I am on a similar search for accuracy. What I am thinking of doing is using a GPS module with TCXO as the clock signal source to replace the crystal on a Teensy. Then when it isn't locked, I still get TCXO accuracy, but when it is, ppb GPS timing accuracy. I was thinking I'd program it for 8MHz output and double it.
Thank you for the tip on the clock multiplier. Much simpler than how I was planning on doubling the clock.
I have a DIY Mall branded USB GPS module labeled G28UF8FUSB, which came up in a lot of Chinese language searches. They seem to indicate that this does have a TCXO on board. There are pads on the board labeled for a separate connector including PPS. This will do for now and is simpler to reprogram with the built-in USB.
I was just about to follow up that last post of mine to refine various of my then untested hypotheses before discovering your follow up post. If you're planning on buying a 3N502 clock multiplier chip as per my suggestion, don't bother (unless you need to add an external 10MHz reference socket to something like an FY6600 which uses an internal 50MHz reference).
Sadly, it doesn't solve the issue of the "Leap 48MHz TCXO clock period adjustments" it uses to keep the PPS synchronised to GPS or UTC time (regardless of the programmed output frequency). If you can monitor the third harmonic of an 8MHz PPS clock setting using a multimode HF receiver or transceiver using the narrow band FM mode, you can clearly hear these "leap cycles" as ticks every few seconds to 3 or 4 a second, depending on the actual frequency of the 48MHz TCXO clock. The 3N502, good as it is for providing very low jitter clock frequency multiplication, can do nothing about this peculiar source of phase noise as I eventually learned from my experiments using the 3N502 chip after finally getting round to setting one up in a SOT8 to 8 pin DIP adapter board to plug into a solderless breadboard lashup yesterday.
It was only this afternoon, after completing the last tests with this rather fragile lash up, that I finally came to realise the scale of the problem and started a fresh search for a suitable VCOCXO module, not because "It's the done thing" but because it's the only sensible way to proceed in this case if you want to go beyond substituting for an absent 10MHz WWV frequency calibration source.
Whilst the raw 10MHz PPS output option is a very good substitute for the now impossible to receive WWV 10MHz broadcasts (sunspot minima period) to effectively trim an HF rig's internal reference oscillator to within half a Hertz, it's certainly no use as an external 10MHz clock reference for communications gear and arbitrary wave function generators and the like. The plain fact is that there's no viable alternative to that of using a VCOCXO, disciplined by the GPS module as your low phase noise carrier source.
You might get away with the much cheaper option of a VCTCXO but the issue of "retrace" rears its ugly head in this case. Still, a VCTCXO might be worthy of consideration since, other than spending real money on a new VCOCXO (or lucking out on a cheap used unit), it should be a straight forward upgrade option, assuming you've made allowances, spacewise, in your home brewed GPSDO to take the older more bulky VCOCXO modules. You'll have already dealt with the issues of designing and assembling the long time constant PLL circuitry by which to discipline your VCTCXO module which will certainly be just as effective a way to deal with those pesky phase correcting jumps in the raw PPS output.
I had no Idea what a "Teensy" was until I thought to try googling it just now. :-) If, as I gathered from the results, you only need it for precision timing where these "leap phase adjustments" shouldn't be an issue, then you can use the raw PPS output. If you'd used a u-blox NEO 7N or M8N, you could even have done away with the clock multiplier chip since they can be programmed to output 16MHz directly.
However, as I've just discovered by testing this, with a divide by 3 factor involved, it lands up alternating between a duty cycle of 33.333 and 66.666 percent on every leap phase adjustment! I tried it with an 18MHz setting which works fine, giving a similarly jittery trace to the 10MHz one (ISTR Scully of scull.com fame discovering the limit being just over the 19MHz mark, hence my bothering to test the 16MHz option).
