What is the provenance of your triangle? Since you started with D/A before A/D, it's worth asking whether you just generated a triangle as data without consideration for aliasing. Instead, if you created the triangle as a sum of sine waves, all below Nyquist, then your tests were valid. I haven't actually seen the results of your test, but I still wonder whether your initial waveform data was valid.
You can't anti-alias filter after the fact - it simply isn't mathematically possible.
The research that has been passed on to me is that 18 or 19 bits is the most you will ever need for playback in a living room. I suppose only an audiophile would be interested in going beyond 16-bit, but at least we can agree that 20-bit is more than enough.
I completely agree with the recording requirement for 24-bit. Having made several hundred live multitrack digital recordings, I can't possibly imagine working at 16-bit (although a heavy mix with 8 or more tracks is not always so bad when each individual track is 16-bit).
Sorry, but it seems your calculations are wrong. If the single error is 10 LSB, then it's really +/-10 LSB (or +/-5 LSB). More importantly, if the "fuzzy distortion" is uncorrelated Gaussian noise, then two +/-10 LSB errors will add to only +/-14 LSB (or +/-7 LSB). Keep in mind that one error might be positive while another is negative, and thus the total will not be as high as you think.
I suppose you can say that the peak-to-peak range will double, but since we don't really hear peaks so much but rather hear something akin to RMS, then the square root of two combination is more accurate for predicting listening results.
Hard for me to accept that a small linear ramp portion of a 1KHz triangular waveform is not a valid input.
But I'm sure I'll find out now...
Thanks
-Antonio
Not so much the ramp as the right angle turn at the top and bottom. Those are infinite in bandwidth, although the magnitudes of the harmonics are only 1/(N*N). If you create the triangle as a linear ramp, you'll have aliasing that cannot be filtered out.
It's fairly easy to sum sine wave harmonics until you have the best triangle wave that is possible at any particular sample rate. Higher sample rates will be 'better' of course.
Scott,
Well, the one I have does have near workstation performance on AC, with a high graphics accelerator and a very fast dual core CPU.
I routinely run quite heavyweight simulation and related electronics packages on it (plus a full installation of a large scale ERP System, stand alone, server & client for my other dayjob).
On battery it shuts off the graphics accelerator and the super fast CPU gets clocked down to minimum and so is not the best platform for complex sims, but can handle playing video or recording music etc.
It seems a rather limited application, but yes, for that portables are still needed. I was actually more talking about "in field" (as opposed to "in studio") music recording. If you want to record jungle noises while walking around it would be limiting...
Ciao T
Well, the one I have does have near workstation performance on AC, with a high graphics accelerator and a very fast dual core CPU.
I routinely run quite heavyweight simulation and related electronics packages on it (plus a full installation of a large scale ERP System, stand alone, server & client for my other dayjob).
On battery it shuts off the graphics accelerator and the super fast CPU gets clocked down to minimum and so is not the best platform for complex sims, but can handle playing video or recording music etc.
It seems a rather limited application, but yes, for that portables are still needed. I was actually more talking about "in field" (as opposed to "in studio") music recording. If you want to record jungle noises while walking around it would be limiting...
Ciao T
Hi,
Well, it removes that one, but it adds massive amounts of "FD"...
It is really instructive to compare a classic analog amp (needs to be nothing better than a gainclone) against a Class D design running on the same rail voltages, both measured and by listening...
Ciao T
Well, it removes that one, but it adds massive amounts of "FD"...
It is really instructive to compare a classic analog amp (needs to be nothing better than a gainclone) against a Class D design running on the same rail voltages, both measured and by listening...
Ciao T
Don Moses proposed such a system to me years ago for measuring the amount of powerline noise that makes it through the power supply. In reality you need a much wider bandwidth since there can be significant energy up to 30 MHz on the power line. I have used a system consisting of a neat 10 Hz to 40 MHz vector analyzer OMICRON Lab "Smart Measurement Solutions": Home and two line isolation couplers that allow me to inject into the line at one place and see what comes out somewhere else. It allows me to measure ac line filters in operation. You could also use a spectrum analyzer and an isolator to protect it.
