Jacco,
Yes that is what I remember, it did make you think at the time, I was used to just showing up at an airport 20 minutes before a flight and just walking to the gate to get on. Now it is a pain in the rear and you can't roam the airports. but at the same time you still don't see those guns here, just police with handguns most of the time. I'm sure they are somewhere around, but they are out of sight in that respect.
Yes that is what I remember, it did make you think at the time, I was used to just showing up at an airport 20 minutes before a flight and just walking to the gate to get on. Now it is a pain in the rear and you can't roam the airports. but at the same time you still don't see those guns here, just police with handguns most of the time. I'm sure they are somewhere around, but they are out of sight in that respect.
I recently built a preamp that features a little bit more than 200 pV/sqrt Hz
(10 pairs of ADA4898-2 averaging in the style of the LT1028 data sheet,
0.1/10Hz low 1k/100K/1MHz high, 20/40/60/80 dB,
good only for very low impedance sources)
and measured some things with a SNA-33 spectrum analyzer and/or a
Agilent vector analyzer as FFT engine...
Biggest disappointment was my HP6633B system dc power supply.
NiMHs show enormous differences. Sanyo Eneloops were really good,
no additional noise visible. Ansmanns AA 2500mAh otoh were 40 dB
above 1nV/sqrt Hz. That does not make them bad batteries for this
& that, but not so good for metrology.
I'm doing a writeup for a lot of bench power supplies, batteries, regulators
and references, but that will take some time. Summer has finally arrived
here and I had a motorbike test drive this morning, that may shift my
priorities 🙂
Gerhard
Last edited:
Ed, Dick, Demian, etc.
Very nice tutorial here on FFT errors.
http://www.bores.com/courses/advanced/windows/files/windows.pdf
Very nice tutorial here on FFT errors.
http://www.bores.com/courses/advanced/windows/files/windows.pdf
Interesting. Did you look at the dependence of the noise on amount of loading?
This one is a nice summary of the many issues which need to be carefully handled or get misleading results.
My starting assumption is that those involved in building test equipment which uses DFT or more commonly FFT take these many issues into account. With that assumption, I can go ahead and upgrade older (more affordable) commercial models. But still know the potential 'blind spots'.
BUT, I have measured a difference of up to 14dB between 2H and 3H amplitudes with commercially available test equipment. However, most of them (so far) fall within a 4dB difference/range.
The comparatively very low cost ADC are not accurate enough to measure ultra low THD levels-- though they will show something. There are some work-arounds for the non-linearity when approaching 0dBv levels. But the low end is full of digital artifacts and build issues (crosstalk, grounding). And the low levels are just not measured accurate enough below -90dBv to be certain and no inferences about a DUT should be made with these testers down there. They arent as good nor better than commercial testers.
Thx-RNMarsh
Last edited:
Almost everyone here probably already knows this but just for the record:
When measuring harmonics or harmonic distortion, by using spectrum analysis, whether the spectrum analysis is implemented with analog hardware or digital/software, there is typically a mixer very near the signal input (usually preceded by only an internal attenuator). Mixers create harmonics.
So the problem is how to know if the amplitudes of the harmonics in the output of the analyzer are only from the input signal or are also partially from the internal mixer.
One practical solution is to increase either the internal attenuation or some external attenuation of the signal (i.e. lower the signal amplitude at the mixer input) until the amplitudes of the harmonics in the analyzer output stop changing. This ensures that the mixer is not being over-driven and is not producing significant harmonics internally and thus the analyzer output's harmonic amplitudes are accurately characterizing the harmonic content of the input signal.
When measuring harmonics or harmonic distortion, by using spectrum analysis, whether the spectrum analysis is implemented with analog hardware or digital/software, there is typically a mixer very near the signal input (usually preceded by only an internal attenuator). Mixers create harmonics.
So the problem is how to know if the amplitudes of the harmonics in the output of the analyzer are only from the input signal or are also partially from the internal mixer.
One practical solution is to increase either the internal attenuation or some external attenuation of the signal (i.e. lower the signal amplitude at the mixer input) until the amplitudes of the harmonics in the analyzer output stop changing. This ensures that the mixer is not being over-driven and is not producing significant harmonics internally and thus the analyzer output's harmonic amplitudes are accurately characterizing the harmonic content of the input signal.
Last edited:
I think there is some confusion here with a classic sweep LO and down converter analog spectrum analyzer (they actually usually use double conversion and two mixers). Neither an AP nor an A/D in your sound card have a mixer.
