For a speaker test, with stuff being carted in & out behind a curtain, the ABX boxes can't be used.
You have to prepare a random list of 'presentations' beforehand. The same technique is used for ABC
Quick statistical significance is important cos a single 'presentation' takes AT LEAST 30min, maybe up to half a day.
The person conducting the test can't be in the same room as the person tested. Each person is tested individually.
One of the most important rules is they have to listen to at least 5 pieces of music, from THEIR OWN MUSIC COLLECTION. These have to include simple instrumental, simple vocal, complex instrumental, complex instrumental & vocal, loud & soft but the genre is what THEY like to listen to.
Perhaps the most surprising result is that the Heavy Metal fans like the same speakers as the snob who designs his own mikes and records big & small orchestras. In my book, these are 'good' speakers.
There are a small number of free analyzers which might be useful for amplifier testing but I don't think TrueRTA is one of them, even in its 'expensive' form.
____________________________
The 'noise level' in these displays is a direct result of the FFT, windows etc. used.
As Marshy says, it has very little to do with the resolution of these 'instruments'
Though I have serious reservations about many of its 'features', RMAA is probably the free software most suited for amp testing.
Have a look at its large library of tests on soundcards to see the ultimate level of THD measurable with these.
_____________________
THD20k using a 96kHz soundcard would ONLY pickup the 2nd harm.
ostripper, you should take up Marshy's offer to test one of your amps.
_____________________
Mr. Marsh, thank you for highlighting, yet again, the effects of grounding, layout & decoupling.
These points can't be emphasized enough.
I think that ARTA software is very good, nor free but not expensive and it's from Croatia.
Can't use the name Richard here as now there are two of you to answer!
RNM,
So you confirmed in one instance that a speaker did sound like the reviewer said it did and you identified the problem that must have been noted. I have just read so many reviews where everything was just so convoluted by all the source components and need for special wires and all the other crazy stuff to just lose interest in reading those types of reviews.
kgrlee,
What I have noticed is that even if a speaker is not set up optimally in a room within a few seconds of hearing the speaker you know if you are going to like it or not. Perhaps it could be improved through proper setup but the overall tonality and the quality is something you tune into very quickly. I can accept a bit of thin bass or perhaps needing a bit of top end extension but the overall sound is either good or bad in a short time. I have no problem with using some tone control, I think a bit of bass boost or high end cut or boost can make a speaker sound right to me and very satisfying without worrying that it isn't perfect out of the box with no eq. To bad so many audiophiles have such a disregard for tone controls, many times they were bypass-able anyway, so why they freak out about including them I just don't understand.
RNM,
So you confirmed in one instance that a speaker did sound like the reviewer said it did and you identified the problem that must have been noted. I have just read so many reviews where everything was just so convoluted by all the source components and need for special wires and all the other crazy stuff to just lose interest in reading those types of reviews.
kgrlee,
What I have noticed is that even if a speaker is not set up optimally in a room within a few seconds of hearing the speaker you know if you are going to like it or not. Perhaps it could be improved through proper setup but the overall tonality and the quality is something you tune into very quickly. I can accept a bit of thin bass or perhaps needing a bit of top end extension but the overall sound is either good or bad in a short time. I have no problem with using some tone control, I think a bit of bass boost or high end cut or boost can make a speaker sound right to me and very satisfying without worrying that it isn't perfect out of the box with no eq. To bad so many audiophiles have such a disregard for tone controls, many times they were bypass-able anyway, so why they freak out about including them I just don't understand.
This is precisely what I look for, except I don't see this as a speaker issue - this is about system performance being adequate or not.
I always remember a high end demonstration 10 years ago or so where an LP was played - brilliant sound, perhaps the best I've heard from vinyl; then immediately after a CD which I personally use for testing was played through probably the most expensive DAC setup available at the time, same amplifier and speakers of course. And it was awful, I cringed inside from the unpleasantness of it! Were the speakers possessed of special powers of detection to change their behaviour, depending on where the source waveform originated - or was it something more ... ?
Absolutely agree with all these points.
But , a 96khz sound card is a 40$ walmart one.
This brings us back to the point of measurement.
We measure because we DON'T trust the simulator and
don't know what level of degradation our layout and grounding will have.
Edmond stuart , Syn08 , and a few other (former members) sure had it right.
That's where I learned to "cripple" a simulation to reflect the worst case scenerio.
As an example , the "blameless" is fully sub-ppm with ideal current sources , class
A and a solid mathematically constant voltage source. In fact , all the classic designs
are , as well. To use this as a guideline would be too optimistic.
If we DO know the rules of "grounding, layout , and decoupling" , our finished
product should test as good (or better) than the "crippled simulation".
Edit - Windows has no "noise" ?? noise is at the A/D or D/A (soundcard + buffers).
windows is just software !!
OS
Last edited:
ARTA is very good. My experience with it is with speakers rather than amps.
This has been the case with my true golden pinnae panel. They produce reliable result in 30min whereas the pseudo Golden Pinnae Audiophools take days only to get garbage results.Kindhornman said:
That's why Room Interface Profile as discussed in my AES paper may be the most important characteristic of a speaker.
