GPSDO Phase noise

[VK2JDS]

Chinese GPSDO units on ebay are cheap and look good, but how do they perform for us on the higher microwave bands.
I have been struggling to get some of the under 200 dollar units to work on my 122GHz gear and similar problems on 76 , 47 and 24.
So, whats the problem ?
Close-in noise right next the the 10MHz carrier, when mutliplied up, causes the resulting signal from your PLL to look terrible. So much so with the poor quality units and (some cheapy pll pcbs) that the actual carrier is lost in a hump of noise that spreads 20 kilohertz or more in width.
What does this mean in practice, in the field trying to make contacts ?
When the local oscillator signal is noisy feeding your microwave mixer block or transverter then the downconverted signal you are looking for is difficult to tune in to. Typically a beacon signal or carrier from a sig-gen will be a weak but wide series of carriers spread over several kilohertz, instead of a single clean carrier. Your transmitted signal will be similarly affected.
A weak beacon tuned in when you are using a nice clean 10 MHz source will completely dissappear into the noise when you plug in a noisy 10 MHz GPSDO reference instead.

Lets have a detailed look at the setup I am working on for 122 GHz as an example. 76GHz 47, 24, 10 and below are all affected too, but as their mutiplication factor is less the severity of the problem decreases.
I am using a PLL from ZL2BKC outputting at 13.5 GHz, feeding into a CMA module and multiplying by 3 to 40.5GHz, then tripling in the mixer diode to give a resulting Local Oscillator on 122250-433 MHz = 121817 MHz.
As you can see the multiplication factor from 10 MHz is about 12,000 . so if there is noise extending only One Hertz either side of the 10 MHz carrier from your reference oscillator that means there will be 12 KHz of noise either side of the resulting Local oscillator signal !
Lets look at some spectrum analyser screenshots.......

[VK2JDS]

PLL output at 13.5GHz using $149 GPSDO

Devices tested, obtained from ebay sellers

As you can see from the spectrum analyser screenshot the carrier level above the noise hump is not good at all. and it was the same for both of the tested units.
Now lets look at exactly the same test setup using an ISOTEMP 10 MHz oven controlled oscillator.
All I did was unplug the GPSDO units and plug in the Isotemp.

PLL output using Isotemp source

Much better, this is what it looks like with the wrapping I added around it removed for the photo:

Isotemp in its box

[VK2JDS]

Our objective is to have a GPS locked Local Oscillator feeding our transverter/mixer block so when we are working out in the field trying to find a weak station on the other end of a long path we know exactly what frequency his similarly locked signal will be on. (then we just have to aim !)

So what can be done to solve the problem :
Firstly, opening up the cased unit and comparing it with the PCB unit shows they are basically the same circuit and only the type of oscillator module differs. The oscillator block signal is fed into a chip that does the dividing and phase comparison etc, also it buffers the signal before sending it to the output socket. I tried buffering the 10MHz output pin directly at the Bliley ocxo and feeding that to the PLL. No improvement.
Having a look at the supply end of things there is a switchmode power supply taking 11 to 16 volts in and converting that to 5 volts, which runs the lot except for the opamp that drives the Vadjust to the OCXO, so I disconnected the switchmode and fed 5 volts from a quiet linear supply. No improvement.
Then I tried adding capacitance from the (Vadjust)adjustment voltage pin of the Bliley to slow down the loop and see if there was any change. No improvement.

Next I tried a different GPSDO offering from the same builder BG7TBL. It is a larger unit I purchased some time ago, but they are still available. It lands here for between 260 and 300AUD from ebay.

BG7TBL larger unit

Testing it on the ZLPLL setup unplugging the cheap unit doing a plug swap to this unit gave this result :

unsmoothed output of larger unit

smoothed output to show improvement clearly :

spectrum analyser smoothed display

[VK2JDS]

As you can see the more expensive ebay unit does a good job and is nearly as good as the Isotemp. For about a hundred dollars more than a cheapy these are the easiest option. With a 10 MHz splitter they can feed several transverters.

The cheaper units will probably be suitable for the low bands as they are, but I am experimenting with using the pcb to lock an Isotemp or Trimble OCXO, in place of the Bliley and noname blocks they came with, and it looks like the pcb may be able to steer it into lock.

Another thought was to use the Isotemp as the actual transverter reference, and have the cheap GPSDO as a test marker to tweak the Isotemp exactly, like a calibration source, but lugging test gear out into the field to do this is impractical.

