There is no problem with "colored" noise being statistically similar to white Gaussian amplitude distributed noise. There are several very good tutorials on the basics on the web. With few exceptions noise of any kind will have a continuous spectrum. White is flat, "colored" noise rises at low frequencies as some power of f usually near -1, random telegraph or generation-recombination noise will rise like a low pass R/C at some frequency depending on the source and eventually flatten or turn 1/f at some very low frequency. These are all additive.
I had the VP of Engineering for a company selling switched-capacitor-based analog processing hardware tell me that there was no excess noise in their products, because you only got that with current flowing. FAIL.
I pointed out that their products might find a market among those who were lousy at analog design, but I wasn't one of them in this case. They were also quite expensive.
I pointed out that their products might find a market among those who were lousy at analog design, but I wasn't one of them in this case. They were also quite expensive.
And if they are the ones I tried very noisy!
Bill W.
I have consulted with the higher powers. They suggest that using the low frequency content of an audio signal to dither the high frequency content is what you mean. As the content amplitude is frequency dependent this results in sum and difference frequencies. So it may be what you hear as brightness is actually an increase in second harmonic distortion.
I also ran a simple test using noise on my FFT. When the noise drops in level to fewer bits it does indeed show the high frequency loss. Image attached.
Demian,
My noise level as measured with a calibrated sound level meter is 10 dB "A" weighted re 20 uPa. As long as I hold my breath during the measurement.
It is common for a small town to be among the quietest places. In the country you have noise sources such as crickets. Just a simple thickness of drywall properly mounted will provide 20 dB of attenuation. My walls are masonry so give better than 40 dB in addition to the drywall. Ceilings and floors all have fiberglass fill. (Done by unsupervised folks resulting in it everywhere rather than just for thermal insulation.) Now the windows are small and only in three of the walls. They are mostly double glazed with some triple. Virtually all noise is coming in through the windows.
As the ambient noise is only about 45 dB "A" weighted re 20 uPa it is not that hard to keep the house quiet.
In more normal buildings the usual complaints of noise coming in is through the ceiling which may be acoustical tile to a common space above. However the worst case I ever had to deal with was a TV production studio. The machine room noise came into the studio. The window between was double glazed but with a 2" gap above and below!
I was recently on a trip and measured the airplane cabins all at around 75 dB "A" re 20 uPa. Except for one where the air nozzle didn't seal properly and was producing a high pitched whistle.
Derfy,
Just for you. A 10,000 uF radial electrolytic capacitor may have 200 nH of inductance. What is it's resonant frequency? If I have a power supply where the filtering begins to fail above 1,000 hertz, what will the remaining noise if it gets into the audio path sound like?
ES
Attachments
That just looks like you are uncovering 1/f noise of something in your chain. Maybe Bill can refer you to some appropriate reading material on dither and arithmetic gain adjustment, the basic premise here is nonsense. Dithering has nothing at all to do with modulation and creating harmonics it's an additive process.
1125 Hz. And? I asked a simple question--if it's on the output, that's a problem, but if it doesn't, or it's a bajillion (technical term) dB down from the music signal, then there's something else to worry about. If your mic preamp is picking it up, then, yes, it needs to be addressed. Never suggested otherwise. But that's a far cry from "magical differences" between various different rectifier diodes.
Also--bypass caps local to your circuit are important.
http://www.murata.com/en-us/products/emiconfun/capacitor/2013/02/14/en-20130214-p1
Also--bypass caps local to your circuit are important.
http://www.murata.com/en-us/products/emiconfun/capacitor/2013/02/14/en-20130214-p1
As usual... The claim was the low frequency energy which does exceed 10 LSB would dither the HF. It does but not the same as a proper dither signal.
The issue is not gain adjustment. The test is above the noise floor. Shorted input shows that. You can try it yourself, just be sure ranging is fixed.
Your comments are simply not about the issues at hand.
Last edited:
This has been my understanding. It's a lump sum for various different noise mechanisms (depending on the material / device) that are in excess of the component's expected Johnson noise.
