[time-nuts] Do we need something like this ?

Tom Van Baak tvb at leapsecond.com
Wed Jun 15 22:38:15 EDT 2005

Hi Ulrich,

Welcome to the group and thanks for your long
message. It's a very nice project and I bet you're
having lots of fun designing and prototyping it.

I'll be very interested to hear what LO you plan
to use and to see early performance results.

I have a number of questions or comments on
your project which I'll address over the next few
messages. But in summary, I very much like
the modern idea of using a DDS to close the
loop rather than a traditional VCO approach.

I suspect the additional phase noise from this
technique would be problematic in some cases
but for the majority of GPSDO applications,
including most of the ones we experimenters
use, it is a simple and useful approach.

Note that some Rb products use a DDS in this
way. See: http://www.temextime.com/ .

On the other hand, others are proud not to use
a DDS, using a SC-cut OCXO instead. See:

Several modern Cs standards use a DDS to
generate the 9.192 GHz probe frequency. I think
the HP 5071A even uses a SAW and DDS to
generate the main 5/10 MHz output. See the
design papers at http://www.karlquist.com/
which go into this in more detail. Or perhaps
Rick himself would like to comment on the
relative advantages/disadvantages of using a
DDS for this purpose in a GPSDO.

Anyway, go for it. You might also want to read
the specs and manuals at the Novatech site --
since many of their cool T&F products are
based on the AD9852. It may give you an
idea what specs you'll approach. The site is:
http://www.novatech-instr.com/ . Note also
several of their products allow a convenient and
precise 1 uHz resolution in frequency (that's
1e-13 at 10 MHz, exactly) rather than 1/2^48
(which I think comes out to something like
3.55271e-14, right?).

Lastly, there was a discussion some months
ago in time-nuts about phase microsteppers
and, if nothing else, part 2 of your project (the
DDS) could be quite useful to many of us as
an inexpensive, stand-alone stepper. Perhaps
it could even evolve into another TAPR board.


----- Original Message ----- 
From: "Ulrich Bangert" <df6jb at ulrich-bangert.de>
To: "Time nuts" <time-nuts at febo.com>
Sent: Sunday, June 12, 2005 05:47
Subject: [time-nuts] Do we need something like this ?

