Ringing on digitized pulse, is this due to bandwidth limitation?

[bbarnes]

I am using a PXI-5922, and I am having problems with the digitized signal.  As a test I digitized a 20uS pulse that is very clean on my oscilloscope, but the digitized pulse has ringing before and after the edges.  This can be seen in the attached file.  What is causing this?  I think it may be a result of a bandwidth limit.  If not, is there something I can do to improve this situation?

Thanks,

Bruce Barnes

[Matt_E.]

Hello Bruce,

If the rise time of your signal is 20 us then your 5922 should have enough bandwidth for your signal.  As a rule of thumb you can estimate the risetime of the digitizer as: risetime = 0.35 / Bandwidth.  Your 5922 has an analog bandwidth of about 6.5MHz which would give a risetime of about 0.054 us.  You want to have a digitizer with a risetime of less tha 1/3 to 1/5 of your signal's risetime which in this case you do.

From your screenshot it looks like you are using the 50 ohm input impedance mode so one thing you could check is that you have a good matched 50 ohm impedance throughout your cabling.  Make sure your cable is 50 ohm and not 75 ohm coax intended for video applications.

When you compared the signal on your oscilloscope were you using the 50 ohm input mode or the 1M mode?  Could you also describe your setup a little more?  Are you measuring the pulse out of a system intended to drive a 50 ohm load or are you directly connected to the output of a digital IC?

-Matt

Message Edited by Matt E. on 01-11-2008 02:07 PM

[EBL]

Hi, Bruce.

Indeed, it is a bandwidth limitation. The 5922 performs its magic by oversampling at 120 MS/s and digitally decimating the data. Inherent in that process is a digital low-pass "brick-wall" filter, which is flat to 0.4 times the output sample rate, is down -6 dB at 0.5 times the sample rate, and completely rejects any signal above 0.6 times the sample rate.

A by-product of the brick-wall filter is poor transient response. Your 20 us still has some fast edges - if you looked at your signal on a spectrum analyzer you'd see a rich set of harmonic content extending up to perhaps 100 MHz and beyond. With your sample rate set to 5 MS/s, the resulting spectrum contains a rich spectrum of power up to 2.5 MHz and suddenly nothing above that. Having lots of power in one region of the spectrum and suddenly very little power in an adjacent region is characteristic of a resonant system, and with such a digital lowpass filter, you have that in spades. The result is a resonant transient behavior, including the familiar "Gibbs Phenomenon" pre- and post-transient ringing. See, for example:

http://en.wikipedia.org/wiki/Gibbs_phenomenon

Now to the question of what can you do about it. The answer is: not a lot. The only way to eliminate the ringing is with a properly-tapered frequency response, typical of a Gaussian or Bessel filter. I suggest you increase the sample rate as much as possible (I believe the max is 15 MS/s) and apply a digital Gaussian or Bessel filter (say 4th or 5th order) to get the highest bandwidth you can while reducing the ringing until you find it acceptable. Do keep in mind that these filters cause significant roll-off in the passband (several dB), and so they are generally not desirable for frequency-domain measurements. On they other hand, because they have good transient response, they are especially suitable for time-domain measurements.

Hope this helps,
Ed L.

[Michael_S.]

EBL is right, the PXI-5922 over samples and uses digital decimation which includes a "brick wall" digital filter.  The the brick wall filter is the default setting, which is ideal for frequency domain applications where it is optimized for passband flatness and stopband rejection (brick wall).  Good news is that you can select from a few other filter settings that are more ideal for time domain applications like yours, and are optimized for parameters such as settling time.  If you are using LabVIEW you can use the LabVIEW NI-SCOPE property node with the property Flex FIR Antialias Filter Type or in C you can use the NISCOPE_ATTR_FLEX_FIR_ANTIALIAS_FILTER_TYPE property.  I would recommend using the 48 or 16 tap Hanning filter setting for your application. 

You can get more information on these settings and their specifications in the NI-SCOPE help file or specifications (http://www.ni.com/pdf/manuals/374049d.pdf  page 12).

Let us know if this helps and good luck on your project.

Michael S.

National Instruments

[EBL]

Thanks for pointing that out, Michael - I wasn't aware of the alternative filters, even though I use the 5922 almost every day. Good to know.

Bruce, the Bessel filter I would have recommended would look similar to the "8 Tap Hanning" filter shown on page 12 of the specifications.

Cheers,
Ed L.

[caz]

Since there seems to be some very knowledgeable people on this thread I figured I would ask these questions here. Why does this board oversample at 120 MS/s to get to 16 bits resolution at the max rate of 15 MS/s instead of just using 16 bit ADC at 15 MS/s? I guess I don't know what the tradeoff is between bit depth and speed and what resolution it is using for the 120 MS/s rate (I'm guessing 8 bit?). It seems to me that it could still get to 24 bits at the 500 kS/s using the 16 bit ADC at 15 Mhz. Also, does this board use a FPGA for its digital filtering and decimation and is there anyway to reprogram what exactly it is doing? This is to answer this previous post http://forums.ni.com/ni/board/message?board.id=150&thread.id=1746

[Jennifer O]

Hi Caz,

The 5922 is unique in the way it digitizes the signal.  It uses delta-sigma technology to digitize the signal which enables sampling at higher rates without sacrificing resolution.  The technology is ideal for dynamic signals due to a very low noise floor and high dynamic range.  Still there is some trade-off between sampling rate and resolution.  So instead of limiting the resolution to 16 bits by using a 16 bit ADC running at 15Ms/s, this "flex" technology returns data with the maximum resoloution possible (16-24bits) for a given sampling rate (rates vs resolution can be found in the specifications document for this digitizer).

By over sampling and filtering along with additional signal processing techniques the noise is greatly reduced.  There is a good whitepaper (http://zone.ni.com/devzone/cda/tut/p/id/3670) which illustrates some of these techniques. Unfortunately the FPGA is not exposed to the user beyond the ability to set the acquisition settings and to choose your antialias filter so I do not think it will help to provide you with a hardware-based solution for summing your signal and the R-Series devices which have a user-programmable FPGA will be too slow to meet your bandwidth requirements.

[jaced]

Message Edited by jaced on 02-05-2008 03:38 PM

[caz]

Thanks Jennifer for your response. You really cleared up my understanding of the 5922. I must ask, however, what do you see winning, more user functionality with FPGAs like in the R series boards or more software processing at higher bandwidths due to faster busses, ie PXI express?

Thanks,

caz

[Jennifer O]

I don't think it is a clear case of one or the other "winning".  There will always be a need for both - Express technology & RAID systems enable streaming at high rates for post-processing while the FPGA solutions are necessary for in-line processing where decisions must be made in pseudo real time.