Tektronix CSA907T Pattern generator(700Mbit) Teardown and repair

[Berni]

So i happened to come across one of these on ebay and there was not much interest on the bidding so i gave it a shot and ended up scoring it for 40 bucks!

If you are looking up information on its good to know that this was originaly made my Microwave Logic Inc and sold as model GigaBert GB660. The manual for that is a bit better than what you get from Tek

So anyway two weeks later it shows up at my door. It indeed looks in good condition. So i give it some juice and turn it on, makes quite a fan noise and as it boots it shows "RAM CORRUPTED" on the LCD before it finished booting normally, but turning it off and back on didn't show up that message again. Things are not looking that great anymore.



So i hook up a scope to it and twiddle some knobs to see if it works or not. Everything seamed to work great but then i noticed something odd. I could never get the clock signals to be above ground. The clock voltage offset adjustment seamed to work fine for negative voltages but when going positive it just doesn't want to go higher above ground. This is the most positive that it lets me go.



Okay technically this could be solved by simply capacitively coupling the clock lines but hey i have to do something about the noisy fan anyway. So i take off the Tek blue covers, all exited to see whats inside it and i get this....



Okay another internal cover.....wait... why does this thing need like a milion frigin screws to hold it on?!

Also noticed some writing on the side, not sure what it means, probably quality control or something. Also someone was in here before me since the warranty sticker is skillfully bypassed.



A whole lot of screws later... (Yes that is JUST to take the cover off)



Aaaaand were in like flin!

[Berni]

So lets have a look at what boards we got in here.



So here we have the CPU board. Nothing special here, just a 8088 CPU with a bunch of memory. Tho one of the ROMs looks like it might have been upgraded, il also replace the sticker on it to cover it up.

[Berni]

Movin on from there we get to the first actual interesting board.

So this is the board that generates a 1 to 700MHz clock signal. The small addon board is an option for 1KHz clock resolution, certainly welcome to have.



Now things are getting rather annoying as the boards are trapped in a tangle of coax cables, but this is one of the most important boards in the whole unit. This is where the patterns get generated, not quite sure how but they do.

[Berni]

Now the board before was tangled in coax but this one is like 10 times worse. There is coax everywhere, a lot of it soldered directly to the board even. After some wrestling with it this is as good of a picture i could get.



And here is the PSU to end the tour. Nothing special really just a switching module, but it does have nice quality caps on it and is rated for 170W so quite beefy too.

[Berni]

So checking the menus for options we indeed have the 1KHz clock resolution and we seam to have 32Kbits of memory. As for PRM i have no idea what that is. These options had to actually be looked up in the old Microwave Logic Inc manuals as the Tek ones do not mention them anywhere.



So on to fixing this thing. Easy thing first the fan...well its just as annoying to get to as everything else in this thing. You need to pull out all the cards next to it to get to it, then you need juuuuust the right kind of screwdriver to get in since if its too long it will hit the backplane, if its too thick it won't reach trough the fans hole and to top it off one of the board supports is in the way and hasto be removed to get to the last screw on the fan. This is ridiculous.



Okay so here is the trouble maker. Spinning the fan by hand makes it pretty obvious that the bearings in here are shot. No wonder its making such a horrible noise. Turns out it doesn't really need two fans. There are only two openings on the entire case so as long as one fan is blowing the air in, it makes it around the case and blows out the other hole. Tested it running with just one fan and its fine.



Next thing is the RAM CORRUPTED error that shows up every so offten. Going trough the manual tells me that this is caused by the backup battery being dead, but its not as bad as it first might seam since the battery backed ram only stores the previous setup rather than any calibration data. So lets see the battery and yup that's totally dead.



Since it doesn't seam to need the battery to work properly i simply removed it just in case it leaks. Its dead anyway. It seams like someone already replaced the battery in here in 1994 and by the looks of it this is a 3 in series Ni-Cd battery. No wonder it dies, this sort of operation is not what this chemistry likes to do.

[Berni]

Okay now we get to the not so easy fix. Why is the clock output always negative? Best place to look seamed to be the last board that is of course by far the most tangled in coax everywhere.



Look at the mess of coax behind it too. Makes it look like its a delay line or something, but doesn't seam like it is. It has already taken quite a bit of coax wrestling to get it out this far.



So by disconnecting and untangling more cables i finally managed to get a look at what is under the two boards. Not a whole lot realy. I was expecting to see some sort of DAC in here but that does not seam to be the case unless its one of those resistor ladders.



The most interesting part now is this so called "TX CLOCK OUTPUT AMP" board, to get a good look at it i had to disconect the clock ports from the front panel, threading them trough this passtrough was a nightmare as it barely fit and there was no room to properly work with it.



Then i also wanted to bring that board out enugh so that i could probe it while powered on. For that i needed a longer ribbon cable. I needed one that is 2x7 pin, but the only ones i could find in my parts boxes are 2x8. So to make it work i filed off one side of it so that it would fit in to a 2x7 pin connector.

