Benefit to go with 96k?

  • Thread starter Thread starter frist44
  • Start date Start date
if you really want to know what the sampling rate means for digital music this little page gives a good example:

how sampling rate works

even better explanation

the long and short is this:

the sampling rate has no direct effect on how high a signal you can hear. the sampling rate tells how many times per second a snapshot of the original sound is taken; however, because a higher frequency changes its sound faster, taking more frequent snapshots will help to more accurately portray that upper range...
hence an even greater perceived improvement in clarity in the upper range with a higher sampling rate.

[edited note] a higher sampling rate will in fact enable the recording of higher frequencies, but you can't hear them.

48khz vs. 44.1khz: 3,900 samples per second difference. which makes your recording of the original sound 8.125% more accurate.

96khz vs 48khz: 96khz is twice as accurate at representing the original sound.

16bit = 2**16 = 65,536 variations in sound pressure that can be measured.

24bit = 2**24 = 16,777,216 variations in sound pressure that can be measured.

that means that 24bit recording represents the sensitivity of the sound with 256 times more accuracy.

although we are MOST DEFINITELY comparing X to Y, you could make the argument that an improvement multiple of 256 in the Y is better than a 2x improvement in the X where X is time and Y is sound pressure at some X-time.

think of it this way:
a computer monitor that is .25 dot pitch and uses SVGA is going to be a better looking monitor than one that is .28 dot pitch and is VGA because the former has more pixels per inch and more color variation... digital music is the same as digital video but the "screen" is linear.
 
Last edited:
one last thing. if you are recording at 44.1khz and your equipment is capable of recording frequencies above 22khz then the frequencies that are above 22khz have the potential to be recorded into your tracks as an octave lower than the actual sound... thereby clouding up your upper frequencies... that may well be the harshness people attribute to digital recording.

so there is definit benefit to recording at 96khz.
 
96khz cards and plugins to boot?

I have a few questions here, but first off i'd like to say this is an awesome thread. Every once and a while you get a really great one like this, and it makes me happy everytime i see one. I was planning on getting a delta 44 soon, but after all this i think i might hold out for a better quality card with 88.2 and 96khz recording capabilities... wait that card has 96khz i think, well maybe just better quality... Im using a fairly fast computer with logic audio 5 and the waves plugins right now.

pipelineaudio - So the waves plugins do some damage then? I've always thought they were really good, but i have nothing to really compare them to other than the build in effects in logic and sound forge. Could you or anyone else recomend another good set of plugins like them?

Can anyone recomend any cards with 96khz capabilities that won't break the bank totaly? I don't need a lot of inputs, at least 4. More converter quality i supose, and less inputs/outputs.

I don't mean to ask questions off topic from the current conversation, it's very interesting and i don't want to throw it off topic. I just figured these questions were related and you all seem to know what your talking about.
 
waves probably does less damage than most other plugins. and every plugin or fx insert (if you are analog) is going to 'damage' the original signal to some extent.

i also like the compressor, EQ, and wah fx from ultrafunk's sonitus fx package, but the 3rd release is more expensive than it use to be.

i also like the reverb and chorus from DSPFX.

but Waves is still the best suite that i have come across in DirectX so far.
 
"Also, you say "all of the waves stuff." Is that wavelab stuff, or are you talking about some specific parts of Nuendo or Cubase?
"'

the brand known as " waves "

"pipelineaudio - So the waves plugins do some damage then? I've always thought they were really good, but i have nothing to really compare them to other than the build in effects in logic and sound forge. Could you or anyone else recomend another good set of plugins like them"

no not damage per se, but in many cases, offline processing isnt as " off line": as you would think. It still behaves as if it were running in real time

this is the DUBEST thing about computer recording! Since the data is already stored, apps could EASILY ready ahead, or behind, they already KNOW what is there!!! SO WHAT THE HELL the idiotic "look ahead" plugins that we have actually delay the signal, so that the key input will know what to do before the audio gets hit


HELLO stupid programers, this is NON LINEAR editing...anaog devices happening in REAL time in the real world can already do this...you MORONS!!!! use the power of the random acess to make this shit work BETTEER than the real world!!!! god knows your algos will never be half as good, least you could offer is SOMETHING unique to the PC world!!!!


sorry, rant against the morons who buold our software
 
so back to the original topic!
Whats the verdict? Should i record in 88.2khz when i get my new soundcard, if space isn't an issue? Should i just go with 92khz, or should i just stick with 44.1khz.

