real time demodulation with pxi-5660

[lanbaba]

Hi all,

I'm considering to use  NI PXI-5660 to do some real-timel demodulation of signals off the air. To do a continuous demodulation I need to acquire signals off the air seamlessly frame by frame.  Suppose my frame length is T and my demodulation scheme running in the controller is fast enough to process a frame if it is acquired and buffered somewhere in the RAM of the controller. The questions is

Is it possible to acquire signals in period T alternately to two separate buffers while the acquisition is continuous?

Thank you for any advice.

LBB

[Andy_Hinde]

Hello LBB,
What you are describing is possible. The PXI-5660 can operate in a mode where the signal coming out of the PXI-5600 RF downconverter is streamed to the PXI-5620 digitizer module onboard memory. The PXI-5620 is the digitizer component of the PXI-5660. This streaming is only feasable with signal acquisition bandwidths of 1.25 MHz, as this is the point at which the Digial Downconverter (DDC) on the 5620 turns on, lowering the effective sample rate and making streaming possible.

The data from the PXI-5620 can be fetched from onboard memory very flexibly, and while the data streaming continues in the background. You just must avoid having the onboard memory overflow from not fetching data out as fast as it is coming in. This is dependent on the processing being done in the loop, which leads to the point that although streaming is possible, some applications can't always keep up.

If the frame is pulsed in time, triggering each acquisition can save memory space by eliminating the storing of 'dead time'.

So in summary, this is possible but very dependent on the application specifics.

Regards,
Andy Hinde
National Instruments

[lanbaba]

Hi Andy,

thank you for answer. 
We want to do software-defined radio with your PXI-5660 and need to  stream signals with a real-time bandwidth up to 20MHz. Suppose we can implement the demodulation fast enough by means of extra hardwares do you  mean a continuous acquisition of 20MHz signals itself is a problem?

Regards,
LBB

[lanbaba]

Hi all,

I read following words in the NI product description for PXI-5660 that confused me a lot.

"This digital downconversion chip performs real-time decimation and downconversion to baseband of any signals with a span of up to 1.25 MHz, ideal for capturing many cellular and other communications channels. The downconversion ASIC also generates complex I/Q data from the spectrum." (http://zone.ni.com/devzone/conceptd.nsf/webmain/C4541E1DAF5E42C986256C6E000C2619#4)

If my RF signal has a bandwidth of 20 MHz I am expecting a baseband complex signal of 20 MHz bandwidth too after digitizing, and not "up to 1.25MHz".  What is exactly this 1.25 MHz signal span?

I believe complex I/Q data are worked out by the digitizer in the time domain first, then the spectrum can be calculated. How / why can the digitizer work another way round, generating I/Q data from the spectrum?

Cheers, LBB

Message Edited by lanbaba on 07-21-2006 02:46 AM

[AlexxxR]

Hi,

this Code is running in our branch:

http://sine.ni.com/apps/we/niepd_web_display.display_epd4?p_guid=D1613205BE003209E034080020E74861&p_...

Perhaps this may help you a bit!

Best regards,

A Rudolph

NI Switzerland

[lanbaba]

Hi,

FM radio has a bandwidth of 200 Khz, lower than 1.25 MHz. As Andy said, for this signal streaming is no problem. How about DAB radio with bandwidht of about 2 MHz? Is streaming still possible? How about streaming of WiFi, WiMax, DVB and so on?

Cheers, LBB

Message Edited by lanbaba on 07-21-2006 04:06 AM

[JerryAL]

Hi Lanbaba

Your question has a few details which doesn’t make 20 MHz bandwidth easy.

The NI 5620 digitizer is 14 bits, effectively two bytes per sample.  For a 20 MHz bandwidth, Nyquist says the sample rate should be twice the BW, or at least 40 MHz, that is 80 MBytes/s.

Another issue is the IF bandpass range of the NI 5600 RF Downconverter, which is 20 MHz, but ranges from 5 MHz to 25 MHz, which then goes to the digitizer.  So, to get 20 MHz, the actual highest frequency is 25 MHz.  So taking into Nyquist and two bytes/sample, we are talking about a minimum transfer rate of 100 MBytes/s.  That is minimum, and probably not possible on many computers.   Even though the theoretical PCI bus bandwidth is 133 MBytes/s, when crossing the bridges and/or chipsets, the bandwidth is much smaller, at least in terms of continuous streaming.  Short bursts of data can be quite fast.

So, streaming raw IF sampled data will depend on what the streaming data rate is for your computer.  Your usable continuous streaming bandwidth depends on this.