[Berni]

So finnaly after all this effort the board is out and accessible, envelopes are there to keep the airflow going and to keep the board from falling back in. This thing is getting really really annoying to work on.



Because there is no service manual to be found means that i had to figure this board out myself. The fact that i could power it on and probe it meant that i could trial and error find the signals of interest by changing the clock offset voltage up and down while probing about. Here is what i found so far.



So after some more probing i noticed that one of the opamps is not behaving as it should. To test that out i injected some voltage from my PSU and sure enugh the clock signals offset moved above ground! Seams like this opamp had this weird failure mode where it forgot how to source current. When sinking current it worked perfectly fine, but when it was asked to source current it just sat there at 0V.



Now i need to find something to replace it with. What makes this more difficult is that the one opamp that was dead is a special high current output amplifier with fixed gain of 1 and a non standard pinout. This means my replacement opamp had to have at least 100mA output drive and work from +/- 12V supply. Closest thing i could is this bulky part.



Because the tab on this part is connected to Vee means that i can't just solder it down to the exposed ground plane. To fix this i cut out a small piece of copper clad board and soldered it to that instead, then superglued the board down. Generous dose of solder also helps with heat dissipation since this is a fairly power hungry opamp.



So now lets see if she smokes or if she works. and BINGO! We can now adjust the clock offset from the front panel all the way up to +1V just fine. Just like i bought one!



Just to make sure everything works its good to also check the data output and yup that looks like dataish stuff.

[Berni]

And there we have it. All put back together and working perfectly!



So now that its working lets have some fun with it. Because of the kind of data rates it works with means that the outputs have very fast rise times. We can use this fast edge to do TDR (Time Domain Reflectomerty). So as an example here is the clock output fed in to a SMA T with about 1m of coax cable hanging off one end that is not terminated. The scope internally terminates the other leg in to 50 Ohm. So here we can see that the reflection takes 10ns to travel down that cable. If we calculate the speed for typical 66% velocity factor cable we get almost exactly 1 meter.



Now for something that this pattern generator is more suited for. It is set up to generate a PRBS23 sequence that is fed directly in to the scope. There the scope does clock recovery and draws a eye diagram of the data. Considering this is 700Mbit/s of data makes this a very nice looking eye. Very open middle, no overshoots, just a tiny bit of undershoot.



So to mess the eye up a bit i decided to turn off the internal 50 Ohm termination in the scope. This makes for a much worse looking eye, but considering its not terminated at all makes it look a lot better than i was expecting to see.

[kcbrown]

You managed to do all that in an hour?!?   

Just kidding (though it would indeed be quite impressive if it only took you an hour to do all that and the writeup!) ...

Regardless, that was cool!   

[Berni]

You managed to do all that in an hour?!?   

Just kidding (though it would indeed be quite impressive if it only took you an hour to do all that and the writeup!) ...

Regardless, that was cool!   

Over an hour alone was probably spent just with a screwdriver to get the thing apart and back together due to how many screws there are and how difficult some of them are to get to.

It has taken a little more than 1 day to do all this. I like to take photos in general when taking apart stuff in case i forget where something goes on stubborn repairs that stretch out for weeks. If it is something interesting i might make a writeup at the end like this one. These can often be useful for someone who is also fixing one, or maybe they are just curious what it looks like inside without taking apart there own unit (I like to know whats inside my test gear).

I recently also recently fixed a Gigatronics 6100 synthesizer 10Mhz to 8GHz. Tho its not as interesting since all i fixed was a 13GHz relay step attenuator module by adding a paper shim inside so it didn't get stuck as it used to.

[TiN]

Elantec's.. They are prone points of failures in other gear, like HP 3458A DMM as well. And hard to replace, due to obsolescence and lack of modern high-speed analogs with high VCC/VEE voltage ratings. 

Thanks for a tour! What you plan to do with this thing? 700Mbit BERT is barely useful for anything modern..

[Berni]

Elantec's.. They are prone points of failures in other gear, like HP 3458A DMM as well. And hard to replace, due to obsolescence and lack of modern high-speed analogs with high VCC/VEE voltage ratings. 

Thanks for a tour! What you plan to do with this thing? 700Mbit BERT is barely useful for anything modern..

Didn't know that these things have such a bad track record. But it did seam to me that a opamp with this much output current at such a high supply voltage would get rather toasty on the die with a regular DIP with no heatsinking. This LME49610 is a pretty nice replacement tho apart from the package.

Well yeah 700Mbit is not all that high speed these days but it can still be useful. A lot of LVDS interfaces run at speeds lower than this and since it has the memory option in there means you can put meaningful data in to them to make them do stuff. It can exercise things that have serial inputs like shift registers to find there max operating frequency and it can be used as a fast clean clock source that is easily adjusted (Normal arb signal generators usually suck at this). So in general i plan to mostly use it to generate stimulus to digital circuits rather than using it as a BERT. I don't have a BERT receiver anyway. Would be nice tho if it had multiple data channels tho so that it could also stimulate serial interfaces that need more than one signal.