VNA number of points

What could be reason that Agilent's network analyzers will only accept such weird record sizes like 201, 401, 801, 1601, etc?

Jeroen Belleman

Reply to
Jeroen
Loading thread data ...

Am 15.04.2012 17:37, schrieb Jeroen:

When you sweep from 10 to 20 MHz, the center frequency will be 15 MHz and not "cannotmakeupmymindwhat2writeinthemiddleofthexaxis,somethingbetween14.996and15.004"

and when you switch to center/span you won't be off-grid.

:-) Gerhard

Reply to
Gerhard Hoffmann

"cannotmakeupmymindwhat2writeinthemiddleofthexaxis,somethingbetween14.996and15.004"

Mmmmh. Maybe, for a linear sweep. For a log sweep, only the first and last point are exactly on gridlines, anyway. Yet the same restriction on the number of points applies. Maybe it's just what people have gotten used to.

Now that we're at it, why do so many VNAs have a lower frequency limit of 9kHz? Another weird number that pops up in Agilent, Rhode & Schwartz and several other VNAs.

Jeroen Belleman

Reply to
Jeroen

"cannotmakeupmymindwhat2writeinthemiddleofthexaxis,somethingbetween14.996and15.004"

Even with a log scale, you have the "fence post" problem.

So that 10kHz is in the range of the instrument?

Reply to
krw

Am 15.04.2012 19:41, schrieb Jeroen:

Some important compliance measurements start at 9 KHz. The 9KHz is more a spectrum analyzer spec.

regards, Gerhard

Reply to
Gerhard Hoffmann

"cannotmakeupmymindwhat2writeinthemiddleofthexaxis,somethingbetween14.996and15.004"

OK, so if I should choose an even number of points, there will be no point spot-on in the centre of the plot. So what? Is that a reason to restrict the possible number of points to a small set of values of the form 100*2^N + 1? It still makes no sense to me.

Yes, that makes sense.

Thanks, Jeroen Belleman

Reply to
Jeroen

"cannotmakeupmymindwhat2writeinthemiddleofthexaxis,somethingbetween14.996and15.004"

It's not the center fence post that's ever the issue, rather the two ends. You must be a programmer. ;-)

Reply to
krw

"cannotmakeupmymindwhat2writeinthemiddleofthexaxis,somethingbetween14.996and15.004"

That's still no reason to restrict the possible choices of the number of points, nor does it explain why it's just *that* weird set of values!

Jeroen Belleman

Reply to
Jeroen Belleman

Normally the width of the output point array is based on the resolution of the target display device, less whatever margin is used for menu text and screen annotation. It's often bumped up to an odd number to establish a well-defined center frequency with symmetrical left and right span boundaries. In the days when hardware state machines were used to implement things like "rosenfell" video detection, this had some economic advantages.

The resolution specs for test equipment displays haven't changed that much in 30 years, and in some cases they've gotten worse. The old

8566/8568 spectrum analyzers had a nice vector display with 10-bit addressability in both X and Y, equivalent to a megapixel display, and anything that used the HP 1349A CRT monitor, like the 8753 VNAs, had a 2048x2048 vector display. But the current Agilent 8560EC portables just have a generic 640x480 VGA LCD, and most of their equipment with larger LCDs seems to be 1024x768.

Obviously with modern digital video detection and processing hardware, the choice of display record widths isn't a big deal... except that there is still some demand for code compatibility with the 8505As and

8753A-Es and 8566A/Bs of the world. That alone could influence companies to stick with the old-school record sizes. When you consider that the displays haven't gotten any sharper, there's little incentive to change.

-- john

Reply to
John Miles, KE5FX

"cannotmakeupmymindwhat2writeinthemiddleofthexaxis,somethingbetween14.996and15.004"

You want real hardware with an infinite number of choices of measurement points? I don't see that those values are in any way "weird". You worry to much.

Reply to
krw

"cannotmakeupmymindwhat2writeinthemiddleofthexaxis,somethingbetween14.996and15.004"

You

What, me? Worry? I'm just curious. That, and annoyed when it coerces the number of points I ask for to one of these weird values.

Jeroen Belleman

Reply to
Jeroen

The "weird" number of samples is chosen to greatly simplify the instrument's FFT calculations. Numerical errors (ie. systemic noise) are introduced - or can be introduced - if other values are chosen. (though no instrument I've ever used would allow them!)

I don't believe there is any compelling reason to select numbers outside the range you've stated. The tiny increase in resolution is not nearly worth the time it would take to calibrate and make such a measurement - particularly when converting to time domain. Also, for a greater number of samples, (though I've never done the math personally) it is my understanding the measurement calibration itself is not repeatable, and at any rate, is more than swamped by the numerical noise of the FFT itself (read: significant digits of the error matrix), so that you're left with an unavoidable rounding error anyway. So why bother?

Not to mention, in time domain, you can always offset the measurement reference plane anyway, if you felt you needed greater resolution or percision.

On the low end, 201 points seems to be the minimum number of calibration and measurement points needed (recommended) to make even the most basic of measurements. In between those values, the choice is simplay speed vs. resolution - all the while, trying to keep numerical noise in check

-mpm .

Reply to
mpm

I believe that FFTs work best with record lengths some power of two. The limitation I'm asking about predates the availability of FFTs on VNAs by quite a few years.

Jeroen Belleman

Reply to
Jeroen

Power of two, plus one (DC), no?

Reply to
krw

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.