I can't speak to common usage, I can speak to correct usage. "Resolution" is the ability to distinguish between two signals. If the amplitude resolution is 0.001, that means that one can distinguish between a signal of, say, 2.000 and one of 1.999. If the frequency resolution is 1 Hz, one can distinguish between 1000 Hz and 1001 Hz.
Audio bit depth - Wikipedia, the free encyclopedia
When in Rome...
http://www.analog.com/static/imported-files/tutorials/MT-001.pdf for further reading!
So if you prefer you can call it processing gain.
Last edited:
There were three fellows walking down the street, one of them said "Seventy-two." They all started laughing. Then another said "forty-six." Again they all broke up laughing. A bystander watching this inquire about why they were laughing. It was explained to him they had spent so much time with each other they just gave their jokes numbers.
The bystander asked if he could try, they agreed and he said "Twelve." No one laughed. The bystander inquired why and was told "Well you see it is all in how you tell the joke!"
The bystander asked if he could try, they agreed and he said "Twelve." No one laughed. The bystander inquired why and was told "Well you see it is all in how you tell the joke!"
"res·o·lu·tion (rz-lshn)
n.
1. The state or quality of being resolute; firm determination.
2. A resolving to do something.
3. A course of action determined or decided on.
4. A formal statement of a decision or expression of opinion put before or adopted by an assembly such as the U.S. Congress.
5. Physics & Chemistry The act or process of separating or reducing something into its constituent parts: the prismatic resolution of sunlight into its spectral colors.
6. The fineness of detail that can be distinguished in an image, as on a video display terminal.
7. Medicine The subsiding or termination of an abnormal condition, such as a fever or an inflammation.
8. Law A court decision.
9.
a. An explanation, as of a problem or puzzle; a solution.
b. The part of a literary work in which the complications of the plot are resolved or simplified.
10. Music
a. The progression of a dissonant tone or chord to a consonant tone or chord.
b. The tone or chord to which such a progression is made.
11. The substitution of one metrical unit for another, especially the substitution of two short syllables for one long syllable in quantitative verse."
SY
I can see with your background you will use it differently than DSP folks, but chill!
n.
1. The state or quality of being resolute; firm determination.
2. A resolving to do something.
3. A course of action determined or decided on.
4. A formal statement of a decision or expression of opinion put before or adopted by an assembly such as the U.S. Congress.
5. Physics & Chemistry The act or process of separating or reducing something into its constituent parts: the prismatic resolution of sunlight into its spectral colors.
6. The fineness of detail that can be distinguished in an image, as on a video display terminal.
7. Medicine The subsiding or termination of an abnormal condition, such as a fever or an inflammation.
8. Law A court decision.
9.
a. An explanation, as of a problem or puzzle; a solution.
b. The part of a literary work in which the complications of the plot are resolved or simplified.
10. Music
a. The progression of a dissonant tone or chord to a consonant tone or chord.
b. The tone or chord to which such a progression is made.
11. The substitution of one metrical unit for another, especially the substitution of two short syllables for one long syllable in quantitative verse."
SY
I can see with your background you will use it differently than DSP folks, but chill!
The version I know was the Comedian's Club. Everyone in the audience is a professional comedian, who are so professional, they just number the jokes. When the guy on the stage says, "Forty seven!" one comedian in the audience is just convulsed with laughter. A visitor asks his comedian host, "Why did that guy crack up so badly?" His host responded, "He hadn't heard that one before."
Not really. My background in this was from my days as an apps specialist at Nicolet. One of the things we had to teach engineer/customers about was the difference between accuracy, precision, resolution, dynamic range, and S/N. As a non-engineer, I often wondered what it is they taught in engineering school. Nowadays, that confusion is exploited for propaganda purposes in certain niches of audio. That doesn't mean that it should be accepted.
At secondary school, and in a university science department for my first degree, all experimental data had to be accompanied by an error estimate. For my second and third degrees, in an engineering department at a different university, such things were not required. I have no idea how typical this is, as my sample size is so small.
Perhaps Stuart could assist us by giving brief definitions of the various concepts which people confuse, to highlight the differences. I find I quickly forget them, but then I was always a poor experimenter!
Perhaps Stuart could assist us by giving brief definitions of the various concepts which people confuse, to highlight the differences. I find I quickly forget them, but then I was always a poor experimenter!
Hi,
If the Johnson noise fills the lower bits (say 7 out of 16) and the transients fall into that amplitude range, absolutely. But if we look at the signal with a sinewave and large FFT window averaging ton's of samples we will not see this.
