I measure all the time. I can easily resolve 1 part in 1 million, up to 100KHz. What else do I need?
50-50 Right, PMA?
50-50 Right, PMA?
john curl said:
Is that when measuring with your ears or what?
One part in one million of what?
Details... ?
I can measure with a single tone to 1 part in 1 million. I usually prefer 5KHz as a test tone. I can do SMPTE IM to maybe 1 part in 100,000. I bought the test equipment in order to do this, long ago.
GRollins said:
I feel that THD should be as low as possible before feedback is applied. I feel the same way about bandwidth, wide, not low. Since it is a trivial matter to get THD figures below 0.005% at high power and frequencies, without feedback, then why not? A 100KHz bandwidth? Why not 1Meg or more? Once again, easy, and without feedback.
You are very fond if inferring what you believe other people say. It seems now that you are saying that good THD and wide bandwidth are somehow detrimental to good sound - possibly even bad for the sound.
Good specs will be the inevitable result of a well thought out and executed design. Failure to get good measured results does not prove audio gear to be sonically superior.
bear said:
That's right. His tests showed that THD had no correlation with sound quality, so the scenarios you mention above fits that pattern.
The description of how the distorted signals were created is in his second article ("Distortion Perception II"). Here it is:
What I was saying is that the GedLee metric is not specifically about the spectrum of the output of a nonlinear device with sinusoidal excitation - except in a very indirect way. It's a criterion based on the shape of the nonlinear transfer characteristic. By that, I mean the (assumed static) nonlinear relationship between input and output in figure 1 of his first paper (PDF). This function is called T(x), where x is the input and T(x) is the output. He shows several examples of nonlinear curves he used in figure 1 of his second paper (PDF). He created explicit formulas for these nonlinear characteristics and implemented them as DSP, taking as input WAV files of music, and creating WAV files of distorted music at the output.
Because he created these nonlinear characteristics mathematically in software, he knew the exact mathematical form of T(x). Now the GedLee metric is equation 5 of his first paper. To evaluate that integral, one needs the mathematical form of the nonlinear transfer characteristic T(x). He has that because he generated those curves himself. So you can see from that formula that the metric depends on the second derivative of T(x), which relates to how abruptly its slope changes as the input changes. For a characteristic that's a perfect straight line, the second derivative is zero, so the GedLee metric is also zero - a perfect device. So the metric is about T(x) and how smooth it is, not specifically the output spectrum with a sinusoidal excitation. The metric is not signal-based. It's based on T(x) only.
So how does this relate to real amplifiers? IMO, not well. First, the metric assumes the nonlinear transfer characteristic is static. But real amplifiers have distortion that depends on frequency, so they violate this requirement. Second, the nonlinear characteristic must be known. That is not really practical to do for a real amplifier. So I think this metric is mostly academic in application. If I make distorted WAV files using software, I can predict fairly well with his metric how subjectively bad they will be. But I can't see a way of applying this to real amplifiers.
PMA said:
Ahh yes, I see what you mean. Of course, with the FFT-based measurements you have discrete freq bins related to the carrier freq components and the FFT resolution. For instance, if harmonic products can occur at say 5.1223Hz and 10.2446Hz, the concept of 1Hz RBW is inappropriate.
Jan Didden
Joshua_G said:
Joshua,
for Fourier (and Laplace) transforms you need the formal education.
If you look at the wave crests of a sea/ocean, it's an irregular pattern.
The sea spectrum is a superposition of waves with different amplitudes, different frequencies, different propagation speeds and direction.
If periodic measurements of a sea spectrum are taken and placed in a matrix, the individual wave contributions can be obtained through matrix operations.
Solving the matrix of a 3-dimensional spectrum takes an endless number of operations with a regular transform.
The Fast Fourier Transform is a special type of matrix operation that splits the matrix in two parts, by solving each one individually a lot of calculations/operations can be saved.
Mechanical vibrations are also waves and 3-dimensional, FFT is a blessing there as well.
Combine high processor speeds with FFT : pushbutton results.
EE's are the lucky stiffs, their wire thingy is 2-dimensional, saves even more calculations.
Hi Joshua_G,
You are an engineer of sorts. You do the best you can without the required equipment to be sure about things and this has left you in the mindset that you don't need more to do your job. That's okay, because I understand where you are coming from. For many years I was also in a similar position. Doesn't make it valid though.
I knew I needed more, and I could accept I didn't know everything. I think that is the biggest difference between you and several others around here. Yes, you can do a job. No, you can not do that job as well as you could with the proper equipment and knowledge. Just as there is a big difference between "making something go" and servicing it to original specifications. A repair isn't fixed until it's adjusted properly as well. That is something I taught my technicians for years, and it's an important differentiation. I do respect that you work with what you have, the best you can. There is no excuse for someone who has the ability to do better, but doesn't. I have fired many technicians for that reason.
