I have a cal need to verify an oddball PXI card (a newer, warmed over version of VXI). It's made by Applicos (from the Netherlands) model PG24192, and works alongside National Instruments PXI cards. There is no published procedure, so I'm going to have to write one.
However, here is the quandary. The harmonic distortion spec on this card is -107 DBC at 1 KHZ / 10 Vpp. The HP 8903B is the workhorse of the industry for measuring THD, and is good only to -99.99 DB. The user said he will be temporarily fine if we certify it only to -99.99DB (which all things considered, I don't believe readings will be quite that good due to inherent internal noise/distortion in the 8903B). Some of the audio analyzers by Audio Precision (such as model 2700) are good enough for the measurement. Through some research, I've found a method for measuring ultra pure audio distortion. The method involves use of an LF spec an (such as an HP 3585A). You measure carrier amplitude without the filter to establish an absolute reference. Then insert the filter and determine insertion losses at each of the harmonics you want to measure (you all can figure out that method). Then null the carrier out with the notch filter and measure the remaining harmonics, and use RSS to calculate actual THD. The person who wrote the article even used an Audio Precision (which has an extremely pure source to test the hypothesis, and he got readings right on the money.
So I am in process of spec'ing out a notch filter. I am having a heck of a time finding vendors to supply a BNC in/out stick filter. I've gone to KR FILTERS, ALLEN AVIONICS, TELONIC BERKELY, FREQDEV. Anyone know any other suppliers where I can get a 1 KHz notch (band reject) filter? OR... does anyone know schematically how I could build one? I could figure it out, but don't have the time to do all the research as to how many stages, etc I would need. The article recommends carrier suppression of -70 DB.
Thoughts or ideas?
I got bad news for you man, there is no way you are going to get down to -107 without a EMC chamber, and some real nice line power filters. Then being that this is a pxi based card you won't be able to have the computer used to control it inside the camber with you. The 8903 can only get down to -99 reading its own signal because it than carries the exact same noise signiture as the reference signal so they cancel out. Reading a UUT you will be lucky to get better than the high 80's.
Standard Noise in the air, and on power lines, will prevent you from measuring any better than 80-90 db down without some very specialized equipement. I really doubt a notch filter is going to get you there.
Hell the 10 MHz signal coming out of the bak of you GPS/cesium/rubium frequency standard is by far the cleanest signal in most labs, and most of them aren't spec'd at < 107dBc.
If you still want to try to head down that road though check out Standford reseaches site. They use the same meathod you are describing, and have schematics for high order notch filters like that.
Power everything with a unpluged UPS system to prevent line noise, and get as far away from everything as you can.
Good Luck?
I am not familiar with that unit. But there are a lot of newer instruments that digitize the signal. Once a signal is digitized the distortion is a mathematical equation (like an inverse FFT). There if that is the case there is no reason calibrate the distortion. Also, the noise floor is based on the digitizer's resolution, and again there is now reason to check the noise floor.
The theory is that once you have verified the digitizer's ability to digitize the signal, distortion accuracy and the noise floor is math. If the digitizer is able to measure min, 0 and max voltage, then the distortion should be able to make specifications.
I know, I know but this is how computers are taking over.
Mike Schwartz
http://www.callabsolutions.com (http://www.callabsolutions.com)
See if you can dig up a copy of the Agilent 339A test procedure, they use another 339A as a notch filter and a low freq spec analyzer to test THD for the oscillator in the 339 UUT, the test was written around some pretty ancient equipment but I have been using a 3585A and had good results.
Some good responses.. Thanks. Let me reply to some of it.
As for being able to get below -99 dB THD, The Audio Precision 2700 can do it. It uses balanced, shielded XLR inputs to isolate. The AP model 2722 THD+N spec is <=-112DB.
Regarding alternate methods for getting that low, According to the researcher in the article I found, introducing a good notch filter at the fundamental frequency kills 70 dB worth of the junk you have to deal with at source and load in the measurement. I believe the AP 2700 series and others like that are what is used in high end audio equipment. The HP 8903B although its a good solid instrument for general test equipment, it's not up to state of the art professional high end audio.
Thanks for the idea about the 339A. I did seem to recall that the 339A had pretty good purity specs, but I hadn't gone down that road. I'll check that out.
The big answer I was looking for, but can't get any documentation about was from CALLABSOLUTIONS, regarding the digitized waveform. I had some similar thoughts along that line: that because it is a synthesized waveform, it is algorithmic, won't periodically drift as an analogue waveform would. If I can come up with something to legitimately support that, I believe my customer would be satisfied.