> Hi Folks,
> I am absolutely new to time nuts, but having studied the last month's
> topics i feel i finally found a place to discuss some of my ideas
> concerning frequency standards for use in amateur radio & elsewhere.
> To be concise: It is more than just ideas. Almost everything that i
> thought it has to be done has been realized in a breadboarded running
> prototype. I am currently at a point where everything needs to be
> ordered and a (multilayer) pcb has to be designed. Before i do, i would
> like to present the basic ideas and features of the design to you and
> leave it to your jugdement. I urgently need some feedback on it by some
> qualified guys and i believe to have found them. For reasons that become
> clearer in a few seconds, i call the device AOS, like A(rbitrary)
> O(scillator) S(ynchronizer). So, do we need something like this: 
> The design basically consists of 4 building blocks.
> 1) A Motorola M12+ GPS receiver. No need to talk about this one. The
> only perhaps extraordinary is the fact that the receiver's sawtooth
> correction value is used in the regulation loop which effectivley lowers
> the receiver's pps's allan deviation to some 2E-9 @ 1 s. This allows to
> use integrating time constants a order of magnitude smaller than without
> using the sawtooth correction. This in turn makes even some not so good
> suited oscillators candidates for use in the standard. Note: The double
> oven 10811 in the Z3801 is NOT because HP tried to make everything as
> perfect as possible! It is a NECESSARY prerequisite for the operation of
> the Z3801. With the ancient receiver technology used these days in
> conjunction with gps's SA the receiver would easily have a Allan
> deviaton of 100E-9 @ 1 s. That noisy signal needs 50 times longer
> integration time constants compared to the M12+ to give the same
> accuracy, resulting in the need for much better LO specifications than a
> single oven 10811 could give. So a double oven design became necessary.
> >From a today point of view the Z3801 is by far not as good as a lot of
> people think. As my design shows, current amateur designs can achieve
> superior results.
> 2) A Analog Devices AD9852 DDS. Basically i do not steer the LO by
> analogue electronis but use it to clock a DDS. This has (so it seems to
> me) the following advantages: 
> A) You need no voltage controlled oscillator, instead any stable source
> of frequency will do. The LO is not part of the project. Thanks to the
> DDS the user of this device supplies whatever he feels is a adequate
> source of short time stability, may be quarz, may be rubidium, may be
> cesium just whatever you like even if you thought it was not steerable
> at all.
> B) The DDS features a built in clock multiplier, so you can get out 10
> MHz on a 10 MHz clock input signal with only a slight increase in phase
> noise. In fact, due to the clock multiplier and the DDS principle the
> clock signal does not NEED to be 10 MHz but may be anything between 1
> and 20 MHz will do for a 10 MHz output. Do not hesitate to buy a good
> oscillator having a absolutely odd frequency, it will be perfect for use
> in this design. Using a all digital DDS removes a lot of the problems
> building a frequency standard that relate to the need for high precision
> low noise analogue electronics which would otherwise be necessary. With
> analogue electronics the thermo-voltages of simple solder joints can get
> a serious design issue.
> C) The AD9852 is one of the few DDSs giving 48 bit frequency resolution.
> You may easily calculate that this results in frequency steps of some
> 4E-14 relative to the desired output frequency. I thought that was fine
> enough. Agreed? Surely it is better than anything i have seen realized
> with analogue parts.
> 3) A Altera gate array of the 10K class holds almost all of the logic.
> Beneath controlling a rotation shaft encoder and a 2X40 lcd display the
> main logic necessary is of course the TIC and the 1E7 divider chain to
> generate a pps out of the locally generated 10 MHz. The TIC realized
> here has a resolution of app. 110 ps and uses a delay element scheme.
> The delay in a single element is temperature and supply voltage
> dependand. For that reason the delay chain calibrates itself against the
> gps signal on a regular basis. I studied Shera's design very intensive
> and one of the really big design clues is also one of the big design
> flaws: By limiting the delay between gps pps and a locally derived
> signal to the millisecond range it is indeed possible to use a low class
> timebase like the ordinary canned oscillator that Brooke uses without
> any limitation on accuracy of the measurement. (This topic has already
> been discussed elsewhere in the group) However, with a phase measurement
> range in the millisecond region, the LO has to be extreme close to the
> desired frequency to make the pll lock and results in the oscillator
> setup procedure that is necessary with this design. With pps signals a
> delay of 0.5 s gives the maximum phase tolerance in both directions
> early / late and that is why my dpll's 'center phase' is 0.5 s. With
> discrete devices a 10 MHz binary counter chain for 0.5 s would be
> unpractical long in terms of number of devices used, but no problem
> inside a fpga. For that reason the LO may be far off initially in my
> design because the phase measurement range is 0 to 1 s. 
> 4) A Beck SC12 microcontroller. Most possibly you have not heard about
> it because it is a German product. Google for "Beck" and "SC12" to learn
> more about it. Imagine a part with a DIL-32 footprint that contains a 20
> MHz clocked 80186, a lot of flash memory, a interrupt controller, 2
> UARTs, a network controller with a ethernet interface and some other
> goodies. The part comes with a DOS like real time operating system and
> Borland C or Borland Pascal 7 is used to write programs for it. As if
> this alone were not astonishing enough, the part comes with a built in
> web server (!) a built in FTP server (!) and a built in Telnet server
> (!), you name it. Due to the multitasking environment all this stuff
> runs concurrent and parallel to your own application. The gate array
> works as a memory mapped i/o device on the controller's bus. The
> application software is a 80 K byte Pascal program. Regulation is done
> by means of a digital pll. The pll's natural time constant can be set
> from seconds to days and a pre-filter for the samples having 1/6 the
> time constant of the pll's natural time constant can be switched on/off.
> If that reminds you to something you remember about the Stanford
> Research PRS10 loop: Not by chance! All (and i mean really all)
> parameters of the regulation are stored in a non-volatile ram on a
> second-by-second base. I case of power loss these values are loaded back
> at restart, so chances are, you still have a "lock condition" after
> restart if only the LO has a UPS. Regulation can even be switched off,
> giving you a open-loop system that is ideal to characterize the LO and
> to determine the integration time constant that is ideal for this
> specific LO and gps receiver combination. 
> All information from the gps receiver (of course in a decoded form..) is
> available at the display, that includes the information about the status
> of all used receiver channels as well as the status of all currently
> received sats. All current values of the regulation process are
> available at the display. All relevant parameters can by changed by
> means of a rotation shaft encoder and are stored non-volatile. The
> receiver may be completely reset or a 'site survey' may be initiated.
> The device works as a NTP time server in networks and can send UDP
> packets containing all relevant process information to a pc in the same
> network segment or send broadcast to all members of the segment. A pc
> software for the realtime display of this data as well as longtime
> storage for later analysis exists. Output is of course Stable32
> compatible. Note, that generating 48 bit control word for the DDS
> involves using arithmetic BETTER than 48 bit resolution. The software
> uses Borland's 64 bit 'Extended' data type for all calculations.
> The M12's serial output as well as the pps is routed to a standard RS232
> port so that the device may serve as the hardware for CNS's well known
> TAC32 software. To drive things to the top, the device can mimic the
> serial output of a Agilent 53131 counter in TIC mode on a second serial
> port. That includes the second by second messages as well as the 100 s
> averaged TIA messages. That saves you thousand's of dollars for a real
> 53131 in case you want to use the TAC32+ with the additional TIC module
> in conjunction with this device.
> In a locked condition the DDS's frequency control word is a direct and
> error free measure of how far the LO is currently off. This being due to
> the fact that in a locked condition this control word is exactly what is
> needed to compensate for the LO's offset. That makes the control word a
> extreme good measure for oscillator characterizing.
> Due to the internal phase comparison i can tell the device's Allan
> deviation for all observation times where the gps receiver's phase noise
> is the dominant factor. Due to a lack of measurememt equipment (this is
> a second project of mine) i currently can not tell exactly how far the
> LO's phase noise is deteriored by the DDS for observation times up to 1
> h or so.  
> Any comment from you on this project is highly appreciated, no matter it
> is critics or suggestion for improvements. If you feel, a lot has
> already been done, let me inform you that i am working on this project
> since some 3 years now.
> Thanks in advance for your comment. English is not my natural language,
> so sorry for typos and other errors.
> Ulrich Bangert, DF6JB

More information about the time-nuts mailing list