CMU200 Rx level issue / failed loop test

[_Wim_]

Hi, after installing some firmware updates to allow AMPS functions on my CMU200, the unit asked to perform an FM Modulation calibration. This calibration failed, and upon further testing I discovered that also all loopback tests failed.
This problem existed probably from before the firmware update, but I only noticed it now. 

Issue 1 & 2 description:
-   Rx level is down -10dB between 10 MHz and 1200.1 MHz, -3dB down to 2200.1 MHz, again -10dB down to 2700MHz. (see FreRes plots)
-   Rx level is down exactly the same on RF4, RF2 and RF1 (so not channel dependent)
-   Rx level is down exactly the same on all signal levels (so not attenuator dependent)
-   Tx output level has been checked with an SDR dongle, and seems fairly accurate (so level drop is not related to Tx part)
-   The wideband power meter measures the correct level on RF2 and RF4 (remark: a 2nd separate issue is that the wideband power meter on RF1 does not work at all and that RF1 is +-12dB  more noisy than RF2)

Attached are the plots measured with FreRes sweeping the entire frequency range at different signal levels. At first glance this appears a similar issue as reported here: https://www.eevblog.com/forum/repair/cmu200-failed-loop-tests/msg3810332/#msg3810332

Remark: I have removed of the remaining RF foam absorber (see attachment), but for me it seems unlikely that this little copper corrosion could be the culprit. I did check the conductivity of the corroded top surface, that the copper plating is still completely conductive everywhere on the top surface.

Rootcause 1: level issue on RF1,2 and 4
For the level issue I suspect some damage around IF1 on the RxTx board because:
-   Its identical on all 3 input channels
-   The wide band power meter (which is included in the RF Front end) indicates the correct power
-   From the block diagram I would expect the different IF1 Frequencies do not affect the front end nor the ADC (and the level shift matches exactly with the IF1 switch points)

Rootcause 2: broken RF1 frontend
For the 2nd issue my guess is that RF1 got severely overloaded at one point causing damage to the power meter and creating the additional noise (my unit has the additional RF generator, so RF1 is limited to 2 Watt instead of 50 Watt, although I am unsure when and how I overloaded it).  This seems however not related with the level issue and is for the moment less important for me, as I can use RF2 instead without losing any functionality. I will work further on this once the first issue is solved.

What I have checked so far for issue 1:
-   Checked the RxTx board for visual damage (quite a few bodges on the RxTx board, indicating a previous repair, but no obviously damaged parts)
-   Ran all unit self-tests. Only failures are the  loopback tests (1=>1; 2=>2,1=>4;3=>2) and the FM Modulation calibration.
-   Switched back to the previous firmware configuration (no effect)
-   Reseat all cards, plugin boards and connectors (no effect)

I am however not sure how to debug the RxTx board, because it is not easily accessible for probing, and without the Rf housing efficient cooling could be an issue. The block diagram in the service manual is not very detailed , so a lot of reverse engineering will be required. I have made some detailed pictures for this purpose, these will be attached in the following posts. Any tips how to proceed are certainly welcome!

Temporary Fix:
Currently I have configured a frequency dependent attenuation correction that flattens the response quite perfectly, even when running a sweep with a 0.001MHz step. This off course is not ideal but will do for now.

[_Wim_]

Detailed picture of Rx side of the RxTx board

[_Wim_]

Detailed picture of Tx side of the RxTx board

[_Wim_]

So I would expect this is the main region of interest. Before appears mainly the attenuator section, and further on we are back to a single IF so it must be in between those (I think  )

[_Wim_]

I started with a little reverse engineering. What Shahriar probably does in 5 minutes took me the whole afternoon.

What I find strange is that the block diagram in the service manual shows a shared LO (between Rx & Tx) for the 2nd mixer, but as far that I can see, this is not the case (only mixer 1a shares an LO with the Tx side).

Mixer 1 in the block diagram also appears to be made up of 2 stages, where mixer1b is used in both directions (up & down conversion).

Disclaimer: I have almost no knowledge of RF, I only watch "The signal path", so attached could be full with mistakes...

Any hints were the issue might be with this extra information?

[_Wim_]

For those who do not know: if you right click the above image (when it is large) and select "open image in new tab" you can zoom in to great detail. I have tried to make the max resolution I could (original image size was almost 80Mb)

[Michael-VK2BEA]

Wim,
   I did respond to your question here https://www.eevblog.com/forum/repair/cmu200-failed-loop-tests/
It looks remarkably like the problem I fixed with my CMU200.