Yes to the tone controls too. Today, we have DSP power from the cheapest PC that I only dreamt about in my early efforts at DSP. But the correct use of this power is still VERY poorly understood.
kgrlee,
I am very interest in what can be done with dsp and how to incorporate it into a self powered speaker. I don't want to hijack the thread but I did start a thread to address this. For a little while people were interested in the thread until I revealed that I wasn't trying to create a $10,000 audiophile monitor speaker, it was as if I was a heretic for wanting to do it in a real consumer speaker.
An Active loudspeaker UNIFICATION thread
It seems if you aren't trying to make things complex or using the most expensive parts there is little interest.
I am very interest in what can be done with dsp and how to incorporate it into a self powered speaker. I don't want to hijack the thread but I did start a thread to address this. For a little while people were interested in the thread until I revealed that I wasn't trying to create a $10,000 audiophile monitor speaker, it was as if I was a heretic for wanting to do it in a real consumer speaker.
An Active loudspeaker UNIFICATION thread
It seems if you aren't trying to make things complex or using the most expensive parts there is little interest.
ostripper, you need to do DSP 101 to understand where the 'noise' comes from and how FFTs deal with the 'noise'. (err.rh! Maybe DSP 102
)There's loadsa 'analysis' software written by people who really don't understand this issue and even more people who don't know how to interpret the wonky results.
But the practical action to take Marshy up on his offer.
ostripper, you need to do DSP 101 to understand where the 'noise' comes from and how FFTs deal with the 'noise'. (err.rh! Maybe DSP 102
There's loadsa 'analysis' software written by people who really don't understand this issue and even more people who don't know how to interpret the wonky results.
But the practical action to take Marshy up on his offer.
You made my point from within your own reply !
Since it is the 3rd party software that is poorly written , it is not windows.
Windows is just the kernel inteface/GUI .
Also , instead of that condescending "DSP101" comment ,
offer up some bloody links. Looked up software noise , DSP as related to software ... results were
mostly about the ability to reduce noise through DSP/software. Errata was poorly
coded drivers and 3rd party coding of the actual FFT GUI.
"Mathematical noise" ... maybe ??
So if you have some solid article or paper to back up your statement , post yer' links.
or something .....
OS
FFT's by nature are averaging; The calculus math used is intrinsically averaging. But as for the noise reduction seen on your display after an FFT run -- it is the technique of averaging random noise thru several data gatherings which causes the random noise to essentially average out to zero (or - zillion dB).
THx-RNMarsh
THx-RNMarsh
Last edited:
Without being condescending, your remark suggests you really need DSP 101. The noise in FFT stuff is really DSP 102.
The 'windows' I'm referring too are https://en.wikipedia.org/wiki/Window_function
You'll excuse me if I don't attempt to take you there today from where you are.
I recommend 'Digital Signal Processing' - Oppenheim & Schafer as a good introductory textbook.
I'm old fashion & like textbooks but web warriors might like https://ccrma.stanford.edu/~jos/
You are quite welcome to regard any software you find on the web as providing good info and to ignore anything I say.
But on a practical & useful note instead of a "mine is bigger than yours" line, I'll just repeat my advice to take up Marshy's offer.
There's nothing like 'real life' to put some context to stuff dreamt up or analysed by computers.
Last edited:
Hi Bob,
Just thought, given the title of the thread, we should say something about the book occasionally!
As you know many people including myself found the first edition a great addition to our libraries. You have received much constructive feedback and no doubt had further thoughts of your own for the second edition. So, how’s the new edition coming along?
Just thought, given the title of the thread, we should say something about the book occasionally!
As you know many people including myself found the first edition a great addition to our libraries. You have received much constructive feedback and no doubt had further thoughts of your own for the second edition. So, how’s the new edition coming along?
If properly done, I don’t think this should happen.
The noise power in each FFT frequency point is the integral of noise existing in that FFT frequency bin.
If an FFT plot is made based on different data gatherings, then it should be the average of the individuals FFTs made for each data gathering. Doing it in this way will average the noise power in each bin, which will not result in vanishingly small noise results.
Of course, if one performs several data gatherings, average them (in time domain), and then calculate the FFT, will result in what you said. But that is the wrong way to do it…
Regards
And what happens to the “noise floor” when you increase the resolution of the FFT? How is the frequency resolution of the FFT increased?
.. just flipped through my copy and though it has hints about 'noise in FFTs' it doesn't really explain it in words of 1 syllable. Maybe the topic is DSP103 and not 102. Which is probably why so many software packages don't explain the true significance of their 'noise floor' cos their authors don't know either.
But you need to do 101 & 102 first which O&S and JOS do very well.
Last edited:
When you perform an FFT the spectrum from 0 to fs is divided in N small regions - frequency bins (where N is the number of points of the FFT).
The level of each FFT point indicates the power in the bin (i.e. small region) around that frequency.
For example if you double the number of points of the FFT, the frequency bins become half the size. That's how the "frequency resolution" increases.
So the level of each FFT point now indicates the power of the signal existing on a region that just became half of what it was.
Therefore the noise integrated in each frequency bin becomes half (assuming the noise is white in that part of the spectrum).
Regards
Exactly. So for a constant sampling frequency, the way one “double
Indeed. So increasing your frequency resolution will make it “look” like you have “less” noise (“noise floor” is lower) even though you don’t.
Another question - to get decent low-frequency resolution of the FFT, how many cycles of a 20 kHz signal will be captured?