I don't think they use windows or FFT for that matter. There system is noise reduction.
The signal is sampled at a multiple of the input frequency. They use a moving average on
multiple sample frames and then reconstruct the signal at a fixed frequency. That's about all I can figure out about them.
They all seem to share the same core FFT algorhythms
from what I understand of it.
Are windows4.pdf
and windows5.pdf the same?
Not fully understanding all of this,
Flat Top has better amplitude accuracy but only on the first three bins
on either side of the Frequency being tested?
If that is the case, then if we are also using a notch
aren't those being handled by the notch?
So they would'n't be used?
Reading further, Flat Top appears to have less sensitivies
to the harmonics....
Haven't you gents been spending a lot of time looking at
the harmonics to minimize them in the source so that you
can find them in the DUT?
If so, then doesn't it make sense to use the windowing with the
largest amplitude and the most sensitivity to harmonics?
Isn't it like a microscope? The greater the magnification
the more details you can see?
Which If I read that correctly is Hann.
Maybe there are multiple uses of the windowing.
For example,
1. without a notch, use Flat Head.
2. with a notch, use Hann.
Which reminds me,
Outside of a dog, man's best friend is a book.
Inside of a dog, it's too dark to read.
--Groucho Marx (not Karl).
The next question is how far or what is the quantity of BINs?
BIN - Buy It Now?
As the San Franciscan said, "Go ahead, make my day."
from what I understand of it.
Are windows4.pdf
and windows5.pdf the same?
Not fully understanding all of this,
Flat Top has better amplitude accuracy but only on the first three bins
on either side of the Frequency being tested?
If that is the case, then if we are also using a notch
aren't those being handled by the notch?
So they would'n't be used?
Reading further, Flat Top appears to have less sensitivies
to the harmonics....
Haven't you gents been spending a lot of time looking at
the harmonics to minimize them in the source so that you
can find them in the DUT?
If so, then doesn't it make sense to use the windowing with the
largest amplitude and the most sensitivity to harmonics?
Isn't it like a microscope? The greater the magnification
the more details you can see?
Which If I read that correctly is Hann.
Maybe there are multiple uses of the windowing.
For example,
1. without a notch, use Flat Head.
2. with a notch, use Hann.
Which reminds me,
Outside of a dog, man's best friend is a book.
Inside of a dog, it's too dark to read.
--Groucho Marx (not Karl).
The next question is how far or what is the quantity of BINs?
BIN - Buy It Now?
As the San Franciscan said, "Go ahead, make my day."
Last edited:
The Shibasoku is a form of synchronous sampling and averaging. They have a PLL that locks to the incoming signal and controls the sampling. The synchronous sample and average will reduce the random and non-synchronous noise a lot very quickly. I think they use a 16 sample averaging. The output is read back at a much slower rate and displayed at the "analysis" output. They can display the time waveform less fundamental and noise with that. Its a pretty neat trick.
The 725 uses a collection of chips to make the ADC and DAC. The 725D uses single chip solutions. The PLL is not too complex.
I just tried Flat Top and Hodie 7 on the monitor output of the Shibasoku (both) and the differences in the measured levels were in the 1 dB range with the different windows. Its late so I won't catch and post pictures tonight.
I was incorrect earlier. I had confused synchronous sampling with synchronous averaging. The latter is more powerful in removing random noise but still needs to be sync'd with the incoming signal. The Melt Noise feature in the QA400 is the latter.
The 725 uses a collection of chips to make the ADC and DAC. The 725D uses single chip solutions. The PLL is not too complex.
I just tried Flat Top and Hodie 7 on the monitor output of the Shibasoku (both) and the differences in the measured levels were in the 1 dB range with the different windows. Its late so I won't catch and post pictures tonight.
I was incorrect earlier. I had confused synchronous sampling with synchronous averaging. The latter is more powerful in removing random noise but still needs to be sync'd with the incoming signal. The Melt Noise feature in the QA400 is the latter.
APs Nice Feaure
From the AP pdf:
Once you've selected and submitted for the frequency
and you have aquired your data points
just change from Flat Top to Hann or vice versa
and it should be quick because it says it uses the same
data it already acquired so you dont' have to wait around.
Food for thought.
From the AP pdf:
Once you've selected and submitted for the frequency
and you have aquired your data points
just change from Flat Top to Hann or vice versa
and it should be quick because it says it uses the same
data it already acquired so you dont' have to wait around.
Food for thought.
Last edited:
I'll be playing with synchronous averaging once the oscillator is complete with post amplifier.
It's been on my to do list for a while.
Also been looking at a digital approach to a lock-in amplifier. The digital handles the maths.
From the AP pdf:
Once you've selected and submitted for the frequency
and you have aquired your data points
just change from Flat Top to Hann or vice versa
and it should be quick because it says it uses the same
data it already acquired so you dont' have to wait around.
