Agilent 4395A 500MHz Network/Spectrum / Impedance Analyzer
Power Supply Repair
Also should apply to 4396A 1.8GHz
YouTube Video (but not yet)

Yet another $40,000 Power Supply repair story 

Why $40,000? I am responsible for a few Agilent 4395A 500MHz Network/spectrum/impedance analyzers. 2 of them are dead or recently defunct. These are ~$20,000 (new) machines, both with dead 24V power supplies. These are great instruments used at my day job for many tasks, particularly measuring ultrasound transducer impedance in the 10-100MHz range. One has been hacked back into operation, the other one is dead.

4395A

These instruments are no longer built. The power supplies are built by Artesyn, with no schematics available, of course. Agilent (now Keysight) might be able to repair it, I didn't look into that option. In fact, I notice that these instruments have even less repair information in their owner and service manuals than other HP / Agilent RF equipment. Not even a system block diagram! Come on, Agilent!

When one unit died a few years ago, I was in a hurry to get it back up and running. Its power supply is a fairly generic 24V switcher, I'm guessing about 150-200 Watts. A second board in the instrument takes the 24V and converts it to all the instrument operating voltages. I had a few 24V switchers lying around, and hacked in a 24V, 115W Condor supply. The instrument came right up. I measured the 24V DC current which was close to 5A. Since the first power supply I chose was only 115W, or 115W/24V = 4.8A max, I didn't feel comfortable having the supply operate so close to its maximum. I found a larger, 250W supply and mounted it in the unit. For cabling I adapted the existing cables to the new supply. I built a quickie, flat aluminum bracket to adapt the power supply to the instrument. It worked swell. See the photos below.   I put the dead power supply aside and put the instrument back into service. 

Then in early 2020, a second 4395A acted up. It was intermittent and would work for a few months and then die for a week. Finally it died for good in early 2021. I opened it up and sure enough, no +24V. Deja-vu all over again. Time to repair the old supply, and if I'm real lucky, the other unit will have the same problem. You know, kill two stones with one bird.

on bench

The AC input connector is a 5 pin 0.156" Molex / Amp connector with only 3 pins loaded. I wired up a power cable to power it up on the bench. To do proper proper measurements, and to see the circuit traces which are all on the back of the board, I removed the PC board. This was straightforward: remove 5 screws, one snap-on plastic spacer, and 5 transistor heat sink mounting clips. For safety, I should have used an isolation transformer, but did not have one handy. So I couldn't use a scope to view signals, just a DMM. 

First, I (again) visually inspected the components and PCB traces. No visible damage, capacitor leakage, or bulging capacitors. No darkened components or PCB areas. No bad smell.

There are only 3 ICs on the power supply board: UC3854, and UC3843, and an LM399 (comparator). The UC3854 is a Power Factor controller. The UC3843 is a current-mode switching supply controller. Both of these drive 800V power FETs. I pulled the data sheets of the controllers to see the basic circuitry. I assumed that the general schematic of the power supply would be close to the block diagrams on the data sheet. 

Here is the typical UCx854 Power factor controller circuit. Its job is to charge the main filter capacitor without the high current, fast pulses that would result from a simple diode-capacitor circuit. It cleverly controls the PWM of a flyback circuit to make the input current approximate a sine wave.  European regulatory standard IEC 61000-3-2 limits line current harmonics, in order to prevent distorting the overall power grid. To meet the standard, line voltage current harmonics must be limited, particularly on power supplies > 80Watts. See https://en.wikipedia.org/wiki/IEC_61000-3-2

These two "typical application" schematics from the IC data sheets are great. Combined, they pretty much show the major components of this power supply.

uc3854

Below is a 'typical' UC3843 Current mode controller circuit for a +12V 50W power supply. It controls the DC-DC converter which converts line-connected +400VDC on the main filter cap to an isolated (grounded) +24VDC. It does so by controlling the pulse-width of the PowerFET that switches the +400V to the main transformer.

In the 4395A supply, there are 2 power-FETS. The circuit looks like a conventional half-bridge; there is a large film capacitor. But I don't see a high-side gate-drive transformer. So the precise circuit is a mystery to me.  But hey, my job here is to fix this thing, not to over-analyze it. Some level of understanding is necessary to get the job done.
uc3843

A few measurements: Fault Found!

I measured the voltage on the main filter capacitor, and it was about +160V This means the AC input fuse, line filter, rectifier and main 450V filter capacitor are basically working. It did not occur to me that the low voltage meant that the UC1854 and its FET were not working. If they were, the voltage on the main filter cap would be +400V, not +156V. But I moved on to the current mode controller anyway. Ignorance is bliss.

I next measured UC3843 VCC (VCC pin 7 to GND pin 5) and it was about +0.6V. It should be about 15V. This meant the controller IC was not getting either the startup voltage via Rstart or the running voltage via Rvcc and Dbias. So either Rstart was open or Cvcc (120uF) or the UC3843 was shorted. I found two 220K ohm 1/2W resistors on the board, figuring one of them was Rstart. One measured 220K ohms in circuit and the other measured > 5Meg. I un-soldered the open 220K from the board and it measured open. I checked the circuit and these two 220K resistors are in series between the +400V and the UC3843, so Rstart is 2 x 220K or 440K ohms. Each resistor burns P = V^2 / R = 200V * 200V / 220K = 0.18W. I replaced the dead resistor with two 100K power resistors in series (200K) and the power supply powered up, providing +24V. Hooray!

At this point the main capacitor measured 400V, meaning that the two controller ICs are somehow connected: without the Current controller working the Power factor controller also doesn't work. I don't understand it, but I don't care. Problem solved! I also don't understand why the 0.5W resistor burning 0.18W failed. Bad resistor I assume. I ordered a handful of 220K 2Watt resistors so I can replace both of them in both power supplies. Now on to troubleshoot Supply #2

Supply 2: Deja-vu all over again

The repaired supply was installed in the unit, and it powered up and operates fine. I removed the power supply from the 2nd unit, immediately went to the 220K resistors, and as I surmised, one was open. I replaced both resistors and the 24V supply came up. Happy day! I removed the hacked power supply and wiring adapters,  installed the original supply, and all is well. These 220K resistors appear to be 'normal' low-cost carbon film 1/2 watt resistors. Body size is 9mm x 3mm. They don't appear to be power rated, not flame retardant, not surge rated. I suspect that the problem with them is that they did not like having 200V applied, and ultimately failed. I checked several resistor data sheets and all 1/2 watt resistors are rated 300V or more. Moral of the story? Don't apply 200V to a cheap resistor.

While I had the supply on the bench, I tested it at full load. It puts out 10A and goes into current limit at 10.3A.  I suspect the power supply is designed for 7-8 amps continuously. As I stated above, the instrument draws about 5A.

Hacked 24V power supply.

Here is the power supply that I hacked into the first broken unit. It is a Condor GPMF250-24, a 24V, 10A, 250W power supply. It worked very well. Of course, you don't have to do this hack if you can repair the original supply.

hack1

Note the flat sheet metal bracket that adapts the Condor supply mounting holes to the 4395A mounting holes. Also note the 2 adapter cables, custom made to adapt the Molex 0.156" pin connections of the original supply to the Condor screw terminals. I didn't want to modify the instrument so that the original supply could be easily reinstalled. Glad I didn't.

hack2



   Dave's Home Page

Last Updated: 3/20/2021