What's all the buzz about an Agilent E3620A power supply anyway?

[ArcticGeek]

So I recently purchased an Agilent E3620A lab power supply on Ebay.   This particular model is a dual output 0-25V 1A power supply.  It was sold as-is and I was able to get it fairly cheap.

Unit arrives and is mostly working.  The first problem I notice is the voltmeter does not switch ranges properly.  It works fine up to 19.99V, but when it switches to the higher range it goes wonky.  I will chronicle that repair in another thread at a later date.

Other than the voltmeter, the unit seems to work properly.  Both channels are clean all the way from 0 up to its max of about 28V.  Both outputs can be loaded up to 1A with no issues, output is clean on an AC coupled scope.  Transient response is good when I connect an electronic load switching from 100mA to 1A.  So far so good.

The surprise comes when I short the outputs.  Everything seems fine and both outputs show a current of about ~1.06A when shorted.  However, output channel 1 seems to be somewhat unusual - I notice a very audible buzzing sound when that output is shorted.  This doesn't happen when channel 2 is shorted. 

I'm attaching a couple of pictures and the schematic to this unit.  This particular unit has a pre-regulator that uses triacs and SCRs to switch the AC output of the transformer to reduce power dissipation.  I suspect the problem is in the pre-regulator circuit, so I will be investigating this.  I have already found and fixed the problem, I will update this post later with more of the details.

[anotherlin]

I'm attaching a couple of pictures and the schematic to this unit.  This particular unit has a pre-regulator that uses triacs and SCRs to switch the AC output of the transformer to reduce power dissipation.  I suspect the problem is in the pre-regulator circuit, so I will be investigating this.  I have already found and fixed the problem, I will update this post later with more of the details.

At first glance of the schematics, it seems the triacs are rather here to switch between secondary taps of the transformer so the linear supply has less to dissipate.
Rather than being a "true" pre-regulator, as used in the HP 6209B for instance.
Maybe the triacs of channel 1 is switching on/off constantly when shorted, whereas it should have just switched once to the correct tap.
Or maybe, it is just the transformer that is buzzing, this happens.

Please update, I'm quite curious as what is the problem.

[anotherlin]

Oops, missed page 2 of the schematics.
The triacs control circuit is kinda complex, so yes, that's really so kind of pre-regulator, not just a transformer tap selector.

[bitseeker]

Looking forward to seeing what you found, AG.

[ArcticGeek]

Anotherlin,

I think I used a poor choice of words when I said it was "switching" the AC output of the transformer, because there really isn't any high speed switching that is occurring at the transformer.  Rather, it is simply selecting the particular tap(s) of the transformer to be used in full wave rectification.  I have seen the pre-regulator circuits of some of the older HP supplies, and in those power supplies they actually phase control the AC waveform to control the voltage at the filter caps.  These power supplies are nowhere near that complex, they are simply choosing the appropriate taps of the transformer secondary to use for rectification. 

The manual describes the triac/scr control with a little bit of detail, and goes on to say that output 1 has 2 distinct regions of operations and that output 2 has 4 distinct regions of operation.  They give no reason why output 1 is controlled differently than output 2, even though both outputs are identical.

So using a DMM I measured the voltage across the filter cap just prior to the pass transistor on both output 1 and output 2.  Output 2 does indeed have 4 different ranges that it operates in, depending on the output voltage is set to.  Loading the output with a 100mA load, this is what I found:

Output Voltage #2          Cap Voltage
0-8V                                21.0V
8-15.9V                           26.2V
15.9-23.8V                      37.5V
23.8V and up                   42.3V

This is what I would expect based upon what the manual described.  Next I fully loaded output 2 with a 1A load and repeated the test with similar results:

Output Voltage #2          Cap Voltage
0-6.4V                             18.7V
6.5-14.4V                        23.5V
14.4-22.3V                      33.8V
22.3V and up                   38.2V

Notice that the cap voltages are now lower because of the heavier load, and consequently the switching point occurs at a lower output setting.  This is because the circuit is monitoring the DC voltage prior to the current sense resistor at the output.  The current sense resistor is 1.78 ohms, so at 1A the voltage drop is 1.8V across the sense resistor.  Consequently, the switching point is about 1.6V lower than it is with a 100mA load (at 100mA the drop would be 178mV).  This is pretty much what I had measured.  I wasn't measuring the switchover voltages that accurately when I took the measurement, I was simply turning the knob slowly and trying to note when the cap voltage jumped up....so my measurements can easily be off by 100mV or so.   There is also about ~800mV of hysteresis in the range switching going up and coming down.  So output #2 is working just as I would expect.

