• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). Come here to have fun, be ready to be teased and not take online life too seriously. We now measure and review equipment for free! Click here for details.

Noise Floor Modulation

March Audio

Major Contributor
Manufacturer
Joined
Mar 1, 2016
Messages
4,730
Likes
5,332
Location
Albany Western Australia
#1
There has been discussion elsewhere in the forum regarding the effects of noise floor modulation in DACS. Unfortunately its got a little buried in other side nonsense, so this is a thread to discuss the issue.

To apply the descriptive terms used previously, this effect is a change in the noise floor that is correlated to the dynamic music signal in frequency and amplitude.

So, what are the audible effects of such a distortion? One poster has described the audible effect as a loss of dynamics. Otherwise there is little else to go on. As such I decided to create simple listening test which is described as follows.

This was conducted using modified versions of a 24bit, 96kHz track - Saint Seans Danse Macabre.

I used Adobe Auditions distortion effect to add a pathological level of harmonic distortion to the original file.

I then inverted and subtracted the original file from the distorted version, trying to leave only the distortion behind.

This distorted file was then adjusted in level, two versions one peaking -**dB and one peaking -**dB (overall level).

This was then mixed back into the original to create 2 files with different levels of "noise floor modulation".

This distortion is correlated to the original both in terms of level and frequency content. I am not saying this is a definitive NFM test, but I think its a good start to understand if people can actually hear anything going on.

The files are below. If people can PM me their observations rather than post them in the thread we can avoid influencing each other. I will anonymise the comments when I report back. I will tell you the levels and which file is which then.

Have fun


http://gofile.me/2vnEF/hO4u99Ykj
 
Last edited:

Opus111

Addicted to Fun and Learning
Joined
Mar 2, 2016
Messages
666
Likes
18
Location
Zhejiang
#2
Last edited:

fas42

Major Contributor
Joined
Mar 21, 2016
Messages
2,818
Likes
147
Location
Australia
#3
Thanks for doing that, BE718 - is this something new beyond the last posted link?

(hmmm ... I feel like a air traffic controller, querying the new 747 that's popped up on the screen ... )
 

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
24,395
Likes
47,203
Location
Seattle Area
#4
I like this thread to be a true investigation of this topic, not debate. BE, thanks for creating the thread and discussion topic. Posts must be constructive from here on.
 

March Audio

Major Contributor
Manufacturer
Joined
Mar 1, 2016
Messages
4,730
Likes
5,332
Location
Albany Western Australia
#6
What leads you to believe 'pathological harmonic distortion' is going to be noise-like? Or have I missed something vital here?

Acquainting yourself with the result of the work of Brockbank and Wass looks to me to be a necessary first step - you'll find they're mentioned in this paper by Belcher : https://www.researchgate.net/publication/271501230_A_new_distortion_measurement
This test has 2 purposes. It clearly is modulating the noise floor in a correlated way as you previously described as a criteria, and secondly allows us to experiment and see what sort of levels people can discern.

If you have a specific suggestion as to what test signals (music related not tones) we can use to investigate this, please describe and I can see what can be generated.
 
Last edited:

March Audio

Major Contributor
Manufacturer
Joined
Mar 1, 2016
Messages
4,730
Likes
5,332
Location
Albany Western Australia
#7
Thanks for doing that, BE718 - is this something new beyond the last posted link?

(hmmm ... I feel like a air traffic controller, querying the new 747 that's popped up on the screen ... )
no, its the same link so no need to download again.
 

Opus111

Addicted to Fun and Learning
Joined
Mar 2, 2016
Messages
666
Likes
18
Location
Zhejiang
#9
It clearly is modulating the noise floor in a correlated way as you previously described as a criteria, and secondly lets experiment and see what sort of levels people can discern.
My concern here is that Adobe Auditions might not be generating a noise-like error residual. Seeing as I'm unfamiliar with how it performs its 'distortion' function I'm guessing here. Do you have a link to where I might be able to gain more details about how it does what it does? Simply applying a distorted transfer function to a music waveform (if that's what it does) may not be sufficiently similar to what happens in reality.
 

March Audio

Major Contributor
Manufacturer
Joined
Mar 1, 2016
Messages
4,730
Likes
5,332
Location
Albany Western Australia
#10
My concern here is that Adobe Auditions might not be generating a noise-like error residual. Seeing as I'm unfamiliar with how it performs its 'distortion' function I'm guessing here. Do you have a link to where I might be able to gain more details about how it does what it does? Simply applying a distorted transfer function to a music waveform (if that's what it does) may not be sufficiently similar to what happens in reality.
You havent defined "noise" or what happens in "reality". This test is, by any standards, a relatively high level of noise floor modulation, its harmonically and level related to the original signal. Lets just see what people can hear.

As I said, tell me precisely what you want in the signal and lets see if we can simulate it.
 

Opus111

Addicted to Fun and Learning
Joined
Mar 2, 2016
Messages
666
Likes
18
Location
Zhejiang
#11
Well that's why I cited the paper I linked to - because a lot of the background is in there. No point in me just repeating what Belcher points to.

If the error power isn't noise like (key aspects - no discrete tones discernable, shaped like white noise with equal power per Hz) then it won't reflect IMD in practice as worked out by Brockbank and Wass.
 

March Audio

Major Contributor
Manufacturer
Joined
Mar 1, 2016
Messages
4,730
Likes
5,332
Location
Albany Western Australia
#12
Well that's why I cited the paper I linked to - because a lot of the background is in there. No point in me just repeating what Belcher points to.