The issue is not that noise is present, but rather the amount of it.
As an extreme example, DSD has a REAL dynamic range of around 7.76dB.
To get 120dB of averaged and severely bandwidth limited "SNR" around 112dB noiseshaping are applied, precisely what I call nearly 50% fuzzy distortion at digital full scale, rising rapidly as levels fall way past 1,000% Fuzzy Distortion...
We are talking about the same process, except I chose to view it for what it actually is, gross distortion, not something that some hand-waving magic turns suddenly into something else.
BTW, we are talking strictly in either analog electronic terms or in terms of information theory. I am unwilling to treat "digital audio" any different because it is suddenly "digital". The rules only change because some people say so, not because something inherent has changed.
We can (and have) debated the sonic results and I know some people actually like for example SACD and keep insisting that they prefer it to CD, that however is a whole other kettle of fish.
First, dither is often not fixed Johnson noise, but something else, often signal correlated. Second, of course, pure noise like dither is just that. But for measurements we can conventienly remove it from the visible FFT with enough averaging.
In reality a single sample is a single sample. It has one value derived at precisely this sample's instant, not by averaging thousands of samples...
Ciao T
If the Johnson noise fills the lower bits (say 7 out of 16) and the transients fall into that amplitude range, absolutely. But if we look at the signal with a sinewave and large FFT window averaging ton's of samples we will not see this.
The issue is not that noise is present, but rather the amount of it.
As an extreme example, DSD has a REAL dynamic range of around 7.76dB.
To get 120dB of averaged and severely bandwidth limited "SNR" around 112dB noiseshaping are applied, precisely what I call nearly 50% fuzzy distortion at digital full scale, rising rapidly as levels fall way past 1,000% Fuzzy Distortion...
We are talking about the same process, except I chose to view it for what it actually is, gross distortion, not something that some hand-waving magic turns suddenly into something else.
BTW, we are talking strictly in either analog electronic terms or in terms of information theory. I am unwilling to treat "digital audio" any different because it is suddenly "digital". The rules only change because some people say so, not because something inherent has changed.
We can (and have) debated the sonic results and I know some people actually like for example SACD and keep insisting that they prefer it to CD, that however is a whole other kettle of fish.
First, dither is often not fixed Johnson noise, but something else, often signal correlated. Second, of course, pure noise like dither is just that. But for measurements we can conventienly remove it from the visible FFT with enough averaging.
In reality a single sample is a single sample. It has one value derived at precisely this sample's instant, not by averaging thousands of samples...
Ciao T
Well, it was a JAES Paper, so it should have been known amongst professionals:
Not sure why there is needed evidence that it had to be an "industry wide" discussion.
I think JC was in the front rows regarding recording that time?
Hi,
First, I either have a bad case of dejavu or we had precisely the same debate before, more than once.
This is one set of opinions and ideas presented as fact. Because it is said so does not make it true in any absolute sense.
Like microphone self noise (which for recording mikes is often quite pink BTW)? Or just the noise in the room in which re record?
Let me blunt, with most (but not all) modern recording systems (Microphones, Microphone preamplifers etc) we are sorely struggling to deliver 16 Bit's worth of actual SNR to maximum peaks.
Well, given that for all intents and purposes all modern ADC's are very low Bit hybrid DS architectures, I am talking about all converters, excluding some prehistoric relics.
I am familiar with what Lipshitz, Vanderkoy et al have written on the subject. I repeat, just because they have written it is not neccesarily true.
I would suggest that they have shown that hybrid (low bit multibit) cores in ADC's and DAC's need less dither than pure single bit cores to achieve a certain highly averaged, band width limited SNR/DR, which from where I stand is something we have known for ages AND if the logic is carried to it's conclusion suggests that at a certain point dither can be completely dispensed with.
A further extension of the logic in their paper could be that in fact if single bit with dither is very bad, then low bit with dither is the "lesser evil" but true multibit with minimal, if any dither applied is the gold standard.
So, I agree with Lipshitz/Vanderkoy/et al that "the multibit converter is essentially infinitely perfectible", however, in my books this done by increasing the numbers of real bits in the core and increasing the ENOB available for a single sample, so I am odds with Messers Lipshitz & Vanderkoy, who prefer to use more dither.
Ciao T
First, I either have a bad case of dejavu or we had precisely the same debate before, more than once.
This is one set of opinions and ideas presented as fact. Because it is said so does not make it true in any absolute sense.
Like microphone self noise (which for recording mikes is often quite pink BTW)? Or just the noise in the room in which re record?
