A couple of weeks back I stumbled across a thing on ebay. The seller described it as an "Aircraft stable oscillator" but knew nothing more about it. I was interested for two reasons: firstly, it was made by GEC Marconi - a predecessor to the company I now work for. And secondly, from the name it sounded like it might contain an OCXO - which would have been handy to use as a frequency reference. Anyway, it was £20 delivered so I figured I may as well go ahead. Here it is: IMG_20130330_231411.jpg

Let's crack it open! Not much to see yet. The main board has some 74HC series logic, a 7805 linear regulator, a P-channel mosfet and a couple of 2n2222 NPN transistors on it. On the far right you can see a power supply decoupling board - not sure exactly what the filter topology's called but it looks like two cascaded pi filters. You can also see marks on the inside of the case which show that it's been machined. IMG_20130330_231437.jpg

From another angle you can see a thermostatic switch underneath the board. IMG_20130330_231453.jpg

Removing the PCB and thermostat and taking out the mounting posts, I could then extract this: IMG_20130330_232647.jpg

An atomic clock for £20! Excellent! I had a look at the datasheet to find the pinout. Tracing the tracks on the original circuit, it seems that this variant was a single-supply one, needing only +15V (EDIT: actually 19.5-38v, see addendum). Most of the circuit seemed to be dedicated to just switching the green lamp on once the frequency had stabilised (these units take a few minutes to warm up) and the red lamp on otherwise - so nothing particuarly interesting here.

Some variants of these could have their frequency set by RS232 but this particular model didn't seem to be one. I hooked it up to a bench power supply and measured the output - an 800kHz square wave. IMG_20130419_231742.jpg

This was slightly annoying - I was hoping for a 10MHz reference for my test equipment. I wondered how I could change the frequency. Perhaps I could multiply the output frequency by 25 somehow and divide by 2? Maybe by filtering out the 5th harmonic of the square wave twice, or using a PLL? The datasheet hints that these devices can be set to a frequency of your choice out of the factory, so perhaps i could change the frequency of mine in the same way that they set them initially? I opened it up and found that all of the interesting circuits inside were potted with what seemed like some sort of polyurethane foam. IMG_20130407_103131.jpg

Just as I was running out of ideas - being too afraid of damaging it to try removing the foam - a friend recommended that i try the time-nuts mailing list. I asked on there if anyone knew how to change the frequency of these and was met with a number of very helpful responses! Unfortunately most of the information they had was on the FE-5650A-58 option with the RS232 control but i did extract a number of useful bits of information here. Firstly, all of these devices have a 50.255MHz clock at their core (the big metal can in the picture above!) and mine most likely uses a DDS chip to create the output from that. Also, changing the output frequency isn't quite as easy as just changing the DDS tuning word since they tend to use crystals on the output for filtering.

As an interesting piece of background, i also found out from G8RPI that these oscillators are from a frequency hopping system called HAVE QUICK.

Anyway, with this information i was sufficiently intrigued to risk picking away at the foam to find out what was going on. IMG_20130419_193634.jpg With the majority of the foam removed I could see the DDS chip - an AD9955. I'm not very patient when it comes to reading things but the gist i got from skimming the datasheet was that a signal goes into CLK, a 32-bit word F is input in parallel and the output frequency is Fin*F/2^32. On this board the 32 DIP switches are clearly to set F, Another thing I noticed was that the DDS isn't actually set to 800kHz - at the top left is a 4040B ripple counter, which divided a 12.8MHz clock down by 16 to produce the output.

I wanted to verify my theory that the AD9955 was dividing down the 50.255MHz clock down to 12.8MHz. Reading the DIP switches (assuming that in to the board=1 and towards the edge=0), the tuning word was set to x4134111F. Dividing this by 2^32 and dividng 12.8MHz by this gives 50.255058MHz... pretty much spot on! From now on I'll take this as the frequency of the reference clock.

