As would I.
The rate of HF rolloff is a function of where the ear is relative to the HF beam. It's not smooth or slow outside of the HF beam width.
Can you provide a link?
The cart L and load C define a 2-pole low-pass.
Mechanically the stylus arm is "stiff" only up to a ~~~20kHz resonance and then decouples, roughly as another 2-pole.
So that's 24dB/octave above the audio band. In old days that was pretty "brick". In the brave new world of digital filters it sounds slack. But you can NOT hear anything past the corner of a 24dB/oct roll-off.
Tha's not even getting to the problem of a >20kHz vinyl-wiggle being smaller than the stylus tip.
Well known since the late 1950s, when the Pickering edged the GE cartridge.
And ignored that much longer.
Vivaldi's Nisi Dominus, as performed by Le Banquet Celeste, seems to have the opposite effect on my cat. Maybe you should try it.
No, but I can provide a reference:
Tsutomu Oohashi, Emi Nishina, Norie Kawai, Yoshitaka Fuwamoto and Hiroshi Imai, "High-frequency sound above the audible range affects brain electric activity and sound perception", Audio Engineering Society preprint 3207, presented at the 91st Convention, October 1991
It isn't clear to me from the description in their article whether their test was double-blind or only single-blind - which would make it quite unreliable.
There are ways to largely get rid of the electrical part, such as using a moving-coil cartridge or using a moving-magnet cartridge in damped mode with or without an inverse resonance filter to correct for the highish-Q mechanical resonance. Hans van Maanen has done some pioneering work on that in the 1970's and billshurv can tell you all about it. See Steven van Raalte, "Correcting transducer response with an inverse resonance filter", Linear Audio volume 3, April 2012, pages 69...90, or if you can read Dutch, H.R.E. van Maanen, "Compensatie van mechanische resonantie bij pick-up elementen", Radio Elektronica 15/16 1979, pages 25...29 and 17 1979, pages 35...41.
Anyway, assuming a fourth-order 20 kHz Butterworth alignment for the cartridge, the magnitude response would be:
20 kHz (typical cut-off frequency of a 44.1 kHz sample rate audio ADC): -3.01 dB
22.05 kHz (Nyquist frequency at 44.1 kHz sample rate): -5.03 dB
21.769 kHz (typical cut-off frequency of a 48 kHz sample rate audio ADC): -4.73 dB
24 kHz (Nyquist frequency at 48 kHz sample rate): -7.24 dB
I may be off by a dB or two because the alignment is often Chebyshev rather than Butterworth, but at least this shows the ballpark.
It seems to me that if you want to be absolutely sure that neither your cat nor any visiting Japanese gamelan players hear any difference at all between the vinyl record and the digital recording, 44.1 kHz and 48 kHz sample rates are both awfully tight. 88.2 kHz and 96 kHz are probably overkill, but with the standard audio sample rates being as they are, those are the next choices.
Last edited:
I just checked and the record is actually a 12 inch single rather than an LP - which means a 35 % larger velocity of the vinyl. Sorry about that.
The tip mass/ groove compliance resonance is a major cause of high frequency loss in vinyl discs. Many years ago when I was involved in the fitting and testing of replacement pickup movements for AM broadcast stations (mono), it was found that with repeated plays the frequency test record showed a rise in output at the resonance frequency (12 t0 14 kHz) and a loss of the high frequencies above that point. A better version of the cartridge which had a lower VTF but was considered too delicate for the broadcast stations had its resonance at 18 kHz and the high frequencies of another test record followed the same trend.
And no precautions to rule out sub harmonic 'oildrum" resonances which many speakers do at high frequencies. These make audible sound from ultrasonic source
Thanks, got it, downloaded. I'm slogging through the paper slowly...