Lastly we could try to steer a TXCO using the cheap GPSDO and rely on the purity of the TXCO to give us a decent output on our Local Oscillator frequency. I believe some Kuhne gear uses this principle.

I hope this article has been food for thought, and please add any suggestions you may have so everyone can benefit.
(putting it on facebook is a waste of time)

73 Dave vk2jds

[VK2JDS]

Cheap GPSDO Pcb experiment to lock an Isotemp.
I removed the Bliley OCXO from the green pcb unit pictured in the earlier post.
After shifting the internal adjustment of the Isotemp to give a centre of lock range of 2.5 volts to be spot on frequency i connected it to the cheap GPSDO pcb.

I didnt even bother to see if it locked, the output was horrible. So much noise.
Disconnecting the Vadjust line reduced the noise by a bit over 10dB, disconnecting the sample port off the Isotemp 10MHz output improved it some more, but unplugging the dc supply to the pcb got the noise back to good again.

So its the pcb. The onboard OCXO it comes with is probably quite good, its the other circuitry letting these units down. First to replace is the switchmode regulator section, but still allowing 12volts to the Vadjust opamp. Lift the last inductor on the 5 volt rail from the switchmode and feed a linear supply into the rest. Remove the 12v to the switchmode input too. I suspect it might need the 4 volts to bias the opamp that comes out of the OCXO, so i cobbled the Bliley in on wires to generate this voltage. Now I am seeing very low phase noise and its looking like its steering the Isotemp towards lock.
work will progress tomorrow
73 Dave

[VK3QI]

Dave,

I have been using a number of those cheaper units successfully for quite a while now up to and including 47 GHz.

I obtain them from Ebay seller RFEXPERTS

https://www.ebay.com.au/itm/NEW-10MHz-S ... 2749.l2649.

The interesting is that the although the external case looks the same, the internals and in particular the OCXO are different.

I looked up the specs on the Vectron C4600 and they certainly spec up much better than the results you are seeing with yours.

On the speccy they look as good as the higher value Trimble and Z3805 units.

Maybe different versions?

Cheers

Peter VK3QI

[VK2JDS]

Hi Peter,
that is the exact pcb thats in the case in the photo i put up but mine doesnt have a vectron in it. Its not real great, same as the green pcb one. The noise looks like it is coming from the pcb rather than the OCXO.
To do the Isotemp test I did this:
Remove the OCXO.
To feed in the 5 volt linear supply i lifted the left hand inductor on your pcb photo and connected 5v across the 1000uF cap closest to the OCXO. Disconnect the 12v input of the 5 volt switchmode near the right hand inductor, but make sure 12 volts still gets around a looped track into pin8 of the 8 pin chip you can see, which is the Vadjust driver to the OCXO.
I sampled the 10 MHz output of the Isotemp with a 50k trimpot about half rotated and fed that (from the pot wiper) into the pcb on the OCXO pin closest to the middle inductor.
The Vadjust dc error voltage to the Isotemp comes from the OCXO pin thats closest to the mega328 chip(square chip with legs all the way around it).
I am still unsure about the 4 volts from the OCXO, the middle pin of the 3 closest to the IC end of the pcb.
More work to be done on it tomorrow, but the phase noise is dramatically improved. Just have to confirm it locks the external Isotemp or Trimble now.
73 Dave

[VK2JDS]

I can confirm the experiment locking the Isotemp does work and solves the phase noise problem.
Next is to test a Trimble brick OCXO and then the Blileys.
To fix the as-suppled GPSDO :
I suspect that removing the supplied OCXO, powering it from 5 volts from a separate linear regulator then applying the modified green gps pcb to discipline it will work.
So, while the Vectron OCXO has great specs, it is being let down by the pcb. Now we know how to fix them !
pictures to follow...
73 Dave

[VK2JDS]

More testing :
The original small OCXO it was supplied with doesn't have enough output to drive the PLL and still have enough voltage to drive the phase detection circuit on the pcb, so it wont lock. I am not using the noisy buffered signal at the socket on the pcb, but connecting directly to the OCXO. More work to be done on that.
The Trimble wouldnt lock.

Isotemp seems to be perfect though, perfect output.
closeup photo :

modified PCB, same circuit as the cased units

overview:

needs a tidy up then into the case

73 Dave

[VK3QI]

Dave,

Although not exactly related to what we were discussing before, I came across an interesting effect yesterday whilst setting up a rover station.

