmic chopping or slapping the guitar is clearly present up to 48 kHz. As one would expect, the level of the music signal relative to noise declines with greater frequencies, though the signal is clearly visible in many places.

Essentially it appears to be a non-faked continuation of real-world audio data above the audible frequency band to 48 kHz.

On the other hand, I still do not know whether my actual desired recording rate, 44.1 kSamples/S gets resampled with this device.

Leaving aside whether this device can actually achieve this for one moment...

      (apparently with a good on-board microphone)





      Essentially it appears to be a non-faked continuation of real-world audio
      data above the audible frequency band to 48 kHz.




So what mic is it? Even with very good mics, I don't get any sort of response like that at any sort of meaningful levels (ie they aren't harmonics, but modulated noise). The chances are that your 'non-faked' recording is very much faked, in fact. The very best that human hearing ever achieves is to be able to hear sounds at about 22kHz if they are pretty loud - and that ability falls away rapidly in your late teens as a rule. So if you see artefacts that appear to move above 20kHz then a) you are looking at noise modulation probably, and b) you won't be able to hear any of it, because it will be at a level too low to detect by human hearing.

Incidentally, I have a pair of DPA 4006 mics, and these are noted for having a pretty extended response at both ends of the spectrum. With the near-field grid on the front, which gives the most extended HF response, the best they can achieve is to be about 8dB down at 40kHz. And with orchestral recording, you get very little, if anything, above about 25kHz from them. And in carefully conducted tests nobody has ever been able to tell the difference between a response that I've rolled off smoothly above 20kHz, and the response allowed to extend to its full range - simply because they can't. And that's with monitors that have a response that extends to about 35kHz.

And all this has been extensively tested by others as well - read this <http://www.audiomastersforum.net/amforum/index.php/topic,6535.0.html>, where you will also find out a bit more about cheating!

Okay, I am reading through that thread you referenced, titled "SACD and DVD-A proven no better the CD in a year of listening tests", to find something about the recording of modulated noise. But to be quite honest I am not really interested in main question that tends to be discussed in threads like that: whether that type modulation, (let alone pure harmonics) above the audible range are actually audible. They're not. So discussions like that tend to be rather tedious to me. :)

So my concern about the reference to that thread is that I already am aware that recording above the audible frequency range is wasteful. Much of that thread's discussion is about the "news" that high-definition recordings aren't inherently better than CDs, based purely on sampling rate. I agreed with that ages ago. I did mention in message #1 above that my target recording rate is 44.1 kSamples/S.

My experiment with recording at 96k was to examine whether the device is doing some resampling to 48k. Poor resampling from 44.1k to 48k might have the potential to be audible. On the other hand, with resampling from 96kSamples/S to 48kSamples/S, I would expect little activity above 24kHz in a spectral analysis. Note: not listening for artifacts at this point, just viewing spectra. So your hypothesis about noise modulation is interesting. The specific (narrow) issue to me is whether the device is recording modulated noise or "real data", but not whether any of this is audible.

For efficient recording of audible data, I would aim for 44.1k or, if necessary due to the quirks of this particular device, 48k.

      Thus the specific (narrow) issue to me regarding my recording at 96 kSamples/S
      is whether the device recorded modulated noise or "real data" between
      24 kHz and 48 kHz, given that it appears visibly in a spectral view to
      be non-synthetic frequency information.





      My experiment with recording at 96k was to examine whether the device
      is doing some resampling to 48k. If resampling were performed from 44.1k
      to 48k, and done so poorly, it could have the potential to be audible
      -- to be determined. On the other hand, with resampling from 96kSamples/S
      to 48kSamples/S, I would expect little activity above 24kHz in a spectral
      analysis. Note: not listening for artifacts at this point, just viewing
      spectra. So your hypothesis about noise modulation is interesting.




Well, even in the best regulated systems, some noise modulation is quite normal - and I think that it can easily confuse some people into thinking that there's something 'real' there when there isn't.

If the system really is passing everything up to 48k, then inputting discrete tones into the input should establish without any doubt what is happening in terms of processing - but you would need a means of generating appropriate tones. If it's full-duplex, then perhaps you could try generating 40kHz in Audition at 96k, playing it out and re-recording it. If it's there as a discrete tone at the correct frequency, then the system indeed is not resampling.

That would mean that it stands alone in the entire Creative output of recent years, with the exception of the devices that have a specific pass-through at 96k, even though they resample 88.2! Which just goes to prove how important a couple of small words are - "recording at 24-bit precision up to 96 kHz" should, I think, more accurately say at, IIRC.