To debug this I used an RF probe (one of the cheap ones from eBAY) and a spectrum analyzer, since you don't care so much about the actual level ... just the difference between the input and output of the filter.
Remove the heat sinks and you can do some probing toward the top half of the board (it's awkard but possible). For sections toward the bottom you might have to solder wires from the board to the probe.
The filters are at the top, so I would try and check that. Since I found this problem, I have heard from three other people who have the exact same failing filters. I think eventually they will all suffer the same fate.
I replaced the IF filters and the unit is restored.

Michael

[_Wim_]

Wim,
   I did respond to your question here https://www.eevblog.com/forum/repair/cmu200-failed-loop-tests/
It looks remarkably like the problem I fixed with my CMU200.

To debug this I used an RF probe (one of the cheap ones from eBAY) and a spectrum analyzer, since you don't care so much about the actual level ... just the difference between the input and output of the filter.
Remove the heat sinks and you can do some probing toward the top half of the board (it's awkard but possible). For sections toward the bottom you might have to solder wires from the board to the probe.
The filters are at the top, so I would try and check that. Since I found this problem, I have heard from three other people who have the exact same failing filters. I think eventually they will all suffer the same fate.
I replaced the IF filters and the unit is restored.

Michael

Thanks. Will try to measure them first. Good point about about the RF probe, I remember I bought one of those a while back when they were discussed here on the forum.

Did you end up buying the replacement filter here?
https://www.ectransistors.com/10-PCS-INDUCTOR-TRANSFORMER-MMCL-DFC3R836P025BHD-3R836P025-DFC3R836P025BH.html

[_Wim_]

Thanks. Will try to measure them first.

I managed to measure the 836MHz bandpass filter, and you are indeed correct, loss is much higher than it should be (8.55dbm instead of something around 3dbm). The 1842MHz filter I did not check , as the RTL-SDR dongle that I am using tops out at 1766MHz (but I am now sure it will be similar )

I must say I am very surprised by this, as the corrosion did not look so bad (see picture from after they were cleaned), and the copper plating is still conductive over the entire surface of the filter. It seems like some corrosion must have migrated through the copper affecting the internal ceramic structure? 

Principle used (might be useful for others who want to perform this test):
Equipment:
- Low cost Chinese RTL-SDRv3 dongle with Satsagen software to make the SDR dongle work as a spectrum analyzer
- Low cost RF Active Probe 0.1 - 1500 MHz from Ebay (similar to this: https://www.ebay.com/itm/332253695460?hash=item4d5bdd0de4:g:CzsAAOSwWlNZzA4E)

Setup
- generated a 1500Mhz 0dBm signal via RF3, looped back to RF4 in (1500MHz to be in the band used by the 843MHz IF)
- CMU200 was put in narrow band power mode, with a center frequency of 1500Mhz (to ensure a continuous IF-signal)
- RxTx card installed in CMU without its shielding/heatsink so probing is possible (after the test I checked with an IR camera, and the max temperature I could see was 65°C)

Thanks again for pointing this out 

[Michael-VK2BEA]

I tried purchasing filters here but with no luck...  after 3 weeks the order is still 'pending' and they do not answer my email. I'm not sure if its a scam but it doesn't look good 8-)

The filters ar Murata monoblock (Gigafil) devices DFC3R836P025HHD (836MHz) & DFC31R84P075HHA (1842.5 MHz)..  (see datasheet)
The filter design is described in the attached IEEE MTT paper. It is effectively a combline filter using a ceramic dielectric.

From what I can tell, the corrosion is just on the top surface (and top edge). If push comes to shove, it may be possible to electroplate copper back on to that surface. I bought an RXTX board on eBay and used the filter from the TX side (The same filters were corroded on that board but there is an identical filter on the TX side which is not in contact with the RF absorber ... and thus in good shape).

Michael

[_Wim_]

From what I can tell, the corrosion is just on the top surface (and top edge). If push comes to shove, it may be possible to electroplate copper back on to that surface.

I did a "quick & dirty" test with some copper tape as a proof of concept. I think the results speak for themselves. This is still with the shield/heatsink of the board removed. Some more careful tape-cutting and maybe soldering it in place (to prevent the glue from coming loose in the hot environment with the shield installed) is definitely a worthwhile improvement! And I would add a layer of capton tape on top to prevent future corrosion for the RF absorber foam.