Assuming i have 180 gigs of hard drive space, at least 100 gigs free for audio projects at any given time.
 
I'd like some other people's opinions as well. I've done a lot of studying on the subject, and can't see how 88.2 kHz downsampled to 44.1 kHz could possibly sound worse than using 44.1 kHz from beginning to end.

There is the possibility that it could sound better, but with decent oversampling A/D's perhaps that isn't very common?
 
crosstudio said:
the sampling rate has no direct effect on how high a signal you can hear. the sampling rate tells how many times per second a snapshot of the original sound is taken; however, because a higher frequency changes its sound faster, taking more frequent snapshots will help to more accurately portray that upper range...
hence an even greater perceived improvement in clarity in the upper range with a higher sampling rate.

[edited note] a higher sampling rate will in fact enable the recording of higher frequencies, but you can't hear them.

48khz vs. 44.1khz: 3,900 samples per second difference. which makes your recording of the original sound 8.125% more accurate.

96khz vs 48khz: 96khz is twice as accurate at representing the original sound.
Click to expand...

See my posts earlier. This isn't quite right. Sampling rate is entirely about frequency, and where you need to limit your input.

... that means that 24bit recording represents the sensitivity of the sound with 256 times more accuracy.
Click to expand...

Again, see the earlier discussion. It sort of depends on what you mean by "accuracy" and "256 times." A perhaps clearer way to think of it is dynamic range; a perhaps more meaningful number is 48; as in 48 decibels (which is the same as "256 times").
 
when people talk about bits, manufacturers and people get ALL SCREWED up confusing 1:Dynamic Range, 2: Signal to Noise Ratio
3: Signal to error ratio, 4: Resolution as all meaning the same thing
 
Going back a few days to BD's earlier question about the relationship between anti-aliasing and downsampling:

I think you need to impose an anti-aliasing filter when you do a digital-to-digital downsample, exactly as you do when you do an analog-to-digital conversion.

Consider, for example, what happens if you downsample from a high sample rate to a lower sample rate, and the signal you're working with is entirely above the Nyquist threshold of the lower sample rate. You flat out cannot represent this signal with the lower sample rate. Any data that is preserved will produce some signal other than the signal you want -- "aliasing."

Here's an example. Your high sample rate is double your lower sample rate. The signal is a simple sine wave that has a frequency exactly 1/3 of your high sample rate (3 samples per cycle), and thus 2/3 of your lower sample rate (3 samples every two cycles). Say the amplitude is 1.

At your high sample rate you would record three values per cycle. Let's say one hits right at the point where the wave crosses 0. The next point would be a bit after the peak, the next would be a bit before the next peak, and the next would be at 0 again. The values would be 0, .86, -.86, 0, .86, -.86, 0 ... There is exactly one sine wave with a frequency lower than 1/2 the high sampling rate that fits these values: the sine wave with a frequncy equal to 1/3 the sample rate.

Let's say you downsample by skipping alternate values:

0, -.86, .86, 0, -.86 .... OR .86, 0, -.86, .86, 0 ....

That looks familar. What fits that sequence of values at the lower sample rate? Unfortunately, it is 1/3 the lower sample rate (3 samples every cycle)! This is, of course 1/2 the original high sample rate. So the original signal has been "aliased" to a lower frequency signal. The phase is also changed.

To be even more specific: say you are doing downsampling from 88.2k to 44.1k. If you have 29.4k signal, it will be represented just fine at the 88.2k sample rate. But, if you downsample you will wind up with a spurious 14.7k signal in your new data.

The only way to deal with this signal is to ignore it when you downsample. Which you do by applying an anti-aliasing filter.

This will apply generally to any signal that has components above the lower sample rate's Nyquist threshold. If you take a complex signal like this and break it up into its component sine waves, those under the lower Nyquist threshold will map perfectly when you do the downsample. Those above the lower Nyquist threshold will alias as some different, lower frequency, producing distortion.

You may still come out ahead by recording at 88.2, then downsampling. In fact, I suspect you would, because you could use a digital brickwall anti-aliasing filter when doing the downsample which would produce less distortion below the cutoff frequency than an analog filter would.
 
Last edited:
sjjohnston said:

This will apply generally to any signal that has components above the lower
You may still come out ahead by recording at 88.2, then downsampling. In fact, I suspect you would, because you could use a digital brickwall anti-aliasing filter when doing the downsample which would produce less distortion below the cutoff frequency than an analog filter would.
Click to expand...