So next I check output 1.  It's behavior is very different.  According the manual and the schematic, there should be 2 distinct ranges of operation.  But that's not what I see.  At 100mA load, the capacitor voltage doesn't change.  It's stuck in the "high" range with the raw DC voltage of about 42.3V, regardless of the output voltage setting.   Hmmm, this is not what it should be.  Loading it at 1A results in similar behavior, with the raw DC measuring about 37V regardless of the output voltage setting.  Furthermore, when the output is low, around ~3V or less, I get this buzzing sound I described.  What's going on?

Taking a closer look at the schematic for output 1, the range switching is controlled by U18 as shown in this marked up schematic.  It looks like this is working as it should, when the output is less than 12V or so, the comparator is high.  This SHOULD shut off the triac and keep the voltage lower.  When the output voltage is high, above 13V, that comparator goes low.  That would turn on the triac and allow a higher voltage on the filter cap.  So it appears that the comparator circuit is working.  Is the triac bad?  Is Q9 defective?  Neither are shorted and Q9 checks good.   What's going on here?   

Not to keep everyone in suspense...but more to come later - I need to get back to my REAL job now.

[Armadillo]

If you have not found the problems, then;

I reckon the buzz has to do with the transformer core, adhesive and the laminated layers separates and breakaway thus vibrates. It will get worse with the loads and the power line harmonics.
To alleviate the harmonics problem, disconnect all other switching loads from any of the the same circuit looping power points like computers, TV, and see if it soften the buzz somewhat.
Also try tightening Up the Shin Woo laminates, if it helps.
To prove it, you may try experimentally swopping the channel for testing since only 1 channel buzz, for now.

[anotherlin]

Taking a closer look at the schematic for output 1, the range switching is controlled by U18 as shown in this marked up schematic.  It looks like this is working as it should, when the output is less than 12V or so, the comparator is high.  This SHOULD shut off the triac and keep the voltage lower.  When the output voltage is high, above 13V, that comparator goes low.  That would turn on the triac and allow a higher voltage on the filter cap.  So it appears that the comparator circuit is working.  Is the triac bad?  Is Q9 defective?  Neither are shorted and Q9 checks good.   What's going on here?   

Measure the capacitor voltage as you have done for output #2 and we'll all know!

I've taken a look at the E3615/E3616/E3617 service manual, which has more detailed explanation.
The circuits are very similar, the triacs are indeed switching taps of transformer secondaries.
The only goal seems to reduce the power to be dissipated by the pass transistor.
So all taps are 1A rated. And even if the triacs for channel 1 don't work as intended, the supply will work fine, albeit not efficiently.

My HP 6209B can supply 100mA at 320V (32W) and its transformer buzzes quite loud if I short it at max load.
Maybe your 3620 is working as intended (low range taps selected) but your transformer just happens to be noisy.

[macboy]

Spoiler alert
I suspect it has everything do to with the pre-regulator's LM393 comparator and its open-collector output. I found this flaw years ago and posted about it at least a couple of times. I fixed mine up nearly 8 years ago.
p.s. good work finding the flaw and fixing it. Most people wouldn't look for a blatant design flaw in an HP/Agilent device.

[anotherlin]

I suspect it has everything do to with the pre-regulator's LM393 comparator and its open-collector output. I found this flaw years ago and posted about it at least a couple of times. I fixed mine up nearly 8 years ago.
p.s. good work finding the flaw and fixing it. Most people wouldn't look for a blatant design flaw in an HP/Agilent device.

I've read your link and re-read the schematic of the TRIAC control of output #1.
From what I understand, I wouldn't call it a "blatant design flaw" but I may be wrong of course.
And I would be very interested to learn what is your fix.

Here's my analysis: it's just one LM393 comparator (U18B) that turn on a PNP transistor (Q9), which turns on the TRIAC (Q5) through an opto-isolator, if a threshold is crossed.
The reference voltage (threshold) at pin 6 (2IN-) should be -2V (R57/R58 voltage divider).
The output voltage to compare (2IN+ at pin 7) is multiplied by a ratio of about 0.23 through voltage divider R59/R60.
So threshold voltage should be -2 / 0.23 = 8.69V.
U18B output (pin 7) sinks current if 2IN- > 2IN+, so TRIAC should be on if output is above 8.7V, makes sense.
Otherwise the open-collector stays open (high-impedance) and Q9 is off.
In a perfect world, we would stop here.

