[time-nuts] "Best" GPSDO

[Rex]

On 9/28/2012 8:31 PM, Chris Albertson wrote:
> You can guess the real question here: "how good does the 10MHz
> reference need to be to test real-world receivers?
>

Pretty sure the answer is good enough, but... Depends on the receiver 
and what it is receiving.

I got interested in time-nuttyness because I am a ham. My interests are 
mainly in the microwave bands, typically 10 GHz and higher. I got a 
couple GPSDOs primarily to make accurate measurements in the many-GHz 
range by feeding ref inputs to my test equipment. Mostly I wasn't 
interested in the nitty-gritty, just being good enough for good 
"trustworty" results. Before the GPSDOs, getting accuracy to 100's of Hz 
at 10 GHz was an act of faith. Now, to Hz is pretty easy.

Recently I got involved in building some boards for an intermediate IF 
(is that redundant?) for a 24 GHz radio that have their LO at 3600 MHz 
and are locked to 10 MHz. (Details to be presented at Microwave Update 
2012 in a couple weeks -- http://microwaveupdate.org/ . Feel free to 
sign up and attend if you are interested.)

Around 2006, John Miles shared with us some measurements he did using 
microwave "brick" phase locked oscillators to get the phase noise from 
OCXOs multiplied up enough to see the differences on a decent spectrum 
analyzer. ( http://www.ke5fx.com/brick/brick.htm ) He used an 8566b SA 
and his own PN  software ( http://www.ke5fx.com/gpib/pn.htm ) to drive 
it with GPIB to make the measurements. Many thanks to John for all he 
has shared with us.

I hadn't made any measurements like this before, but this seemed like a 
good way to get a feel for the quality of the 3600 MHz boards (which 
turned out good.) I also have an 8566b SA, so that with John's PN 
software seemed like a good setup to try. The 3600 board has a loop 
filter about 10 KHz wide so in the audio range the output PN is related 
to the quality of the 10 MHz reference.

My two main frequency references are two GPSDOs, an HP Z3816A and (few 
years ago added) a Z3805A Sumsung. Both have an MTI 260 OCXO as their 
internal locked source. Testing the 3600 MHz board using these two 
references, the best phase noise came with the 3805 at about -90 dBc. 
The 3816 was about 7 dB higher. Not sure why. The MTI 260 oscillators 
are 5 MHz so are doubled in the GPSDOs to 10 MHz. Maybe that is part of 
it, or maybe the two MTI 260s are that much different.

I also measured with two small eBay oscillators from China -- all in 
equivalent small packages about 2 inch square by 1.5 inch high. A CIC 
STP2145A gave results similar to the 3816. A Morion MV89A was the worst 
so far, about 10 dB higher than the 3805. Clearly, the affects of the 
the oscillator PN are quite visible when multiplied by 360 to 3600 MHz. 
(20 log 360 = 51 dB.) I'm not sure about the exact accuracy of my 
measurements, but I am certain I am seeing the relative effects of the 
PN from the OCXOs.

I have a bunch of 10 and 5 MHz OCXOs I have accumulated and now that I 
have this tool for evaluating, I need to take the time to fire them up 
and sort them by PN quality. I guess I need to build a trustworthy 
doubler too, for the 5 to 10 MHz like the ones on Bruce's pages at KO4BB.