As for the 1.25 MHz IQ bandwidth on the NI 5620, when the frequency bandwidth is below this number, the IQ down-conversion is done in hardware on the NI 5620.  This is the bandwidth limit of the down conversion circuit on the module.  This make it easy to transfer data of a smaller bandwidth even if the actual IF from the NI 5600 RF downconverter is at 23 MHz.  The IQ down conversion in hardware means that only a minimum of ~1.6 IQ MSps (6.4 MBytes/s). 

If wider, the digitized IF time domain data is returned directly to the computer where IQ down conversion takes place in software.

As you say, the NI 5620 hardware based IQ down conversion is too narrow for your application.  And your bandwidth is limited due to the PCI bus bandwidth.

I suggest you contact your local National Instruments Field Sales Engineer.  He will be able to work with you on a few options and what you may achieve with them.

Out of interest, what type of signal are you demodulating that you need 20 MHz bandwidth?

Jerry

[lanbaba]

Hi Jerry

 
Thank you for your detailed analysis. Real-time stuff makes the real life harder and the devil is in the details. At this point I would very much appreciate NI to provide this platform for technical discussions and many thanks to those who answered my questions. I have not tried PXI-5660, have very limited knowlede about implementation details of PXI-5660. In some NI websites I even see MIMO-OFDM testbed implemented with PXI-5660. So I am trying to find out what this PXI-5660 is really capable of.

OK, let's come back to technical details.  Forget about streaming first and talk about Nyquist.I aggree that 40 MHz is the minimum feasible frequency to sample a real bandpass signal with 20MHz bandwidth. Therefore, 80MBytes/s is  the maximal data rate going through PCI-bus theoretically. Further, I believe this 40MHz sampling frequency is independent of the IF position. Expressed another way, must I sample at 2*(15+1)= 32MHz if the signal bandwidth is only 2 MHz? I hope not. In order to sample 20 MHz signals I should be able to choose a sampling frequency of 40 MHz. Please drop a line if it is not true in PXI-5660.

Now, regarding signals wider than 1.25MHz,  the digital down conversion is supposed to be done by software in the controller PC. What is the delay of this operation compared to the hardware decimation?

My current applications are DAB (2MHz BW) and DVB (8MHz BW). Upcomming applications are UMTS (5MHz), WiFi (20MHz) and WiMax (20MHz). GSM is lower than 1.25 MHz but we are not interested in it any more.

Cheers,
LBB.

Message Edited by lanbaba on 07-21-2006 12:23 PM

[JerryAL]

Hi

I’m only going to address the sampling issue.

I assume your numbers are based on a 2 MHz signal centered at 15 MHz.  Yes, it would be a good idea to sample at least 32 MSps, but even higher to avoid complications.

If I were to sample the IF signal at 8 MSps, I would be under sampling the signal.  In some special cases, it may be OK to under sample, but this isn’t one of them.  A multiple of the sample rate falls at 16 MHz, right at the edge of the signal bandwidth.  Taking an FFT of the signal would place it from 0 to 2 MHz in the FFT.  So far, so good.  Here is the issue, the signals from 6 MHz to 8 MHz, 8 MHz to 10 MHz, 16 MHz to 18 MHz, 22 MHz to 24 MHz, and 24 MHz to 26 MHz will all be mapped into that frequency space. 

The only time under sampling works is if you were to have a band pass filter around the signal of interest, attenuating all the other signals that can distort the acquired signal.  Although a filter that truly attenuates the signals just higher than 16 MHz will be hard to create in hardware.

If there is no filtering of undesired signals, the Nyquist frequency should be higher than the actual frequency the signal is at. So yes, the 2 MHz signal centered at 15 MHz should be sampled at 32 MHz at least.

This is true for the NI PXI 5660.  This should be true for all other hardware as well unless they use some kind of filtering.  IQ digital down conversion does use filter in its process, so that is one reason you can use an IQ sample rate at least 1.25 times the bandwidth.

There may be an option that may help you, but I need you to contact your local National Instruments Field Sales Engineer.

Jerry

[lanbaba]

Hi Jerry,

thank you again. I'll surely contact my local sales engineers, but as I've asked so much here let me ask a last question to finish this topic.

Are the following statements true for PXI-5660?

1. To get alias free complex baseband signals with DDC the sampleing frequency fs can be set to fs=1.25B , where 0<B<20MHz is the real-time bandwidth of the real signal of interest and adjacent channels are loaded.

2. To get IF samples without aliasing,  the sampling frequency should be fs > 30MHz + B if no extra hardware IF filtering is added. Moreover, I should write a program to do the digitial down conversion and rate conversion to transform the IF samples to baseband samples.

3. Real time streaming of signals is possible if B<1.25MHz, otherwise the software implementation of DDC and PCI-bus throughput may limit the aquisition speed.

Cheers, LBB

Message Edited by lanbaba on 07-22-2006 01:23 AM