High-Speed Digitizers

cancel
Showing results for 
Search instead for 
Did you mean: 

High sampling rate data stream to PC

Solved!
Go to solution

I have PXIe-5160 PXI Oscilloscope, 500 MHz, 10 bits, 2.5 GS/s, 2 Channels, 64 MB digitizer. If I want to continuously acquire data at full sampling rate for just a single channel i.e. 2.5GS/sec, what steps should I take care and be aware of to handle this high speed ? My understanding is:

  1. 2.5 Gig Samples/sec with each sample of 10 bits resolution = (2.5*10^9*10)/8 = 3.125 GB of data stream. What kind of PC should I use to handle such a great volume of data streaming? I intend to collect data continuously for 10 secs - 30 secs.
  2. Where does all the data go once it reaches the PCIe card that connects to the chassis containing the  5160? Is it to the RAM or directly to the Hardware?

I am trying to understand what kind of system requirements I should be looking at to handle this scenario.

Any advise is appreciated. Thank you.

NI System Configuration:
- NI PXIe-1071, 4-Slot 3U PXI Express Chassis , 1 GB/Slot throughput, Part Number: 781368-01
- NI PXIe-PCIe8381,x8 Gen2 MXI-Express for PXI Express Interface,3m, Part Number: 782522-01
- PXIe-5160 PXI Oscilloscope, 500 MHz, 10 bits, 2.5 GS/s, 2 Channels, 64 MB, Part Number: 782621-01
- Astronics PXIe-1209 2-Channel, 100 MHz PXI Pulse Generator, Part Number: 785033-01
0 Kudos
Message 1 of 8
(5,250 Views)

Bad news: You can usually read only in 8/16/(32) bit .. so unless you discard 2 bit resolution you look for a 5GByte/s data streaming solution.

 

There are some data streaming papers here, but I don't know the actual numbers, however 80Gbit/s (Gbd) seems a quite high number.  According to

https://de.wikipedia.org/wiki/PCI_Express

even a PCIe 5.0 with 16 lanes is 'just' about ~63 Gbd

Assuming that the memory is fast enough .. the hard drive seems to be the bottleneck.

(or have a ?Tbit? network connection?...  CERN has solutions for this ;))

 

That's why you often have a FPGA nearby for data reduction 😉

 

Greetings from Germany
Henrik

LV since v3.1

“ground” is a convenient fantasy

'˙˙˙˙uıɐƃɐ lɐıp puɐ °06 ǝuoɥd ɹnoʎ uɹnʇ ǝsɐǝld 'ʎɹɐuıƃɐɯı sı pǝlɐıp ǝʌɐɥ noʎ ɹǝqɯnu ǝɥʇ'


0 Kudos
Message 2 of 8
(5,214 Views)
  1. Why can I read just powers of 2 resolution bits and not what is claimed in the spec-sheet of the digitizer?
  2. If one cannot use the entire sampling rate then selling such high sample rate digitizers seems to be off isn't it? What kind of applications usually make use of these high speed digitizers ? and do they look for continuous sampling like my original question ?
NI System Configuration:
- NI PXIe-1071, 4-Slot 3U PXI Express Chassis , 1 GB/Slot throughput, Part Number: 781368-01
- NI PXIe-PCIe8381,x8 Gen2 MXI-Express for PXI Express Interface,3m, Part Number: 782522-01
- PXIe-5160 PXI Oscilloscope, 500 MHz, 10 bits, 2.5 GS/s, 2 Channels, 64 MB, Part Number: 782621-01
- Astronics PXIe-1209 2-Channel, 100 MHz PXI Pulse Generator, Part Number: 785033-01
0 Kudos
Message 3 of 8
(5,208 Views)

wrote:
  1. Why can I read just powers of 2 resolution bits and not what is claimed in the spec-sheet of the digitizer?

 Don't mix up ADC resolution and memory management. Historically (memory) bit's are organised in chucks of 8/16/32bit. You don't want to spend a lot of slowing down programming .. even a boolean value (1bit) is stored as a 32bit value in memory in modern systems. If you read your spec and manual you migth find that also the sample buffer is organised in smallest chucks of 4 Byte, just because modern architecture can't read and write smaller chucks.

 


wrote:
  1. If one cannot use the entire sampling rate then selling such high sample rate digitizers seems to be off isn't it? What kind of applications usually make use of these high speed digitizers ? and do they look for continuous sampling like my original question ?

 A lot of applications/events that need such high samplerates (high frequency) doesn't last that long. 😄  (Pulses... LIDAR/RADAR ... )

If you want to display something.. even with modern screens you need 4k samples... want to zoom: even at 100000% (x1000) it's still 'only' 4M points. 

If you want to measure something in such high resolution over a longer periode one would use clever triggers, multiple buffer, fast preprocessing.

 

May I ask what you want to measure? What do you want to do with 64MSample .. XX Tbyte of data?

 

I 'only' have some 200MSPS  cards, but they are fine for me to capture heterodyne laser signals  (usual capture times 2ms to 200ms)  

 

Wait another 5-10 years and  you can have such bandwidths and storage (maybe not in you phone , but close :D) .Current systems with such a performance exists, but they are pricy.

