HP8595E spectrum Analyzer: readings at -40dB and CAL SIGNAL NOT FOUND

[auato]

Hi,
After a long disuse period I turn it on and after a while it turns off definetively. Once the PSU has been fixed, it runs again but measures almost 40dB less. If I try to calibrate it or try the CONF TEST, it returns "CAL signal not found".

-checked the voltages (all LEDs on PSU are on) and I have TP403=+5.14V,  TP404=+12.10V,  TP401=+14.94V,  TP402=-15.00V but I am not able to find where I can check +24Vdc (it looks that 8590A service manual doesn't mention any 24Vdc rail);
-checked the output calibration signal with another SA and it is OK (300MHz @-20dBm);
-the backup battery still seems to be good (3.01V);
-at the top left, the year has two question marks (attached video);
-in the attached photo I don't remember whether the connector with a purple thin wire starting from the PSU is connected somewhere ;
-if CAL AMPTD runs after entering the deep debug mode (Frequency --> -37Hz) the calibration starts but when it is into the EMI-test part it goes into a loop and no longer exits until I abort the procedure.

But what happened to it?

https://youtu.be/CKte98jvFmI?si=KDvNotXRYJM_psud

edited: I would like to measure +24Vdc

[MarkL]

I would try to characterize the problem a little more.

The first thing I would try is different attenuator settings.  You might have a bad step in your attenuator.

Do you have a signal generator to test the high band (2.9GHz - 6.5GHz)?  Could be a problem with just the low band.

You could try defaulting the calibration constants using the procedure in the Assembly Service Guide.  See "If the analyzer displays a low signal level" on page 234.

On the date/time issue, can you set the date to a sane value, and after unplugging the analyzer for a minute then plugging it back in, does it it keep time?  How long was your "long disuse period"?

I'm not sure if the goofy year could point to possible NVRAM corruption, but it's something to keep in mind.

I'm not seeing a +24V supply on the schematic.  Is there one?

According to the Assembly Service Guide, the purple wire supplies -24V to the A7 analog control board on 8592L and 8593E analyzers.  (I believe it's used to extend the YIG tuning range for those models.)  On the 8595E, it doesn't go anywhere and just terminates in a 3-pin housing that's left hanging.

[DavidAlfa]

You definitely should have 24V, or the switches won't work, just check how the attenuator clearly says 24V, if missing it would explain it all.

Once you have the 24V tested working, if the signal keeps failing:
The attenuator block has four elements (-5, -10, -20, -40dB) and makes a combination of them in series and/or bypassed.
If one of them is burned, you'll get an open circuit.
So try finding a signal by manually setting attenuation to -5, -10, -20 and -40dB.
It's pretty easy to disassemble, but there're a lot of screws behind the cover, where the sma connectors are.
This is it:


One 8592 had that one burned, other had a burned transistor array which activated a rf switch, whose coil was completely roasted, can't remember what it was for.
It was a latching switch, so it only needed a small pulse to change, but the switch contact had fallen (Plastic rivets...), so my guessing is the system did not detect it switching and kept energyzing the coil until it overheated, then everything burned down in a chain.

[MarkL]

You definitely should have 24V, or the switches won't work, just check how the attenuator clearly says 24V, if missing it would explain it all.
...

The label doesn't mean there's a 24V supply.  The attenuator is supplied with +/-15V on J5 pins 3 and 10.  See schematic snippet below.

But the point remains it could still be a bad attenuator step.  I think we both agree on that.

[DavidAlfa]

The label means the coils in the attenuator are meant for 24V, thus the logic tells it'll be powered at that voltage.
These are also available in 5 and 15V versions IIRC.

Or maybe someone replaced the attenuator block with a wrong one?
There're also 5 and 15V versions, so it's strange HP mounted a 24V part there.
Wether HP used them like this, I don't know.