Let me blunt, with most (but not all) modern recording systems (Microphones, Microphone preamplifers etc) we are sorely struggling to deliver 16 Bit's worth of actual SNR to maximum peaks.
Well, given that for all intents and purposes all modern ADC's are very low Bit hybrid DS architectures, I am talking about all converters, excluding some prehistoric relics.
I am familiar with what Lipshitz, Vanderkoy et al have written on the subject. I repeat, just because they have written it is not neccesarily true.
I would suggest that they have shown that hybrid (low bit multibit) cores in ADC's and DAC's need less dither than pure single bit cores to achieve a certain highly averaged, band width limited SNR/DR, which from where I stand is something we have known for ages AND if the logic is carried to it's conclusion suggests that at a certain point dither can be completely dispensed with.
A further extension of the logic in their paper could be that in fact if single bit with dither is very bad, then low bit with dither is the "lesser evil" but true multibit with minimal, if any dither applied is the gold standard.
So, I agree with Lipshitz/Vanderkoy/et al that "the multibit converter is essentially infinitely perfectible", however, in my books this done by increasing the numbers of real bits in the core and increasing the ENOB available for a single sample, so I am odds with Messers Lipshitz & Vanderkoy, who prefer to use more dither.
Ciao T
I suspect a bit of false modesty.
The concepts are actually very simple and easy to illustrate.
Resolution: the ability to distinguish between two quantities. I gave a couple examples above.
Accuracy: the difference between a measured value and the true value.
Precision: the repeatability of a measurement.
Let me give an example to distinguish the last two terms, which are often confused. We have a signal whose true value is 14.100V. I have a two digit meter, which shows the value to be 14V. I have a five digit meter which shows the value to be 16.073V. The latter has great precision but poor accuracy. The former has great accuracy but poor precision. These quantities are not the same as resolution- for example, if my five digit meter shows 16.073V for a 14.100V signal, but then shows 16.074V for a 14.101V signal, it has a 0.001V resolution despite the poor accuracy. If my five digit meter stubbornly shows 16.073V for both 14.100 and 14.101, it has 0.001V of precision, but worse than that for resolution.
Now, let's make our five digit meter noisy. The needle wiggles constantly with an RMS variation of 0.5V. That's about 4% noise (29dB S/N). But... if the average needle position is 16.073V for a 14.100V signal and 16.074V for a 14.101V signal, it has a 0.001V resolution despite the poor noise. If we express the resolution in bits, then try to use the bits to characterize S/N, we'll get a wrong result- in that case bits ain't bits.
Dynamic range: the ratio between the largest possible measurement and the smallest resolvable measurement. Again, you can express this in ratios, dB, bits, but don't try equating those bits to bits of resolution or bits of noise or S/N ratio, which is the technique of High End Audio Numerology.
The Walt Kester article that Ed linked to gives an excellent overview of S/N and noise floor.
R&R analysis: Repeatability and Reproducibility
Repeatability is the variability of the measurements obtained by one person while measuring the same item repeatedly. This is also known as the inherent precision of the measurement equipment.
(Just for information: Reproducibility is the variability of the measurement system caused by differences in operator behavior.)
SY is correct regarding "precision" = "repeatability" if commonly used Quality Management terms (automotive industry eg.) are used.
I do not agree that the first instrument has worse precision than the second though.
Repeatability is the variability of the measurements obtained by one person while measuring the same item repeatedly. This is also known as the inherent precision of the measurement equipment.
(Just for information: Reproducibility is the variability of the measurement system caused by differences in operator behavior.)
SY is correct regarding "precision" = "repeatability" if commonly used Quality Management terms (automotive industry eg.) are used.
I do not agree that the first instrument has worse precision than the second though.
Yeah, R&R are distinct terms, though commonly confused. If I measure the same thing six times and get 5.000, 5.100, 5.050, 5.025, 5.030, and 5.045, I have repeatability to within 0.100. If someone else measures the same thing and gets 5.500, 5.600, 5.550, 5.525, 5.530, and 5.545, he also has 0.100 repeatability, but the reproducibility is 0.500.
1982 AES Premier Conference on Digital Audio Barry Blesser is editor, his kickoff paper on the basic aspects of digital audio does not mention dither. It is mentioned in passing later noting some of the problems with his add/subtract method. It is not mentioned anywhere else I could find. It was clearly very much in its infancy in 1980 and I don't see how any definative discussions on techniques for appying it were going on, perhaps except to discredit digital audio.
Last edited:
- Status
- Not open for further replies.