Connecting Multiple PXI-boards for faster data aquisition speed. Is it feasible?

[ZeroNegative]

I have a PXI board sampleing at 10 Ms/s.  In principle I am wondering if it is feasible to make an array of a couple of the same boards, (even if in different chassis) and create some sort of offset between sampling triggers, which if all the boards are sampling at the same rate, (say the maximum 10 Ms/s), then I am getting a sampling rate of (number of boards * 10 Ms/s). 

I know that there has to be some sort of device at the end that can pick up all the samples and so, for example, if I connected these boards/chassis together so that I had 10 cards all together, I could theoretically have 100 Ms/s, but I would need something at the end that could grab all the data.  Does anyone know anything about this or have any ideas what so-ever?

I am thinking at worst, I could have some sort of device that writes this information to a buffer, (a really large one), and then whatever it is at the end could then re-combine everything.  I however wonder if there is a better solution, have looked around but think that I just don't have the right keywords or knowledge.  So I thought that perhaps someone could give me a heads up or a keyword or something.  Or, if there even is a methodology or something that has been developed that does what I am thinking.

I am trying to create this in part because I need a lot of channels, (like 32), at 100 Ms/s.  I am trying to make the hardware scalable since say some day I need to go to 200 Ms/s, instead of 100 Ms/s.  Instead of buying a whole new Chassis/boards, perhaps I can just use the same ones that I have and create a raid like system.

Any thoughts are helpful,

thank you,

Brian

[Aaron_W.]

Brian,


Thank you for posting on NI Forums. I have a couple of questions that will better help me answer this question for you. What PXI board are you currently using to take measurements and how many of those boards do you currently have? What type of PXI-Chassis do you have? What is the signal that you will be measuring which should help answer the question of why do you need to sample at 100 MS/s. If you were going to sample the same signal across 10 different device, splitting that signal 10 ways would cause some signal deterioration, would this even be acceptable for your measurements? Also assuming this is a mixed signal what range of frequency components (bandwidth) does your signal contain. Also, how long are we wanting to acquire data for? This will better help me determine how much data we will be dealing with. You also mentioned that you would be taking this project to a final sampling rate of 200MS/s, which depending on how feasible it is to put multiple cards together in an array would mean putting 20 cards together and splitting the signal 20 different ways, which could present problems as well.

 One possible-solution (no guarantees) would be to take one of our HSDIO cards and make it run at 100 MS/s and then divide the pulses across 10 lines of the HSDIO card. This would then provide a 10 MHz clock on each of the HSDIO lines, but they would all have a slight phase shift from one another. Then with cables that are the same length give your PXI card an external clock that would be provided with the HSDIO board. Then each of your PXI cards would be sampling at a phase shift from one another. Then the data could have to be processed and put in the correct location. However, depending on the signal you would be giving all ten cards splitting that signal 10 times could deteriorate the signal. There are alot of different considerations to make with this type of system, if you could tell me a little more about your system requirements it would give me a better insight into possible solutions.

 Another solution would be to look into some of our digitizers such as the NI-5114 which would be able to sample up to 250 MS/s. The drawback as you mentioned is that you would only be able to acquire two channels on the same card and therefore would have to have 16 of these devices. I feel as though this option would be easier to implement, but once again it all depends on exactly what you want to do with the system.

These are very initial ideas on my part and I would like to talk to you more about some of them. 

[ZeroNegative]

Thank you Aaron for the in depth response.  I am currently using a PXI-6115 board which samples at 10Ms/s.  I have only 1 of these.  I have a PXI-1002 Chassis with a PXI-8174 board in it.  I am needing to sample a dc voltage.  The signal that I am trying to capture is approximatly 10ns, (nano seconds).  So I figure I will need at least 200Mhz sampling as 200Mhz/2 = 100 Mhz, (nyquist), which would give me my 10ns resolution.  I have a chip, (and am going to get another), that has 16 detectors, (all with pre-amplifiers), on it.  So I need 16*2 channels for all of them.  So each of the signals is at 10ns so I need some pretty fast hardware.  I had talked to NI sales rep and found out about the PXI-5114 board/s and some of the other boards, (5124, 5154,5105, and 2593).  I was thinking of using the PXI-5105 board, well actually 12 of them for 96 inputs.  I could use 3 inputs on the PXI for each of the 32 elements I want to get data from.  Then by sampling at 60 Ms/s, (the 5105 board samples at 60Ms/s and has 12 bit resolution), for each input, I could over all have 180 Ms/s, (or 90Ms/s after Nyquist).  I think I could do this if I had an offset sampling time of 3-4ns for each other input relative to the first triggered input, (for each group of 3 inputs for each of the 32 elements).  I know that the 200Ms/s is slightly better but I think I could use a little bit of the random sampling to get what I need.  So in the end, I would have a 12 bit resolution system at about the same price as that of the 8-bit system. 