[rf-loop]

HP8594E, 8595E... etc  Attenuator is 24V version ( 33321-60036 ) It have dc bloc, 10, 20 and 40dB attenuator chips

A long time ago I repaired many of these spectrum analyzer attenuators. When it's done well, renovated equals new. But then it is not even opened on a dusty table. Namely, you can't get dust out of there if it gets there or least it is difficult. The dust-sensitive area is the coaxial attenuator tunnel. The outside where the relays are is a normal dirty area.
 Nylon pins and contact strips are very easily damaged, especially the end that is in the attenuator tunnel moving contacts. (the glue on the contact strips may also break and repositioning is the work of a "watchmaker" if you intend to get it in such condition that it has the same specs as new) Most common problems have been burned  attenuator chip (some burned atten chip also visible in @DavidAlfa image) ( and/or) O-rings crumbled broken. The size and material must be correct. I myself have used some Seiko O-rings (spare parts for watches) which in the end cost... yes, quality costs.. The attenuator have 16 0-rings.
For change O-rings work is best to do in right order what is most safe.
In all cases keep all kind of oil far away from these low pressure contact surfaces. They must be absolutely dry and clean. Material is such that it does not require lubrication, even though the surfaces slide just a little bit when contact is made. Instead, there must be a microscopic amount of lubrication at the end of the nylon pin, which must not spread to the contact surfaces under any circumstances, not even when the attenuator changes at a high rate without pause 24/7/365 in some automatic production test line.

[Bicurico]

That was a nice reading, rf-loop. Thank you.

Makes me better understand why these devices were so incredibly expensive.

[auato]

First of all, thank you all for the replies received so far.
I haven't disassembled anything yet. I just did a few more tests that I want to share with you before disassembling the parts.
I performed some measurements having a good signal generator (R&S SMJ100) and another Rigol spectrum analyzer (but with a maximum band of 1.5GHz). What I have noticed is that from 300MHz up to 2GHz, with a signal of 0dBm, generated br SMJ100, I always have a reading of approximately -38dBm (at the default attenuation of 10dB but also at -20 and -30dB but if I move the attenuation from - 40dB up to -70 it becomes a value approximately +2dBm or in any case slightly positive). From 3 to 6GHz I have a reading around -44 / -45dBm which is stable in the various attenuation steps -10, -20 and -30 and which becomes -3 or around -5dBm in the attenuation steps starting from -40dB onwards. The summay is in the picture.

[MarkL]

The label means the coils in the attenuator are meant for 24V, thus the logic tells it'll be powered at that voltage.

Yes, that's not a bad assumption, but it's not the case here.  The attenuator is powered with 30VDC from J5-3 and J5-10 (+/-15V).  There's no ground or other power connections on J5.  The switching is done by pulling the desired input pin to -15V (look at the driver).

The attenuators on both of my 8595E are labeled with "24V".  I've measured the voltages shown on the schematic and they are correct (30VDC supplied as +/-15V).

My point is that there is no +24V supply in the 859xE units for the OP to worry about checking, which was one of the initial questions.

[MarkL]

First of all, thank you all for the replies received so far.
I haven't disassembled anything yet. I just did a few more tests that I want to share with you before disassembling the parts.
I performed some measurements having a good signal generator (R&S SMJ100) and another Rigol spectrum analyzer (but with a maximum band of 1.5GHz). What I have noticed is that from 300MHz up to 2GHz, with a signal of 0dBm, generated br SMJ100, I always have a reading of approximately -38dBm (at the default attenuation of 10dB but also at -20 and -30dB but if I move the attenuation from - 40dB up to -70 it becomes a value approximately +2dBm or in any case slightly positive). From 3 to 6GHz I have a reading around -44 / -45dBm which is stable in the various attenuation steps -10, -20 and -30 and which becomes -3 or around -5dBm in the attenuation steps starting from -40dB onwards. The summay is in the picture.

Hmmm... that's very interesting.  A problem with the 40dB thru path?

Try turning off correction (under CAL --> CORRECT ON/OFF).

[MarkL]

One other thing that might be worthwhile is to look at your calibration constants (CAL --> Service Diag --> DISPLAY CAL DATA) to see if anything is way off.  Typical values are described on page 584, and/or share here.  The cal data and correction constants are stored in NVRAM, and the only reason I bring this up is the unexplained weird (corrupted?) date.

If CORRECT OFF gives you pretty much the same results as before, meaning approx 40dB down when the 40dB attenuator step is supposed to be out, I would say it's time to isolate the attenuator and take a closer look at what it's doing.  The other dB differences you're seeing are relatively much smaller in magnitude, and I wouldn't worry about them yet.  Successful self-calibration might take care of them.

Since you have a signal source and another SA, I would connect the source to the 8595E input (say 300MHz @ 0dBm), and the other SA to watch the signal coming out of the 8595E attenuator.  The attenuator output is the SMA connector on the top (on the end with the ribbon cable).