Food for thought.
There you go ! This is
I'll try it with the A-P. Regarding true noise level using the A-P. A-P says you just put it into No window and the result is the true noise level.... no correction factors for the windowing needed.
Boy, they make everything seem so complicated... then give you the easy answer.
THx-RNMarsh
There you go ! This is
Regarding true noise level using the A-P. A-P says you just put it into No window and the result is the true noise level.... no correction factors for the windowing needed.
Boy, they make everything seem so complicated... then give you the easy answer.
THx-RNMarsh
Learning curve. You know how it goes.
I recall Samuel posted something on rather obscure windows that haven't seen there way into audio testing. The windows mentioned far exceed the performance of what we see used now. They are way off in another corner of physics.
I'll see if I can dig something up on it. I did save some material or links.
I'll see if I can dig something up on it. I did save some material or links.
You will nonetheless need to figure out the equivalent noise bandwidth in order to arrive at a meaningful figure.
I'm not sure why they have not yet decided to offer the option to do plots in V/rtHz... Doesn't take more than a few lines of code.
Samuel
Windowing compare
Right. One reason I got various noise levels shown was due to my changing the BW settings. But like I said, i dont want to know the noise.... but why the harmonic levels weren't closer. I think I have zero'ed in on it.
Originally, I wanted to see what was a rough estimate of the harmonic level.and was fighting EM noise intrusion. I said it showed about -146 (1KHz) using Flat-Top. When I try for more accurate testing, i get closer numbers to the Hann. But there is still a small difference with Hann (Von Hann, if one prefers). That difference has been narrowed to about 2-3dB. Hann is always lower, still. At these low signal levels and noise nearby, I can get variations from test to test. Not much but its there.
On absolute terms, there may be differences from my use of higher sampling rates and larger FFT size and/or step size etc. Not to mention the peak freq detector algorithm.
Flat-Top
Hann
This may be the most accurate method at locating/measuring the peak level:
View attachment Peak det algor.pdf
THx-RNMarsh
Right. One reason I got various noise levels shown was due to my changing the BW settings. But like I said, i dont want to know the noise.... but why the harmonic levels weren't closer. I think I have zero'ed in on it.
Originally, I wanted to see what was a rough estimate of the harmonic level.and was fighting EM noise intrusion. I said it showed about -146 (1KHz) using Flat-Top. When I try for more accurate testing, i get closer numbers to the Hann. But there is still a small difference with Hann (Von Hann, if one prefers). That difference has been narrowed to about 2-3dB. Hann is always lower, still. At these low signal levels and noise nearby, I can get variations from test to test. Not much but its there.
On absolute terms, there may be differences from my use of higher sampling rates and larger FFT size and/or step size etc. Not to mention the peak freq detector algorithm.
Flat-Top
HannThis may be the most accurate method at locating/measuring the peak level:
View attachment Peak det algor.pdf
THx-RNMarsh
Last edited:
One of the reasons I keep the Wavetek/FFT analyzer on the bench is that it has the noise analysis as a standard measurement. It agrees precisely with my Quan-Tech for noise measurement (as precisely as its possible to measure noise. . . )
I don't understand why they have not included noise analysis in many of the FFT programs. HpW-Works does include noise analysis. I find HPW-Works is a steep learning curve but I will soon need to overcome it since the program has many capabilities.
The lights are coming on.
I'm starting to take baby steps.
As I observe mine I am reminded by
her actions that she will fall, might,
stumble, cry, look around, be okay,
crawl for a time...but she will stand up again.
So I take comfort in seeing her struggle,
knowing that just as I struggle,
just as we all struggle,
we look around,
we might crawl,
but we don't quit,
we continue to stand,
we continue to struggle,
and then we don't any longer,
it is that we've accomplished something.
And then we look in another direction
where it starts all over again, we don't quit.
So reading here I take comfort in knowing
bits of knowledge are coming together in me.
It isn't the noise we want to see, it is the harmonic content.
To get there, we have to lower the noise so the harmonic
content and artifacts remain. To differentiate between the two.
Then if we remove the noise floor completely we can only
have harmonic content and their artifacts.
Of course, this only works if there is a fundimental frequency.
If only a fundimental frequency, then we'd have a hard time
of it because it would dominate the spectrum.
So we use a notch to remove it so we can study the harmonics
and the artifacts.
What will these data tell us of the DUT?
AND
It reminds me of that song in Paint Your Wagon...
"Where are we going? I don't know.
When will we get there? I ain't certain.
All's that I know is I am on my way."
I'm starting to take baby steps.
As I observe mine I am reminded by
her actions that she will fall, might,
stumble, cry, look around, be okay,
crawl for a time...but she will stand up again.