Wow! This GPS module never ceases to surprise me. :-) This issue with alternating duty cycles every few seconds to 3 or 4 times per second, may, or may not, be a significant factor in your application. I think, if your GPS module works similarly (and with a 48MHz TCXO), you might be needing that 3N502 mulitplier chip anyway. The resulting phase jumps correspond to 30 deg at 8MHz which, using the 3N502, will become 60deg jumps at 16MHz. Again, this could prove problematic in your usage case (but less so than alternating between 33/66 duty cycle on every corrective phase jump when the module is programmed to generate the 16MHz directly on its PPS pin).
Continuing, for the benefit of anyone else interested in this particular NEO M8N module, I did eventually find out why the built in patch antenna wouldn't deliver the goods. Basically, it amounted to the manufacturer neglecting the presence of circuit traces scattered within the ground plane upon which the antenna had been glued, using the very common (as I eventually discovered) method of a very thin adhesive pad.
It was only when I started searching for cheap passive patch antennas to test with, that I spotted the protective peel away backing tab displayed in all the illustrations. It was this which inspired me to desolder the feed pin joint so I could use a thin craft knife blade to prise my "prize", the free patch antenna, away from the module. I hadn't been completely successful with the desoldering tool but the only collateral damage was to the through plated hole used by the feed pin which is of no further consequence anyway, with my having discovered the true cause of the antenna failure in what lay out of sight beneath it (pictures to follow).
I soldered a short 15cm length of RG174 type co-ax to my prize, formed the other end of this co-ax into a co-ax plug shape, plugged it into the antenna socket and got even stronger signals than my 3/4 wave wire antenna had given me. Indeed, after temporarily soldering it to the socket terminals, it was almost impossible to lose satellite lock by shielding it with both hands in my first floor office/workshop, now 'man cave'. To do that I had to go to a ground floor room in this three storey Victorian house and it now locks back onto the satellites in a matter of seconds as opposed to quarter of an hour when I'd bodged a solution to the 'deaf LNA' with a 33K bias resistor to raise the RF input pin bias voltage from its low of 0.61v to 0.81v to approximate the 0.83v 'typical' figure given in the data sheet, which bodge had allowed the GPS module to just about manage to receive usable satellite signals from the built in patch antenna.
This bodge solution still couldn't match the results I was able to get just by sticking a 14cm 3/4 wave piece of wire into the antenna socket so I landed up removing the suspect LNA chip altogether. I had hoped to discover a dry joint but gave up trying to solder it back onto the board when I realised the utter hopelessness of my achieving such an impossible task in the face of such insane micro-miniaturisation (the damn thing was barely bigger than the smallest smd resistor FFS!).
I'm guessing that previous revisions of this module hadn't been cursed by such signal carrying circuit traces on the 'back side' of the board, allowing such a misplaced patch antenna to work tolerably well. Manufacturing inertia seems to have led to this egregious error in mounting a patch antenna right on top of strong sources of RFI, rendering it as nothing more than an ornamental adornment.
The patch antenna base plate relies on capacitive coupling alone to the module's ground plane which works well provided that it doesn't
also capacitively couple to strong sources of RFI as was obviously the case here. In all likelihood, there wasn't anything wrong with the microscopic LNA chip or any of the other components involved but I'm unlikely to know despite keeping hold of that much abused LNA chip in a small glass jar as testament to the folly of such extreme miniaturisation.
Having done pretty well all of the research on this particular GPS module that I'd originally been seeking advice on, my conclusion is that, aside from the waste of a perfectly good patch antenna and, most likely, the microscopic LNA chip as well, since it's using the NEO M8N GPS module which is very well specified for the task. It's RF sensitivity is such that the use of an on board LNA with the patch antenna is hardly necessary since it can retain lock just from the stray pickup on the strip line connection to the EXT ANT socket, receiving almost as many satellites with a passive patch antenna indoors as it does with an externally placed active patch antenna.
I think the only reason why the RFI was only upsetting the built in patch antenna and not the receiver in the u-blox module itself was simply because it had overloaded the LNA chip so severely it had totally desensitised it to the point where very little signal was able to get passed onto the output pin. The low bias voltage on the RF input pin was very likely just the side effect of this overloaded condition. It seems just dumb luck that this mistake was so huge as to effect damage limitation to just that of the patch antenna feed, preserving the main receiver section for externally connected signal sources.
