How Distortion Free are the Distortion Measurers?

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Meaning, how sensitive are the supposedly high resolution distortion measurement tools to their environment?

As an example, Audio Precision is the name to mention for audio measurement, and are spec'd to resolve very low levels of distortion. But if we deliberately inject some rather nasty waveform distortion into the mains supply that the analyser is running off while taking a particularly subtle reading, will the figures and graphs remain rock steady? Or, introduce some unpleasant RF signals in the vicinity, same question ...

There is no mention of the robustness of the equipment against such factors on the AP website, I wonder if anyone has tried to assess this sort of behaviour in any way, at any time?

Or does everyone just assume the test equipment is perfect ...;)?

Frank
 
I have an HP 8903B, not the best, but good. It's distortion floor (measuring its own oscillator output) is about 0.0015%. If I turn on a bench power supply on the same circuit, the reading goes up to about 0.0022% or more. That supply is a good Agilent one, you wouldn't think that it would create that much noise, but it does. I don't know if it is power line noise or airborne EMI/RFI, but it is there.
 
Meaning, how sensitive are the supposedly high resolution distortion measurement tools to their environment?

As an example, Audio Precision is the name to mention for audio measurement, and are spec'd to resolve very low levels of distortion. But if we deliberately inject some rather nasty waveform distortion into the mains supply that the analyser is running off while taking a particularly subtle reading, will the figures and graphs remain rock steady? Or, introduce some unpleasant RF signals in the vicinity, same question ...

There is no mention of the robustness of the equipment against such factors on the AP website, I wonder if anyone has tried to assess this sort of behaviour in any way, at any time?

Or does everyone just assume the test equipment is perfect ...;)?

That's why you do a loop-back test (i.e. running the analyzers output directly into its input) to check the baseline of the equipment before doing any measurements.

se
 
fair point though, a well laid out design using modern techniques and modern parts is becoming very difficult to measure with anything but the best equipment. you can do as Steve suggests so you know the floor and then measure the design under conditions that stress it in a quantifiable way to produce more distortion and noise than it would under operating conditions. After this apply math to get the predicted result. measuring such things directly is becoming near impossible for even those that make the parts.
 
That's why you do a loop-back test (i.e. running the analyzers output directly into its input) to check the baseline of the equipment before doing any measurements.

se

You can and should do a baseline measurement, but it only tells you the lower limit of your measurements. For something as complex as a distortion measurement, you can't do somthing like subtract the baseline from the measurement to try to get a "better" result.

For example, if the THD reading is 0.0035% and the baseline is 0.0015%, you must not conclude that the THD of the DUT is only 0.0020%. It would be more accurate to say that the THD of DUT is 0.0035% +/-0.015%. This is an important distinction. Realize that certain harmonics of the DUT may be out of phase with the residual harmonics present in the instrument, and may partially cancel each other (they will add as vectors, not arithmetically).

Baseline measurements with other equipment like an AC voltmeter can be equally deceiving. I have one TRMS AC meter that will display about 0.00150 V (on the 3 volt scale) with the input leads shorted. This is due to noise in the RMS converter. It is tempting to zero/null this out, or to manually subtract it from the reading. That would be a big mistake. The input voltage and the offset voltage add like this:
Vreading = sqrt( Vinput2 + Vnoise2 )
It is tempting to think that when measuring say 0.10000 V, that the reading would be 0.10150 V due to the offset noise observed when taking a null/baseline measurement. But using the above you can see that the reading will be 0.10001 V. So the input noise of 1.50 mV contributes only 0.01 mV of error to that measurement.

So while you can and should do baseline measurements, you need to be careful about what you do with that information.
 
In my experience, most commercial equipment is pretty well-shielded electrostatically, but is not well-shielded electromagnetically -- that is a *lot* harder to do. So EMI is always a possible issue.

Fortunately for us, most of the magnetic noise is power-line related, and the simple expedient of high-pass filtering can alleviate a lot of junk, as long as you're not trying to see any stuff that's filtered out. That's why most distortion analyzers have a 400Hz HP filter.

High-res spectrum analysis really helps with this sorting out process, since it lets you see the signal components of interest mixed in among the offending noise.

