Re: [Thumbgps-devel] the serial alternative/radio noise

[Patrick Maupin <pmaupin@gmail.com>]

On Mon, Mar 12, 2012 at 6:22 PM, Dave Taht <dave.taht@gmail.com> wrote:

> I direct you at the 'smile plug', which is also an ideal device for
> deployment in places without reliable power.

Is checking for good network connectivity where there isn't power that useful?

> Earlier versions of this hardware (see the open-rd) had some
> interesting connectors available in addition to usb. I don't know if
> the dreamplug does - the guruplug had such lousy thermals as to rule
> it out for any usage....

Agreed.  I thought about that and then discarded it.

> I'm not wedded to the the wndr3800 as the 'reference time ticker
> beyond the internet edge' for this project.

Well, I kind of like the idea of a braindead ethernet thing, too.  That
would require a router ethernet port, but a lot of routers don't have USB,
so that could be OK.

> we certainly seem to have identified a market and price gap between
> shipped devices with bad time, and devices with good time, but we
> haven't established that usb with PPS on 'dcd emulation' can be made
> 'more right', as yet.

Right, which is why the first thing we need to build is precise
measurement capability for what we want.  (You can buy precise
measurement capability for tens of thousands of dollars, but you still
have to wire up your own stuff over to it, so I don't really see the
point.)

> So I'm kind of looking for ways to get the right parts and evaluate
> the following without having to do much soldiering, to try and isolate
> where each of the latency problems are.
>
> PPS serial
> PPS usb
> kernel pps
> gpsd w/wo r/t privs under load

We're definitely on the same page.  I'd like to be able to do this in
hardware rather than software.  Or rather, do the lowest level stuff
in hardware, and collect data for the software to analyze.

The two major pieces of this are an FPGA and a really stable frequency
reference.  For example, you can get a 5ppb OCXO that will drift less
than 4 ms in a day, and a lot less than that for short term
disturbances.  Without such a thing, you can't say for sure that what
Hal saw wasn't times when his computer got so loaded down that tick
interrupts didn't happen.

Even though the Linux drivers and firmware aren't open source, my
favorite FPGA board at the moment is the 1.2M gate version of the
Digilent Nexys2.  I've probably bought around 40 of the Nexys and then
the Nexys2 for the lab at work.  You can easily build a big board that
plugs right into it, and a few little boards that plug into it too.
If there are no objections, this is what I plan on using for the base
of the timing tool:

http://digilentinc.com/Products/Detail.cfm?NavPath=2,400,789&Prod=NEXYS2

The big FPGA version is $189.

I can easily build a board to attach that has the following:

1) Provision for modules for a few different FTDI chips (FT232H, FT232R, FT232x)
2) Provision for modules for a few different GPS devices from different vendors
3) Provision for connectors for a few different GPS devices (e.g. Garmin 18x)
4) Provision for an OCXO
5) Provision for input from a timing reference such as a rubidium clock

I can fab 4 identical boards for ca. $200.  Probably another
$200-300/board for modules depending on how crazy we get.  The
question comes with the OCXO.  Do we want to get used ones off EBAY or
buy really expensive new ones?  I'm contemplating that perhaps we
should make it where we can use multiple ones and then check them
against each other.

> The venus really does look like a nice device.

Agreed.  Nice but expensive.

> A better way to put my question is: what evidence do we have that a PPS signal over usb can provide better time?

That's evidence we need to collect.