pci 5153 data transfering speed

[mingjie]

Hi All,

We are planning to buy a NI PCI 5153 digitizer to perform data acquisition.

In our application, we use a photodiode to recieve reflected light of a pulsed laser which has a repetition rate of 7KHz and pulse duration of 0.74ns.

We hope to use PCI 5153 to acquire data from the photodiode continuously and transfer these data to host computer for data processing in realtime with Labview.

We know that for PCI 5153, one channel can reach 1GS/s sampling realtime, so we want to know if installed on a proper desktop, can PCI 5153 transfer the data to host computer as fast as it samples?

Thanks.

Mingjie

University Glasgow

[Nathan-P]

Streaming using the NI System can be complicated due to the complexity of the system. For streaming, you have variables to consider for the module, the chassis, the controller, and the Hard Drive/RAID. Since you are asking about a PCI device, we don't need to worry about the chassis (PXI).

1. The PCI-5153 Module
The module is an 8-bit digitizer, that can simultaneously sample at 1GS/s on two channels, or 2GS/s on one channel. The card can produce up to 2GB/s of data. This is a problem since the max bandwidth of the PCI bus is less than this (discussed below). This brings up the importance of onboard memory, which acts as a circular buffer on the hardware while data is waiting to be transferred to the host computer. More memory means larger and longer acquisitions before needing to transfer data.

2. PCI Bus (http://en.wikipedia.org/wiki/Conventional_PCI)
I've included a link to the PCI Bus Wiki page. One limitation of the PCI bus is the maximum bandwidth, or amount of data it can transfer. It is theoretically limited to 133MB/s. NI PCI devices will see an actual performance around 70-90MB/s. Therefore, for NI digitizers, you would only be able to sample at 90MS/s max, in order to stream data continuously (forever). In your case, your not streaming continuously, but going to stream records.

3. Records Streaming and Calculations
For your application, you will acquire records at a rate of 7kHz. If your record length is 1024 samples, then you will be generating ~7MB/s (7kHz * 1024 samples). This is well within the streaming capabilities of the PCI bus. This also means that the 8MB/Channel option of the PCI-5153 will likely not be enough, as it would only be able to fit one record in onboard memory at a time, and might overflow while it is being transferred to the host computer while the next record is being acquired. The PCI-5153 will write data to the onboard memory at the rate it acquires data (1GB/s), but the PCI bus can only empty it at a rate (theoretically) at 133MB/s.

4. Application Considerations
Unless I'm not understanding your application correctly, the PCI-5153 will not work for you.  For your application, your laser pulse duration is 0.74ns. If you are attempting to acquire this pulse on the PCI-5153, then it is not fast enough. With a sampling at 1GS/s, this corresponds to 1ns per sample, which is more than your pulse. You might miss a pulse between samples! If you want to capture the shape of your pulse, you will need at least 10 points on the pulse, or 10x the frequency, or ~13GS/s. This would match better with the PXIe-5186. 

I hope this helps!

-Nathan

[mingjie]

Hi Nathan, 

Thanks for your  very specific information. It clarified many things for us.

There are some further information about our application.

1. Though the pulsed laser is 0.74ns duration, our photodiode has a rise time of 1-2ns. Therefore as we are using a 500MHz (5GS/s) oscilloscope now, we can see direct pulses of 2-4ns FWHM. If the light is reflected, then the signal will be broadened to even 10ns. These reflected lights are what we concerned most.

In this case, is PCI 5153's bandwidth large enough?

2. In our application, we are interested in those relevent points, is it possible for the digitizer to acquire only relevent points and dump all useless data?

In this case, considering 90M/s PCI transfer speed and 7KHz laser repetition, the maximum points for each pulse should be about 13K points, right?

3. The whole acquisition and streaming process is a) the digitizer samples the data, b) these data are written into the onboard memory, c) host computer fetch these data from the onboard memory once Labview gives a "Fetch"/"Read" order. Is it right?

Thanks

[Nathan-P]

Hi mingjie,

1. To answer your question about the relationship between required bandwidth and expected rise-time, that can be found here: https://knowledge.ni.com/KnowledgeArticleDetails?id=kA00Z000000P8qGSAS&l=en-US

2. The PCI-5153, like all or digitizers, operate similarly to oscilloscopes.  They don't transfer all the data from the ADC to the host, but take snapshots, and transfer those.  You use triggering to configure when and where the snapshots are taken. (Snapshot = Record)  Therefore, only "Relevant" data is sent to the host for your real-time processing. 

3.  The process is a little more complicated than that, but that is a good overview.  

     3a/b. How and when samples are written to the onboard memory is detailed in the firmware state diagram

     3c.  Yes, niScope Fetch or Read.vi.  In a multi-record acquisition, you will actually use niScope Fetch.vi instead.  You can see all the examples in the LabVIEW Example Finder.  (Highly Reccomended!!!!)

Again, to get a good shape on your waveforms, you will need to have more than 1GS/s, since you would have 1ns of time between each sample.  Compare this to 5GS/s, where you are getting 200 ps between each sample.  It sounds like the PXIe-5162 is comparable to the oscilloscope that you are already using:   https://www.ni.com/en-us/shop/model/pxie-5162.html

I hope this helps,

Nathan