[time-nuts] GPS Locked and Unlocked Performance Comparison

[Bruce Griffiths]

Tom Van Baak wrote:
> Thanks to all of you so far that made comments on the
> GPSDO plots; I'll handle this one first.
>   
Tom
> Both the Thunderbolt and Miller designs appear to measure at least 10x
> better than the most optimistic expectations.
>   
>
> Bruce,
>
> Interesting thought. What calculations did you use to set
> your expectations? The free OCXO in both those GPSDO
> are well within reason, no? And both GPSDO follow a fairly
> standard couple of ns over root tau line, yes?
>
>   
The Thunderbolt GPDSO appears to have an asymptotic performance more
like a few ns/Tau not a few ns /SQRT(Tau).
The Miller GPSDO appears to have a similar performance.

Comparing the performance of the locked and unlocked OCXO suggests that
a loop response time of perhaps a few tens of seconds at most may be in
use, do you have any idea what the actual filter time constant is?
Are all GPS receivers really that good, even an old design like the 5V
(non timing - at least there is no 0D mode) Jupiter receivers?
Since the receiver determines its location every second the  timing
solution will inevitably be noisier than  if a 0D mode were available.

However these receivers do make the GPS carrier phase data available.
They also offer serial port baud rates to 115kBaud.
All one needs is to lock the 10.95MHz signal to one's local standard to
facilitate high resolution (5 psec) measurement of the local standard's
phase error with respect to the GPS carrier.

The 5V (and 3V) Jupiter receivers have 3 crystals a 10.95MHz crystal
used by the RF section, a 29MHz crystal used by the DSP chip, and a
32KHz crystal used by the RTC.
If the 10KHz and PPS signals are  derived from the 10.95MHz crystal
oscillator then the 1PPS sawtooth timing error amplitude (and the 10KHz
phase jerk amplitude) will be either 22.83ns or 45.66ns pp depending on
whether the leading edge of the PPS signal (or the leading edge of the
10KHz signal that coincides with the PPS transition) is coincident with
the nearest transition (positive or negative slope) of the 10.95MHz
signal or say just to the nearest positive slope transition of the
10.95MHz signal.
The datasheet timing spec is within [-1us, +1us] of UTC. This was
probably specified when SA was turned on but expecting the error to be
100x better now that SA is off is perhaps somewhat optimistic.
>> One possibility is injection locking of the OCXO via its EFC due to
>> inadequate shielding or via its output due to its buffer amplifier
>> having inadequate isolation.
>>     
>
> To prevent, or at least detect, this effect I allow my 10 MHz
> house reference to drift off-frequency by quite a bit (last
> month it was 1.7e-12 off). That way there are no on-time
> or on-frequency sources near the test setup.
>
>   
A much larger offset would be better (although difficult to easily achieve).
>> What cable types did you use to connect the GPSDO's and the reference
>> signal to the Allen deviation measurement instrument?
>>     
>
> For these tests I used a TSC 5120A; a 3 meter mil-spec
> RG58C cable to the UUT; the reference comes into the
> lab from another room over Andrew heliax FSJ1-50A.
>
>   
RG58C has stranded inner conductor plus 95% coverage braided outer
conductor with no foil shields.
With the single 95% coverage braided outer conductor it is a bit leaky.
Heliax has a solid corrugated copper outer conductor and consequently is
well shielded but somewhat inflexible.
> I get the same performance with and without 10 dB pads,
> and with or without RF isolators. Between the deliberate
> frequency offset, the cables, and the pads/isolators I'm
> pretty confident of the test setup. If you have additional
> ideas, let me know, and I'll give them a try.
>
>   
Pads and isolators wont help much if coupling is through the air.
A 10dB pad probably wont have enough attenuation.
At least 120dB (preferably more) reverse isolation is desirable.
>> Is there any way to measure the Allen deviation of the 10kHz GPS
>> receiver output, or at least verify that the 10KHz phase is jerked every
>> second.
>> The jerk may be done by adjusting the length of the 10kHz cycle just
>> before the second marker.
>>     
>
> Yeah, I will try to measure the 10 kHz directly for you. I was
> curious also, having never examined a Jupiter GPS in detail
> before.
> p.s. It's Allan, not Allen.
>
> /tvb
>   
Measuring the PPS jitter would also be useful.

The 5V version of the Jupiter GPS receiver is somewhat scarce, however
the 3.3V versions are less so (I have 5 plus a marine version in a
radome somewhere).
Maybe they could be useful for carrier phase discipling of an OCXO, they
are certainly relatively inexpensive.

Bruce