[_Wim_]

For those who are wondering about the pattern in the tape, my role of copper tape is similar to this: https://www.newark.com/3m/1245-tape-1-2/tape-copper-foil-54ft-x-0-5/dp/66F1197?st=3m%201245

(part number mentioned on mine is 3M 1739-7)

[Sas]

Not all CMU200 units have this problem. Some units, probably newer ones from the end of production, have rx / tx modules which do not have this problem. The originally fitted RF absorber is much thinner and does not touch the filters. The modules can be recognized without opening, because they have the serial number printed on the pcb, not glued on the sticker. I don't know if this is the rule. For about 20 CMU200s, only 3 had such a newer module with a similar serial number without the problem of corrosion of the filters. They passed the test 1-> 4 / 3-> 2 RF Loop Path 1. I attach a significant fragment of the module photo without this problem.
The copper foil should be smooth, thin, without glue, adhere as well as possible to the ceramics and not cover the larger BPF holes. For me, using the foil, it was not possible to get the correct test 1-> 4 / 3-> 2 RF Loop Path 1, i.e. +/- 2 dBm in the entire range. Later I will attach photos of modules and filters with problems and I will try to explain more. The original 836.5 MHz murata filter is probably impossible to get, I have been looking for it for 2 years. It can be desoldered from a damaged FSH3 board.
I am looking for a method of home copper coating of ceramics without the use of electricity and an initial layer of carbon or palladium. Service companies dealing with catalytic copper plating, after learning about the size of the object and the problem of covering the holes, just laughed at me :-)

[Michael-VK2BEA]

Interesting..  The number to the right of the S/N is probably the date code. Yours shows 17/18, mine shows 14/09 (maybe 2014 week 9).
Its likely that they began seeing the problem after several years and fixed the design.

The 1842 MHz filters are available from one source I found in Italy. The 836 MHz filters are unobtanium in the original pinout but can be found in a slightly different package (BHD).

[_Wim_]

The copper foil should be smooth, thin, without glue, adhere as well as possible to the ceramics and not cover the larger BPF holes. For me, using the foil, it was not possible to get the correct test 1-> 4 / 3-> 2 RF Loop Path 1, i.e. +/- 2 dBm in the entire range.

Would conductive paint be an option? Something like this: https://www.ebay.com/itm/194247218349

I only have copper tape with structure currently available. Another option might be to this the surface with a layer of solder (for less badly corroded parts at least, so the solder has something to adhere to)

[Sas]

This module 1718 is from CMU200 calibrated in 2018. Based on the dates on the power supply (January 2005), hard drive (October 2004), fan (week 33 2004), R&S Electrical Safety Inspected sticker (March 2006), I assume the unit was manufactured in 2005 and sold or safety tested in 2006 year. There are numbers 0301 on the power supply (production January 2005, i.e. 0301 is not a date). Other numbers on the modules are 0309 0608 4000 0208 1104 0400 0300 1500. If it were weeks and dates, the device would be rebuilt many times. In this device I am almost sure that if something was replaced it was only the txrx module, although it may be the original module.
I have another CMU200 calibrated in 2015 and it has a sticker 1607 on the rxtx module. The rxtx module is almost certainly not from this unit. I bought this unit incomplete, without the dsp module and repaired it. Then, in 2020, a buyer cheat opened it, replaced the parts (keyboard, ccfl lcd, rxtx, dsp) and returned the damaged one. Now he is trying to fix it again. Based on the dates on the power supply (September 2004), hard drive (July 2004), fan (week 12 of 2003), R&S Electrical Safety Inspected sticker (March 2006), I assume the unit was manufactured in 2004 and sold or safety tested in 2006 year. There are numbers 0301 on the power supply (production September 2004). Other numbers on the modules are 1300 0400 0200 3800 0203 0208.
Another CRTU calibrated in November 2006 by R&S has a sticker of 1026 on the rxtx module. I think it is the second calibration after a year from sale, traces of removing the first glue on the housing. R&S Electrical Safety Inspected sticker for testing performed in November 2006. Based on the dates on the power supply (August 2001, no additional sticker), fan (week 38 of 2000) and hard drive (August 2005, numbers 0100), calibration and stickers, I assume that the unit was produced at the end of 2005. Other numbers are 2700 (cpu) 0300 (main board) 0409 0500 0100 (hdd) and 1026 (rxtx). By suggesting that this date, it would be difficult to manufacture a device in January 2000 from parts from future years.
EDIT: The numbers 1026 on the module rxtx are: Part Number, Serial Number and Modification Index (e.g. the number 10.26 can be read after pressing the INFO key on the CRTU). It has nothing to do with the production date.

I have 1842 MHz filters. I ordered 836 MHz filters in the BHD version a month ago, when the topic of cmu / crtu repair came back.

Painting cavities with the Chemtronics CW2000 marker pen does not give positive results, it additionally damages the copper layer. Painting with a conductive fabric based on graphite does nothing. Painting with copper-based paint does not do anything, the layer is not electrically conductive. Painting with Armack Kupfer Spray advertised by the manufacturer as powdered copper in an aerosol (does not contain paint) gives little, the layer is not electrically conductive. I added various copper strips, it is better but it is not good, fails the 1-> 4 / 3-> 2 RF Loop Path 1 test. I have not tried to solder the surface, it will probably damage the filter irreversibly. Each piece of the filter is individually tuned on a ceramic core. The core has defects on the ceramic surface prior to the copper plating process. These surface defects are filled with a copper layer and affect the filter tuning. Perhaps later the filters are measured, grouped and marked. I will add photos from the microscope soon.

