Can You Provide an Objective Defintion of "Noise" ? ...

[Detail Man]

... which (necessarily) requires that one first provide a clear, objective defintion of what is "signal".

 

[JimKasson]

... which (necessarily) requires that one first provide a clear, objective [definition] of what is "signal".

Defining noise doesn't to me require defining signal. One definition of noise is the stochastic component of the data stream. I'm not a martinet on the subject; I allow pseudo-randomness to stand in for true randomness. I think there is utility in some contexts for treating chaotic signals as random as well.

In general, I don't like the definition of noise that parallels one definition of a weed: a plant out of place. I have built systems over the years whose proper functioning depended on noise, if the above definition is accepted. Here's one:

Loading…

www.google.com

You'll note that the patent lawyer didn't like the above definition of noise. and used something like what I think you're looking for.

I recognize that there is utility in a weed-like definition, it just seems to me to be a rabbit hole.

Jim

 

[Detail Man]

... which (necessarily) requires that one first provide a clear, objective [definition] of what is "signal".

Defining noise doesn't to me require defining signal. One definition of noise is the stochastic component of the data stream. I'm not a martinet on the subject; I allow pseudo-randomness to stand in for true randomness. I think there is utility in some contexts for treating chaotic signals as random as well.

In general, I don't like the definition of noise that parallels one definition of a weed: a plant out of place. I have built systems over the years whose proper functioning depended on noise, if the above definition is accepted. Here's one:

http://www.google.com/patents/US4187466

You'll note that the patent lawyer didn't like the above definition of noise. and used something like what I think you're looking for.

I recognize that there is utility in a weed-like definition, it just seems to me to be a rabbit hole.

My understanding is that dithering can decrease quantization errors correlated with the data quantized - but at the same time necessarily increases errors that are uncorrelated with such data.

How exactly would a clear differentiation between input and output data constitute a "rabbit hole" ?

 

[JimKasson]

How exactly would a clear differentiation between input and output data constitute a "rabbit hole"

It wouldn't. I'm just not expecting one to come from this discussion, having seen similar efforts fail, even when the participants were all engineers in the same room. I think what you're looking for is only possible in tightly constrained situations.

Jim

 

[Detail Man]

How exactly would a clear differentiation between input and output data constitute a "rabbit hole" ?

It wouldn't. I'm just not expecting one to come from this discussion, having seen similar efforts fail, even when the participants were all engineers in the same room. I think what you're looking for is only possible in tightly constrained situations.

I do believe that you have made an important point, Jim. Objectively defining something such as noise necessarily requires the objective definition and specification of that which is its opposite.

 

[JimKasson]

My understanding is that dithering can decrease quantization errors correlated with the data quantized - but at the same time necessarily increases errors that are uncorrelated with such data.

Yes, but you'll notice in that patent that the post-quantization dither signal got filtered out.

Jim

 

[JimKasson]

I do believe that you have made an important point, Jim. Objectively defining something such as noise necessarily requires the objective definition and specification of that which is its opposite.

Unless you go with my suggested definition. Then some discussions that we've agonized over on this board are pretty simple. For example, is photon noise really noise? Yes.

Jim

 

[Detail Man]

My understanding is that dithering can decrease quantization errors correlated with the data quantized - but at the same time necessarily increases errors that are uncorrelated with such data.

Yes, but you'll notice in that patent that the post-quantization dither signal got filtered out.

That's nice - but the trade-off (of an increase in uncorrelated noise relative to the input-data in return for a decrease in correlated noise relative to the input-data) remains the case. All such conceptualizations necessarily requires that a clear, objective definition of the input-data be provided.

 

[Detail Man]

I do believe that you have made an important point, Jim. Objectively defining something such as noise necessarily requires the objective definition and specification of that which is its opposite.

Unless you go with my suggested definition. Then some discussions that we've agonized over on this board are pretty simple. For example, is photon noise really noise? Yes.

Have searched all occurances of the term "noise" existing with the document that you referenced here, and I do not findy any particular definition of "noise" - only usages of the term within the text.

Would you mind providing the definition that you speak of in words within a post on this thread ?

 

[JimKasson]

Have searched all occurances of the term "noise" existing with the document that you referenced here, and I do not findy any particular definition of "noise" - only usages of the term within the text.