It's neither DBT or SBT. They take direct shots as what they call "paired comparisons" (DBT), with their objection being that they found ultrasonic stimulus results in changes in the brain's EEG that has a long residual effect, which they claim confuses the results from short paired comparisons. So their testing was not done by collecting listeners judgmental responses, it was collected by observing changes in "brain electrical activity mapping" using electroencephalogram data in response to differing bandwidth stimuli. All of the test stimulus material was Gamelan music from Bali, India. I wonder if they found any correlation between that music and EEG data indicating extreme annoyance. The claimed correlation was when their ultrasonic radiating tweeters were turned on and a notable and persistent change in brain electrical activity mapping, with a connection to that of pleasure inducing stimulus.
I do recall the paper a bit, also that there was something a bit amiss in the methodology and conclusions, but until I re-digest a bit, it's not jumping out. Might be the introduction of ultrasonic information using dedicated (and special) tweeters and amps that may not be calibrated relative to the audio-band speakers. But I really don't recall.
What I think is a somewhat better approach to figuring out the results of high bandwidth in audio was the AES paper by Deane Jensen and Gary Sokolich titled "Spectral Contamination", which draws parallels with previously unmeasured intermodulation (spectral contamination) and band-limited systems of the time. Their measurement technique is difficult to replicate exactly, but showed that nonlinearities of severely band-limited systems near their upper cut-off frequency caused intermod products to be "folded" down into the upper midrange. The problem, of course, would largely go away with the exit of multi-pole analog filters, but that doesn't mean it doesn't have validity today. But it showed a reason people may perceive that audio systems with beyond 20kHz response may sound better that had nothing to do with the presence of content above 20khz.
Last edited:
The precaution was to only pass their ultrasonic content through a set of dedicated tweeters and amps with response up to I think 30kHz.
A few problems with the paper so far:
1. No calibration method shown to match the "super tweeter" channel gain to the full range channel.
2. No standardization of playback level - the test subjects could adjust to preference. This introduces a serious uncontrolled error source.
3. Of the stimulus above 26kHz, more than half of that spectrum was down 60dB re: mid band.
4. No attempt to isolate the means of neurological stimulus, i.e. auditory only vs skin
5. Of a total claimed test group of 40 subjects, analysis focuses only on 16 subjects. There is no information on why and how the presented subject data was chosen. Subject #1, for example, had a Full Range/High Cut success rate of +134%. It is common to reject extreme data samples in full statistical analysis. These 16 look quite "cherry-picked", and the balance of the subjects are never presented.
6. The main criticism of "pair comparison" testing relates to the duration of the comparison being too short with the potential carry-over of the positive effects of ultrasonic stimulation confusing the results when, in actual fact, a real DBT comparison is not time-limited at all, and comparisons may be as long or short as the subject desires, with total subject testing capability stretching to months. Long-duration pair comparisons have always been a possibility, but they claim they are limited to very short, rapid switched, time limited tests, which is incorrect.
7. The claim is made early on that the choices of sampling rates of 32kHz and 44kHz were chosen solely on he basis of paired-comparison testing. While PCT was certainly used, the specific sampling rates were chosen for other practical reasons. Initial digital audio recordings by Tom Stockham (SoundStream) were done at 50kHz. 44.1 was chosen because of the use of standard video recorders as the data storage method, and 44.1kHz x 2 resulted in evenly blocked data and error correction information within the lines and frames of that system. Had 44.1 not been chosen, the production path would have included far more expensive data storage methods, and the CD would have had to wait for a disc with higher packing density. While their statements re: choice of sampling frequencies does not directly impact the validity of the paper, it is one of the earlier of a string of "spin" statements that are presented.
I probably won't spend any more time on this, for obvious reasons.
1. No calibration method shown to match the "super tweeter" channel gain to the full range channel.
2. No standardization of playback level - the test subjects could adjust to preference. This introduces a serious uncontrolled error source.
