[ntp:questions] ESR looking for good GPS clocks

A C agcarver+ntp at acarver.net
Sat Mar 3 22:50:12 UTC 2012

On 3/3/2012 14:08, Michael Deutschmann wrote:

> Although that brings up one other idea -- a wireless GPS clock.  That
> would be basically a relay that listens for long-range-line-of-sight
> signals, and then synthesizes a short-range-building-piercing signal for
> its owner to use.  Only a simple pulse modulated signal would be needed --
> the seconds could be labeled with NTP over 802-whatever wireless.

You could pull that off as long as you had a good modulation scheme that 
could ensure the pulse wasn't distorted.  Mostly likely the solution 
would be to just borrow from the UWB/SS design of GPS signals and create 
a pair of orthogonal chipped signals.  One signal represents the low 
side of the pulse and the other is the high side.  Flip the pulse at the 
PPS transition and the autocorrelator on the other side should spot it 
pretty easily even in a noisy environment.

> Back to ESR's plans, reading the comments I think he may actually not be
> adventurous enough.  He estimates the budget of the volunteers who will
> run his sensors at under $200 and absolutely zero hardware work -- not
> even building an enclosure for a bare circuit board.  Of that budget, $100
> is taken by a linux router that will run his surveying software.  So he
> figures a $75 maximum price for his GPS fob.  (He could easily make this
> if not for the PPS problem.)
> It might make more sense to ask the hardware designers to invent a linux
> router and timing-grade GPS *in one box*, making the entire budget
> available to the custom part.  Also, the USB conversion overhead would be
> gone -- you could even weld the PPS signal to a dedicated IRQ.
> Such a box would be just as popular with NTP hobbyists outside ESR's
> project as a USB fob.  The sort of NTP hobbyist who rejects "router"
> boxes in favor of a customized old desktop will be holding out for RS-232
> anyway.

Most of the routers that are available already have GPIO pins hiding in 
them.  If not a router then one of the new low-power plug computers with 
some GPIO in them as well.  The time stamps can be read in via whatever 
convenient serial method is available but a GPIO could be bound to an 
interrupt servicing routine in the kernel then pump PPS onto GPIO.

My current router is a version 1 Linksys WRT54G and it has 8 GPIO, four 
of which go mostly unused.  Lots of drivers have been made for 
bit-banging the GPIO pins to do things like add SD cards for additional 

More information about the questions mailing list