Yes, as I said, we went with the lawyer's idea of noise. And, you're right, it's not defined.

Would you mind providing the definition that you speak of in words within a post on this thread ?

The stochastic component of the data stream.

Jim

 

[Detail Man]

Have searched all occurances of the term "noise" existing with the document that you referenced here, and I do not findy any particular definition of "noise" - only usages of the term within the text.

Yes, as I said, we went with the lawyer's idea of noise. And, you're right, it's not defined.

Would you mind providing the definition that you speak of in words within a post on this thread ?

The stochastic component of the data stream.

Use of the term "stochastic" completely ignores all periodic components. Following that line of thinking, it would appear that any/all components of the output-data (that are unrelated to the input-data) which are periodic in nature do not constitute "noise". Are they "signal", then ? ...

 

[JimKasson]

Use of the term "stochastic" completely ignores all periodic components. Following that line of thinking, it would appear that any/all components of the output-data (that are unrelated to the input-data) which are periodic in nature do not constitute "noise".

Right, if purely periodic.

Are they "signal", then?

I'm not taking a position on that. I don't think there's a universal answer to that question. (Perhaps I haven't been making myself clear here. If that's the case, I apologize.)

Jim

 

[Detail Man]

Use of the term "stochastic" completely ignores all periodic components. Following that line of thinking, it would appear that any/all components of the output-data (that are unrelated to the input-data) which are periodic in nature do not constitute "noise".

Right, if purely periodic.

Your statement appears to exclude what are "almost periodic functions". Are they "signal", then ?

Are they "signal", then?

I'm not taking a position on that. I don't think there's a universal answer to that question. (Perhaps I haven't been making myself clear here. If that's the case, I apologize.)

No need to apoligize when sincere individuals attempt to exchange thoughts and ideas, my friend !

[If] it is your position that "signal" cannot be objectively defined, [then] how could or would any thinker(s) go about endeavoring to attempt to objectively define what the term "noise" means ?

 

[JimKasson]

Use of the term "stochastic" completely ignores all periodic components. Following that line of thinking, it would appear that any/all components of the output-data (that are unrelated to the input-data) which are periodic in nature do not constitute "noise".

Right, if purely periodic.

Your statement appears to clearly exclude "almost periodic functions".

If there are stochastic elements, those are excluded.

Are they "signal", then ?

Again, I'm not going there.

Are they "signal", then?

I'm not taking a position on that. I don't think there's a universal answer to that question. (Perhaps I haven't been making myself clear here. If that's the case, I apologize.)

No need to apoligize when sincere individuals attempt to exchange ideas and opinions, my friend !

(If) it is your position that "signal" cannot be clearly, objectively defined, (then) how could/would any thinker(s) endeavor to attempt to clearly, objectively define what the term "noise" means ?

I just did so, without defining "signal". Maybe you don't like the definition. Maybe I don't either, in some contexts.

But let me give you an example in everyday, non-engineering, usage. I once moved my engineering department into temporary facilities while a new building was being constructed. The temp space was open plan, and everyone had had private offices up to then. I got complaints about people having difficulty concentrating. I dealt with that by having sound generators installed to mask low-level distant conversations. Everybody, including me, called the sound producers "white noise generators", even though, from another perspective, their output was the desired "signal".

Jim

 

[Detail Man]

Use of the term "stochastic" completely ignores all periodic components. Following that line of thinking, it would appear that any/all components of the output-data (that are unrelated to the input-data) which are periodic in nature do not constitute "noise".

Right, if purely periodic.

Your statement appears to clearly exclude "almost periodic functions".

If there are stochastic elements, those are excluded.

Are they "signal", then ?

Again, I'm not going there.

Are they "signal", then?

I'm not taking a position on that. I don't think there's a universal answer to that question. (Perhaps I haven't been making myself clear here. If that's the case, I apologize.)

No need to apoligize when sincere individuals attempt to exchange ideas and opinions, my friend !

(If) it is your position that "signal" cannot be clearly, objectively defined, (then) how could/would any thinker(s) endeavor to attempt to clearly, objectively define what the term "noise" means ?

I just did so, without defining "signal". Maybe you don't like the definition. Maybe I don't either, in some contexts.

.