Being careful not to comment about my customers applications (proprietary), they need to know it is at least -90 DB THD or better. It's spec'd at -107. I've explained to them that I can't measure that good, but that I hopefully could get a number to satisfy them (honestly) that they are better than that. If I could come up with a method where I could somehow use the notch filter in conjunction with an 8903B (manually lock on fundamental frequency (1 KHZ), then add the filter and measure whats left), that might do it for them. They're fine with me wriiting a custom procedure. I did find a notch filter from KR Filters (have to build to order, not off-the-shelf). I ordered it yesterday.
Let me know if anyone wants the link to the technical article and I'll find it. It was pretty fascinating. Only drawback is, you have to have a notch filter for each test frequency.
Have you thought of comparing a known RF carrier through a SSBPN without and with this 1 Khz signal at say, 12 Mhz ( I hate anything near an oscillator beat frequency)..
The brains fried a bit after dealing with the grandkids..but if you put a known clean signal in, then add the 1 kHz through a mixer, I do believe you should see the harmonics down to the level of your noise floor..
I will give it a try next week, as soon as the dudes are done with the lab's floor ( a painful work in progress :-D )
Have a great one..
Mike
I don't think a SSB Phase Noise measurement will cut it either. There is too much noise from the fans of the generators around 1 kHz to get anything close to the levels he's looking for. I'm almost sure this technique is a non-starter.
Heh..my solution occasionally, to eliminate fan noise, is to stick a screwdriver in the fan :-D Maybe not the most scientific method, but if necessary, I'll replace the fan...
But, seriously, you can customize the sweep increments in the FFT table to lower the resolution BW to the point where you can identify individual spurs vs harmonics..takes some effort and time, and it will lengthen the process, but it is worth it on those generators that have the "hump" between 2 kHz and 5 kHz..and it is also a valuable troubleshooting technique.
But anyways..the new system is pretty clean, and if you have two of the generators with dividers, the mixing should work..
I will try it after my fifth grandchild is born tomorrow.
Mike
I think you guys may be barking up the wrong tree on this one. Digitizers can lower the SNR on a signal using some fancy signal filtering, but this process will actually slightly raise THD. This is due to the fact that because computer based anything cannot handle true random filtering, you end up with some kind of pattern. This pattern unfortantly will always end up as some mutliple of the original. Sorry, but thats just a law of mathematics and there is no way around it.
Digital filtering is heavily used by the audio guys becuase it works when dealing with human hearing capabilities. Not when dealing with test equipment though. This is why analog generaters will always outpreform systhisizers in THD based off of a simple CW signal.
The notch filter meathod kinda works based off of the fact that it lowers the CW signal power level to a point where you can set the reference level of the 3585 down low enough to see -110 dB without the original CW level causing the mixers to go into compression. The 40-50 db loss at the CW frequency will also have minor effects to the amplitude of the harmonics as shown on the 3585. This is do the the resolution bandwidth filter settings on the 3585 interacting with the high order slopes of the notch of your filter. This can all be corrected with some basic math formulas though.
The real kicker here is that THD means different things to calibration techs than it does of those dealing with digital audio. They rely almost soley on the math of what their dacs, and filters can do, and don't loook at everything else involved in the process of getting signals from A to B. There is a DAC in that unit that can theorically do -107 DB THD. The problem is that number is only any good and the pin on that chip. You will never measure that far down in a real world enviroment. Think about it. Where are you going to find a cable with that kind of isolation to hook to this thing. The real funny part about that spec to me is the fact that this is a computer based VXI card. That is about the noisiest possible enviorment you could come up with. Go find a EMI loop. I bet the spurious noise it radiating is higher than -107 dBc.
This artical here should sum up the formulas used commonly by the audio, and dacs superfreaks to get the unrealistic numbers they like to use.
Warning make sure you bring your sientific calculator with you.
http://www.hindawi.com/GetArticle.aspx?doi=10.1155/2008/497187&e=html
edit: fixed link
hoopty
As it is Sunday evening, and I don't intend any further technical thought tonight, I'll just keep the pot stirred until work tomorrow.
I contacted the OEM (Applicos in the Netherlands). They did tell me they use the Audio Precision 2700 for their outgoing testing. So supposedly, the card can be verified with the 2700.
Regarding the notch filter method, it did include in the article that I needed to calculate the insertion loss at the various harmonics, as they are likely on the slope of the notch curve. I know how to do that. The filter I'm having built per spec the OEM sent me on my quote, is 350 HZ (3db point on each of the slopes), and suppression at Fo of at least -70 dB.