      Regarding the violin microphone: a small, on-board condenser microphone.
      I don't have the model number, but will inquire if it turns out that it
      really was producing a response that high in the frequency range.




If it's very small, then it has a much better chance of producing a response that extends way beyond anything reasonable, yes - but there's not a lot of point in keeping it (as you acknowledge) and it would probably be an internal resonance causing the effects, which wouldn't be radiated very far anyway. You have to watch it with mics mounted directly on acoustic instruments - they can be misleading!

As AA is capable of generating test tones up to the sort of frequencies being discussed, would it not be better to use these for this type of exercise rather than the unknown microphone?

NB I'm not volunteering to get involved.

      If it's full-duplex, then perhaps you could try generating 40kHz in Audition
      at 96k, playing it out and re-recording it.




Looks like a plan to me, with a possible pitfall: I haven't yet made a solid conclusion from perusing online chitchat about this device regarding output above 48 kSamp/S getting resampled to 48k. It may be that output at 96k with no DSP effects enabled is not resampled. If so, the device may perform well as a tone generator at 40 kHz. But if the card does downsample the output to 48 kSamp/S, then no 40 kHz tones.

Conversely, I'm trying to figure out how to generate intentionally an intermodulation artifact to view in Audition. I haven't actually tried to do that on purpose before. I'm familiar with viewing spectra, but haven't focused much attention intermodulation, so I don't know quite what to look for.

      Looks like a plan to me, with a possible pitfall: I haven't yet made a
      solid conclusion from perusing online chitchat about this device regarding
      output above 48 kSamp/S getting resampled to 48k. It may be that output
      at 96k with no DSP effects enabled is not resampled. If so, the device
      may perform well as a tone generator at 40 kHz. But if the card does downsample
      the output to 48 kSamp/S, then no 40 kHz tones.




I had a look at the online reviews, and there don't seem to be many even vaguely technical ones around - but Tom's Hardware has one, sort-of. Here is a link to page 8 <http://tau12.pair.com/2003/11/14/usb_2/page8.html> of it, and it's worth reading on from this point. The gist of it appears to be that 44.1k is indeed resampled to 48k, and although they haven't stated this explicitly, it appears that the frequency response is limited to 20-20kHz, certainly in terms of digitising. My guess is that the Sigmatel D-A converters, which can run up to 192k may well be providing an output that is exactly what it says it is, but that the input is indeed resampled. Although without them exactly specifying chapter and verse what they got, it's hard to say definitively without actually trying it.

So my guess, based on what Tom's Hardware say, would be that you can generate the tone, and that the card may even play it out - but that it won't re-record it at the same rate. And that's the first experiment to try, I think...

      I had a look at the online reviews, and there don't seem to be many even
      vaguely technical ones around




That's the thing. I would think at least one engineer would have dug in, but this unit may have been too uncommon.

I had actually read the Tom's Hardware article, but via a different URL (which has includes graphs):

 <http://www.tomshardware.com/reviews/usb-2,711-13.html>

Unfortunately, even with test graphs, they did not clearly define precisely what is being measured. "Frequency response" I assume refers to output. It does show energy output above 24 kHz in 96 kSamp/S mode.

So, even the Tom's Hardware review does not absolutely clarify the situation. I think what you may have noticed on page 8 of that review was this: "Naturally, reading 5.1 tracks at 24 bits/96 kHz is something that the Audigy 2 NX is capable of. However, the card is limited to 16 bits/48 kHz."

That actually jibes with the specs in the user's guide, when using the device with USB 1.1. The full table from the user's guide:

Speaker mode	Function	USB 1.1			USB 2.0
2.0		Playback	Up to 24-bit/96 kHz	Up to 24-bit/96 kHz
		#Record		Up to 24-bit/96 kHz	Up to 24-bit/96 kHz
4/5.1		Playback	Up to 24-bit/48 kHz	Up to 24-bit/96 kHz
		#Record		Up to 24-bit/96 kHz	Up to 24-bit/96 kHz
7.1		Playback	Up to 16-bit/48 kHz	Up to 24-bit/96 kHz
		#Record		Up to 24-bit/96 kHz	Up to 24-bit/96 kHz

# Note that configurations are available only if there is enough bandwidth. Simultaneous playback and recording at 24-bit/96 kHz on USB 1.1 is not possible

------------------------------------------

In other words: playback through 4+ speakers using USB 1.1 is limited to 48 kHz. With USB 2.0: limited 96 kHz for any number of speakers.

