Tom, I for one would like to see what you have down that "rabbit hole" regarding the new architecture of the CNS clock, that does not do variable delay line sawtooth corrections. Dana On Sat, Aug 15, 2020 at 1:41 PM Tom Van Baak <tvb@leapsecond.com> wrote: > Hi Ed, > > > I'm also aware that some people have come up with external > > sawtooth correction, using programmable digital delay lines > > and special control circuits. > > For decades fellow time nut Rick Hambly has done hands-on work with GPS > receiver timing, including sawtooth correction. Here's a recent version > of his classic series of presentations: > > > https://www.cnssys.com/files/TOW/High-accuracy_Time_and_Frequency_in_VLBI_2019_sem.pdf > > His "CNS clock" was an example of a GPS/1PPS source that used the delay > technique to reduce sawtooth effects. The paper is worth a look; lots of > good info packed along with plenty of plots and photos. > > Note that his most recent version of the CNS clock no longer uses the > delay technique. I can explain more if you want to go down that rabbit > hole. > > > But, why do all that if you can just fix the clock instead? > > Because it is not possible for you or me to "fix the clock instead". > That's a complex business and design decision inside the GPS chip > itself. Almost every DIY or commercial GPSDO uses an off-the-shelf GPS > receiver chip or PCB, so you work with what you can get. Those GPS > modules all use some tiny carefully-specified free-running internal > oscillator for signal sampling, computing, and output pulse generation. > > Trimble was in a unique position when the GPSDO era began because they > made both GPS chips and a GPSDO product. For them the integrated OCXO > and GPS receiver design was possible. Even hp couldn't do that. > > > I think the clock is a Motorola brand odd looking TCXO, labeled "19096" > > or something like that, probably 19.096 MHz, as I recall. > > That sounds right. See http://www.leapsecond.com/pages/vp/sawtooth.htm > and note Dr Clark's mention of 9.54 MHz, which would be half your 19.096. > > > 1. If the GPS RX module's internal clock is synthesized (to the same > > nominal frequency) from the 10 MHz output of the GPSDO, can that > > alone eliminate or substantially reduce the sawtooth effect? > > Yes. But it would also create a list of other technical concerns. I'd > probably rather have to deal with sawtooth than having to deal with DDS > artifacts or phase noise requirements. And now your GPS chip won't even > boot or work right if the user's 10 MHz isn't there, or isn't accurate > enough, or is still warming up, or if the user's 10 MHz has too much > phase noise or instability. > > > 2. Does the T-bolt actually do this, and if so, is that all it takes? > > Yes, we think so. But remember they had control over everything; the GPS > chip, the firmware, and the specification of the OCXO. > > > If either answer is yes, then I would think the GPS RX makers would > have provisions > > for external clock reference, at least for certain high-grade timing > type models. > > Yes, very high-grade timing receivers require, or allow, external > frequency inputs. I've seen some in operation at national timing labs. > When I heard they cost $20k I decided sawtooth correction wasn't so bad > after all. ;-) > > It seems sawtooth bothers you. There are sawtooth effects in many parts > of technology; from printer dots to PC clocks; from calendars to leap > seconds. The goal isn't always to eliminate but rather to fully > understand the effects. The fact that GPS timing receivers output > quantization corrections should be seen as a good thing; it gives the > user maximum information. > > /tvb > > > On 8/13/2020 3:35 PM, ed breya wrote: > > I have often wondered about all this sawtooth correction stuff, and I > > think I've asked here too, but never got a definitive answer. Every > > time this comes up, there are all sorts of explanations of the > > characteristic, and inevitably someone mentions the T-bolt having its > > internal GPS clock synchronized with the desired ideal 10 MHz output, > > either eliminating or greatly reducing the sawtooth effect. But, to my > > knowledge, nobody has said for sure if this is indeed the case, or > > whether that's all it takes to achieve perfection, or if more magic is > > needed besides synchronizing. I'm also aware that some people have > > come up with external sawtooth correction, using programmable digital > > delay lines and special control circuits. But, why do all that if you > > can just fix the clock instead? > > > > Many years ago I looked at my Motorola Oncore VP (or whatever model is > > used in the HPZ3801A), to see what its clock was, and if it was > > reasonably possible to synthesize it from the 10 MHz. I think the > > clock is a Motorola brand odd looking TCXO, labeled "19096" or > > something like that, probably 19.096 MHz, as I recall. That's about as > > far as I got. Since then I've just wondered what would happen if it > > was synthesized from the 10 MHz, or if it was even worth trying, or if > > it would result in other problems. Maybe a certain amount of dither is > > necessary for proper operation. > > > > So, here are some questions, that if answered, may go a long way > > toward possible improvements in our GPS stuff. > > > > 1. If the GPS RX module's internal clock is synthesized (to the same > > nominal frequency) from the 10 MHz output of the GPSDO, can that alone > > eliminate or substantially reduce the sawtooth effect? > > > > 2. Does the T-bolt actually do this, and if so, is that all it takes? > > > > If either answer is yes, then I would think the GPS RX makers would > > have provisions for external clock reference, at least for certain > > high-grade timing type models. > > > > 3. Do any GPS RX modules have such provision? > > > > That's all I can think of for now. > > > > Ed > > > > > > > > _______________________________________________ > > time-nuts mailing list -- time-nuts@lists.febo.com > > To unsubscribe, go to > > http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com > > and follow the instructions there. > > > > > _______________________________________________ > time-nuts mailing list -- time-nuts@lists.febo.com > To unsubscribe, go to > http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com > and follow the instructions there. >