So, i now know how to change the DDS divider, but what about the crystal filter? I figured this must be underneath the board I was looking at, so I undid the nuts holding it in place and gently pryed it off. (Be careful of the header and loose wires connecting the boards near the SMA connector end!) Underneath, sure enough, was a crystal. It seemed to be connected (possibly via a resistor?) to the COUT pin of the DDS. IMG_20130419_195735.jpg The markings (156/Y-1294-2) didn't seem to indicate its frequency or manufacturer and googling them didn't help. Regardless, I desoldered it to unlock the frequency range of the DDS chip.

By some simple maths i calculated the tuning word for 20MHz was x65E15AC0. Why 20MHz? Well, i was going to have to put a crystal back in eventually, but i couldn't find any 10MHz ones in my junk box! I had plenty of 20MHz ones though so i decided to improvise... i'd make a 20MHz clock, then rewire the 4040B counter to divide by 2 rather than 16. This also has the advantage of giving me a CMOS logic buffered output. Anyway, i switched it on and probed it to find my 20MHz signal i was after. It was a bit ugly because of the lack of filtering though.


Here you can see i've bodged in a replacement 20MHz crystal. I've wrapped it in kapton tape and grounded the end, just like the original one. This one was a smaller package though, so I had to scratch away some of the soldermask from the PCB to tie the can to ground.

I fired it up again for a smoke test. 125mhzsig.jpeg As hoped, i saw a 1.25MHz square wave on the output - the by-16 divider still in place. Next job was to change the divider to by-2. This simply involved disconnecting the output from Q4 of the ripple counter and connecting it to Q1 instead. IMG_20130419_215735.jpg In doing so, I noticed that the output was decoupled with a 330pF and 100pF capacitor across the SMA connector. At 10MHz, this amount of capacitance won't do, so i... ripped the capacitors off with a pair of pliers. IMG_20130419_213241.jpg

With this done, i put it back together and revelled in my success! IMG_20130419_223419.jpg The error in frequency here is almost certainly due to the innacuracy of the counter (bought for a tenner at a radio rally) rather than of the source.

Thanks to Herbert Poetzl, Robert G8RPI Atkinson and Bill WB6BNQ - all from time-nuts - for their invaluable advice with these frequency standards! After I found one of these inside the Mysterious Aircraft Part, i decided to buy the seller's remaining stock (another 6 of them) since a rubidium frequency standard must be worth more than £20. I'm now selling them on ebay and I'm considering offering them with 10MHz modifications. Once I've extracted the frequency standards, I'm left with a very nice machined aluminium casing which will make a great project box - it's aluminium with fins so will be good for high power dissipation stuff. Audio amplifier project maybe? The cases also seem to be IP67, with the power switch and connectors rated thus and a rubber gasket around the edge where the two sections come together. Perhaps it's time for an underwater electronics project!

ADDENDUM 1: Input Voltages

Since i wrote this post, Robert G8RPI has pointed out that these units aren't actually designed for a 15v input voltage! They actually have a switching pre-regulator on board, as you can see below. IMG_20130407_102937.jpg This buck regulator is set with a 91kohm resistor to give a nominal output voltage of 15.3v. I measure it to actually be nearer 15.5v. With a dropout voltage of 4V, this gives a permissible input voltage range of 19.5-38v. This is apparently so it could be used directly from an aircraft's 28v supply. I could have easily removed this board by undoing the screw, prising off the edge which is potted to the Rb package, desoldering it from the main power supply board and placing a short between where the ECW and the yellow wire used to go. I decided to keep it for now though - it will be quite handy to be able to run this thing from an old laptop charger!


So, some people have gotten a bit upset that I bought all of these and am selling them on for more. I appreciate it's frustrating but... hear me out. For £20 I bought a completely undocumented part on a hunch - for all I knew it could have been completely worthless. I took on that risk of losing out. Once I'd found that it was actually worth something, I spent hours of my time dissecting and documenting it here. So I'm now selling on a documented part - you know what's inside it and you don't have that risk of losing out! Also, I have college bills to pay :) Tl;dr: I charge a premium for the elimination of risk and for my time.

I've listed a complete unit on ebay since it seems like some people would be interested in getting hold of the whole thing.