Edit: Here are two photos of a working filter. I am posting two more extremely corroded filters after being returned from a buyer cheat. He will post a few more photos soon.

[Michael-VK2BEA]

Would conductive paint be an option? Something like this: https://www.ebay.com/itm/194247218349

I tried silver paint but it only partially restored the loss. I assumed because of the higher resistivity (compared to the original electroplated copper).

[Sas]

Photos of the 836 MHz filter before and after cleaning. The enlarged view shows the ceramic clearances without the copper layer. The ceramics are dark gray (inverted orange).

[Sas]

Photos of the 1842 MHz filter before and after cleaning. The enlarged view shows the ceramic clearances without the copper layer. The ceramics are dark gray (inverted orange).

[Sas]

Photos of the 1842 MHz filter from CMU in very bad condition, corroded, scratched by the buyer (from my previously edited post). The filter was soaked for one day in isopropanol and washed in an ultrasonic cleaner. As you can see, there is not much left of the copper layer. The core is tuned by cutting the sleeve from the inside.
Photo of another 1842 MHz filter, which shows well the damage to the ceramic core, which has been copper-plated. Each ceramic core has more or less damage before copper plating. The damage to the ceramic core is mainly at the edges. The tuning takes into account the copper-bonded faults of the ceramic core.
Re-alignment of the filter may be possible after supplementing the copper layer, taking into account the imperfections of the ceramic core.

[_Wim_]

Thanks for all the excellent contributions so far! 

For the sake of forum knowledge (and to satisfy my curiosity), I went a little overboard in testing. 

First I tried so remove the glue from the copper tape, and then solder it to the surface of the filters. Results were worse than with the copper tape applied directly, both for the 1842 as for the 836 filter.  Probably due to the fact the tape is to bendy and I could not get it to solder perfectly flat.

Then I removed the soldered copper tape again (which probably overstressed the 836 filter, see below), and solder a tiny copper plate on top  (plate pressed flat on the surface filter, only soldered the sides while pressing down). This improved the result on the 1842 filter to only -3dB (was -10dB before), my best result so far. Unfortunately no more output on the 836...

So, it is possible to restore them to some degree, but not without any risk. Corrosion on my filters was probably not the worst to begin with, so this also could be a reason for a partial success.

(pictures were taken before clean up, so some debris is still visible in the pictures)

[Michael-VK2BEA]

Re-alignment of the filter may be possible after supplementing the copper layer, taking into account the imperfections of the ceramic core.

From the explaination in the IEEE paper and by examining the 'gaps' in the filter posts, it seems that the entire filter is plated and then the gap created by grinding off the copper. I'm amazed that this can be done in a cost effective way, considering how cheap these originally were. I find it hard to believe that they were individually tuned; more likely to be mass produced from calibrated tools (and testing samples).

[Sas]

I find it hard to believe that they were individually tuned; more likely to be mass produced from calibrated tools (and testing samples).

Probably you are right. They produced it in a large number of pieces at the factory. Do you think they cut the filters and sold it like that, or rather they cut, measured and marked the group, what they got from the measurement?

I don't know what the difference between the two new filters might be. The manufacturer states a loss of 2.6dB in BW for the 836MHz BPF (2.8dB for the 1842MHz BPF). No two filters are alike, and the entire device is calibrated for differences in filters (and all the weight of other parts). Replacing the filter with the same second filter will not recalibrate the entire device. Maybe it will be a small brick to improve your results. The fact is that even minor damage to the surface causes a significant loss of signal. Hence my suggestion about trying to save the original, especially the 836MHz BPF filter, that you can't buy it anywhere.

[Michael-VK2BEA]

The manufacturer states a loss of 2.6dB in BW for the 836MHz BPF (2.8dB for the 1842MHz BPF). No two filters are alike, and the entire device is calibrated for differences in filters (and all the weight of other parts).

Yes, the whole chain is used for the factory calibration. The unit I repaired was within 0.1dB so there was little difference in the filters I swapped.
Has anyonw cracked the code on the calibration file? (where is it stored and what is it's structure?)

Michael

[RF_fanatic]

Nice work and good research.

I gave up on repairing CMU;s after countless units checked with reference units,and cleaning up my shack now (see for sale part of this forum) this specific filter is indeed a culprit, cleaning or better finding a donor filter is best option.

Most of the time a bad overal respons can be pinpointed to the digital units but the specific band drop more point to the rxtx or frontend, in this case the rxtx was a good conclusion, well done!