Another good article is: Practical Considerations of ADC circuits. Located at ---> www.allaboutcircuits.com
If you look at the deviation from true - graph I extracted and put up... the Nyquist freq is 64KHz. Fine for measuring harmonics of the ac power line. But not so good for audio. Especially not for measuring harmonics of a fundamental of 10KHz.... which some like to do.
For measurement accuracy, the highest practical freq to be measured needs to be 1/5 to 1/10th the Nyquest freq.
I would imagine the same holds true for audio accuracy/listening to digital products.... thus, sampling at 192KHz would barely be a minimum for accurate reproduction. More like double that rate.
Thx-RNMarsh
If you look at the deviation from true - graph I extracted and put up... the Nyquist freq is 64KHz. Fine for measuring harmonics of the ac power line. But not so good for audio. Especially not for measuring harmonics of a fundamental of 10KHz.... which some like to do.
For measurement accuracy, the highest practical freq to be measured needs to be 1/5 to 1/10th the Nyquest freq.
I would imagine the same holds true for audio accuracy/listening to digital products.... thus, sampling at 192KHz would barely be a minimum for accurate reproduction. More like double that rate.
Thx-RNMarsh
Last edited:
"thus, sampling at 192KHz would barely be a minimum for accurate reproduction. More like double that rate". RNMarsh
I am going to just wait and see how many jump on this statement. There are so many competitive thoughts on this and I know that many have shown that even the Red Book Standards should be good enough for music reproduction.
I am going to just wait and see how many jump on this statement. There are so many competitive thoughts on this and I know that many have shown that even the Red Book Standards should be good enough for music reproduction.
No, not yet, but that clearly must be done. It was more a
reality check for the preamp; I needed it to chase a bump
in the phase noise of a precision oscillator for a space app
(@daytime job)
Also I want to measure the PSRR of the regulators. I have built
a "dirty power supply" on base of BUF634 that can inject
AF/RF from the network analyzer and feed the stimulus plus DC
to the device under test. Kind of active bias tee.
Also the load pulse response according to the Jim Williams app note.
Quite a lot of work.
Gerhard
"Wer will klares Wasser haben, der muss tiefe Brunnen graben"
(Rammstein, Rosenrot)
I think Richard means the higher bandwidth is needed to accurately capture harmonics and resonances etc created by audio frequency content, not so much to capture the audio frequency itself.
qusp,
You are more than likely correct qusp, I don't feel like reading those Nyquist papers again about the audio frequency reproduction. that is what I thought he was eluding to here.
You are more than likely correct qusp, I don't feel like reading those Nyquist papers again about the audio frequency reproduction. that is what I thought he was eluding to here.
Well, if getting to the 40th harmonic for midband is an issue, then there's an issue.
Dick's comments were possibly valid 15 years ago. We can routinely and repeatable do sub-100dBV measurements to 85kHz with sound cards these days.
Dick's comments were possibly valid 15 years ago. We can routinely and repeatable do sub-100dBV measurements to 85kHz with sound cards these days.
One of my EE professors in the late 1970s mentioned "10x or more" of the highest frequency as being desirable, for sampling rate.
One of my EE professors in the late 1970s mentioned "Now you understand
that CMOS will never be fast."
He did not know that nobody escapes the CMOS steamroller.
🙂 Gerhard
Fast sampling (i.e. much faster than the Nyquist rate) is useful in situations where you don't have an antialiasing filter before the ADC or a reconstruction filter after the DAC, but instead just rely on the source signal having little energy at higher frequencies - such as some data sampling applications. It is a mistake to extrapolate from this to digital audio, unless you really don't have an antialias filter!
When taking (or propagating) someone's advice, always ensure you know exactly under what conditions their advice is valid. Otherwise, you could create confusion in yourself and others.
When taking (or propagating) someone's advice, always ensure you know exactly under what conditions their advice is valid. Otherwise, you could create confusion in yourself and others.
DF96, I was not suggesting it was required for audio reproduction, simply my interpretation of what I figured he meant wrt audio sampling for diagnostic type operations
Don't see the graph. There is nothing to support this statement except flawed interpretations of answers to poorly posed questions. Yes, jump! One flaw might be the meaning of accurate, by ED's definition (<140dB) it can never work 🙂.
EDIT - found it, that was just an off the cuff comment with no data to support it and just plain wrong.
Last edited:
- Status
- Not open for further replies.