If the error power isn't noise like (key aspects - no discrete tones discernable, shaped like white noise with equal power per Hz) then it won't reflect IMD in practice as worked out by Brockbank and Wass.
OK, so you want white noise level and spectrally related to the original signal?

That may well be possible, I have a few thoughts, sounds like a vocoder effect.
 
Last edited:

Opus111

Addicted to Fun and Learning
Joined
Mar 2, 2016
Messages
666
Likes
18
Location
Zhejiang
#13
I didn't mention anything about spectrally related to the original. And its not 'what I want' rather its based on my understanding of B&W's work in Belcher's paper. I could be wrong about how I read that paper so I'd appreciate additional pairs of eyes on it. Have you digested it?
 

March Audio

Major Contributor
Manufacturer
Joined
Mar 1, 2016
Messages
4,730
Likes
5,332
Location
Albany Western Australia
#14
I didn't mention anything about spectrally related to the original. And its not 'what I want' rather its based on my understanding of B&W's work in Belcher's paper. I could be wrong about how I read that paper so I'd appreciate additional pairs of eyes on it. Have you digested it?

No sorry I havent had the time yet. Perhaps in the meantime you can just describe precisely what would satisfy you and your understanding? This will stop a lot of messing around. We can try all sorts of tests with different parameters.

In any case here is another set of files. Similar to the first with essentially white noise (10 bands) modulated by the original music and mixed back into the original at two different levels.

http://gofile.me/2vnEF/1fLIN7AwA


As a complete aside to this I noted in the original recording a continuous 15.6 / 7 kHz tone which is very close to PAL NTSC CRT TV line frequency. Could be anything of course :)

DM LF.PNG
 
Last edited:

TBone

Major Contributor
Joined
Mar 16, 2016
Messages
1,001
Likes
160
#15
As a complete aside to this I noted in the original recording a continuous 15.6 / 7 kHz tone which is very close to PAL NTSC CRT TV line frequency. Could be anything of course
strange ... right after pre-amp mod, ripped a Traveling Wilburys Vol1 LP and it showed a similar large spike at around 11khz - on every track. I initially blamed the pre-amp ...

btw, still struggling to get a complete download. 3 partial attempts failed. Most likely my end, will try again ...
 

John Kenny

Addicted to Fun and Learning
Joined
Mar 25, 2016
Messages
568
Likes
17
#16
Hi all - most of you know me already - it appears some of you think Opus111 & I are one & the same individual - that was fun & entertaining for a while :)

Anyway, this thread attracted my participation here & I hope that it can provide both learning & fun.

So let's state up front that the audibility of noise floor modulation is an hypothesis so let's deal with this hypothesis in a spirit of true scientific investigation wherever it leads.

The Belcher paper already referenced refers back to the work of Brockbank and Wass & what Belcher seems to indicate is that in a signal of >30 individual tones, the cumulative IMD distortion will be two orders of magnitude greater than cumulative harmonic distortion "One interesting result given by Brockbank and Wass is that if the programme signal is assumed to be represented by n tones, each of equal power, and if n is greater than 30, then the contribution to the total distortion power due to harmonic products is at least two orders of magnitude less than that due to intermodulation products" I believe Opus & others interpret this to mean that given a music signal, the IMD products of each tone in the signal will also intermodulate to the extent that a "fluctuating grass" of noise is produced. The composition (spectrum, modulation frequency & amplitude) of such "fluctuating grass" is probably very complex

Remember that the Belcher paper is defining a test signal that will allow a measurement technique that when used on audio equipment provides better correlation to the subjective evaluation of the equipment. It doesn't specify a signal which if added to a music signal will be audible.

If we want to make some progress on this thread then we need to discuss what might be the composition of such a noise signal & also be prepared to be wrong.

Just as an aside to this - can I mention something recently brought to my attention - that 0.1dB amplitude difference between two audio devices will be perceived as a quality difference rather than a loudness difference. I've seen it stated before but never looked for the proof of this claim - if it turns out to be true this is a situation where a 0.1dB signal, below the accepted JND level (1dB?) for loudness, is actually perceivable with music.
 

John Kenny

Addicted to Fun and Learning
Joined
Mar 25, 2016
Messages
568
Likes
17
#18
My take on this noise floor modulation is to try to understand what could possibly be causing it to be perceivable o i will be referring to psychoacoustics & auditory processing in my replies.

For instance, the way I understand how 0.1dB amplitude difference might be audible could be explained if the test signal was NOISE. Noise being a broadband signal will span many ERB filter banks & the power in each of these filter banks is additive - hence 0.1dB spread across 5 such ERB filter banks would add to the loudness equivalent of 0.5dB

As support for this I quote JJ Johnston
Regarding that complex relationship between loudness and intensity, Johnston reviewed a few essential points:

- When the frequency content of a signal is unchanged, loudness is approximately proportional to the 1/3.5 power of the signal power, or the 1/1.75 power of amplitude.
- The ear has a mechanical filtering system that splits the signal into "critical bands" or "ERBs (Effective Rectangular Bandwidth)."
- Signals in the same critical band convert intensity to loudness with approximately the power law relationship stated above.
- Signals in different bands add linearly in loudness."​
 

John Kenny

Addicted to Fun and Learning
Joined
Mar 25, 2016
Messages
568
Likes
17
#20
Just to follow on from the ERBs, amplitude modulation filterbanks are also considered a crucial element in auditory processing. It is tuned to speech interpretation where syllabic (~5 Hz) and phonemic rates (~20 Hz) are processed by different modulation filterbank elements. I just throw that in there to give some psychoacoustic underpinning sto the idea that the audibility of noise modulation isn't a blue sky notion.
 
Top Bottom