Hi Keep in mind that a GPS receiver is *very* sensitive to close in phase noise on the clock source. If you dig into the data sheets on the chip sets, they are quite explicit about this. Any approach you use to steer the TCXO will need to be very quiet. Next, the internal firmware *assumes* that the oscillator does not move (much). It needs to take care of things like doppler on the each of the sat’s it is tracking. Doing so with a clock that moves too much likely will not work well. One of the issues with the early clock sources for these devices was perturbations in the frequency vs temperature curve. They spent a lot of time re-designing crystals to “clean them up”. The issue was simply that a perturbation moved the frequency faster than the firmware could keep up. Lots of fun ….. Bob > On Aug 17, 2020, at 4:53 PM, ed breya <eb@telight.com> wrote: > > Thanks all, for the info on this issue. It does indeed look doable to experiment with phase-locking the GPS RX module's internal clock to the ultimate 10 MHz GPSDO output. That is, doable, but not necessarily easy or without problems. > > I dug up my old notes and info from about ten years ago, and found I had studied it quite extensively, and had some various schemes sketched out already - I forgot about all this. In the notes I found some pages from US Patents 4,785,463, and 5,745,741, which are pertinent to the Motorola Oncore RX units. I also found page 26 of a document "Remote Frequency Calibration: The NIST Frequency Measurement and Analysis Service," which talks about it too. > > The internal clock appears to be 19.096 MHz nominal, +/- 2 ppm, from a TCXO, that may or may not have voltage tuning too, depending on the model. The '463 patent appears to be about the original overall design, while the '741 is about reducing the sawtooth error by doing coarse digital corrections in the counter system, while also fine tuning the VCTXO, to get an integer clock frequency under all conditions. It mentions the Oncore model as prior art, and the resulting sawtooth error. > > As I understand it, the overall process disclosed is to allow for the TCXO to drift to any frequency in range, but automatically tweak it slightly to make sure it's an integer (Hz ) value, and set the counter system to divide by that same integer value, so the 1 PPS output is consistent, without sawtooth error. So, if the clock is nearly exactly right on, the counter divides by 19,096,000, and it figures out how to fine tune the clock to keep it there. If the drift goes beyond the fine range to say all the way to the upper stated limit of +38 Hz, the counter is set to divide by 19,096,038, and the clock is again tweaked to keep it close to that integer Hz. How it does all this is disclosed in the patent, but I haven't studied it enough to say any more. > > It looks like this improvement was in a later model, or was perhaps never actually used, since this was around the time that Motorola was departing the GPS business. I assume the older Oncores like mine do not have any of this improvement, so are subject to both the clock drift and the sawtooth. But, one thing I got from this, is that if the nominal clock and divide numbers match, and are fixed at 19,096,000, then replacing the original clock with a sufficiently clean synthesized external clock should work too. > > I have a number of possible options, depending on the actual original TCXO. If it's also voltage-tuned, but that isn't utilized, then it can stay, but needs circuit mods to release and access the tune signal. The clock signal is accessible at the TCXO, or possibly less risky in 2X form 38.192 MHz from the downconverter IC. This would be the best option, to make an external PLL to tweak it. BTW I have a spare GPS RX unit - I would not risk taking the Z3801A out of commission and messing around with it. > > If the TCXO has no tuning ability, then a new clock signal needs to be made. Of all the schemes I sketched out back then, the most straightforward seems to be simply adding 4.096 MHz from a VCXO, PLLed to 10 MHz, and 15 MHz derived directly from the 10. These can all be scaled up in frequency in various arrangements, and use standard binary frequency XOs. The next notch up for instance, is 8.192 MHz + 30 MHz giving twice the clock, and so forth. Unfortunately, this method is additive in frequency, which I hate - I prefer to take the difference of two much higher frequencies, which is so much easier to filter. I don't yet see any ways to do a difference method without using special frequencies, so for now I assume regular old standard XO frequencies will have to do, and the filter designs will need to be fancier. The PLLing seems to be straightforward. I can get a decent 16 kHz comparison frequency with simple integer dividing, like 4.096 MHz/256 = 10 MHz/625, but would like to get it as high as possible, without resorting to fancier schemes. Fractional-N is not in the cards here, I don't think. > > It would be nice if for some reason a 19.096 MHz VCXO module - or even just a crystal - was available for cheap, but I doubt such exists outside of the Oncore line. Maybe a TCXO from a junker would do, but again, it has to have voltage tuning too, and I don't know if any had that. (One of my more far fetched schemes even pictured thermal tuning - TEC heating and cooling the TCXO module - but imagine the nightmare of characterizing the part and the dynamics, and the mechanical and control loop issues.) > > So anyway, I think I can do it with the additive VCXO combo, but doubt I ever will - there are too many projects to worry about without reviving this one, but it's fun to ponder. > > BTW one last thing is that in my collection, I have parts of an ancient (ca 1990), huge Trimble L1/L2 GPSDO. I was going to share some info on it, but it will have to wait for another time. I find it very interesting. > > Ed > > > > _______________________________________________ > time-nuts mailing list -- time-nuts@lists.febo.com > To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com > and follow the instructions there.