24-Bit / 48kHz or 96kHz

[aaronji]

For anyone interested, here is a good resource on the measured output of instruments >20kHz-

"There's Life Above 20 Kilohertz: A Survey of Musical-Instrument Spectra to 102.4 kHz"

At least one member of each instrument family (strings, woodwinds, brass and percussion) produces energy to 40 kHz or above, and the spectra of some instruments reach the measurement limit of 102.4 kHz. Survey includes French horn, trumpet, violin, oboe, crash cymbals, sibilant speech, claves, a drum rimshot, triangle, jangling keys, and piano. Includes short description of others' work on perception of air- and bone-conducted ultrasound, and points out that even if ultrasound be taken as having no effect on perception of live sound, yet its presence may still pose a problem to the audio equipment designer and recording engineer.

That's an interesting paper.  Together with the Lavry white papers and an excellent physiology course in college, it forms the basis of my understanding of this whole issue (although I have read a lot of other stuff at this point).  One thing I would note about it, however, is that it has not been peer-reviewed.  He says that people involved in standards-setting wanted it to be available as soon as possible.  It seems to me, however, that he would still want to submit it to a professional journal (and I don't think a preliminary release of data precludes that).  Perhaps unfair of me, but I tend to think there is some flaw in his methodology that he fears would undermine his results and/or conclusions if subjected to review...

EDIT:

I have no stake in this and don't understand the physics in the least so its only barely comprehensible, but am curious what people have to say about this souce (http://people.xiph.org/~xiphmont/demo/neil-young.html) posted in another thread but didn't seem to get much attention.

It says recording the ultrasonics at the higher frequency is actually very slightly detrimental to average playback systems.  Though, I didn't try to really  understand what intermodulation distortion is and how its reproduced on normal playback above 48khz.

Thanks for the article! I will read it later in depth, but took a quick glance and it's definitely interesting. I think I already pointed this out in this thread, but...

The 24/192kHz format has been universally panned by critics. Many will tell you that it is clearly inferior to 24/96kHz and there is some science (or at least intarwebz science) that supports the claim that 192kHz is actually pushing and exceeding the theoretical limits of the modern ADC. Metaphorically, we're red-lining the RPMs and the engine is starting to overheat thus creating errors in the audio data.

24/96kHz does not suffer from this real-world performance limitation and is thought to be the superior format for anything that can hear issues with a 192kHz sampling rate.

As I read it, this author thinks there is no benefit to going over 44.1 or 48 kHz.  Obviously, given the almost ten pages we have devoted to it this thread (and many similar ones), that's debatable.  His line of reasoning does seem pretty solid, though...

[Gutbucket]

..curious what people have to say about this souce (http://people.xiph.org/~xiphmont/demo/neil-young.html)

Just got a chance to take a look.. interesing and well thought out, need to review it more in depth, some good links in there.  Thanks for posting this.

[TimeBandit]

I'm getting more motivated to setup a proper ABX test, but that's somewhat challenging to do as I'd need to run identical recorders at seperate rates.

Maybe the easier way choose a band which releases in 24/96 (as FLAC download), and other formats like normal 16/44 for CD, and maybe lossy 320 Mp3. Buy one track in all that formats and do the test.

[datbrad]

I have done that with Phish sbds. The 24/96 flacs are compressed and have a dynamic range around 50-60db. They also don't show information above 22khz when analyzed in Wavelab.  Basically an unecessarily large file that costs more, but sonically not worth the extra costs. Marketing genius, but not worth it in reality. 48khz sampling already captures up to 24khz of high end, and no recorders I know of have analog capability above that, so what is 96khz capturing?       

[hi and lo]

I have done that with Phish sbds. The 24/96 flacs are compressed and have a dynamic range around 50-60db. They also don't show information above 22khz when analyzed in Wavelab.  Basically an unecessarily large file that costs more, but sonically not worth the extra costs. Marketing genius, but not worth it in reality.

I think the same thing about most music today... especially vinyl's pressed from the digital masters. 

[Gutbucket]

I'm getting more motivated to setup a proper ABX test, but that's somewhat challenging to do as I'd need to run identical recorders at seperate rates.

