[time-nuts] Changes in commercial GPS clocks over the decades

[Magnus Danielson]

Hi,

On 2020-10-30 21:37, The Fiber Guru wrote:
> During my telco career I was responsible for Network Synchronization and
> witnessed several generations of clock designs.  Post-telco I now
> manufacture and sell Network synchronization systems.  Here are a few
> observations from legacy and modern topologies:
>
>  
>
> 1. BITS clocks used to consume an entire 8 ft rack in a large central office
For those not in the Telco world. BITS is the US/ANSI term for the
international ETSI/ITU-T term SASE. In telco system you have a central
clock in the heart of the telephone station, so you route any
synchronization signal incoming to the station through the SASE/BITS
interface over to the SASE/BITS which then selected amongst available
sourced and configuration locked up it's OCXO or rubdium to that and
then distributed that out back. A modern SASE/BITS is a 6 U box.
Applicable standards are ITU-T G.781 for routing, ITU-T G.812 for clock
qualities and ANSI T1.101 for the specifics of the US system.
> 2. Legacy clocks easily cost $35k to $50k
>
> 3. The most critical part of clock installation is the antenna......this has
> never changed.  If you get this wrong the clock will flop around like a fish
> out of water
In traditional SASE/BITS installations, they never had a GPS antenna to
start with, the system was designed to have analog cesiums as clock
source, and frequency errors of 1E-11 was tolerated for the Primary
Reference Clock (PRC). The PRC is specified in ITU-T G.811. Now the
companies is competing in inventing harder and harder specs on PRCs
based on what they modern cesium technology can deliver, but with
marginal benefit to operators, which tend to follow the ITU-T a little
too much because that used to be a safe route and a recepy for working
solutions.
>
> 4. Most critical part of antenna installation is to have unobstructed view
> of the sky, but not be the highest electrical element (for lightning
> protection....come of protection).  Of equal importance is selecting the
> proper size antenna cable for the required distance (RG58 up to 100 ft,
> RG213 up to 300 ft, LMR400 up to 600 ft)
DHS has a good guide for GPS/GNSS antenna installations. Some of it will
be overdoing it for the hobbyist, but it could be a good read anyway. On
the latest upgrade I was surprised how much details they chose to show
on effects of jamming.
>
> 5. Always, always, always install a lightning arrester.  Get away from me
> with, "But we don't get lightning strikes."  Arresters are WAY less
> expensive than GPS receiver modules in commercial clocks
Well, a problem doing that is that contrary to telecom installations,
which have clear grounding stategies, see ITU-T K.27, most of us do not
really have a suitable grounding point to do that. The lightning
arrester as such is indeed not expensive, but for it to be efficient you
want a low-inductance path for the lightning current into ground, and
this is usually not available in a hobbyist setting. Other than that I
agree completely.
> 6. The latest "smart ocxo's meet the performance of legacy Rb oscillators at
> a fraction of the cost, and last much longer with very low heat dissipation
Competing with rubidiums have been a challenge for some OCXO vendors to
take a piece of the market. It's a place where there have been a market
for stable OCXOs, and we should be thankful for that. Many BVAs found
their way to the market in telecom, in fact I have an Oscilloquartz SSU
(Synchronization Supply Unit) with double BVAs here, the SSU do the
SASE/BITS function from above. Other vendors did other developments of
their designs, but BVAs got famed.
>
> 7. Since they don't use Composite Clock signals in Europe (64kbps with an
> 8kHz error rate), almost none of the Network Synchronization systems outside
> the USA provide these TDM outputs, but thousands of small carriers and
> government sites still require it
Remember that this is an international list, so you do not write from
inside US about the outside. In Europe we chose to normalize all
synchronization to use the 2048 kHz rate, which is the base-rate of the
European PDH multiplexing structure. The US PDH multiplexing structure
has a several different paths, some of which the European did not have
nor felt it needed. They have different history.
>
> 8. IEEE 1588 has been really slow to catch on.  Our first PTP product was
> available for a decade with extremely low sales
The teleco standards for synchronization used o build on a set of
standards and transports which could be engineered from the top down and
it worked. You had standardized products that multiple vendors was
competing to deliver and you could buy from these, combine into a
network and the system would work. It was a nice eco-system. However,
with packet networks the basic bounds of PDH/SDH is broken and when
trying to redo the trick with PTP the ITU-T standards have been an
uphill struggle and latest development does not even work, which have
made the operators very nervous. PTP when done correctly over fully
supported networks is nice, but the networks does not support it fully,
the implementations is full of bugs and the bounds of contributions from
the network you used to have is no longer there. Only locally you get
PTP to work, and this means only cheaper GPS/GNSS clocks to PTP is
meaningful as it turns out to be hard and expensive to build a classic
synchronization network, and therefore the PRC type clocks is fairly
useless and overpriced.
>
> 9. We started with grandmasters, but now also have boundary and transparent
> clocks to cut down on segments
As needed, but separate clocks is troublesome as they need to be
integrated everywhere.
> 10.  I've sold clocks for $50k, then we got them down to $30k, then $15k, to
> $7k and we now sell $millions of clocks sub-$5k.  That's all in a 15-year
> span.
>
> When I present material on Network Synchronization, I always point out the
> irony that multi-gigabit transmissions systems are synchronized by clocks
> that rely entirely on a 
>
> 1 PPS signal (derived from GPS to discipline an oscillator).  So our fastest
> networks are dependent on our slowest (but highly accurate) signal!
Which many times is not really true. For most interfaces you do not lock
up the bit clock at all, but it is free-running. It will be true when
you do Synchronous Ethernet (which is replicating the old SDH frequency
distribution over Ethernet by locking up the bit clock and recover the
phase from that), but very few do that. I know dominant vendors which
for a long time could not lock up their links, and now only a handful of
the boxes can. Most of them still keep +/- 100 ppm and think that is
high precision. This is then just part of the modern mess that makes the
good old telco stuff actually look kind of good even if we suffered from
other limitations.
>  
>
> I enjoy reading your comments and learning about your projects.
>
>  
>
> Daniel B. Burch (The Fiber Guru)
>
>  <http://www.fiber.guru/> www.fiber.guru

Welcome.

You are however far from the only one with telco background here.

Cheers,
Magnus