If this hadn't been the case, the whole module would have been 'deaf' to signal reception on the EXT ANT socket as well, leaving the only (reversable) solution as removal of the output coupling capacitor link to the main ANT input line to break the signal path. Luckily, this wasn't the case, allowing a simple experiment with a 14cm 3/4 wavelength wire antenna to prove the rest of the GPS module as being fit for purpose.
There may well be other brands and/or models of GPS modules similarly afflicted by this "Deaf Integrated Patch Antenna" syndrome so my experience of the problem might well apply to those as well. With care, such modules could equally benefit from a "Patchectomy", allowing the excised patch antenna to be repurposed as a separate plug in "Test antenna" via a short 15 to 30 cm fly-lead with co-ax plug of your choosing.
The trickiest part of the patchectomy operation ime being the successful desoldering of the one and only through plated hole connection to the antenna feed pin. As long as the antenna itself isn't damaged, some damage to this through plated hole in the PCB is of no consequence since you'd have to be insane to even contemplate fitting a replacement antenna back onto the board once you've tested the extracted antenna with a soldered on fly-lead connection (aside from gross physical damage, there's little that can go wrong in such a simple design of antenna).
Rather intriguingly with the u-blox control centre software, I noticed a section on disciplined oscillator options (external and internal oscillators) with half a dozen DAC types and protocols on offer, suggesting that, with the right hardware in place to control a VCXO (TC or OC type) it might be possible to create a simplified GPSDO without resorting to separate microcontroller and PLL hardware. Unfortunately, the "Help" file has been no help in explaining this or any of the other, many and varied options available in the configuration menu. I guess some more search engine activity is called for.
One other curious observation I've made relates to the additional 5 metres of RG174 antenna extension cable which arrived a few days ago to supplement the 5 metre external active mag mount patch antenna I've got placed on the flat roofed bay window of my 'man cave'.
I'd checked out the dB loss rates per hundred feet/metres and had extrapolated from the 1GHz loss figures an attenuation figure of some 5 to 6 dB which I'd thought was on the optimistic side. When I did comparative tests, the extra five metres of RG174 seemed to be only offering an additional 2 to 3dB of loss. This was a far smaller loss figure than I was expecting. I'm absolutely certain that a 5 metre length of even the finest RG174 can't have less than 5dB of loss at 1.6GHz which rather suggests that the input impedance is significantly higher than the nominal 50 ohm rating of RG174 cable, allowing for a mismatch induced voltage magnification at the receiver front end. Possibly the active antenna's five metre feeder mismatch loss was causing a modest voltage drop of around a dB or two with the additional 5 metre cable introducing a mismatch induced voltage magnification of another dB or two which could account for the overall reduction in observed loss compared to the expected 5 or 6 dB based on a reasonably close impedance match between the receiver input and the co-ax.
I suppose it's just possible that the apparently haphazard LNA chip T connection to the antenna input line may have been the result of a design to keep the impdances close to the nominal 50 ohm match to the antenna feeder cable, possibly putting it at the lower impedance tolerance limit until I disconnected the LNA from the antenna input circuit, raising the impedance sufficiently for just such an unexpected result to arise. I guess I'll be able to get a more reliable assessment of the true cable loss when I can get hold of a 10 to 16 dB SMA attenuator pad to plug into the module's antenna socket to provide a more consistent 50 ohm termination for testing the antenna cable losses.
Normally, such additional attenuation would never be deemed a benefit in a GPS antenna setup hence my not having such items to hand. I don't think my collection of F plug/socket attenuator pads are rated beyond 1GHz so rather than buy F to SMA connector adapters, I think I'd be better off ordering some SMA pads instead if I were to be so inclined as to do such testing (I might if the price is right - it's not vital, just a case of satisfying my curiosity over this mystery of a seemingly ultra low loss antenna extension cable).
Pictures as promised. The first three are self explanatory. The last two IMG_9484 and IMG_9486 are the "Before" images for comparison.
JBG