So I take comfort in seeing her struggle,
knowing that just as I struggle,
just as we all struggle,
we look around,
we might crawl,
but we don't quit,
we continue to stand,
we continue to struggle,
and then we don't any longer,
it is that we've accomplished something.
And then we look in another direction
where it starts all over again, we don't quit.
So reading here I take comfort in knowing
bits of knowledge are coming together in me.
It isn't the noise we want to see, it is the harmonic content.
To get there, we have to lower the noise so the harmonic
content and artifacts remain. To differentiate between the two.
Then if we remove the noise floor completely we can only
have harmonic content and their artifacts.
Of course, this only works if there is a fundimental frequency.
If only a fundimental frequency, then we'd have a hard time
of it because it would dominate the spectrum.
So we use a notch to remove it so we can study the harmonics
and the artifacts.
What will these data tell us of the DUT?
AND
It reminds me of that song in Paint Your Wagon...
"Where are we going? I don't know.
When will we get there? I ain't certain.
All's that I know is I am on my way."
Well that is very close and I'll take you a step further.
We are reducing the amplitude of the fundamental to reduce the distortion of our instrument.
We want to see the distortion of the device under test and not the harmonics generated by test equipment. The measured harmonics are very small and won't generate a lot of distortion in the amplifier(s) of our measuring device. The fundamental is large enough to do this so we make it smaller.
We are reducing the amplitude of the fundamental to reduce the distortion of our instrument.
We want to see the distortion of the device under test and not the harmonics generated by test equipment. The measured harmonics are very small and won't generate a lot of distortion in the amplifier(s) of our measuring device. The fundamental is large enough to do this so we make it smaller.
See Tech Note -- TN115 from Audio-Precision. This is their view on the measurement of noise. Nothing there about using v/sqrt hz. Maybe a SW window shows that number?
THx-RNMarsh
Last edited:
Уважаемый Виктор, добрый день!
Как Вы понимаете, я пытаюсь повторить Ваш замечательный генератор. До запуска (т.е. до первого включения) осталось совсем немного. Не будет ли очень нахально с моей стороны попросить Вас хотя бы коротко описать процесс или методику наладки. Приборы у меня самые обыкновенные - осциллограф (двухлучевой), спектроанализатор на базе компа, звуковой генератор - стандартный с высокими искажениями, цифровой тестер.
Очень хотелось бы самостоятельно пройти этот интересный процесс. Но понимаю, что это не просто.
Игорь.
As you know, I'm trying to retrace your wonderful generator. Prior to start-up (ie, before the first turn) there are very few. Would not it be very cheeky of me to ask you to at least briefly describe the process or technique adjustment. Devices I have the most common - Oscilloscope (dual beam), spectrum analaser -based computer, a sound generator - standard (HEATHKIT IG-18) with high distortion, digital tester.
I'd love to yourself to go through this interesting process. But I understand the difficulty.
Igor.
Как Вы понимаете, я пытаюсь повторить Ваш замечательный генератор. До запуска (т.е. до первого включения) осталось совсем немного. Не будет ли очень нахально с моей стороны попросить Вас хотя бы коротко описать процесс или методику наладки. Приборы у меня самые обыкновенные - осциллограф (двухлучевой), спектроанализатор на базе компа, звуковой генератор - стандартный с высокими искажениями, цифровой тестер.
Очень хотелось бы самостоятельно пройти этот интересный процесс. Но понимаю, что это не просто.
Игорь.
As you know, I'm trying to retrace your wonderful generator. Prior to start-up (ie, before the first turn) there are very few. Would not it be very cheeky of me to ask you to at least briefly describe the process or technique adjustment. Devices I have the most common - Oscilloscope (dual beam), spectrum analaser -based computer, a sound generator - standard (HEATHKIT IG-18) with high distortion, digital tester.
I'd love to yourself to go through this interesting process. But I understand the difficulty.
Igor.
Ну, как и предполагалось После первого включения и тщательной проверки собранной схемы на выходе нет генерации. Что-то посоветуете?
Well, as expected after the first time and a thorough check of collected circuit at the output of no generation. What will advise?
Hi all -- I happened to pick up a Fluke 510A 1kHz AC standard unit. These are designed for high stability of amplitude, and this one seems able to hold 10VRMS within 100ppm after warm up, as read by my HP 3456A. My HP 3458A is coming back to me from Keysight (nee Agilent) in Loveland, having passed its incoming inspection there completely -- no "cal" needed -- met specs, 5 years after a full cal by them. of course there has been some drift -- about 5ppm on DCV or about 1ppm per year -- gotta love those HP measurement guys.
Anyhoo -- I will check the 510A with the 3458A when it gets here. But here's the point of this post -- I wanted to use the T-T, et al on the 510A. So I decided to get out all the stuff -- the HP 339A, the Twin-T box, and the EMU 0204, and get ARTA running with them on my Win7 box. It was a PITA, but it is working.
