Synchronization of three USRP-2932

[tmy540]

hy everyone,

  I am using three usrp-2932 to build a system of three synchronized Receivers

  Referring to the example of 8x8 MIMO system, it use one gigabit ethernet interface and through switch to connect the eight USRP

  But, in my application,  i use 12.5MHz as my sampling rate.  Besides,  due to  the concern of the throughput, i use three individual gigabit ethernet interface to connect the 3 usrps.

  So, I am wondering that using 3  gigabit ethernet interface will influence the Synchronization of three USRP-2932 or not? 

  Because it is hard to confirm that the 3  gigabit ethernet interface will send the message to the 3 usrp at same time.

 Besides,  i was using MIMO cabel to do Synchronization of two USRP-2932.

 And now i am  using GPSDO to do Synchronization of  three USRP-2932.

 But, in my experiments, it seems that the performance of Synchronization via MIMO cable is better than using the GPSDO.

 So, i am wondering that is there any better way to setup the system. Using external clock and PPS ?

[Kevinye]

It seems like external clock with REF IN is not used for 2932. You must change jumper position in 2932, but that's not recommanded.

[ErikL]

Synchronization options with best performance listed first:

---- Very Best

0. NI FlexRIO SDR and VST offer sharing of local oscillator among devices giving both time, frequency, and phase alignment among multiple channels. (cabling them together is required.)

----- Best USRP Sync Options

1. MIMO cable (2 devices).  It synchronizes off of the 10 MHz reference and shares the time base over the link.

2. Clock Distribution (>2 devices) best is using an octoclock or other clock distrubiton method:  https://www.ettus.com/product/details/OctoClock-G

 (Using the 2932 you will have to power down the internal GPSDO to use this method, as well as move a jumper because it shares the same line as the external SMA inputs.)

------- Better USRP Sync Options

3. GPS Disciplined Clock with GPS Lock (>2 devices) can provide you up to 50ns of synchronization. For best performance all devices need to be able to lock to the same satellite constellation and be stationary.

------- Good USRP Sync Options

4. GPS Disciplined clock with NO GPS Lock (>2 devices)  gives you better frequency accuract but does not synchrnoize between devices.

5. Standard USRP with internal TCXO (>2 devices) is equivalent to no synchronization and you will need to consider frequency offsets between devices.

[tmy540]

Thanks for the response, Kevinye and ErikL.

      I want to confirm my  of thoughts about the GPSDO synchronization.

      The GPSDO use the PPS(1Hz) derived from the received GPS signal to adjust the frequency and PPS signal of the USRPs.

      It seems that it will do the calibration among the USRPs automatically every second.

      So,  the Frequency or the time difference(- 50ns~+50ns)  between different USRP is not a constant and will vary with the time??

      Because in my system, i want the USRPs can start to record the signal at the same time.

      However, even use the MIMO cable to do the synchronization, it still has time difference between two USRPs.

      But, in the MIMO cable case, the time difference is nearly a constant.  

      So, I can use the measured constant to calibrate my system.

      Hence,  when I use the GPSDO to do synchronization, it seems that the method I used in the MIMO cable case doesn't work.

      Is my thought about the GPSDO correct?

[ErikL]

A GPS Disciplined Clock as 2 methods of synchronization, a 10 MHz Ref and PPS.  The 10 MHz is constantly disciplined by the GPS to be as accurate as possible, commonly measured in ppb (parts per billion). The PPS is simply used to align the clocks every second (PPS = Pulse Per Second).  Because the reference is disciplined and holdover is very good, this can give you 50nS of precision time between devices.

Regardless of the method (1, 2, and in some cases 3) will have perfectly aligned ADC samples.  (phase will be different so I and Q will be weighted differently due to the rotated constellation).  The only exception with 3 is that accuracy is determined by the level of GPS sync up to 50ns which could mean you have aligned samples or they could off by a fraction of a sample or up to a few samples from the ADC.

[tmy540]

Hi ErikL:

      Thank you very much.

      You said that GPSDO can give 50nS of precision time between devices.

      Is the time difference between the devices a constant ?

      Or the time difference between the devices are randomly varying with the time and will  be limited to 50ns by the GPSDO?