You can then exercise the attenuator like you did before and watch the changes in output amplitude.

If you get bad behavior with the 40dB step, I wouldn't jump to disassemble the attenuator just yet.  I would confirm that the 40dB step control input is working.  There is a section in the Assembly Repair Guide on what to expect from the control signals on page 291 and table 6-11.

[MarkL]

And also, if you haven't done it already, backup your correction constants using the procedure described in Chapter 3 in the Assembly Repair Guide.  If something goes wrong and you lose the NVRAM, or your battery dies unexpectedly, you'll be much happier.

[auato]

I'm having trouble extracting the assembly to which the attenuator is attached. In meantime I share with you that CAL --> CORRECT OFF does not produce any change and and calibration constants appear as below. I'll update you when I get the attenuator out the assembly... now I have a coaxial that is pulled like a string and I don't understand where it is hooked up and what else I have to dismantle

[DavidAlfa]

You can easily check the attenuator without fully disassemblying it, only remove the coax connections and measure resistance with the DMM between input and output.
Adjust the attenuation level with the knob, from 0dB all the way to -75 or the max. it allows, if it measures open circuit at some point it's very likely you have a bad element.

That's how I got the definitive hint before opening it, the module either bypasses (shorts) or leaves the elements in series, there's no way to create an open circuit except by broken elements.

[rf-loop]

I'm having trouble extracting the assembly to which the attenuator is attached. In meantime I share with you that CAL --> CORRECT OFF does not produce any change and and calibration constants appear as below. I'll update you when I get the attenuator out the assembly... now I have a coaxial that is pulled like a string and I don't understand where it is hooked up and what else I have to dismantle

Are you serious you do not have service manual. There is all explained including disassembly and assembly procedures.

For take attenuator out... you need take rf frontend out first and after then all is more easy. It is very easy when do it just right way. And note that some filters and some other components are mechanically fragile even when they do not looks like fragile and then also care ESD and repeating again: ESD!  Also I recommend keep your tools and fingers out from YTO YIG tuned oscillator except is absolutely necessary.

And then... never enter any DC after attenuator...
(or DC to RF input when DC bloc is turned off...)

You can find a lot of instructions here as well, if you look carefully at what is there to apply.
https://xdevs.com/doc/HP_Agilent_Keysight/08590-90316.pdf

Also due to its age it is wise to check and replace battery. (and even when service manual do not ask to use support battery.. I have always done it with support battery (but it need also take care how to use it)  for keep Mr Murphy outside from my work bench)

There is also somewhere available full CLIP for these (I have only original paper versions, whole repair service set, long time ago purchased from HP/Agilent)

[MarkL]

So sorry you went through the trouble of trying to remove the RF section, and if I said something that implied to do it.  As you are discovering, it's an involved procedure.  I was trying to leave that as a last resort and eliminate all other possibilities first.

More explicitly, and perhaps you can still do this, disconnect the hard coax going to the top of the attenuator (that's the attenuator out) and run an SMA jumper from the attenuator to your Rigol SA.  The attenuator and whole RF section can stay in place.  Then apply your signal generator to the 8595E input and exercise the attenuator as previously described.  See photo.

As for taking the RF section out, it's likely your attenuator is bad and you will have to do this anyway, but my advice would be to take a lot of photos during the disassembly so you will have a record of the cable routing.  (Be thankful you don't have the tracking generator in there too; it's like sausage getting it back together.)

In addition to the instructions in the Assembly Repair Guide (link above from rf-loop), I've also found that removing the screws holding the front panel in place (qty 5 on the sides) and pulling the front panel forward an inch or two will give you a little more working room.  Be sure to disconnect the hard line coax going to the front panel N connector before moving the front panel.

[MarkL]

I would also reinforce rf-loop's concern about DC on the input, and NOT use a DMM to check the resistance.

In order to do a DC resistance check, you need to disable the DC block in the attenuator which is on by default in the 8595E (under AMPLITUDE --> COUPLE DC).  But by doing so you would expose the RF input stages to a DC voltage from the DMM, and possibly damage components there.  Don't do it.

If you really wanted to do a resistance check, disconnect the attenuator output from the RF section (as in the above photo), and put a 50 ohm terminator on the attenuator output.  But because you have a signal generator and another SA, a signal test as previously described is a better confidence test.

EDIT: Doing a resistance check from the front connector is not what DavidAlfa was describing, but since we already had one misunderstanding I want to make sure the point about DC is clear.