Magical differences are unexplained noted real changes. One skilled in the art may indeed be able to determine the mechanism so it is not magic to them. As mentioned keeping power supply noise out of an audio system so that it is at least 3 dB and maybe even 10 dB below the lowest signal one can perceive (or 30 dB below the lowest signal level) is not quite trivial.
In my listening environment a noise reproduced at -5 dB re 20 uPa in the critical band around 3,000 hertz would be perceived. If I listened at a real symphonic concert level of 90 db re 20 uPa fast weighted "C" scale that would require 125 dB signal to noise ratio to never clip and go dead quiet. Even more if a rock concert. However I don't listen that loud and the hearing level shift from listening to the loud bits would shift the lower level limits. But the system doesn't have prior knowledge of the music so it should have that capability.
Last edited:
I simply have less faith in people's ability to hear things that far down than you do, which puts us at a bit of an impasse wrt your opening sentence. Critical band or not. Occam's razor likes the fewest assumptions, and that lends towards a minute changes making minute (real) effects.
Anything above the Johnson noise of the reviewer should be safely considered excess.
Anything above the Johnson noise of the reviewer should be safely considered excess.
Simon, I wasn't referring to added dither, just that lower level spectra when on the same track with higher level other spectra doesn't just live down near the DAC codes near zero -- it affects even the higher order bits as the signal gets pushed up there, and is already represented from the dither of the original recordings. The loss is SNR, not frequency response.
Did you write down how much you were attenuating the noise for each step? It does look to me like a relative boost in LF noise rather than a loss in HF.
I've been thinking that in reality when we listen to loud music no matter the type that our hearing becomes compressed we don't easily hear the very low level noises once we are listening to loud tracks. So to hear with a range of 125db without a period of time that is much to fast for music we just aren't going to hear really low level noise in a real environment. Not to say we don't have that hearing range, just that within a short time period I don't see that type of acuity of range over 125db or even much less.
Yes I understood that, but in effect without any low frequencies the high frequencies would be below the threshold. So the low frequency energy is acting as dither. When there is little or none of the LF energy there is nothing of the HF high enough in level to hit the LSB. That drops the HF level.
The readings were a bit more than 30 dB above the noise floor. I think the 1/F noise shows on all three traces. It is the HF that only rolls off at low levels. I had that problem on a set of measurements and it took me a while to find the issue. You are welcome to try it yourself.
I did offer the entrails of a goat or some such to a higher power for this insight. (Well a couple of emails...)
Last edited:
I though I said that. But the equipment doesn't know what level is on the source and needs to reproduce whatever is there. Now a 16 bit CD cannot have that kind of dynamic range. But higher resolution formats or "Live" can.
You are confused and/or wrong. As I said the understanding is out there for free, I've made quite an effort to connect folks with good tutorials on dither, etc. hoping it would stop some of the common misconceptions. Not my job anymore.
No, No, No, and No. You do not attenuate by dropping LSB's, you divide in FP (even cheap DSP's now use 56 or more bit accumulators) and dither properly. Resolution is lost only at the 1 LSB level of dither noise. This is easily demonstrated even at 16 bits. Bill maybe you can help out here.
And we are not talking about attenuation either.
But I will take a stab at an analogy.
If I have an 8 bit A/D that the LSB step is 1 volt. What will I see if the signal is .25 Volts peak to peak? Now if it increases to .5 volts peak to peak? Now if I add proper dither to the .5 volt signal? Then what happens if I use a proper dither with a 50% duty cycle on/off?
Last edited:
So what are we talking about? When asking experts garbage in/garbage out is always a possibility.
Yes, this started around your statement of what was going on with Bill's files. You posited that the highs vanished with attenuation because the LSB's were truncated. This concept is hopelessly wrong in the real world.
Why do I care, I'll never know. You know what Ed I don't care, if Bill or someone else wants to discuss this fine I'm done.
Yes Scott, I stated it wrong about the loss. It is not complete attenuation, it is a roll off or reduction in level. Bill posited that it had no effect. It does cause a reduction in level. As there is a dithering effect from the signal that does go full scale there are additional distortions because it is not a clean dither signal (noise or shaped.)
- Status
- Not open for further replies.