How would you use a digital brickwall anti aliasing filter, or any anti aliasing filter? I have logic audio, and i only have a sbcard right now so i never have to dither, but im going to get a delta card very soon. I use logic audio for mixing, so when i bounce, i'd set it for whatever bitrate i wanted, and then i'd bounce it? And how would i set the anti aliasing filter?

in the options it gives me when i click bounce, i can turn dithering OFF, POW-r #1 (dithering), POW-r #2 (noise shaping), or POW-r #3 (noise shaping)

so i need to get a seperate dithering plugin?
 
It's important to distinguish between sample rate and bit rate.

When you downsample, you reduce the number of samples per second. For example, converting a file from a 96 kHz sample rate (current typical high end) to 44.1 kHz (CD standard). You need to use an anti-aliasing filter to avoid aliasing. I expect that any software (or firmware, in a hardware device) that does sample rate conversions is smart enough to use an anti-aliasing filter. You shouldn't have to worry about this unless you're just taking a cable from a device outputting data at one sample rate and just plugging it into a device that's expecting to get data at a different sample rate ... in which case you may well have a number of other problems anyway.

When you reduce the bit rate you reduce the number of bits used to record the value of each sample. For example, converting a file from 24 bit (current typical high end) to 16 bit (CD standard). You can do this just by truncating: that is, by just ignoring the eight least-significant bits from the 24 bit data. This works fine. You can, however, use dither (or noise shaping, which is just a modified form of dither), which should work a little better than truncating.
 
With twice as many samples to use, you are starting with more information in the first place. So unless the downsampling algorythm is just taking every second sample, you are going to end up with a more accurate wave form overall.

I know in graphics that oversampling can blur an image, but generally it still looks better, especially near areas of high frequency change where a limited sampling rate is highlighted (edges of polygons). I imagine the same is true for audio. As a wild guess I'd wager you get softer but more accurate high frequency response after downsampling, as apposed to just tracking at 44 in the first place.

The arguments I've heard for whether the human ear can tell the difference.
1) CD quality targeted the genreal accuity of the human ear, but many people can hear the difference compared to hgiher quality sampling.
2) 16 bit isn't enough to give quiet segments dynamic range.
3) As mentioned in this thread, 44 khz has limited ability to sample high frequencies.

The trailing end of a symbol crash comes to mind.

Anyhow, after all that let me say I really don't know wtf I'm talking about when it comes to audio, I'm just guessing based on what I've heard and what I know of other digital media. :)

Doug
 
sjjohnston said:
I think you need to impose an anti-aliasing filter when you do a digital-to-digital downsample, exactly as you do when you do an analog-to-digital conversion....


You may still come out ahead by recording at 88.2, then downsampling. In fact, I suspect you would, because you could use a digital brickwall anti-aliasing filter when doing the downsample which would produce less distortion below the cutoff frequency than an analog filter would.
Click to expand...

Precisely my thinking. Recording at 96 kHz and downsampling to redbook audio CD would probably have issues, which may or may not outweigh the benefits of less antialiasing filter phase distortion.

However, since 88.2 to 44.1 is a "clean" downsample in which no interpolation error should occur, and has the benefit of ensuring that the antialias filter doesn't cause audible distortion, I can't see any reason not to use it, other than disk space.

Perhaps a good A/D would oversample, digitally filter frequencies above the Nyquist limit, and then downsample cleanly, but since I can't find that in writing I'll record at the higher sampling frequency and make sure I know what is going on.
 
What if i send the mix out of my soundcard, then route it back in through a mixer, and just re-record it, but in 16bit 44.1khz. "dithering" down, without dithering.

i think i read a discussion about this before.
 
you better have an asynchronous clock in your sound card!
 
Well, what you are proposing is to play the music at, say, 24/96, sending it through a mixer, and then recording it at 16/44.

That would mean that your soundcard would need two clocks, so that it could play at 96 kHz and record at 44.1 kHz simultaneously.

That gives me an idea though... I wonder if you could put a second card in your machine to record the 16/44 signal? You would only need two channels, since you would be doing this most likely on the final stereo mix, and only need 16/44 capability. $200 or less, right, even for the best quality?

Hmm... no dither. Wonder if the extra A/D D/A steps would hurt more than dither? I know pro studios don't usually worry about the extra conversion steps, but then many of them are using Apogee and other very fine converters.
 
You must log in or register to reply here.
Back
Top