3. Of the stimulus above 26kHz, more than half of that spectrum was down 60dB re: mid band.
4. No attempt to isolate the means of neurological stimulus, i.e. auditory only vs skin
5. Of a total claimed test group of 40 subjects, analysis focuses only on 16 subjects. There is no information on why and how the presented subject data was chosen. Subject #1, for example, had a Full Range/High Cut success rate of +134%. It is common to reject extreme data samples in full statistical analysis. These 16 look quite "cherry-picked", and the balance of the subjects are never presented.
6. The main criticism of "pair comparison" testing relates to the duration of the comparison being too short with the potential carry-over of the positive effects of ultrasonic stimulation confusing the results when, in actual fact, a real DBT comparison is not time-limited at all, and comparisons may be as long or short as the subject desires, with total subject testing capability stretching to months. Long-duration pair comparisons have always been a possibility, but they claim they are limited to very short, rapid switched, time limited tests, which is incorrect.
7. The claim is made early on that the choices of sampling rates of 32kHz and 44kHz were chosen solely on he basis of paired-comparison testing. While PCT was certainly used, the specific sampling rates were chosen for other practical reasons. Initial digital audio recordings by Tom Stockham (SoundStream) were done at 50kHz. 44.1 was chosen because of the use of standard video recorders as the data storage method, and 44.1kHz x 2 resulted in evenly blocked data and error correction information within the lines and frames of that system. Had 44.1 not been chosen, the production path would have included far more expensive data storage methods, and the CD would have had to wait for a disc with higher packing density. While their statements re: choice of sampling frequencies does not directly impact the validity of the paper, it is one of the earlier of a string of "spin" statements that are presented.
I probably won't spend any more time on this, for obvious reasons.
Actually it is very relevant today, because most DAC chips with built-in digital interpolation filters clip on peak sample normalized music and many recordings are peak sample normalized. When the largest samples are exactly at full scale, the waveform in between the samples may have to exceed full scale, this is the so-called intersample overshoot problem. DAC chip manufacturers compete with each other on dynamic range numbers measured at 0 dBFS, so keeping some headroom for intersample overshoots would reduce their market potential.
That is a very real problem, but unrelated to the discussion.
The only real solution is a DAC with adequate headroom, and there are a few, Benchmark comes to mind.
Really? The OP has a misconception of what is required digitally archiving vinyl. The discussion of high sampling rates and the lack of meaningful contribution to audible result of frequencies above 24kHz directly relates to clarifying that misconception. The discussion of intersample overs is tangential at best.
1
This paper isn't about audibility of sound.
2
Not double blind
3
debunked over and over and over and over and over and over and over......
I wouldn't digitize vinyls below 96kHz@24bit.
AFAIK, the higher the sample rate of the AD-converter, the more flat the frequency response is.
And the Nyqvist theorem (or whatever) works only if the bit depth is infinite (which it is not). Am I right?
Also, you can digitize upto 24kHz with 48kHz sample rate but if and when you want to remove high frequency noise (above 24kHz) from the signal, you may be in trouble.
After noise filtering and other edits has been made, then downsample to 48kHz.
Wrong. Sample frequency has nothing to do with FR flatness. Sample frequency defines the maximum frequency that can be quantized.
No, you are wrong. The Nyquist-Shannon theorm applies at all bit depths. Here you go.
When you attempt to make statements about which you have no understanding, you may be in trouble.
Oh my.
I used to sample at 24/192 but I find 24/96 is good enough these days. I use a Xonar Essence STX right now, but I'm looking for a good deal on a Behringer DEQ2496 which would allow me to get the sound from my tube phono amp into my computer VIA optical (no ground loops, cheaper than good isolation xfmrs).
Personally I prefer vinyl over CD basically every time, but that might have more to do with the vinyl mastering vs the loud as hell with no dynamic range you get on most modern CDs than the actual medium.
Personally I prefer vinyl over CD basically every time, but that might have more to do with the vinyl mastering vs the loud as hell with no dynamic range you get on most modern CDs than the actual medium.