Quantifying degradation may be a challenge given all the sensitive nerves here. Showing increased im distortion or reduced snr in the audio band when out of band noise is injected would be interesting. An old trick, long lost, is to use a CB radio to test for stability and noise rejection. Its closest modern equivalent is the TDMA noise from a GSM phone.
Would the sine sum generated 1kHz triangle be linear even in its middle third?
The 22kHz brickwall filtering is unacceptable for highest fidelity music reproduction. No one has ever proven that this sharp frequency cut-off has no influence to music perception and sound quality. If the roll-off were slow, then there would be no big problem. As the slow roll-off is impossible for the reason of aliases, we get what we get and we can complain to eternity. "We" means probably 0.00000000001% of population, and of course the rest counts for the consumer music industry.
PMA, unfortunately people have been running blind tests since the beginning of digital reproduction and assured everyone that the 'brickwall filter' was not audible. You know, the usual 'scientific tests'. There is some research that preceded digital audio that proved the opposite, of course, in a book called 'Audio Quality' by G. Slot, 1971.
sorry the wavefroms offend your eyballs - but isn't Hearing the subject
G Slot (pseudonym?) isn't found in Author Search at AES or ASA sites - lots of people publish many "soft" speculations, especially in Audio - so without seeing the reference it is hardly more conclusive than "someone once wrote the opposite"
as I read the debate the question is about finding some(any one) who can tell with music, at normal listening levels that it had been passed thru 16/44.1 ADC/DAC in blinded, controlled, pass through testing
and my psychoacoustics handbook's 2006 3rd edition still prints the graph of audible "threshold in quiet" with even the 10% highest resolving fraction of "subjects 20 to 25 years old" going vertical well before 20 kHz
again the text is explicit: "limit above which no sensation is produced even at high levels... between 16-18 kHz at an age of 20-25 years"
from the threshold in quiet graph's top label 80 dB SPL would the start of "high levels"
so why the exaggeration? - I 'm not saying increased sample rates will never be proven to be audible vs 44.1
and with there being some uncertainty, with a couple of very small studies on “the other side” it might make us more comfortable to use higher sample rates since today the BW is cheap and the material is captured at higher rates in the studio
but it simply isn't "proven" that even as many as 10% of the population can tell the difference in typical recorded music listening - we have null results in studies of 100s of subjects, some Audio professionals
G Slot (pseudonym?) isn't found in Author Search at AES or ASA sites - lots of people publish many "soft" speculations, especially in Audio - so without seeing the reference it is hardly more conclusive than "someone once wrote the opposite"
as I read the debate the question is about finding some(any one) who can tell with music, at normal listening levels that it had been passed thru 16/44.1 ADC/DAC in blinded, controlled, pass through testing
and my psychoacoustics handbook's 2006 3rd edition still prints the graph of audible "threshold in quiet" with even the 10% highest resolving fraction of "subjects 20 to 25 years old" going vertical well before 20 kHz
again the text is explicit: "limit above which no sensation is produced even at high levels... between 16-18 kHz at an age of 20-25 years"
from the threshold in quiet graph's top label 80 dB SPL would the start of "high levels"
so why the exaggeration? - I 'm not saying increased sample rates will never be proven to be audible vs 44.1
and with there being some uncertainty, with a couple of very small studies on “the other side” it might make us more comfortable to use higher sample rates since today the BW is cheap and the material is captured at higher rates in the studio
but it simply isn't "proven" that even as many as 10% of the population can tell the difference in typical recorded music listening - we have null results in studies of 100s of subjects, some Audio professionals
The main problem is that 44.1kHz sampling violates conditions of high quality sound reproduction. It is simply unacceptable to cut everything above 22kHz.
This is a disaster of purely engineering approach. They got an order to digitalize the sound in 20Hz-20kHz band, so they cut everything above cca 20kHz and they said they done the job. Oh yes they did, regarding Nyquist. Regarding music reproduction, we have had horrible digital for 30 years at least.
This is a disaster of purely engineering approach. They got an order to digitalize the sound in 20Hz-20kHz band, so they cut everything above cca 20kHz and they said they done the job. Oh yes they did, regarding Nyquist. Regarding music reproduction, we have had horrible digital for 30 years at least.
- Status
- Not open for further replies.