Hardware options for synchronization

[cochenob]

Was wondering if someone could help "rank" these approaches for synchronization.

- TCXO with shared references/PPS (i.e. - MIMO cable or OctoClock)

- OCXO

- OCXO with GPS

The "standard" TCXO models require extra stuff (MIMO or OctoClock or external sources) to phase lock and time synchronize. 

The OCXO gives you better accuracy...but it is unclear the degree to which this has an effect on phase coherency (and over how long a time).  Presumably, you'd still need something like a PPS to time synchronize?

With the GPS disciplined devices, are you basically getting the phase coherency and time synchronization from the GPS signal rather than external 'stuff'?  Or are there instances where you'd still need something like the MIMO cable or OctoClock to further nail down timing and synchronization?

---

Brandon

[cochenob]

...for example...this document from the Ettus site gets into the above synchronization options, but doesn't really go into the relative performance characteristics.

http://www.ettus.com/content/files/kb/mimo_and_sync_with_usrp_updated.pdf

[warEagle12]

Hello!

Here is an article that discusses building a MIMO system and some of the different specifications available. I believe that the Frequency and Time Synchronization section of the Scalable 8x8 MIMO Hardware System Architecture part discusses the information that you are looking for:

http://www.ni.com/white-paper/14311/en/

Thanks!

[cochenob]

Hi Stephanie-

Thanks for the link.  This almost answers my question.  I'm more interested in how well several GPS OCXO's are synchronized themselves.  In other words...for the example in the link...how well are the Tx and Rx OCXO's synchronized to each other?  How does this compare to a situation where I might have been able to run BOTH the Tx and Rx radios with a single external reference?

Or...said another way...take just the Rx radios.  Instead of sharing 1 external reference among the 4 USRP pairs...what if the radios were so far spread apart that I couldn't physically connect them, and had to use 8 GPS OCXO's?  How well would this work compared to sharing a single reference via cables? 

----

Brandon

[JamesB.]

I'm not an expert on the USRP products, but I am pretty familiar with synchronization concepts. I don't know if this will help, but in the spirit of discussion forums, I'll throw out the information I know in hopes it will at least represent progress.

The (civilian) GPS standard is accurate to less than 100 nanoseconds. So, two nodes that are using different GPS antennas can still differ by that amount. In practice, synchronization between two such nodes is usually better than 100 ns (tens of ns), but it isn't guaranteed. But two (or more) nodes sharing the same GPS antenna -- or the reference clock from that antenna -- are going to be better than that.

---

@cochenob wrote:

With the GPS disciplined devices, are you basically getting the phase coherency and time synchronization from the GPS signal rather than external 'stuff'? 

---

Yeah, you can think about it like that. You can (a) distribute a really good reference clock (like a rubidium clock) to several devices using an OctoClock, (b) connect a GPS antenna to an OctoClock-G to get a pretty good reference and distribute it to devices, or (c) share the GPS antenna using a splitter with several devices that have OCXOs that can be disciplined by GPS (like the NI USRP-293x products).

To be honest, I'm not sure whether (b) or (c) would provide better synchronization. It might depend on the specs of the OCXO in an OctoClock-G compared to the OCXOs in each of the nodes.

[warEagle12 · Accepted solution]

Hi Brandon,

Thank you for the extra information! I have been looking further in to this issue, and a lot of your question depends on your hardware and system setup. JamesB, awesome post! This is great information.  To add to this, for our GPS USRPS (NI USRP 293x) unfortunately you cannot use the Ref In inputs, so splitting out the external clock is not going to be an option on this hardware. With this in mind, option (c) will likely be the easiest since that is exactly how our USRP's work. Are you using an NI USRP? If so and you're trying to sync up 8 of these USRP's together, GPS might be the only real option. 

As far as direct cabling vs. GPS for accuracy, direct cabling is more accurate if you are using an external clock that is is more accurate than the onboard clock and if you're using a USRP that can use an external reference clock 

The NI USRP 2930 and 2932 both have OCXO's which alone are more accurate but there is no way to extract the 10MHz source to another device, except for a MIMO cable which is only good to use for 2 devices. Therefore the direct cabling method only exists for specific scenarios:

1. Using MIMO cable for up to 2 devices

2. Using an external clock, and a splitter.

However, the internal OXCO does not do you any good for more than two device synchronization. Using a GPS antenna, clock disciplining and a clear view of the sky will let each USRP utilize it's own OXCO and still be synchronized. So it is not just GPS, it is the act of using GPS information to adjust the already great OCXO time reference which required each device to adjust it's own clock. The following document shows the specifications for a USRP without a GPS antenna (just the OXCO accuracy) vs. the OCXO accuracy in conjunction with the GPS disciplining in section 4 (Synchronization with GPS Disciplined Oscillators on NI USRP 293x):

http://www.ni.com/tutorial/14705/en/

While I was looking in to this, I also found this forum post that had a similiar question:

http://forums.ni.com/t5/USRP-Software-Radio/Synchronization-of-three-USRP-2932/m-p/2835772

I hope this was helpful!

[cochenob]

Stephanie/James-

Thanks for the insightful discussion here.  Both links were excellent...and Erik's postings on the other topic was exactly what I was looking for. 

Thanks!

---

Brandon

[cochenob]

So I thought I was done with this topic...but maybe not!

The one link above gives a good example upfront about stability and accuracy...though it's not entirely clear sometimes whether the different options are addressing improved stability or improved accuracy, or both.

Consider two scenarios where two radios are connected via a MIMO cable.  One scenario has TCXO's in each radio, while the second scenario uses OCXO's.   What are you buying between the two scenarios?  If the TCXO's have lower accuracy, stability, or both...does it matter since both radios are ultimately phase locked?  If my TCXO drifts a few Hz, but they both drift together since they're sharing the reference over the MIMO cable, then what have you really lost?

To add an extra level of complexity to this...if I remember that each radio ultimately synthesizes it's center frequency via an analog PLL...what are the odds that the individual PLLs (with their own component variations) will 1) actually settle at the same frequency and 2) track the shared reference in the same way?  In other words...who cares how accurate (or stable, or both) the XO is if the other components in the PLL have enough variation that you'd never be able to take advantage of the better performance anyway?  Or...am I over-valuing the potential differences between PLLs?

---

Brandon

[psisterhen]

Good questions.

Even if two TCXO devices are tracking each other, you would still benefit from the accuracy and stability of the OCXOs.  Think about communicating to a remote device, which may have an LO not tracking with these.  You want the frequency difference to be minimal and stable, so the freq offset can be easily detected and corrected.

It certainly seems true that the radio PLLs may add another layer of offset and drift.  I don't know that we have hard numbers on this, although the part data sheets may tell us.  However, they will definitely track with the reference frequencies, and I imagine using OCXOs would alleviate these problems further.

Patrick