Greetings from Germany
Henrik

LV since v3.1

“ground” is a convenient fantasy

'˙˙˙˙uıɐƃɐ lɐıp puɐ °06 ǝuoɥd ɹnoʎ uɹnʇ ǝsɐǝld 'ʎɹɐuıƃɐɯı sı pǝlɐıp ǝʌɐɥ noʎ ɹǝqɯnu ǝɥʇ'


0 Kudos
Message 4 of 8
(5,201 Views)

Don't mix up ADC resolution and memory management. Historically (memory) bit's are organised in chucks of 8/16/32bit. You don't want to spend a lot of slowing down programming .. even a boolean value (1bit) is stored as a 32bit value in memory in modern systems. If you read your spec and manual you migth find that also the sample buffer is organised in smallest chucks of 4 Byte, just because modern architecture can't read and write smaller chucks.

 

A lot of applications/events that need such high samplerates (high frequency) doesn't last that long. 😄  (Pulses... LIDAR/RADAR ... )

If you want to display something.. even with modern screens you need 4k samples... want to zoom: even at 100000% (x1000) it's still 'only' 4M points. 

If you want to measure something in such high resolution over a longer periode one would use clever triggers, multiple buffer, fast preprocessing.

 

I would be interested to learn about this techniques of clever triggers, buffers and fast pre-processings, please suggest some sources that might direct me towards it.

 

May I ask what you want to measure? What do you want to do with 64MSample .. XX Tbyte of data?

 

Classic time-correlated single photon counting (TCSPC). So when you are dealing with very low photon counting applications you need to look at long capture time ( in seconds if possible). And having a high sampling rate is beneficial. Especially when you are trying to use it as a part of fluorescence applications.

 


 

NI System Configuration:
- NI PXIe-1071, 4-Slot 3U PXI Express Chassis , 1 GB/Slot throughput, Part Number: 781368-01
- NI PXIe-PCIe8381,x8 Gen2 MXI-Express for PXI Express Interface,3m, Part Number: 782522-01
- PXIe-5160 PXI Oscilloscope, 500 MHz, 10 bits, 2.5 GS/s, 2 Channels, 64 MB, Part Number: 782621-01
- Astronics PXIe-1209 2-Channel, 100 MHz PXI Pulse Generator, Part Number: 785033-01
0 Kudos
Message 5 of 8
(5,195 Views)

If you have single events, and I assume you want the energy spektra, configure your digitizer for a multiple record aquisition. some points pretrigger.. record length according to your max pulse length.

like that, you only record the pulses including an absolute timestamp. 😄  

I don't know if your card has this feature , but chances are high ... RTFM helps and reading papers on that topic and asking others that have done it before.

 

SMC-Based Digitizers Multiple-Record Acquisition.

Copy from my 2012 NI-Scope help

SMC-Based Digitizers Multiple-Record Acquisition

SMC-based digitizers support multiple-record acquisition, which allows the capture of multiple triggered waveforms without software intervention. In this mode, the digitizer automatically begins a new acquisition in a new memory record immediately after finishing the previous one. Multiple-record acquisitions can quickly acquire numerous triggered waveforms because they allow hardware rearming of the digitizer. Between each record, however, there is a dead time during which no triggers are accepted. During this time, the memory controller sets up for the next record. There is also a holdoff between the last trigger in a record and the trigger of a new record. This means that the actual time between triggers is the greater of the between-record dead time plus pretrigger sample time and the holdoff time. There may also be additional dead time while the minimum number of pretrigger samples are being acquired. To increase the minimum time between triggers, use the trigger holdoff feature.

The number of records that can be acquired varies depending on the memory option of the device. NI-SCOPE currently has a limit of 100,000 records that can be acquired without fetching. However, if the records are fetched while they are being acquired, more records can be acquired. Refer to Acquiring More Records Than Fit in Digitizer Memory for more information.

Greetings from Germany
Henrik

LV since v3.1

“ground” is a convenient fantasy

'˙˙˙˙uıɐƃɐ lɐıp puɐ °06 ǝuoɥd ɹnoʎ uɹnʇ ǝsɐǝld 'ʎɹɐuıƃɐɯı sı pǝlɐıp ǝʌɐɥ noʎ ɹǝqɯnu ǝɥʇ'


0 Kudos
Message 6 of 8
(5,190 Views)
Solution
Accepted by topic author asukumari

I had a look into the 5160 spec. If you use multiple record acquisition , you have a dead time of about 720ns,  so as long as your single photons frequency is less than ~1MHz  (I know statistics say another value 😉 )

Happy counting 🙂

 

If you run into detailed questions about your acquisition task, prepare some of you code and post it here and if no NI expert is jumping in (they usually do sooner or later) , send them a mail and point to it 😉

 

Greetings from Germany
Henrik

LV since v3.1

“ground” is a convenient fantasy

'˙˙˙˙uıɐƃɐ lɐıp puɐ °06 ǝuoɥd ɹnoʎ uɹnʇ ǝsɐǝld 'ʎɹɐuıƃɐɯı sı pǝlɐıp ǝʌɐɥ noʎ ɹǝqɯnu ǝɥʇ'


0 Kudos
Message 7 of 8
(5,186 Views)

Thank you very much Henrik, I will definitely post on this forum.

 

Regards,

maria!

NI System Configuration:
- NI PXIe-1071, 4-Slot 3U PXI Express Chassis , 1 GB/Slot throughput, Part Number: 781368-01
- NI PXIe-PCIe8381,x8 Gen2 MXI-Express for PXI Express Interface,3m, Part Number: 782522-01
- PXIe-5160 PXI Oscilloscope, 500 MHz, 10 bits, 2.5 GS/s, 2 Channels, 64 MB, Part Number: 782621-01
- Astronics PXIe-1209 2-Channel, 100 MHz PXI Pulse Generator, Part Number: 785033-01
0 Kudos
Message 8 of 8
(5,163 Views)