[DavidAlfa]

I said doing so between the input and output of the attenuator, which means both sma connectors removed, completely isolated from the analyzer RF input, so no risk of any kind.

[MarkL]

Thanks for posting the CAL DATA DISPLAY.  I'm not an expert on every number in here, but nothing jumps out at me that says the calibration data is invalid.  The AMPLITUDE table looks ok.

The last frequency and amplitude cal is showing as August 11, 2013.  Is that about right?  Has this unit been unused for 10 years?

[MarkL]

I said doing so between the input and output of the attenuator, which means both sma connectors removed, completely isolated from the analyzer RF input.

Yes, I did not mean to say your check was erroneous.  I already edited my post after realizing it could have been interpreted that way.

[DavidAlfa]

Modern DMMs will source pretty low currents, last one I measured did 2mA, was a supercheap meter.
That would put 100mV, 200uW (-7dBm) on the 50Ohm RF input.
I guess good DMMs will provide way lower current.
Would that small DC signal be able to cause any real damage?

[rf-loop]

I said doing so between the input and output of the attenuator, which means both sma connectors removed, completely isolated from the analyzer RF input, so no risk of any kind.

Yes, of course this is a safe way.

However, I wanted to point this out because I have too often seen when something is done slightly different from the instructions, even if just out of curiosity. If by chance such danger factors were not known.

 

[MarkL]

...
There is also somewhere available full CLIP for these (I have only original paper versions, whole repair service set, long time ago purchased from HP/Agilent)

For reference, there is this one that's been copied far and wide, still available from Keysight's server:

  https://www.keysight.com/us/en/assets/9018-05708/user-manuals/9018-05708.pdf

CLIP part number 5963-2951.

It's titled "8590A Component Level Information Packet" (CLIP) on the opening page, but it's actually a CLIP for 859xE/L/C units.  There are nuances for specific E/L/C models and serial number ranges that will appear in other CLIPs, but everything is so similar it doesn't usually matter.

If you ever need to find the exact right CLIP, there is a table on page 521 (Table 12-2) in the Assembly Repair Guide that lists the CLIP part number for specific models and serial number ranges.

[auato]

Hello guys, thank you for the valuable advice you've been giving me. I would like to update you on the test you recommended (especially from @MarkL), which involves directly connecting the attenuator output to a second analyzer and injecting a well known amplitude signal from an external generator. I injected 300MHz at +15dB and here are the results below and in the video:

Att 10dB

-43dBm

Att 20dB

-52dBm

Att 30dB

-60dBm

Att 40dB

-33dBm

Att 50dB

-43dBm

Att 60dB

-53dBm

Att 70dB

-60dBm



https://youtu.be/8tmC3t8NghE?si=OeWIvvGoLs2DWhDO

With attenuation at 40dB, I get the highest reading (-33dBm from +15dBm).
It seems clear that there is something wrong with the attenuator. If I want to exclude the rigid connection and the N connector upstream of the attenuator I would just have to proceed to disassemble the attenuator and try to get my hands on it 

[rf-loop]

Hello guys, thank you for the valuable advice you've been giving me. I would like to update you on the test you recommended (especially from @MarkL), which involves directly connecting the attenuator output to a second analyzer and injecting a well known amplitude signal from an external generator. I injected 300MHz at +15dB and here are the results below and in the video:

Att 10dB

-43dBm

Att 20dB

-52dBm

Att 30dB

-60dBm

Att 40dB

-33dBm

Att 50dB

-43dBm

Att 60dB

-53dBm

Att 70dB

-60dBm



https://youtu.be/8tmC3t8NghE?si=OeWIvvGoLs2DWhDO

With attenuation at 40dB, I get the highest reading (-33dBm from +15dBm).
It seems clear that there is something wrong with the attenuator. If I want to exclude the rigid connection and the N connector upstream of the attenuator I would just have to proceed to disassemble the attenuator and try to get my hands on it 

Once the attenuator is removed and it is on your desk.

Carefully remove the second end piece (4 screws)
Don't open anything else yet!

There is a metal mesh between the end piece and the frame. Note its positioning. also be careful that no dirty / dust of any kind gets into the exposed attenuator cavity.

After the end piece is removed, pull out the aluminum protective case. You can then fasten the end piece back in place (even if only with a couple of screws, just a light finger-tightness) It is also good to be in its place if repair work go forward to next step after detect O rings condition. 