The Tom's Hardware review was written in 2003, apparently in the presence of questionable support for USB 2.0: "It will only be able to give full measure when upgraded USB management through Windows XP becomes available from Microsoft and when Creative upgrades its system accordingly to enable 24-bit/96 kHz data-streams to pass through the USB connection. Hopefully, this should happen before the end of the year."

The page also has this regarding resampling: "Note that the Audigy 2 NX, like the other Creative cards, operates at 48 kHz for EAX effects. Operation at 44.1 kHz thus uses a resampling method that has a negative effect." However, when EAX effects are disabled (as they always are for me), it is unclear to me what happens.

-----------

Another report from 2003 (I think) was David Griesinger's PowerPoint presentation. Very amusing. From slide 8 onward: "All I needed was a good quality sound card that would record and playback 96kHz. Many manufacturers seem to offer such a thing at reasonable prices. I chose one by the most popular manufacturer� let�s call them 'C'.... Although you could set the device to record at 96kHz, signals above 23kHz would not record.... I decided to buy a USB based device that did 96/24, as this would be useful for demos from my laptop � such as at this lecture.... But this external board would also not record or play above 23kHz.... 'Are ANY of your boards actually capable of operating at 96kHz?'.... 'NO.... In fact, none of our competitors do either� We all have 96/24 converters, but we run them at 48kHz.'"

Much of DG's discussions with manufacturers back in 2003 ended up being about USB stuff. I'll have to see what happens with proper USB 2.0 running in the modern era.

I'll take a look at RightMark.

      I'll take a look at RightMark.




When you do, make sure to go through the settings at the start - you can configure the bandwidth you are checking and displaying here.

What type of distortion would this be:

On one of my computers with a RealTek sound chip on the motherboard, I can produce a form of distortion which I assumed was harmonic. To do this, I create a test file of white noise at 96 kSamp/S. I set the RealTek "Stereo Mix" input default format to 44100 Hz. Then play the 96 kSamp/S white noise file in WinAmp and record at 96 kSamp/S in Audition. So the source and destination are 96 kSamp/S, but this is downsampled in the stereo mix along the way. I did this to simulate a what is supposedly happening with the Creative Audigy products.

Result: the recording into Audition shows the average frequency response down by 3 dB at 18.5 kHz, smoothly declining until 22050 Hz. Above 22.1 kHz are several wide lobes, peaking at 23.0 kHz, 25.7 kHz, 28.8 kHz, 32.5 kHz, 36.2 kHz, 40.0 kHz, and 44.0 kHz (with an additional sharp peak at 44.1 kHz).

Too bad we can't post graphs on this forum.

      Too bad we can't post graphs on this forum.




Don't need 'em in this case - you've described pretty well what you got:

      Above 22.1 kHz are several wide lobes, peaking at 23.0 kHz, 25.7 kHz,
      28.8 kHz, 32.5 kHz, 36.2 kHz, 40.0 kHz, 44.0 kHz (with an additional sharp
      peak at 44.1 kHz), and around 48.0 kHz (but cut off due to 96 kSamp/S
      rate).




With any system like this, if you feed signals to it that are above the Nyquist rate, you end up with aliasing - basically parts of the waveform above half of the sample rate (or resample rate) get 'folded back' into the lower half of the spectrum. So, if you feed a wideband signal of any sort through a resampling system which restricts the rate, even though you have a wide bandwidth to store the output, you end up with basically a mess - unless you use an anti-alias filter.

RealTek on-board sound chips and the like are notorious for not bothering with proper filtering - and because in this case you've resampled, you are going to get lobes at the sample rate and sub-multiples, and also lobes at the difference rates folded up and down too. Probably if you measured the frequencies very accurately, you'd find that there is actually 3.9kHz between them; 48-44.1 would give rise to that. But these are all artefacts of resampling.

Even on well-filtered systems the filters aren't perfect - and if you put a wide-bandwidth noise signal into one of these you won't see peaks as such, but much shallower rises - which are effectively what causes noise modulation effects. Incidentally, I think that putting noise up to the Nyquist frequency into a system like that is almost certainly the cruellest thing you can do to it - it will expose every single thing that's wrong with it!

      With any system like this, if you feed signals to it that are above the
      Nyquist rate, you end up with aliasing - basically parts of the waveform
      above half of the sample rate (or resample rate) get 'folded back' into
      the lower half of the spectrum.




