NTP GPS LED Clock

[..] [Introduction] [History] [Mainboard] [Display] [Software] [Known problems] [Acknowledgements] [Copyright/Disclaimers] [Feedback]

Introduction / What is it

It is a clock with a big display unit, that has a GPS receiver from which it gets its timesignal, that can be connected to an ethernet network, where it can serve as a (S)NTP server. All of this based on an Atmel ATMEGA328 microcontroller.
Unfortunately, this isn't a project that I would call properly finished, and considering how many years it has been in the makings, it probably never will be. There are a number of hardware- and software-problems that were never quite worked out. I still publish it, because who knows, somebody else might reuse it or fix it up?

History

This is the successor of the NTP-DCF77-LED-Clock that I built (together with a few others) a few years ago.
It is a new design from scratch. Reasons for the rebuild were a few:

As nerdy-cool as the use of an old ISA-network-card was, it had a few downsides, namely that it took up huge amounts of space, and that those cards haven't been manufactured for many many years, making it harder to come by a working one.
The 8 KB flash on the old design were completely full, literally to the last byte, I had to shorten the command-line prompt by one byte to make it fit, and there was no drop-in replacement for the microcontroller with more flash available, making any extensions like IPv6 support impossible. In fact, it hasn't been possible to compile the firmware with recent avr-gcc and avr-libc for a few years now, as they would generate slightly larger code that wouldn't fit into 8 KB anymore.
The accuracy of DCF77 with the simple receiver used is only about 0.1 seconds, which is way too bad for proper NTP use - GPS should give much better accuracy (less jitter).
SOMEBODY dropped the old clock, causing some damage including breaking one of the boards on which the digits were mounted. It never could be properly repaired, a slack joint always remained.

The Mainboard

This is a single-sided euro-size (160x100 mm) board housing everything that is needed. Mainly the microcontroller (an ATmega328), the networking chip (ENC28J60) and miscellanous other stuff. There was also space for a certain GPS module, it could be mounted onto the mainboard - this module was later replaced by a different one which has different connector positions and is a different size. The board was designed so that the main parts (the microcontroller and the networking) could be reused for other projects, by simple replacing the stuff surrounding them.

The schematics (Eagle files) are included with the software download below.

Display Unit

This consists of three identical single-sided boards, each 200x150 mm, each containing two 100 millimeter high 7-segment digits (SA40-19SRWA) and an SAA1064 chip to control them. They are addressed through an I2C-Bus that runs through all of them (and each of the three boards gets assigned an individual address on the bus by soldering in a jumper or two resistors).

The boards were constructed with Target 3001 (because their size was too large for my low-end version of eagle).
board for two 7-segment digits (PostScript)
board for two 7-segment digits (Target3001 file)

The Software

The Software was written from Scratch in C, and compiled with AVR GCC.
Downloads:
Sourcecode v0.01 (06. Apr 2012)
With a bit of luck, you might find a newer version in this public gitweb repository.

Known problems, or: why I would not call this project finished

There are multiple problems with the hard- and software that were never properly ironed out. These include, but are not limited to:

The software

The software was never quite finished. Functionality is missing. Also, in particular the networking "stack" is a big piece of crappy code that is almost impossible to read or debug and should be rewritten from scratch.

The GPS receiver

There is a fundamental problem here that I did not realize beforehand: While you do get a time signal from GPS as a by-product, support for getting this signal out of the receiver is completely broken in the firmware of most GPS modules, even if the modules have a "PPS" pin for signalling the start of a new second.
The first module that was tried and that you can see in all the schematics was the ST01SP that AFAIK was also the base of one of the first versions of Mr. Lee CatTrack. It does offer a PPS pin, but unfortunately, had a firmware that is not at all suitable for the purpose of getting the time out of it. For starters, it did not support activating the GPZDA message (a message dedicated to getting time) in the NMEA output, so you needed to parse the GPRMC message instead. And it most of the time (but not always) had a constant 2 second offset to the correct time - apparently, because the firmware used a hardcoded value for the number of leap seconds, and there had been two other leap seconds since the module was produced that it did not know about. We also had problems mounting the module so that it would get GPS reception since it did not support connecting an external antenna.
So we finally decided to try a different module, and got the GPS-10921. Of course this did not fit our mainboard, so it is connected through wires and fitted with glue, which looks a bit ugly, but works. Thanks to the external antenna connector, getting a signal was a lot easier, and the firmware was better too, but still not perfect: It does support the GPZDA message, but you do not get the message at the start of a second, only somewhere in between. This unfortunately is very problematic: Assume you get a PPS pulse, and afterwards the GPZDA message with a timestamp of say 12:00:00.900. This can happen, because naturally the serial transmission takes some time, while the PPS signal is instantly delivered. It tells you that the pulse you just got was the pulse for 12:00:01.000, not the one for 12:00:00.000 as you might have originally thought. It means you have to make some guesses as to which second a pulse was actually for, and unfortunately, I never managed to implement a stable heuristic for that.

Bugs

There are some bugs, and it isn't exactly clear if they are in hard- or software. As the clock is now mounted to the wall, debugging them is almost impossible for me and will probably never happen because I'm a lazy bastard.

The clock display sometimes jumps back and forth, i.e. 12:00:00 -> 12:00:01 -> 12:00:00 -> 12:00:01. This is most probably an effect of the GPS problem described above but might also be unrelated.
The network stops responding after some time.

Acknowledgements

Andreas "llandon" Schwarz helped a lot with the clock, and built all of the casing.

Copyright/Disclaimers

All trademarks are property of their respective owners.
Should you try to rebuild anything shown or use anything offered here, you do so at your own risk.
Your mileage may vary.
All plans and software are released under the GNU GPL.

Feedback?

You can post feedback under the Blog Article referring to this.