Now that the side of the relays is exposed, you can see the nylon pins that control the attenuator contacts. Each one is on the moving plate of the relay and there are black O-rings. Check carefully whether all O-rings are completely intact. At this stage, you can also move each relay by hand and it will stay in the position you press because of the magnet. In this case, you can see better that all the O rings are in order and that the pins move through their full range of motion. Move every relay and check every single O-ring when nylon pin moves when you turn relay in and out.

I have a 50% prediction that the 40dB attenuator bypass position will not work.
Let's hope problem is O-rings... and nothing else.

Do not disassemble or proceed further until this is resolved.
If the O-rings have a typical defect, next you need the right O-rings and information on how to change them so that the risk of further damage is minimal. It requires the skills and precision of a fine mechanic work. Those nylon pins and contact strips inside attenuator cavity and their fastenings in the attenuator cavity are quite easily damaged, as are the attenuator chips. There are also screws that should not be turned if you want the attenuator's properties to remain correct throughout its frequency range (and those things are not compensated by selfcalibration). Before you get there, you have to do a few things in the right order.

[auato]

Thanks a lot.
I finally disassembled the RF assy and opened the attenuator. At first glance what I found is the first relay on the right side (near the RF input connector) in a different position than the other relays (as vidible in the short video I published). Could this state be normal? Also, are the o-rings the tiny ones in the photo?

https://youtube.com/shorts/Yz_BRhvwg4g?si=0_UAxk8avDcMJ8Se

[rf-loop]

Thanks a lot.
I finally disassembled the RF assy and opened the attenuator. At first glance what I found is the first relay on the right side (near the RF input connector) in a different position than the other relays (as vidible in the short video I published). Could this state be normal? Also, are the o-rings the tiny ones in the photo?

https://youtube.com/shorts/Yz_BRhvwg4g?si=0_UAxk8avDcMJ8Se

Yes it is normal. Relays are bistable and they stay in position what was last commanded.
You can move these relays manually up or down. They have two position.

In your image can see this one O ring. Each nylon(or whatever material it is) lift-push pin is attached to the metal plate of the relay with two O-rings. There are two of these pins on the every relay plate. (The metal plate of the relay has a U-shaped hole where the pin and the O-ring are. It does not fall out of the slot because the hole through which the pin passes acts as a guide.) The pins go through the thin hole into the attenuator cavity and there these move the contact strips. The structure of the cavities inside is very delicate.
Often the pins are almost white, but in your case they are brown. It is perhaps older version. In my experience, brown pins are more prone to breaking than white ones.  (what is your analyzer serial number prefix what tell manufacturing year)

Move each relay to the lower position and to the upper position so that you can see the condition of the O rings as accurately as possible. Look closely. You  can easy move these up and down... (magnetic construction keep it in each position) 
The movement of that pin must be full. A missing, crumbled, etc. missing O-ring no longer pushes/pulls the full distance and the movable strip does not make contact in the cavity. Sometimes a small chip may have detached from the O carriage, which can prevent the full range of motion.

I can't find the full documents I made a long time ago about them and I don't remember the order of the attenuators.
However, the first one after Input is the DC-block. The next one is pretty sure 10dB, but I'm not sure right now if the order after that was 20 - 40 or 40 - 20.

If all the o-rings are fine and the movement is normal, pin need move exactly same amount than relay plate... then need to think about another possibility. One possibility is that the end of the nylon needle in the attenuator cavity is damaged and it no longer closes the contact that bypasses the attenuator chip in question. However, it can still work in such a way that you can push the contact into the attenuator position. yes, at least in theory, but I've never seen such a case. 

[scopeman]

I have rebuilt these attenuators but sometimes it's worth it to consider other alternatives.

If all else fails you can send the attenuator out for a full rebuild at the company at the link below. I had them totally rebuild an attenuator
for my HP 4407B (9KHz-26.5GHz) a job that I could not easily certify in house so I sent it out.

They did an excellent job and I got certified plots of every attenuator step. I have no personal stake in this company, I am just a satisfied customer.

It's not cheap but you get a 1 year warranty. Something to keep in your back pocket when you need it.

https://www.testequipmentplus.com/agilent-hp-step-attenuator-repair.php

Sam
W3OHM

[Kean]

More info on repairing the 33321 step attenuators can be found in this post:
https://www.eevblog.com/forum/testgear/hp-8644b-attenuator-question/

[bg1wxd]

MY ENGLISH IS NOT GOOD ENOUGH, I FIXED THE ATTENUATOR OF THE HP 8595E, IT WAS THAT LITTLE BLACK RUBBER RING THAT WAS AGED AND BROKEN, AND A BOX OF RUBBER RINGS BOUGHT FROM AMAZON COULD BE REPLACED.

[auato]

very interesting your pictures. Thank you! Fortunately or unfortunately the o-rings problem doesn't seem to be my case. They look intact.

You can easily check the attenuator without fully disassemblying it, only remove the coax connections and measure resistance with the DMM between input and output.

and if I check with a DMM between input and output the resistance is always high (with both the relay positions)

[MarkL]

very interesting your pictures. Thank you! Fortunately or unfortunately the o-rings problem doesn't seem to be my case. They look intact.

You can easily check the attenuator without fully disassemblying it, only remove the coax connections and measure resistance with the DMM between input and output.

and if I check with a DMM between input and output the resistance is always high (with both the relay positions)

Don't forget the first attenuator stage, on the opposite end of the ribbon, is a DC blocking capacitor.  So, with this in circuit you won't see any DC resistance.

In combination with the faulty 40db thru position, if you flip all the actuators in the same direction at the same time (all in or all out), you won't have any DC resistance from end to end.

It seems fairly certain there is an issue in the cavity: dirty contact, broken contact, broken nylon pin, etc.  But before digging into it, I would reconnect the ribbon cable to the 8595E one more time and again exercise the actuators: 1) You should be able to identify which one is the 40dB step and 2) Verify the 40dB actuator has the *full* range of motion like the others (nothing is sticking or lopsided).

From your photos everything looks ok from the bottom.  Figuring out which one is the 40dB step will at least allow you to focus on the right set of contacts inside the cavity.

[auato]

Guys, I don't know what happened but now everything looks OK. There is probably still something mysterious. Now as you can see from the photo, the attenuator is still outside and the HP8595E only controls the relay contacts. The attenuation steps are OK and represented as in the photo. 
frankly I still don't know what to do. I was not so keen on getting my hands inside that attenuator with those tiny mechanisms. Maybe it's dirty inside and I recall that I only sprayed some compressed air on it. In any case I studied what to do watching some videos on YT on how to operate with this stuff and if the problem will appear again, I will proceed with a disassembly of it. I could wait a few more days and then proceed with the assembly. First, however, I would like to replace that 10 ohm resistor R80 in the photo that I clumsily smoked on the analog board A7. Anyway I will keep you updated. In the meantime, Merry Christmas to all of you and thanks again

[MarkL]

Well, glad it's working.  But unexplained, it will of course come back at the most inopportune time.

Thanks for documenting the attenuator dB step positions.

[rf-loop]

@auato

It is possible there have been some small piece of what ever inside attenuator relay side.
But in more worst case there can be some other problem inside attenuator cave. I do not list all what are somehow possible.

I think your some one previous images was not from your attenuator but similar (because it is partially disassembled and bit weird way, perhaps just for image...who knows). I have used this image (part of it) here.





There need always carefully look these positions are clean (marked red arrows). These metal surfaces need be free from some small garbage and also these surfaces of nylon pins, so that it can move perfectly full amount. (I have seen many times example small piece of old brittle O-ring rubber on these surfaces, sometimes these are like attached there and need carefully remove if they exist and if still do not yet want do whole renew work (there is two methods for take nylon needles out from contact strips in cave - very risky "lottery" way  or very safe but lot of more "watchmaker" work. (example I use always dust free air room and and magnifier (microscope in my case dye to my after "best before" aged eyes )

Btw, in some pictures in this thread can see some attenuators where can see bit weird contact strips and also really weird contact strips positions... without further explanations and other images can not tell what are these and why... there is one image where strips are positions I have never seen when nylon needles have removed... really weird. (but because it is not your attenuator ... no need think more these)

If you use compressed air to inside attenuator cave. Be very careful because these thin contact strips. They are fragile. Some times also old glue can brittle (they are glued in perfect positions). Also air need be real clean and oil free, and still it can do more bad than good because some particles attached to the surfaces inside may detach and then remain in contact. The low contact pressure contact is to some extent "self-cleaning", however, because there is a small slip when the contact is made... gold alloy against gold alloy.

If the cavity of the attenuator and the end flanges have never been opened, it is quite unlikely that there is debris there. The holes of the nylon needles have quite a small clearance, so normally no debris gets in there.
But what have I ever found. Rarely. Some kind of "oil" or "grease". It is put in when the witches are made. There, at the point of contact between the nylon needles and the contact strips. Usually it is a very small dose. And it must never come into contact with the surface. Sometimes the attenuators run very hot (production test use, for example, is very tough, the user's HP didn't even give them a warranty for that use).
In such a situation, the fat may heat up and drain, especially if there has been a bit too much of it there. Grease between the contact surface does not prevent contact. But then the typical problem is that during the SelfCal / Self Check phase the contact is created a little too slowly and it fails or sometimes even gives the wrong value (capasitive contact just before galvanic)

Merry Christmas!

[auato]

I still have the HP8595e disassembled on my bench.
The problem seemed to be the third relay (40dB). On some rare occasions it did not release when the instrument was turned on or when the preset button was pressed

But now that I've discovered it furtively, as you can imagine, it's been working perfectly everytime for days. 

@rfloop thanks for the detailed image, I am checking again near the red arrows with a microscope. Honestly before reassembling it I was thinking of spraying something like WD40 or isopropyl alcohol under the magnet of the third relay but I recalled the @rfloop's advice leaving them perfectly dry. I'm attaching some images taken by a microscope and as you can see they are not perfecrly clean

I was also thinking about replacing the battery because I've had the instrument for more than 7 years and I also bought it as a second hand tool and I don't know if it's worth doing this operation now. The battery is a Panasonic CR2477 and measures precisely 3.06V

[MarkL]

...
I was also thinking about replacing the battery because I've had the instrument for more than 7 years and I also bought it as a second hand tool and I don't know if it's worth doing this operation now. The battery is a Panasonic CR2477 and measures precisely 3.06V

Whether you decide to replace the battery or not, I would recommend doing the correction constants backup procedure, mentioned previously in this thread.  Some of the numbers may change slightly depending the repairs needed for the attenuator, but at least you'll have them in case the battery dies unexpectedly.  It's free and takes 10 or 15 minutes.

As far as the bad actuator is concerned, I think it would be worthwhile to compare the driving waveform on the bad actuator against a good actuator when switching in and out.  It could still be a mechanical issue, but maybe there's a problem with the driver.

[rf-loop]

I still have the HP8595e disassembled on my bench.


@rfloop thanks for the detailed image, I am checking again near the red arrows with a microscope. Honestly before reassembling it I was thinking of spraying something like WD40 or isopropyl alcohol under the magnet of the third relay but I recalled the @rfloop's advice leaving them perfectly dry. I'm attaching some images taken by a microscope and as you can see they are not perfecrly clean

Yes, do NOT use something like WD40 there. There is danger that it can also go inside attenuator cave through nylon pin actuators holes.
These low pressure contact surfaces there must be absolutely clean and dry from any kind of oil. Oil on contact surfaces may delay or even prevent contact and also contact surfaces "self cleaning" do not anymore work properly (because oily surface collects and keep all possible small dust particles). Also if open attenuator cave these contact surfaces need keep clean and NO OIL.
There is nothing what need oil after attenuator is manufactured! Some wd40 etc can do only bad things or very bad things.

It need also keep in mind that many RF circuits may have some ceramic etc materials what need keep untouched and newer ever use some WD40 or similar things... after they are absorbed to these materials then high-frequency performance values are even permanently destroyed/changed.

In your images it looks like enough clean.
I have some old day measured these nylon needles movement dimensions and how much it tolerate reduced amount of movement in both ends.
It tolerate ~0.1mm ... I know it because in some emergency case I have temporarily used some ~0.1mm too thin O-rings and check that it works well including still enough "marginal" for reliable contacts. But in this case longevity is perhaps highly reduced if attenuator is under heavy use (continuous stepping in a short period of time, such as in some test applications of product lines, etc) (these O-rings and the upper and lower surfaces of the nylon needle also act as dampers where the kinetic energy turns into heat and can at worst fry these rings - you know when you listen to the buzz of the attenuator when it is driven continuously 24/7 at maximum speed (controlled via HP-IB under test automation system)

[scopeman]

On the cal constants I made an excel spreadsheet for recording those for my HP8593E and 8593E. I have attached a copy along with one I found years ago Hopefully it will be useful to you.

Sam

[scopeman]

Missed the last two.