I think the strange lobes in the frequency response I observed may be something other than run-of-the-mill aliasing artifacts. The lobes in the frequency response are the result of upsampling, rather than downsampling, and appear above the Nyquist frequency.

To avoid the effect of downsampling, I tried another scenario with the Realtek chip: play 44.1 kSamp/S white noise using 44.1k Speaker Output and 44.1k Stereo Mix settings -- but record at 96k in Audition. Thus no downsampling, but one upsampling for the chip to provide the requested sample format to the recording software. (In the scenario I described in my previous message, there had been both downsampling and upsampling: 96k -> 44.1k -> 96k.)

The result of this are the same lobes in the frequency response above 22.1 kHz. So if I am observing noise or distortion above the Nyquist frequency due to upsampling, wouldn't that be something other than aliasing?

On the Infinitewave web site, the SRC comparisons for Audition 2 with and without Pre/Post filter are interesting. Apparently pre/post filter delivers what it should, attenuating the aliasing above 22.05 kHz. But the filtering produces a faint haze of noise in the audible spectrum, looking like uncorrelated noise, plus some faint harmonics.

      The lobes in the frequency response are the result of upsampling, rather
      than downsampling, and appear above the Nyquist frequency of the first
      portion of the chain.




They can occur in either direction - they only need a place to go, after all...

      The "spectral images" above the Nyquist frequency apparently are what
      I am observing with the Realtek chip. But the writer doesn't use the term
      "aliasing" for this artifact.




Any frequency present in the original signal that reacts with the Nyquist frequency will produce an alias frequency within the spectral image. Since you provided noise, every single alias possible will have been present. The significant ones though would have been the images of the sample rate differences - it's as simple as that. I called them aliases because - that's what they are!

      But the filtering produces a faint haze of noise in the audible spectrum,
      looking like uncorrelated noise, plus some faint harmonics.




What actually happens depends very much on the filtering system in use, and there are basically two FIR and IIR. Without looking it all up, I'm not sure exactly what AA does (although I think that I have a pretty good idea) - suffice it to say that getting reflected coefficients back within the passband is not exactly unknown, because these filters model 'real' ones pretty well, and they do that too. An anti-alias filter has to have a pretty steep skirt in order to retain a maximally flat passband, but this is always a compromise; the steeper the skirt sides, the greater the ripple and phase changes, and the greater the chances of ringing occurring. The trick here is to optimise the performance, and the Audition compromise of minimal aliasing and a little bit of noise is the one that was deemed to be most acceptable. It has been changed at least once in the software's history, and there are people who actually prefer the earlier version, I believe. Nvertheless, any implementation at all that Audition uses is way, way better than anything that Creative has ever managed with an Audigy.

On <http://src.infinitewave.ca/> , would the tests performed for "passband" and "transition" be essentially the same? The help page indicates both tests use pulse-train inputs the examine the frequency responses of low-pass filters. The "passband" graphs seem to be scaled-down versions of the "transition" graphs.

I ran a batch of tests with Rightmark and playing/recording a test tone. The Audigy 2 NX really can play and record at 96 kSamp/S.

RightMark shows a respectably flat frequency response well above 30 kHz when playing through line-out looped back to line-in via patch cord. Separately I also played a 40 kHz test tone file from Winamp and recorded it through the same analog loopback. The recording of the 40 kHz test tone shows the 40 kHz peak at -6dB. There are some other narrow peaks, with the strongest reaching -97 dB at 15.7 kHz and 24 kHz.

      would the tests performed for "passband" and "transition" be essentially
      the same?




Yes, it's just a matter of how much they display. Personally I think that the passband graphs could have been a little bigger - just about all of the tested software is flat most of the way up, so it's only really the top section you are interested in.

      I suspect things might not be so good at 44.1k or 48k, though.




What Creative did with some of their cards was to provide a 'pass-through' mode for 96k, but resample everything else to 48k, and this might well be one of those. I would have thought that careful inspection of the frequency response would give you the evidence you'd need to find out exactly what is happening, though.

<TheodoreG@adobeforums.com> wrote in message
news:59b862b8.-1@webcrossing.la2eafNXanI
> I am trying to figure out what has been determined about
> the internal processing and resampling in Creative Labs
> Soundblaster Audigy 2 cards.
>
> I have a Creative Labs Sound Blaster Audigy 2 NX for my
> laptop computer. I am wondering whether this performs
> resampling and processing at 48 kHz, as was found to be
> the case with the first generation of Audigy cards.

You're wasting your time. Spinning your wheels.