Also, perhaps there is some sort of HSDIO card that samples really fast, (like 2 Gigs, I know that the PXI-5154 samples at 1 Gs/s), and then we could divide that up using the technique you are talking about and using LabVIEW to calculate the phase shift and use an amplifier somewhere if needed.  I like this idea of yours as I wouldn't have as many cards in the long run.  What else are you thinking?  Thanks for the help by the way,

Brian

[Keith.Shapiro]

Hi Brian,

I'm not sure I understand what it is you're trying to do.  Are you just looking to capture a short pulse and use it to trigger another device?  Are you needing to actually sample a waveform with some voltage accuracy?

In general, the reason devices are rated a certain number of samples per unit time is because they need that much time of the signal being stable to accurately sample.  I don't believe that by adding additional devices and trying to offset their samples in time that you'll be able to acheive the superposition effect that you're after.  More than likely, you'll just degrade your signal quality to where you're not able to sample the data effectively.

So, first, is the data you're trying to sample just a trigger for your system, or are you trying to characterize the pulse?

Thanks,

Keith Shapiro

National Instruments R&D

[ZeroNegative]

Keith,

           Thank you for the response.  I am trying to look at an input signal.  To try to explain what I am doing in another way:  what if I used a random sampling technique to "sample" at a higher rate?  Well the more samples that I took, the more information obtained about the original wave coming in.  So the more samples, the better representation of the original signal.  What If I had 3 cards sampling the same signal, but at random?  I would get more data points and in the end I would get a better representation of the signal.  Instead of using a random sampling technique, (well I might use it in the future), I want to take more samples of the wave.  I understand that I will not get a better resolution in the amplitude of the incoming signal, I however want to take more samples of the signal so that I might get a better resolution in the frequency domain. 

Brian 

[Keith.Shapiro]

Okay, so you're trying to characterize a signal.  In that case, HSDIO boards are probably not your best approach even though they have a fairly high sampling rate.  Digital input boards are designed to detect either a logic '1' or a logic '0', not to characterize the way a scope does.

I understand what you're getting at,  but (and especially at higher frequencies with higher edge rates) adding more devices that are sampling a signal to that signal affect more than just the amplitude of the waveform.  Adding that much capacitance to the line can affect edge rates as well as the shape of your waveform.  Have you considered the transmission line effects that might come as you try to sample a fast edge?

Do you need to sample many signals in parallel?  Or are you just needing to get a high time resolution capture for many devices but serially?  In that case, you could use a high resolution digitizer and a set of switches to mux in the different data lines you need to sample.

 Does that make sense?  Do you think that is a good fit for your application?

Keith Shapiro

National Instruments R&D

[ZeroNegative]

Keith,

         Thank you for the answer, it is very insightful, especially since I don't know that much about line capacitance....   Do you mean a high resolution digitizer such as the PXI-5154? Then use a mux to switch between the different devices?  I would still need a few of these cards, but I suppose it would be better on the line copacitance then having 12 5105's for 96 channels and then trying to integrate the signals in a way that gave me a max of 180 Ms/s over 16 channels in some crazy home made interleaving technique.  I was suggested that I do all this in parellel, but am unable to see how it would make a difference if I did this serially but really fast.  Is there a big advantage of going parellel?  Either way, perhaps there is some sort of card with many channels which is also high speed.  I suppose having the RIS availability would be good too although, could I just reproduce this effect with some clever algorithm in LabVIEW?  Forgive me if that is a really stupid question.  I am un-aware of the RIS is caused due to some special hardware or just some Random data Algorithm that is implemented.   I know some of these questions, (such as line capacitance), is standard for Engineers, but I am Math and Computer Science.  In other words, as much expounding, (such as problems like the line capacitance), as you can without exhausting yourself is much appreciated.  

Basically a PXIe-1075 18 slot chassis was recommended to me along with 8 PXI-5122 cards were recommended to me to get 16 channels.  This was not recommended by NI, so don't hesitate to tell me if there is a better solution.   I can see the PXIe-1075 chassis as eventually I am going to want 32 channels, which would mean 16 cards at just the 2 channel cards.  Maybe the Serial Idea with the fastest card you got and a mux would be a good idea...  Any more ideas are very much appreciated.

Thank you,

Brian 

[Aaron_W.]

Brian,

To make a better suggestion for what would work for your application let me ask you a few questions: 

You mentioned needing to acquire signals at a high rate, do you need to make sure to acquire all these samples at the exact same time? 

Also how long are you needing to acquire each channel for? 

It seems as you are interested in making your system expandable which is why PXI is a very good option, but as of now what is the approximate channel count and speed that you need to sample. 

The type of system that I would recommend for you will depend on these questions. If you need to sample all channels at the same time at a high rate then using a MUX is not a viable option. If you are able to sample different channels at different times, then maybe a MUX would be an option.