Maybe the easier way choose a band which releases in 24/96 (as FLAC download), and other formats like normal 16/44 for CD, and maybe lossy 320 Mp3. Buy one track in all that formats and do the test.

I could do that with the 24/96 files I've already recorded, which would be a better test as I know my recording situation had instrument created frequencies present >20kHz  and know the setup, gear used, and all processing of the files.  In that case, to eliminate as many other variables as possible, I would take a copy of the original 24/96 file, downsample it to 48 or 44.1 and upsample again to 96kHz.  I'd then ABX test that file with the original.  But that isn't the test I want to make as it would simply be duplicating the already tested and refuted audibility of a standard 16-bit/44.1-kHz A/D/A loop inserted into a high-res playback chain.  I accept the results of that double-blind study, which concluded the insertion is inaudible, so duplicating that doesn't interest me.

What I still wonder about is the original capture conversion of 48 vs 96 kHz using my particular equipment.. not because I think any ultrasonic information may be audible or that the high-quality resampling I can do on the computer may be audible, but because the recording equipment I'm using is modest and the ADCs in the recorders may perform better at 96 than 48 (or vice-versa) for a number of reasons.  As I understand it, the main reason is that the nescesary anti-aliasing filtering can be less steep and easier to implement at 96kHz than 48kHz since the available transition band from 20KHz up to the half-sampling rate frequency is bigger.  That's covered in the blog achalsey linked to above.  Here's the illustrative image posted there-


Above: Whiteboard diagram from A Digital Media Primer for Geeks illustrating the transition band width available for a 48kHz ADC/DAC (left) and a 96kHz ADC/DAC (right).

So it's really more a test of how well the modest ADCs in some of my equipment work.  For example I wonder about the real-world performance of the analog low-pass filter before (or integrated in) the ADC chip used in the Tascam DR2d I use frequently (now selling for ~$100).  I would wonder much less about high-quality ADC's like a Mytek, Lavry, etc. where far more resources are available for top quality analog filtering beore the ADC chip.  It's somewhat ironic I suppose that it's likely to be the inexpensive equipment where recording at 96kHz may make a difference, whereas it's less likey to make a difference with higher end gear.  But it makes sense when considering that the filtering must be done in the analog stage, and good quality high-slope analog filters are more difficult and costly to implement.

..48khz sampling already captures up to 24khz of high end, and no recorders I know of have analog capability above that, so what is 96khz capturing?

See the graphs above.  The anti-aliasing lowpass filter must exclued all signal above half the sampling rate, so response tapers off as you approach that frequency.  Practically, you can never really record all the way up to 24kHz at a 48KHz rate since the filter cannot be infinitely steep.  The advantage of higher rates is easier to implement low pass filters with lower slopes, not capturing higher frequencies- one of the main points being made in the blog achalsey linked above.

[notlance]

"As I understand it, the main reason is that the nescesary anti-aliasing filtering can be less steep and easier to implement at 96kHz than 48kHz since the available transition band from 20KHz up to the half-sampling rate frequency is bigger."

I have read reasoning similar to the above time after time on various audio forums, and it is not applicable to the ADC design used for, oh, about 20 years now.  Many-pole, i.e. "brick wall", analog filters are almost never used anymore since most ADC oversample.  (I hesitate to say "all current audio ADCs oversample" since I don't know how every manufacturer has designed their ADC, but I would like to know why you'd implement and ADC that does NOT oversample.)

Oversampling Benefits
• Almost no stringent requirements imposed on analog building blocks
• Takes advantage of the availability of low cost, low power digital filtering
• Relaxed transition band requirements for analog anti-aliasing filters
• Reduced baseband quantization noise power
• Allows trading speed for resolution

An oversampling ADC can greatly reduce the order of analog anti-aliasing filters.  For example, 44.1 kHz/24 bit implemented by a 256 times oversampling ADC can use a 3-pole analog anti-aliasing filters (18 dB/octave).  3-pole filters are well understood and are cheap and easy to implement.

[Brian Skalinder]

Oversampling Benefits
• Almost no stringent requirements imposed on analog building blocks
• Takes advantage of the availability of low cost, low power digital filtering
• Relaxed transition band requirements for analog anti-aliasing filters
• Reduced baseband quantization noise power
• Allows trading speed for resolution

It's considered common courtesy to cite your source(s) when quoting others.