You don't care what processing your audio interface does, you do care about 
how it works as a record/playback device.

You can evaluate the record/play performance of your audio interface using a 
jumper cable that reaches from the main output to the line input, and the 
Audio Rightmark program.

Here is an example of the report that you will receive:

http://www.prohost.org/~hackie/audio/YP-P2_NWZ-S616F/Comparison.htm

Repeat the test at 24/96 sampling. If the card has flat frequency response 
above 24 Khz, then it doesn't resample for sure.

      You're wasting your time. Spinning your wheels.




No matter what equipment one uses, whether consumer or pro-quality, there is a benefit from understanding how it works, what its optimal settings are, and what quirks need to be worked around. That includes microphones, software, sound cards, etc. In this case, it is a sound card that has plenty of quirks but can actually perform quite well if one avoids its pitfalls.

      You don't care what processing your audio interface does, you do care
      about how it works as a record/playback device.




Indeed I do care about the latter, "how it works as a record/playback device". Wouldn't we expect this to be impacted by the former -- that is, what processing the audio interface does? In particular, if the audio interface performs some poor-quality processing that degrades a recording or playback, then one might want to be aware of the circumstances that lead to that, and avoid those circumstances.

As I described in message #1 above, "I am aiming to record at 44.1 kHz", but am examining the question of whether the device's performance with that sample format is degraded by resampling. If that is the case, the next question is whether that can be worked around, and how.

      You can evaluate the record/play performance of your audio interface using
      a jumper cable that reaches from the main output to the line input, and
      the Audio Rightmark program.




See the batch of messages immediately prior to yours.

      Repeat the test at 24/96 sampling. If the card has flat frequency response
      above 24 Khz, then it doesn't resample for sure.




Two messages prior to yours I posted about that exact test. The device does not resample at that rate. Therefore, there is a workaround to get good frequency data in the audible range (ultimate sample rate 44.1 kSamp/S): record at 96 kSamp/Sec, then downsample during post processing.

      So what mic is it? Even with very good mics, I don't get any sort of response
      like that at any sort of meaningful levels (ie they aren't harmonics,
      but modulated noise).




I've concluded after testing this very carefully that what I was viewing in the spectrum were actual harmonics of the violin going up to and somewhat beyond 46 kHz -- not modulated noise. The musician tells me he uses a DPA 4061, which he describes as by far the best onboard mic you can get.

      I've concluded after testing this very carefully that what I was viewing
      in the spectrum were actual harmonics of the violin going up to and somewhat
      beyond 46 kHz -- not modulated noise. The musician tells me he uses a
      DPA 4061, which he describes as by far the best onboard mic you can get.




I wouldn't go as far as to say that they are 'the best' onboard mics - and that's speaking as a DPA user. These particular ones have quite a pronounced boost at around 18k, and their response falls off quite significantly above 20kHz, and there are definitely better performing instrument mics around than that, that's for sure. They are perfectly competent mics as far as they go, but perhaps that's a bit too far...

Because recording a violin with them isn't exactly going to give you the natural sound of the violin. I wouldn't even like all that unnatural HF getting as far as a mic preamp, quite frankly, never mind being digitised - it won't do much for the intermodulation performance at all, and definitely won't improve the sound quality as a result. There may well be other harmonics generated within a violin that could be picked up by a mic under the bridge, but they sure as hell aren't supposed to escape from the instrument!

I think that those mics are probably fine for providing a signal from an orchestra pit, and I'm sure that they have a niche there, but I don't think that I'd want to be recording anything from them unless I was absolutely desperate, thank you.

      I wouldn't go as far as to say that they are 'the best' onboard mics -
      and that's speaking as a DPA user.




Naturally this depends entirely on the application for which it is used. The musician plays live music at social dances; live music at jazz clubs (acoustic); and might use it or another mic for studio recordings. I gather that he tried a number of microphones over the years, and concluded this model is "it".

Note that priority #1 in the setting in which I made this recording was good-quality live music at a social dance, rather than recording the music in a heavily-controlled environment. This was a close-mic setup, giving the advantage of less ambient noise and potential for feedback from the stage monitors into the microphone.

If one can position a microphone suitably far enough from a violin to make it sound more natural in a recording, that may be the way to go in order to get that type of sound. On the other hand, sometimes that positioning can create problems in a reverberant hall with main speakers, stage monitors, and noisy dancers. That stated, I have heard violins played both through on-board microphones and larger mics on boom stands in various positions, and I did find this one to sound well given the circumstances of the recording.