I suppose R62 (562k) is here to supply enough current to the collector of U18B output NPN transistor so that base current from Q9 cannot leak into it.
My guess would be that the open-collector is not "open" enough, that is its high-impedance is not high enough, so enough current flow from Q9's base to turn on the TRIAC.
And/or that Q9 is too "sensitive", and a very small leakage would be sufficient to turn on the TRIAC.
One way to solve the problem would be to lower the value of R62. This would be a the expense of the "precision" of the R59/R60 voltage divider, but we do not need a particularly threshold point.

So if my analysis is not completely wrong, I would say that HP/Agilent/Keysight design is not completely flawed.
But may be they did all their testing with a certain batch of Q9 and U18, but actual component used in production are different.
And the selected value for R62 is not conservative enough.

[macboy]

Consider what happens when U18 (LM393) output is set to high, which means that the power supply output voltage is below the tap-switch threshold. Its output is an open collector one, so it actually is just a high impedance node. The emitter of Q9 sits at +12V, and there is a string of resistors: R61, R62, R59, and R60 which will pull some current through the base of Q9. Effectively it is about 580k connected to a volt or two below ground. This will result in just enough current through the opto to switch on the TRIAC regardless of the output voltage. (Measure TP1 and TP2, it is always >40 V).

If you simply add a pullup (10k or so) from U18 pin 7 to +12V, then zero current is pulled through Q9 base, and the opto is truly OFF, so the pre-regulator works. Then TP2-TP1 measures about 25 V (not 40 V) when the output is at a low voltage. I also changed R59 so that the threshold is closer to 15 V, not the 8 V or so that it switches at normally. This gives cooler running when charging SLA batteries, but still provides plenty of overhead for regulation. Finally, I changed the sense point (top of R59) to be after the current sense resistor (R55), not before it, so that varying output current can't affect the pre-regulator switch point.

[ArcticGeek]

Sorry for the delayed response, been busy at the real job.

As Macboy indicated, the problem is with the switching of U18.  When U18 is high, it really does not fully turn off the opto-coupler and allows the triac to fire.  That's because of the base leakage current from R61 and R62.   What is actually causing the buzzing sound is the triac is only trigger on every other line cycle.   The buzzing is just the lower frequency of 60Hz (vs 120 Hz), and probably because there is a higher inrush current into the cap so the transformer buzzes more. 

Attached are a couple of waveforms of the bulk filter cap in the high and low range.  Notice on the low range that the ripple frequency is only 60Hz - this is because the triac is only firing every other cycle.  This is most likely because it is harder to fire a triac in one or two of the four possible quadrants.  I don't remember which quadrants are worse, but I do recall that it takes more current to fire a triac in some quadrants than others.

As for a fix, as macboy mentioned you can add a resistor between the base and emitter of the transistor to fully turn it off.  This definitely works, as that was the first thing I did to test the circuit.  But I don't really like adding additional components to the board just because of the ugliness of the rework, so I tried to think of other ideas.  Another option would be to use a pre-biased "digital" transistor such as a FJN4302 in place of the 2N2907.  These transistors would have the resistor built in from the emitter to base.  I didn't have one of those parts on hand though.

So the other idea I had was to use a rail-to-rail output op-amp in place of the comparator at U18.  An LM6132 works nicely here, as that part will saturate within 50mV or so of the positive rail.  I had one of these parts on hand, so I swapped the comparator for the LM6132.   Since it has a push/pull output instead of an open collector output, it pulls the output all the way to the positive rail and fully turns off the transistor.

The funny thing is I would never have suspected a problem with this power supply if it wouldn't have been for the buzzing sound.  While the buzz is not terribly loud, it was a noticeable difference between the output #2, and that's what prompted me to look closer.  The supply seemed to work fine except for that weird buzzing that was only present on output #1.

As macboy also mentioned, it is somewhat troubling to find a mistake like this in an Agilent piece of equipment, especially given their good track record over the years.   But what I find even more troubling is that this is a somewhat bonehead mistake.  This is the kind of error I would expect to find from a rookie designer....a seasoned analog designer would likely not make a mistake like this.  It makes me wonder what kind of engineers Agilent/Keysight has designing nowadays!