The direct way to do this in the analog domain is to synchronize and oscillator sweep to a display sweep. You output a signal and track that with a filter, capturing the amplitude information as you go. In audio, you often input one or more constant frequency tones and sweep the filter across the frequency range of interest, again capturing the amplitude information. This is how the HP 3580A and 3585A works (and others).
-Chris
You are an engineer of sorts. You do the best you can without the required equipment to be sure about things and this has left you in the mindset that you don't need more to do your job. That's okay, because I understand where you are coming from. For many years I was also in a similar position. Doesn't make it valid though.
I knew I needed more, and I could accept I didn't know everything. I think that is the biggest difference between you and several others around here. Yes, you can do a job. No, you can not do that job as well as you could with the proper equipment and knowledge. Just as there is a big difference between "making something go" and servicing it to original specifications. A repair isn't fixed until it's adjusted properly as well. That is something I taught my technicians for years, and it's an important differentiation. I do respect that you work with what you have, the best you can. There is no excuse for someone who has the ability to do better, but doesn't. I have fired many technicians for that reason.
As an engineer, you should know. Maybe not everything, but most f it. These answers are strewn throughout this site, and it's up to you to find them. It isn't up to anyone to hand you the information if you demand it.
Curious question coming from you, as an engineer especially. Are you telling me all you have are tube amps? I read on another forum that you like the sound of Sony mid fi. You would not be alone there.
Well, he is to a point. If you read back, I have publicly thanked him for the positive information he does share. Also, if you look, it's mostly John's attitude towards others that is problematic. From what I see, John is a good designer. However, he isn't the best and I don't know if there is a best. John is one of a group of good designers and treats his contemporaries with disdain and disrespect. He has his viewpoints which are valid for him I guess, but they are not for everyone. Something that John will only accept by putting those with differing opinions "below" his perceived status. I respect the guy, but I choke on the idea that he is better than everyone else here, myself not included. I'm not up to the level of these guys.
Sure, but John himself said the following ...
So, where am I off the track? The only people like this might be to indoctrinate slack-jawed, drooling idiots who have no thoughts of their own. They are much easier to handle, wouldn't you think?
Now that is a very interesting question coming from you. I guess it's best to ask how you would answer that question. Let's see, we have here people who have designed successful commercial products, designed test equipment and even designed some of the top operational amplifiers intended for audio and instrumentation. I'm afraid to ask who of these are not accomplished. As for me, I've only corrected damage from users of equipment and isolated component failures. I have also corrected engineering errors and worked with various design engineers to solve problems showing up in the field with their designs. Still, I am not the one who as designed the entire device. All of this does give me some insight on what good design practices are. What works and what doesn't for long.
You do need some formal education. This gives you the basis to build on, trains you how to learn and teaches you how to set up good experiments. Beyond that, it's self taught. I think most of us here are self taught beyond post secondary education. I find application notes very valuable. Walter Jung's work and papers are also pure gold (yes John, I've read them - and thanked the man). I have enormous respect for people like Walter Jung, and he doesn't have an attitude problem either! 😉
Just a slight correction to the other answers. There are two ways most used for transforming a time vs amplitude type display into a frequency vs amplitude type display. If you capture data points using an amplitude / time method, you can capture segment of data and perform a "Fast Fourier Transform" math procedure to generate a frequency / amplitude display. There are sometimes errors with this having to do with a finite dataset, so different types of windows can be used. A window is how the beginning and end of the data set is treated, it may involve changing amplitude values near the ends of the samples. Digital oscilloscopes use this method to generate spectrum displays, but you have to be careful of "aliasing". Google it.
The direct way to do this in the analog domain is to synchronize and oscillator sweep to a display sweep. You output a signal and track that with a filter, capturing the amplitude information as you go. In audio, you often input one or more constant frequency tones and sweep the filter across the frequency range of interest, again capturing the amplitude information. This is how the HP 3580A and 3585A works (and others).
-Chris
Hi Pavel,
That has been my point, but I have only had to argue it one way in this thread. 😉
Hi Jan,
-Chris
Completely agree with you.
That has been my point, but I have only had to argue it one way in this thread. 😉
Hi Jan,
Here is a PDF from Agilent that seems to have some info on that. Darn, these guys are good! I wish I had the money to buy these toys!!
-Chris
Geddes is applying his metric to speakers and to whole systems but especially to speakers as quality evaluation tool because he believes the usual measurements aren't indicative of subjective experience of quality.
His metric can't tell us what exactly might be wrong with an amplifier and I don't think that was his intention.
All it can tell us is that there is a qualitative problem in the system.
[EDITED]
His metric can't tell us what exactly might be wrong with an amplifier and I don't think that was his intention.