As I was playing around, I confirmed the lower limit of the 339A oscillator at 0.0011% on its analyzer and meter, which is what it's been for some time now, then checked the 339 oscillator through a medium Z attenuator, the T-T, EMU, and ARTA. The second and third product peaks were hanging around -130dB re full output of 6.3VRMS from the 339. This was a few dB better than when last I checked. I set up the EMU for an input level of 0dBu/775mVRMS at 1kHz to be just a couple of dB below the clipping point. So I was happy with that.
Then I wondered about using the T-T between the 339 oscillator and the 339 analyzer input. The 339 will tune and work to spec with a 77.5mVRMS input signal at 1kHz -- 100mVRMS full scale on the input attenuator. I bypassed the T-T and set the ref level to 775mV/0dB, then dialed the notch in for exactly -20dB; then switched the 339 input attenuator to -20dB -- this means that the distortion and noise products were now up 20dB with respect to the fundamental. I switched in the T-T and read the distortion on the 339 at -85dB, meaning the THD+N was rerading at -105dB, or about 0.0006%.
Well why not more? Set the oscillator up for full output of a little over 6VRMS, dialed in40dB of notch, set the 339 to the 100mV attenuator position and read the THD+N as -76dB -- so, subtract the 40dB for the notch and the THD+N is -116dB or 0.00015%, which is a real improvement over the 339 all by itself. Remember, the 339 adds to the second harmonic in its filter amps, so the actual THD+N is bound to be a few dB lower -- skirting near 1ppm -- pretty good.
This will obviously work with any oscillator (or amp or preamp) that is capable of fairly high output or with analyzers that can work with smaller input levels -- the fundamental just has to be large enough for the analyzer to work correctly. And if the fundamental is really large or the analyzer can work 10dB lower than 100mVRMS FS.... Will this help the Shibasoku analyzers read David's oscillator?
Anyhoo -- I will check the 510A with the 3458A when it gets here. But here's the point of this post -- I wanted to use the T-T, et al on the 510A. So I decided to get out all the stuff -- the HP 339A, the Twin-T box, and the EMU 0204, and get ARTA running with them on my Win7 box. It was a PITA, but it is working.
As I was playing around, I confirmed the lower limit of the 339A oscillator at 0.0011% on its analyzer and meter, which is what it's been for some time now, then checked the 339 oscillator through a medium Z attenuator, the T-T, EMU, and ARTA. The second and third product peaks were hanging around -130dB re full output of 6.3VRMS from the 339. This was a few dB better than when last I checked. I set up the EMU for an input level of 0dBu/775mVRMS at 1kHz to be just a couple of dB below the clipping point. So I was happy with that.
Then I wondered about using the T-T between the 339 oscillator and the 339 analyzer input. The 339 will tune and work to spec with a 77.5mVRMS input signal at 1kHz -- 100mVRMS full scale on the input attenuator. I bypassed the T-T and set the ref level to 775mV/0dB, then dialed the notch in for exactly -20dB; then switched the 339 input attenuator to -20dB -- this means that the distortion and noise products were now up 20dB with respect to the fundamental. I switched in the T-T and read the distortion on the 339 at -85dB, meaning the THD+N was rerading at -105dB, or about 0.0006%.
Well why not more? Set the oscillator up for full output of a little over 6VRMS, dialed in40dB of notch, set the 339 to the 100mV attenuator position and read the THD+N as -76dB -- so, subtract the 40dB for the notch and the THD+N is -116dB or 0.00015%, which is a real improvement over the 339 all by itself. Remember, the 339 adds to the second harmonic in its filter amps, so the actual THD+N is bound to be a few dB lower -- skirting near 1ppm -- pretty good.
This will obviously work with any oscillator (or amp or preamp) that is capable of fairly high output or with analyzers that can work with smaller input levels -- the fundamental just has to be large enough for the analyzer to work correctly. And if the fundamental is really large or the analyzer can work 10dB lower than 100mVRMS FS.... Will this help the Shibasoku analyzers read David's oscillator?
SOTA +Plus --
What a great idea !
I own a 725D... at Demian's house now. Maybe we can get him to try it. I wonder if I can use it on my A-P 2722 also? It is also very sensitive. Plus the 339A's I have here. And, the Panasonic 7722A et al. They should all do better.... though the Panasonic needs a higher minimum level and therefore isnt as sensitive as the others.
Thx-RNMarsh
What a great idea !
I own a 725D... at Demian's house now. Maybe we can get him to try it. I wonder if I can use it on my A-P 2722 also? It is also very sensitive. Plus the 339A's I have here. And, the Panasonic 7722A et al. They should all do better.... though the Panasonic needs a higher minimum level and therefore isnt as sensitive as the others.
Thx-RNMarsh
Last edited:
