Agilent 4396B spectrum/network/impedance analyzer repair

[mark03]

This thread will document the repair [I hope] of an Agilent 4396B network/spectrum/impedance analyzer purchased on Ebay.  I have been wanting to delve further into RF experiments and was attracted to this model as it combines a spectrum analyzer with a VNA having similar specs to the 8753 series.  My unit also comes with option 010, a software-only upgrade which works in conjunction with a pricey external test fixture (p/n 43961A) to perform I-V impedance measurements---DUT voltage on one receiver port, DUT current on the other.  The 4396B frequency range tops out at just 1.8 GHz, but that's enough for me.

I believe the 4396B came out in the late 90s, around the same time as the later models in the 8753 series.  Unlike the 8753 VNAs, however, the 4396B and its lower-frequency sibling, the 4395A, were products of HP's Yokogawa Electric division in Japan.  Some of the PCBs bear a "YHP" logo, and Japan-isms are evident in the manuals.  I get the impression HP did not sell many of these analyzers.  Serial numbers only go up to 1000-something in the first 5-10 (?) years before production moved to Malaysia---mine is #431 and it is evident from the various service bulletins that the product had already been on the market for several years by that point.

A significant risk for me as repairman is the relative lack of popularity of this model, with a matching scarcity of replacement parts.  There is also, AFAIK, no component-level documentation (CLIP) available.  The service manual has a detailed troubleshooting procedure, instrument theory of operations, and some block diagrams.  There's an extensive set of self-tests programmed into the firmware, not only the internal tests run at boot-up, but also external tests requiring simple test harnesses outside the unit.

The condition of this particular 4396B was described as "powers on, no screen or LEDs."  It appeared clean in the photos, and was loaded with options (there aren't many: 010 RF I-V impedance, 1D5 oven osc, and 1D6 time-gated analysis).  Could it be a simple power-supply issue?  I decided to roll the dice and find out    Happily, the seller packaged my analyzer exceptionally well and it survived the trip to Seattle in good shape:



As advertised, when the unit is switched on the fan starts running but nothing else happens: no beeps or LEDs.  Looking under the top cover, first hurdle cleared: all of the boards are there   and everything appears clean:



From left to right are the pre-regulator board A40, DC-DC converter A50, graphics processor (GSP) A51, then the post-regulator A2 and the RF sections A3/A4/A5/A6.  Most of the boards have grid-style connectors which plug into a backplane (motherboard A20) through cutouts in the bottom plate, though not all of them are actually used.  The DC-DC converter A50, for example, has one hanging in the air because the backplane has no mating connector in that position!  All of the A50 connections in this instrument are on the rainbow Molex cables: 24V in from the preregulator A40 (outside the field of view), +5V (six-conductor cable) to the CPU board A1 under the backplane, and +/- 18V and +/- 7.8V to the post-regulator A2.

The DC-DC board A50 has a poorly named but clearly labeled green "shutdown" LED.  In normal operation, this LED is on.  If the board detects an over-current condition or the absence of a fan encoder signal (propagated back through the A2 post-regulator on the same connector supplying A2 with its split rails), the power supply is disabled and the LED goes off.  When I turned on the power, I could see the shutdown LED flash briefly, then stay out.

At this point I ran into a small problem:  It turns out the service manual contains quite a few typos, including minor details like which connectors to unplug while troubleshooting   With enough head scratching and cross-referencing, however, it's possible to triangulate around the documentation problems.  So far.

Unplugging the A50 output cables to the CPU board A1 and post-regulator A2 and powering up again showed the shutdown LED still off.  This is fortunate as A50 is one of the simplest boards in the analyzer and should be easy to fix.  (Performing this test requires connecting a function generator to supply a fake FAN LOCK signal and connecting a load resistor to the +7.8V output.)

[mark03]

As there are no schematics available for this system, the first order of business was to reverse-engineer one for the A50 board.  Several hours of work later, here is part of the schematic showing the shutdown logic.  (Component references are arbitrary except for the integrated circuits, which are the only parts with refdes on the actual PCB.)



and a close-up of that section on the PCB:



U1 and U8 (TL594) are PWM controllers for the two switchers on the board, one for the +/- 18V and +/- 7.8V supplies, and another for the +5V digital supply.  U5 (LM393) is a dual comparator which trips on an over-current condition for each switcher.  Its outputs are wire-ORed together with a 100k pullup.  U2 (74AC00) is the shutdown-state flip-flop and a couple of gates for processing the FAN LOCK signal.  (The identities of the SOT-23 components are somewhat uncertain.)

Probing the comparator outputs in the shutdown state showed that both were still high (normal operation).  That left the FAN LOCK signal.  Examining U2 pin 13 during start-up revealed a rapidly rising voltage, much too rapid in fact for the supposed one second time constant of R52 and C43...

At this point another piece of evidence becomes significant.  Agilent issued about a dozen service bulletins for the 4396B over the life of the product.  Bulletin 11A seems relevant (and covers my serial number):



Sure enough, C43 was bad.  I replaced it with a 22 uF junkbox spare, and now the board fires up and all of the output voltages measure within stated tolerances   I haven't returned the board to the chassis yet.  Clearly if one 10 uF capacitor was bad others may be too, so while waiting for my Digikey order to arrive I went ahead and removed the other eight (service bulletin was wrong; there are actually nine total).  About half of those tested bad on my DMM capacitance mode.

If this is the only thing wrong with my 4396B I will be very happy indeed!

Another photo of the RF guts:

[mark03]

Part 2

With the repaired power-supply board A50 re-connected, I took a deep breath, pressed the power button, and... success!  My 4396B booted up and passed all internal self-tests   I performed some sanity checking on the spectrum-analyzer side and everything seemed more-or-less as expected.  Great!

I then started to run through some of the "external" tests (self-test routines requiring external cable connections or other hardware).  The easiest group "EXT 1" only requires an N-to-N cable between the RF OUT port and the spectrum analyzer "S" input.  And unfortunately, there is a problem   The level/flatness test is complaining.  The same problem appears with RF OUT connected to the network-analyzer R input.  Here's a plot of log |R|:



The fact that this strange dip at VHF manifests in both the S and R input channels is suggestive of a problem in the source module, perhaps an issue with the ALC.  I also know it's not my coax cable because it appears the same with two cables of different type and length.  Because of the relatively low frequency, I was able to confirm that the source module was in fact the culprit by observing RF OUT on an oscilloscope with the analyzer's sweep slowed down.  I could also see that the signal was not only weaker, but also distorted.

Here is a block diagram of the source signal path from the service manual:



The source side is just the reverse of the receive side; a 21.42-MHz reference signal is mixed up to 2 GHz and then down to the source's 100 kHz - 1.8 GHz range with the swept 1st LO.  (The 4396B is fully synthesized, no YIGs or other exotica.)  The ALC target output power is controlled by a 12-bit DAC over a 20 dB range.  There is another 4-bit DAC, driven by calibration constants in EEPROM over a 6-dB range, to roughly level the source over frequency and improve the response of the ALC loop.  The service manual doesn't have a lot more to say about it except that there is an error amplifier and an integrator in the loop.

Fortunately the service menus built into the 4396B's firmware allow for quite a bit of experimentation without physically probing the circuits.  You can turn the various EEPROM calibration factors on or off, open the ALC loop, and manually set ALC DAC values.  I was able to make some useful observations:
1) The dip at VHF did not change significantly as the ALC loop was opened and closed, or random values written to the ALC DAC.  The whole curve (power vs frequency) just shifted up and down.
2) The problem turns out to be intermittent!  I was playing with the source power level when suddenly the trace was flat like it should be.  It seemed to have some dependence on power level, because cranking up the power from 0 dBm towards 20 dBm would usually "cure" it.  And it would often stay fixed for a while after reducing the power again.

About this time I was fiddling with the RF OUT connector while the instrument seemed to be balanced on the edge between the two states, and I made another discovery:  Lateral pressure on the RF OUT connector is also affecting the problem

I decided it was time to get the bottom cover off and have a look at the output cabling.  Here you can see the large CPU board A1 at right, 3-1/2" floppy disk drive at lower left, network-analyzer input assembly (R/A/B ports) at upper left, and the step attenuators for the source output (RF OUT, above center) and the spectrum-analyzer input (S, below center):



Here is a close-up of the A1 board.  We can see the 68332 MCU, 56001 DSP, M27C4002 UV-erasable EPROM containing the bootloader, Xicor X28C64 EEPROM containing the calibration constants, and a 5V 1F supercapacitor for short-term memory retention (about three days, according to the manual).  Happily, there is no battery backup.  (Unhappily, it appears there is no way to update the calibration, should the need arise, without a machine running HP-UX and some impossible-to-find HP adjustments software "ADJ4396B".  But the Xicor EEPROM claims a 100-year data retention.)



(As a side note, wouldn't be nice if companies released their schematics and firmware source code when a product goes EOL?  You can't look at an instrument like this without wondering what an enterprising hacker could do in the way of retrofits.  E.g. digitizing the 21.42-MHz IF directly.)

Back to the problem at hand, nothing appeared amiss at RF OUT:



However, I could reproduce the problem with ~ 80-90% success rate by wiggling the short, semi-rigid SMA jumper between the step attenuator and the RF OUT bulkhead connector.  Could the semi-rigid coax have a defect?  And could that manifest as a dropout at VHF, while retaining a largely flat response up to 1.8 GHz?  The attenuator is very rigid mechanically, so the problem seems to be in this coax jumper or the bulkhead connector itself.  Unfortunately I do not have a right-angle adapter to look at the attenuator output directly, but a replacement cable assembly is on its way.

[mark03]

Part 3

I found a fancy Huber+Suhner "minibend" cable of about the right length to replace the semi-rigid coax between the output attenuator and the RF OUT bulkhead, $12 on ebay.  After installation the source output looks good and the level/flatness tests are passing.  Examination of the original semi-rigid coax shows what looks like a break in the shield at one end:



Whether this was the problem, or something just needed tightening, I don't know.

The rest of the "external tests" require a resistive power splitter to check A/R and B/R performance, so I hooked up a cheap ebay part that I thought would work, only to discover that it was a cellular diplexer not a power splitter.  My mistake for not researching the Aeroflex part number more carefully, but for $10, it's an interesting RF gizmo to measure anyway.  I ordered an actual splitter and made use of the delay to take care of some minor front-panel issues.  The top-most soft key required noticeably more force to actuate than the others, and the protective plastic over the LCD was hazy.

The front panel comes off easily as a single unit with cables for the LCD (shiny ribbon), CCFL backlight (white wires), and controls (gray ribbon).  The CCFL inverter shows signs of rework; from the service bulletins it appears there were some retrofits after the initial LCD become obsolete.



The rubber switches were in good shape overall, although the problem soft-key contact was slightly tarnished.  I cleaned all of the contacts with IPA and followed some advice in an old thread to lightly "sand" the carbon disks by dragging them a short distance across a sheet of paper.  All of the keys are working normally now.



I thought a quick swipe with window cleaner would improve the display window, but the haze was permanent and there were some scratches too.  It was not that serious but the rest of the instrument was so clean I just couldn't put it back that way.  I looked for anti-reflective acrylic in the correct thickness but the choice came down to plain acrylic, or anti-reflective glass.  The nice folks at Frame Destination sell "art glass" cut to size, cheap.  It looks great!!  Maybe I should be more concerned about breakage but I intend to be careful.  There are no screw holes, of course, but it was easy enough to silicone the window to the mounting studs, allowing plenty of time for curing with good airflow before re-assembly.

[mark03]

Part 4

A Mini-circuits power splitter (ZFRSC-42) arrived in the mail today, so I set up the apparatus for "external tests 3" (channel A) and "external tests 4" (channel B).  They pass   Everything appears normal.

For those keeping score, the total invested so far with shipping is about $1100 for the as-is analyzer, $50 in repair parts, and $500 in RF plumbing: cables, adapters, terminations, attenuators, DC block, DIY cal-kit ingredients, etc.  An S-parameter test set is still on my shopping list.  I'm not sure if this qualifies as a "steal" because it's getting into the price range of a working 8753 which would go to 3 or even 6 GHz, vs 1.8 GHz.  But for a newer instrument with a decent spectrum analyzer rolled in, I'm happy.

PM me if you've got an 85046a for sale

[0culus]

That's not a horrible amount of money, honestly. If you just want to work in the "normal" ham bands for instance, it's just about perfect! I've enjoyed following the thread, thanks!

...I have the ultimate goal of working in much higher bands...so eventually I may need one hell of a VNA. 

[DaJMasta]

Not like the plumbing or cal kit would have been any cheaper with a higher bandwidth VNA.  Great news that it worked out!  That's quite a gnarly looking little break...

[mark03]

Thanks @DaJMasta, your advice was much appreciated.  I'm sure rooting through the RF sections would have been highly educational, but I feel lucky that the actual faults were simpler than they first appeared.

[mark03]

For the benefit of anyone else who comes here looking for 4396B information:

It appears that the 4396B does not support S-parameter test sets with mechanical transfer switches.  I found an ok deal on a used 85046A with the original logic board 85046-60001 and mechanical switch.  I was aware of this when I bought it, but because the 8753 VNAs get along with either mechanical or electronic switching, it didn't worry me.  Turns out I cannot perform a two-port calibration without the analyzer attempting to destroy the test set by switching forward/reverse every second.  This happens as soon as the two-port calibration is complete, even if, e.g., you are only measuring S11

No one on the HP mailing list has any suggestion how to fix the problem, and a post on the Keysight VNA forum has no replies.  It appears the only recourse is to upgrade the test set to electronic switching, possibly by gutting a cheaper 75-ohm 85046B.  This has turned out to be an expensive mistake; buyer beware!

[LapTop006]

For the benefit of anyone else who comes here looking for 4396B information:

Useful to know. This thread actually lead me to buying a 4395A, but I'm still waiting for a deal on an S-parameter test set and an impedance test set.

It's really not clear which S-parameter test sets are compatible with what.

[don15246]

This three years ago publish is very helpful. I wonder that mark03 is still here to help us? I bought a preowned Agilent 4396B spectrum/network/impedance analyzer has lot of noise when I measure the filter insertion loss using the 85046A together. Therefore, I did trouble shooting. By isolate Faulty Group Troubleshooting it passed "power on self test, 0", "A1 DRAM and Flash Memory test", "A1 Volatile Memory test, 2", "A30 Front Keyboard test, 17", "A53 FDD test, 18", ""A32 I-BASIC Interface and the mini DIN Keyboard test, 1", "A51 GSP test, 3", " A52 LCD test, 48 to 52", "test 11, 5, 6, 7, 8, 9, 13, 16, 10", but failed  "the A7 Output Attenuator test, 20", which is over limit. When I have power splitter I will do more test. I would like to know if this A7 Output Attenuator test failure causes noise on the measurement or anything else? Which part adjustment shall I do? Thanks

[Extrasolar]

Hi All,

I wonder if there is any possibility anyone of you guys has the 4396B-010 or 4396V-010 Impedance Measurement Function upgrade available to share ??

[copyright927]

Similar problems here. I believe it is just some flag bits in ROM to activate 010. Already backup software and searched in it. But not have any progress yet.