I was using one of the GPS units previously discussed feeding into a BG7TBL 8 port distribution amplifier with an in-built OXCO as a backup.

I already use the same setup in my main portable setup, up to and including 47GHz WITH THE EXCEPTION that the 8 port distribution amplifier was without the inbuilt OCXO and with excellent "on air" results.

Monitoring the 1135.88889 MHz 10MHz locked oscillator in the transverter there was a pure signal, BUT when listening to a 10GHz (and lesser extent 5GHz ) received signal, there was a distinct warble on the signal at about 1 second period.

After a lot of mucking around with other GPS units that I have in the shack, various distribution amplifiers and a variety of levels, I concluded that there must be some sort of sampling of the distribution amplifier input to ascertain that there is an external 10Mhz source plugged in and that is introducing a noticeable warble on the signal when multiplied up.

Sure enough, removing the OCXO and disabling the sampling circuit cured the problem.

***********************

So a warning to anyone using one of the BG7TBL 10 MHz distribution amplifiers with the in-built OCXO, don't necessarily blame the GPS unit if there are issues with the resultant signal from your transverter.

Cheers

Peter VK3QI

[VK2JDS]

That is an interesting observation Peter,
the bg7tbl gpsdo designs all seem to run the 10MHz output though a buffer chip, and through that same chip is the 1PPS output. hence why the 10 MHz has a slight modulation of that 1 Hz signal.
It is noticable on my signals on 47GHz upwards.
Running the Isotemp disciplined by the bg7tbl small board gets rid of that, as the 10MHz output is taken directly off the output of the Isotemp brick.
73 Dave vk2jds

[VK5ZD]

Thought I'd add my 2c worth to this discussion. I have one of the BG7TBL GPSDO units the same as Dave's, but with a different OCXO inside (CTS 970-2178-46). I think they must use whatever surplus OCXO units they can get hold when they build these. By way of comparison, I also have an Isotemp 134-10 OCXO, a CIC STP2145A OCXO, a home brew GPSDO (which also uses an Isotemp 134-10) and an Efratom Lpro-101 rubidium oscillator. The following plots were made from my spectrum analyser (HP-8562A) using software from John Miles, KE5FX.

Phase Noise

As you can see, the cheap Chinese GPSDO is the best of the bunch. Note: I suspect the 'lumps' you can see at (roughly) 32kHz, 39kHz, 75kHz, 120kHz & 160kHz (which appear in all four graphs) are some sort of artifact created by the spectrum analyser.

BG7TBL GPSDO

The following plots show the output from each oscillator in a 50kHz span (averaging was turned on).
Note the spike at -16kHz from my home brew GPSDO; something to be investigated one day (maybe).

BG7TBL GPSDO

CIC OCXO

ISOTEMP OCXO

VK5ZD GPSDO

Of the OCXOs, I like the CIC STP2145A best. It only draws 150mA once it's warmed up plus it's very stable. I turned it on this evening after not using it for a couple of months and, after a 30 minute warm-up, the output was still exactly on 10MHz. The Isotemp, which also hadn't been used for a couple of months, was about 2Hz off.

[VK5ZD]

An finally, the Efratom Lpro-101 (the previous post wouldn't let me add any more images).
All I can say about this device is that I don't recommend it if you want a clean 10Mhz reference!

Phase Noise

50kHz Span

500kHz span

[VK5ZD]

More testing...
This time I used the five different 10MHz sources as the reference for a 'Kuhne MKU LO 8-13 PLL' oscillator which was configured to output 11,952MHz (x 2 + 144 = 24,048MHz).

The five sources are as shown in my previous two posts, namely my home brew GPSDO, a BG7TBL GPSDO, a CIC OCXO, an IsoTemp OCXO and an Efratom rubidium oscillator. Strange as it may seem, the signal on 11,952MHz was pretty much the same regardless of which refernce I used. I ws expecting poor results with the rubidium oscillator but it's weird output seems to make no difference.


These plots are all done with a 2MHz span (200kHz/div) with averaging turned on.
Hard to tell them apart.

Using the BG7TBL GPSDO

Using the CIC OCXO

Using the IsoTemp OCXO

Using the Efratom Rubidium Oscillator

Using my home brew GPSDO