But let me give you an example in everyday, non-engineering, usage. I once moved my engineering department into temporary facilities while a new building was being constructed. The temp space was open plan, and everyone had had private offices up to then. I got complaints about people having difficulty concentrating. I dealt with that by having sound generators installed to mask low-level distant conversations. Everybody, including me, called the sound producers "white noise generators", even though, from another perspective, their output was the desired "signal".

Your point [which, to me, implies that clearly described elements of all related context(s) must also be included within any such "knowledge claims"] seems to indicate that your answer to the question posed in the title of the original post in this thread (appears to me) to be "no" ? Please advise ...

 

[JimKasson]

Your point [which, to me, implies that clearly described elements of all related context(s) must also be included within any such "knowledge claims"] seems to indicate that your answer to the question posed in the title of the original post in this thread (appears to me to be) "no" ? Please advise.

You're right; at least not in a broad context. But my definition leaves us somewhere useful, if not in a perfect place.

Consider:

Photon noise is clearly noise.

Most read noise is noise; there is a non-stochastic portion that is often called "pattern noise". We could call it "pattern error".

PRNU is tricky, as it is anyhow. It's not noise pixel-by-pixel, since it doesn't change much image to image. However, if you look at the entire frame, it looks stochastic. I'm OK with calling it noise.

Quantizing error statistics depends on the input signal, but with 14-bit ADCs, there's enough real noise in the system that the quantizing error is itself stochastic. So let's call it noise.

By my definition, the following unwanted ("weedy", in the garden analogy) effects are not noise:

• Flare.
• Motion blur.
• Diffraction.
• Lens aberrations.
• Pixel-level crosstalk.
• Lens falloff.

You get the idea, I hope.

Jim

 

[Detail Man]

Your point [which, to me, implies that clearly described elements of all related context(s) must also be included within any such "knowledge claims"] seems to indicate that your answer to the question posed in the title of the original post in this thread (appears to me to be) "no" ? Please advise.

You're right; at least not in a broad context. But my definition leaves us somewhere useful, if not in a perfect place.

Consider:

Photon noise is clearly noise.

Most read noise is noise; there is a non-stochastic portion that is often called "pattern noise". We could call it "pattern error".

PRNU is tricky, as it is anyhow. It's not noise pixel-by-pixel, since it doesn't change much image to image. However, if you look at the entire frame, it looks stochastic. I'm OK with calling it noise.

Quantizing error statistics depends on the input signal, but with 14-bit ADCs, there's enough real noise in the system that the quantizing error is itself stochastic. So let's call it noise.

Above we have selected components of the output-data interpreted by you to be "undesirable".

By my definition, the following unwanted ("weedy", in the garden analogy) effects are not noise:

• Flare.
• Motion blur.
• Diffraction.
• Lens aberrations.
• Pixel-level crosstalk.
• Lens falloff.

Here we have selected components of the output-data interpreted by you to be "undesirable" - yet considered by you to (in some unrecognized by me way) to fall into a different "category". Why so ?

You get the idea, I hope.

See my inquiry appearing directly above.

 

[JimKasson]

By my definition, the following unwanted ("weedy", in the garden analogy) effects are not noise:

• Flare.
• Motion blur.
• Diffraction.
• Lens aberrations.
• Pixel-level crosstalk.
• Lens falloff.

Here we have selected components of the output-data interpreted by you to be "undesirable" - yet considered by you to (in some unrecognized by me way) to fall into a different "category". Why so ?

Not stochastic.

Jim

 

[Detail Man]

By my definition, the following unwanted ("weedy", in the garden analogy) effects are not noise:

• Flare.
• Motion blur.
• Diffraction.
• Lens aberrations.
• Pixel-level crosstalk.
• Lens falloff.

Here we have selected components of the output-data interpreted by you to be "undesirable" - yet considered by you to (in some unrecognized by me way) to fall into a different "category". Why so ?

Not stochastic.

Your position is made clear. My only quibble with your list above is that (camer/subject) "motion blur" seems to me to be "close enough" to being random (in plenty of actualizations of photography).

DM

 

[mostlyboringphotog]

Nature is nature and all the measurements are to some degree an approximation within some range. And as a definition "noise" is part that measurement which we do not want (nature did not say this is noise) to be included.

BTW, what is your intent? That is, what do you expect to do with the "Objective Definition of Noise"?