The ongoing larger issue for me is the ability to see the self-distortion of the analyzer -- evaluating this requires a source which is much lower in distortion than the analyzer itself, and this is fairly hard to do. A good case in point is the HP 8903B mentioned by macboy. Its analyzer section can auto null to better than -110dB relative to full scale inputs, but the notch filter's self-distortion (mostly 2nd H.) limits its resolution to -100dB, or 0.001% or so, if carefully adjusted. It has a very good oscillator, but you don't know how good because of the analyzer's self-distortion.
 
In my experience, most commercial equipment is pretty well-shielded electrostatically, but is not well-shielded electromagnetically -- that is a *lot* harder to do. So EMI is always a possible issue.
That is not my experience at all.
Consumer equipment isn't shielded at all due to the use of unbalanced connections.
Pro equipment is usually shielded magnetically (twisting of the positive and negative cables) but not electronically because the shield is directly connected to the signal earth.
 
My experience matches Dick's, a spectrum analyzer is far more valuable than a the meter. I see a lot of junk on the spectrum between 40 and 50khz that is from my computer and computer monitor power supply. My hp8903b does not decriminate these extra spikes from harmonic spikes when it calcs thd + N. For a recent 10kHz measurement of a buffer with signal supplied by a KB 4402, I calculated the thd at 0.0009% but the HP 8903 said 0.013%.
)
Ken
 
Administrator
Joined 2004
Paid Member
Good point. Below is a HP 8903B output vs M-Audio soundcard output, both going into the unbalanced input of the soundcard. M-Audio is green, HP 8903B is orange. IIRC, this was 1V RMS.

The 8903B has a monitor out that is the signal after the notch filter, and it can be scaled. It's handy for seeing what's being measured as THD+n. I could post a graph of that on Monday, if anyone is interested.
289510d1340982195-balanced-soundcard-interface-power-amplifier-distortion-measurement-hp-vs-m-audio.png
 
Interesting idea, I haven't tried hooking the monitor out up to the spectrum analyzer.

What most here on the site have said is that the Thd value without the spectrum doesn't have much value. I've come to understand why. The spectrum can tell you a lot; bad power supply, weak psr, bad grounding, pour shielding, where the harmonics are... Good stuff!
 
Hi Ken -- RE hooking up the monitor output to the spectrum analyzer -- The self-distortion of the 8903, as least for the 8903E I used to own, makes it fairly useless in further investigation. I was able to see the somewhat more than 115dB of fundamental null, but as to anything else, no so much.

I also had the problem that the 8903 monitor output gain varied in some way with analyzer gain so, unlike the 339A, in which the monitor output 1VRMS full scale is always referred to the fundamental full-scale of 1VRMS, thanks to the auto set-level, the 8903's output has a hard-to-determine full-scale referent.

There probably is some work-around that I couldn't find in the short time I owned the 8903E -- maybe you can count the clacking of the various relays and figure out the final monitor gain setting from that.
 
Member
Joined 2004
Paid Member
1) analyzer sensitivity to external stuff. if properly boxed up (no covers off) most analyzers (all the ones I have tried) are pretty well shielded and don't pick up noise or hum from the transformers. That won't hold true when you connect to an external device. getting the two boxes to play well is very difficult. When making low noise measurements I shut off the lights on the bench. I don't have a computer near the bench (no room) which helps.

The Boonton 1120 I use has the monitor output, which I connect to both a scope and an FFT box. I also hacked in an input monitor that follows the input signal so I can see what is coming in. Saves lots of time when things are not right. The gain changes so you can't be sure of the levels you see, but the relative content is obvious, which is what you want to know.

Also remember the measurement is THD+N so everything in the band, including harmonics and noise is added. And the Mfr's specs are a clue as to where they stopped fighting with the low level stuff. If its spec's at -95 dB then you may find the input circuit will have a noise level at just below -100 dB. There would be no path to -110 dB in that system.

I have a super oscillator (-120 dB THD). Its a great sanity check. There are several published designs that get to the -120 dB mark. I think a battery powered fixed frequency one with a good attenuator on its output would be a great tool.
 
AX tech editor
Joined 2002
Paid Member
@ RNMarsh -- so am I, and everyone else too. I want 26 bits or more and 768kHz Fs. But right now, I'd settle for 20 bits and 1MHz Fs in a USB box so I can see the products form a 100kHz oscillator. Let us know if you find anything or design something...

Audio Precision has a box that does 24bit/1Mhz FFT.
Maybe they take your car as downpayment ;)

jan
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.