[nfmax]

Ahhh! If the triac is firing only every other mains cycle, it is putting a DC component into the transformer secondary winding current. In turn, this puts a constant flux into the core, which is therefore saturating on on one of the flux peaks (actually at one of the zero voltage points) and that is what is generating the buzz! It would be more noticeable on 50 Hz mains.

I will now have to go and check mine...

[Armadillo]

How would you account for the ramp at the high output [when the opto is fully ON]? and how much voltage ramp Delta would that be [not shown on the scope]?

Edit: Your 1st post said that it when kongky at high output and only 1 channel is affected. If its design flaws, wouldn't it affect both channel and almost all instrument of same design? Also 400mV ripple at half wave rectification will cause transformer humming/buzzing is hard to swallow wouldn't you say?

Edit: CR4 rectifier bridge is independent of the triac and diode CR5, so you should still be seeing CR4 doing its rectification at 120 Hz on the scope. I would say Check CR4 full bridge.

[anotherlin]

Consider what happens when U18 (LM393) output is set to high, which means that the power supply output voltage is below the tap-switch threshold. Its output is an open collector one, so it actually is just a high impedance node. The emitter of Q9 sits at +12V, and there is a string of resistors: R61, R62, R59, and R60 which will pull some current through the base of Q9. Effectively it is about 580k connected to a volt or two below ground. This will result in just enough current through the opto to switch on the TRIAC regardless of the output voltage. (Measure TP1 and TP2, it is always >40 V).

I agree that the simplest and most robust solution would have been to add a 10k pullup resistor connected to +12V at the ouput of U18B.
In fact, they've done it for the comparators' outputs of  the TRIAC control circuit of channel 2.

I thought that R62 was added to act as a pullup resistor: It is connected to R59 of the R59/R60 voltage divider, which should swing between 0V and -5.75V, and U18 is grounded (pin 4) to -12V. Supplying current to the high impedance node of the comparator output, and thus preventing Q9 base to leak through R61 into the output open collector.
But yes, current should be able to flow between Q9 base -> R61 -> R62 -> R60 as it is below ground.

So what is the point of the R62 feedback resistor? In particular, we have similar resistors, for instance R20 on U7A of the TRIAC control of channel 2, it uses the same value of 562k?

[anotherlin]

As macboy also mentioned, it is somewhat troubling to find a mistake like this in an Agilent piece of equipment, especially given their good track record over the years.   But what I find even more troubling is that this is a somewhat bonehead mistake.  This is the kind of error I would expect to find from a rookie designer....a seasoned analog designer would likely not make a mistake like this.

My "real" job is software engineer for a big EDA company. Believe me, the code I go through sometimes has bugs and errors that rookie shouldn't even make. So I'm not surprised at all, shit happens !

The supply is working and meeting the performance specification, except for the excess heat dissipation and the buzzing noise.
The heatsink is probably over-rated so it handled the excess without problem. Some transformer buzzing is expected at high load, so people thought it was normal. And that may not be that much noticeable if you have other test equipments featuring loud fans around. That may explain why that flaw did go unnoticed.

[macboy]

How would you account for the ramp at the high output [when the opto is fully ON]? and how much voltage ramp Delta would that be [not shown on the scope]?

Edit: Your 1st post said that it when kongky at high output and only 1 channel is affected. If its design flaws, wouldn't it affect both channel and almost all instrument of same design? Also 400mV ripple at half wave rectification will cause transformer humming/buzzing is hard to swallow wouldn't you say?

Edit: CR4 rectifier bridge is independent of the triac and diode CR5, so you should still be seeing CR4 doing its rectification at 120 Hz on the scope. I would say Check CR4 full bridge.

The waveform for the "opto fully on" case looks normal. There will always be some ramping up as the AC waveform rides the sine wave shape to the peak, then down (linearly, with a constant load current) as the AC input sine wave falls below the charge level of the capacitors. The scale is shown on the screen captures: 10 V per division. The ripple appears to be about 8 or 9 V for the bad case, and about 3 or 4 V for the good case. The hum is caused by the asymmetric loading of the transformer, resulting in saturation due to the DC component (current flows only one way).

The second channel has a different pre-regulator design with four (not two) different voltages. It has no such design flaw.

The CR4 bridge provides only ~24 V peak compared to the ~40 V peak provided when the TRIAC conducts through CR5. The voltage on the reservoir caps never falls low enough to be charged by the 24 V peak.