All it can tell us is that there is a qualitative problem in the system.
[EDITED]
It's not worth it, Joshua. See what I mean? Next time just Wiki the topic and get up to speed, and then they can't use it against you. It just wastes bandwidth when we let them get away with it.
Hi Chris,
we have several very good Agilents, but our probably best one is Rohde & Schwarz FSP7
Chris,
That's the nice thing with the digital analysis. Your freqs are synchronous with the FFT so you chose your freqs so that there is always exactly an integer numbers of cycles in the FFT length. So you need no windowing and you keep the max resolution.
It's a pain to calculate those freq components so most test software has a utility for that, AP has 'wavemaker' which does this for you. It also adjusts the phase and the crest factor for max dynamic range. These type of things are simply impossible to do in the analog domain.
The ISO31 freqs are also calculated along these principles and they are:
41.025
52.725
64.450
82.025
99.600
123.050
158.200
199.225
252.000
316.500
398.500
498.000
632.750
802.750
1,002.000
1,248.000
1,599.500
1,998.000
2,502.500
3,152.500
4,002.500
4,997.500
6,352.500
7,997.500
10,002.500
12,497.500
16,002.500
19,997.500
I think that is for 48kHz sample rate but I'm not sure.
I also think but am not sure that software test systems like Spectrum Plus can also do this. Anybody has any info on that?
Edit: Last AES Stanford Research debuted an audio analyzer that does this for breakfast and then some, for less than $ 7k. I know, serious money, but an AP 27xx starts at $ 30k without options. And we can dream, can't we?
Jan Didden
That's the nice thing with the digital analysis. Your freqs are synchronous with the FFT so you chose your freqs so that there is always exactly an integer numbers of cycles in the FFT length. So you need no windowing and you keep the max resolution.
It's a pain to calculate those freq components so most test software has a utility for that, AP has 'wavemaker' which does this for you. It also adjusts the phase and the crest factor for max dynamic range. These type of things are simply impossible to do in the analog domain.
The ISO31 freqs are also calculated along these principles and they are:
41.025
52.725
64.450
82.025
99.600
123.050
158.200
199.225
252.000
316.500
398.500
498.000
632.750
802.750
1,002.000
1,248.000
1,599.500
1,998.000
2,502.500
3,152.500
4,002.500
4,997.500
6,352.500
7,997.500
10,002.500
12,497.500
16,002.500
19,997.500
I think that is for 48kHz sample rate but I'm not sure.
I also think but am not sure that software test systems like Spectrum Plus can also do this. Anybody has any info on that?
Edit: Last AES Stanford Research debuted an audio analyzer that does this for breakfast and then some, for less than $ 7k. I know, serious money, but an AP 27xx starts at $ 30k without options. And we can dream, can't we?
Jan Didden
janneman said:
Yep, SpectraPlus 5.0 does up to 10 tonnes.
Jan, RBW and dynamic range are independent variables in any FFT measurement. RBW is inversely proportional with the FFT sample size (e.g. about 0.36Hz for 128k in 20-20KHz), while the dynamic range is given by the number of bytes in the conversion (e.g. 24).
Hi John,
I have a suspicion that you do design using both measurements and listening tests. You are capable of doing this and I don't think someone like you with your income at stake would allow any chance for it to go away. You are very good at throwing fuel on a fire too.
-Chris
Your attitude is not you best friend John. I have no axe to grind, but I always wonder why you treat people so badly. If you have been the victim of bad things in your life, I wonder if some was just karma for the way you can put people down.
I have a suspicion that you do design using both measurements and listening tests. You are capable of doing this and I don't think someone like you with your income at stake would allow any chance for it to go away. You are very good at throwing fuel on a fire too.
-Chris
Hi Pavel,
All I want is something to help me do my thing better, for now, the 3585A is a significant step forward. Too bad most of us can't come close to being able to afford these things.
At least you know how to use them and get full advantage, so I'm happy for you.
Hi Jan,
-Chris
Drool!
All I want is something to help me do my thing better, for now, the 3585A is a significant step forward. Too bad most of us can't come close to being able to afford these things.
At least you know how to use them and get full advantage, so I'm happy for you.
Hi Jan,
$7K is cheap. An Agilent DSO 6000 series runs easily past $10 K, and that's a 'scope. Agilent also has spectrum analyzers that start at $9 K new. That is cheap. It's all relative and I can't complain since the unit I have started life out around $30 K.
-Chris
Stanford Research has certain limitations, consider 24- to 16-Bit, 500 kS/s - 15 MS/s, 2 Ch PCI card from NI
dimitri said:
Hi Dimitri,
Did you see that SR at the AES? I wouldn't think that 500kS/s at 24 bit was a limitation for audio?
Jan Didden
- Status
- Not open for further replies.