[notlance]

OK, here is my source:

http://www-inst.eecs.berkeley.edu/~ee247/fa09/files07/lectures/L23_2_f09.pdf

These are lecture notes from lecture 23 of the UC Berkeley class EE247, which is a graduate level class with this description:

EE247:  Analysis and Design of VLSI Analog-Digital Interface Integrated Circuits. (3)   Three hours of lecture per week. Prerequisites: 140. Architectural and circuit level design and analysis of integrated analog-to-digital and digital-to-analog interfaces in CMOS and BiCMOS VLSI technology. Analog-digital converters, digital-analog converters, sample/hold amplifiers, continuous and switched-capacitor filters. RF integrated electronics including synthesizers, LNA's, and baseband processing. Low power mixed signal design. Data communications functions including clock recovery.

[cashandkerouac]

i've been in the camp that thinks there is little to no benefit to 24/96 vs 24/48 (for live concert recording) without ever having tried it.  

so last night i was going to a one set show and thought "what the heck, i'll give 24/96 a shot and see how it comes out".  i fully expected it to be a one show experiement at 24/96, afterwhich i would quickly retreat to my normal 24/48.

after listeining to the results i think i've been converted.  24/96 is noticeably better than 24/48, but it's somewhat subtle at the same time.  the playback results on an average sound system may not be immediately noticable (this i will acknowledge), but here is where i noticed significant benefit with 24/96. 

before i even started the recording i noticed a significant difference in the behavior of my levels.  i had headroom for days with the recorder set to 24/96 and had to use a much higher gain level compared to 24/48.  i was recording a jazz show (Lee Ritenour & Mike Stern), so the music had a pretty wide dynamic range.  when set at 24/96 the loudest peaks of the set did not cause my levels to spike like they would at 24/48, so setting levels was much easier. 

when i listented to the recording after normalizing the levels the first thing i noticed was the "spaciousness"... in the sense that the loud parts did not sound as if they were nearing the headroom limitations that can be sensed with a 24/48 recording.  the music sounded like it had plenty of room to "breathe" through the full range of dynamics. 

i'm not articulating my comments as well as i should, but i think y'all probably know what i'm trying to say.  there's still nothing wrong with 24/48, but i'll continue to experiement with 24/96 to see if my initial results are consistent. 

[Chuck]

I'm still in the camp that says 16 bit 44.1kHz is enough to capture everything that is hearable.

http://people.xiph.org/~xiphmont/demo/neil-young.html

[F.O.Bean]

My last recording was done at 24/96, just for shits and giggles. I dont know if it sounds BETTER than 24/48, but it does have a spaciousness that I dont often here in that same venue/same location. I just HATE dealing with the HUGEASS 24/96 files, and the LMA ALWAYS gives me problems when UL'ing 24/96 flacs.

With that being said, I think 24/48 is just fine and I'll be sticking with that as usual

[aaronji]

before i even started the recording i noticed a significant difference in the behavior of my levels.  i had headroom for days with the recorder set to 24/96 and had to use a much higher gain level compared to 24/48.  i was recording a jazz show (Lee Ritenour & Mike Stern), so the music had a pretty wide dynamic range.  when set at 24/96 the loudest peaks of the set did not cause my levels to spike like they would at 24/48, so setting levels was much easier. 

when i listented to the recording after normalizing the levels the first thing i noticed was the "spaciousness"... in the sense that the loud parts did not sound as if they were nearing the headroom limitations that can be sensed with a 24/48 recording.  the music sounded like it had plenty of room to "breathe" through the full range of dynamics. 

I don't think that sampling rate has much of an influence on headroom...Maybe something was set differently?  Or just quieter music than what you're used to? 

[page]

I don't think that sampling rate has much of an influence on headroom...Maybe something was set differently?  Or just quieter music than what you're used to?

or confirmation bias.

[Chimney Top]

I've only read the first few pages, but I wanted to add (unless it's already been mentioned) if you record at 24/96 to an SD card, you should use a higher performance/class 10 card.  Some recordings have stopped due to a 'write error' with a class 6 SD card.