So what is the point of the R62 feedback resistor? In particular, we have similar resistors, for instance R20 on U7A of the TRIAC control of channel 2, it uses the same value of 562k?

Note that the feedback resistor is from output to + input, not the - input. Thus, it adds postive feedback. This provides hysteresis on the comparator's trip voltage, which prevents oscillation or glitching when the output voltage is sitting just right at/near the trip point.

[macboy]

Sorry for the delayed response, been busy at the real job.

As Macboy indicated, the problem is with the switching of U18.  When U18 is high, it really does not fully turn off the opto-coupler and allows the triac to fire.  That's because of the base leakage current from R61 and R62.   What is actually causing the buzzing sound is the triac is only trigger on every other line cycle.   The buzzing is just the lower frequency of 60Hz (vs 120 Hz), and probably because there is a higher inrush current into the cap so the transformer buzzes more. 

Attached are a couple of waveforms of the bulk filter cap in the high and low range.  Notice on the low range that the ripple frequency is only 60Hz - this is because the triac is only firing every other cycle.  This is most likely because it is harder to fire a triac in one or two of the four possible quadrants.  I don't remember which quadrants are worse, but I do recall that it takes more current to fire a triac in some quadrants than others.

...

In two the E3620As that I fixed, neither ever hummed. The TRIAC always fired, so the pre-regulated voltage always had 120 Hz ripple not 60 Hz as in your case. In my case, the symptom that I noticed was excess heat. The heatsink got very hot with 1 A output current and low-ish output voltage. I was simply going to change the switch over voltage when I discovered that it never did switch over. Then I hunted down the cause.

[Armadillo]

Like I said, its hard to swallow with that kind of half bake analysis and calling professional engineer a rookie..... LOL   

[ArcticGeek]

Like I said, its hard to swallow with that kind of half bake analysis and calling professional engineer a rookie..... LOL   

I don't believe may analysis was "half-baked" as you indicate - I believe I have a pretty good understanding of what was causing the excessive "buzzing" sound and why the unit wasn't working as it should have.  What part of my analysis do you consider half-baked?

I too noticed the excessive heat at about the same time as I observed the buzzing sound....but the buzzing sound was a noticeable difference from output #2 that caused me to investigate.

And yes, I consider this a rookie mistake.  Not having a pull-up on the output of an op-amp is not something a good analog designer would do.  A good designer would also take into account the leakage from positive feedback resistance and how that would affect transistor operation.  At first glance it might seek like an esoteric oversight, but if you are designing something like this you should understand how the circuit works and make sure that it operates as intended with margin.  You'd be very hard pressed to find an oversight like this in an old HP power supply...those things were built like iron.

[Armadillo]

Like I said, its hard to swallow with that kind of half bake analysis and calling professional engineer a rookie..... LOL   

I don't believe may analysis was "half-baked" as you indicate - I believe I have a pretty good understanding of what was causing the excessive "buzzing" sound and why the unit wasn't working as it should have.  What part of my analysis do you consider half-baked?

I too noticed the excessive heat at about the same time as I observed the buzzing sound....but the buzzing sound was a noticeable difference from output #2 that caused me to investigate.

And yes, I consider this a rookie mistake.  Not having a pull-up on the output of an op-amp is not something a good analog designer would do.  A good designer would also take into account the leakage from positive feedback resistance and how that would affect transistor operation.  At first glance it might seek like an esoteric oversight, but if you are designing something like this you should understand how the circuit works and make sure that it operates as intended with margin.  You'd be very hard pressed to find an oversight like this in an old HP power supply...those things were built like iron.

Well for someone who cannot understand full wave ripple and bridge and always trying to align with others to elevate its position and worse call professional engineer a rookie, I say this is half baked man!.

BTW, they got a name for the pair. Its call the Sziklai pair and not half baked pair.

The way I see it, the little bit more you know, then all others become "Rookies" to you.

Well, I am not the one having the problems and I am mum!.  Cheers!

[ArcticGeek]

Well for someone who cannot understand full wave ripple and bridge and always trying to align with others to elevate its position

I wasn't trying to align with anyone...it was macboy who decided to steal the thunder and take all the glory.  Fair enough, he found the problem a while ago, but never posted any waveforms.  At least I posted some scope shots to back up the analysis.

I resent your tone here....but 'nuff said.  Later

[Armadillo]

it was macboy who decided to steal the thunder and take all the glory.

This is a repair forum, not about "thunder and glory".     

If you have found the "problems" and know all, then please move-on.