Tektronix 2232 scope.... in need of a doctor..

[WastelandTek]

good job Vince!   

[Vince]

Thanks !     But of course the real thank you goes to all the knowledgeable people who helped me through this ! Thanks a million chaps ! 

Fan is sorted : just ordered the two Sunnon models I talked about.... turns out they were the ones that made most sense anyway.

I compared all the fans from the various manufacturers that Farnell would list, looked at a few datasheets.. and in the end they really are pretty much all the same... the laws of physics apply to all of them, so there is no reason why one manufacturer would achieve performances significantly better than others.

The physical dimensions of the fan kinda dictates what can be achieved...  basically, there is no miracle : the "silent" fans are silent simply because they run slower, and therefore give lower airflow. There are 3 speed ratings: "slow" (2000+ rpm), "medium" (3000+ rpm) and "Fast" (4000+ rpm), each speed gives a flow and noise combo figure. 

The old fan was putting out 19cfm at 4600 rpm, fast type then, hence noisy.

Modern units must have improved aerodynamics of the blades a bit, because now the 19cfm class fans fall in the medium speed range. The high speed ones therefore achieve better airflow, 23cfm or so.

I ordered the older "KD" series, being obsoleted, hence cheap, which is the fast kind hence with increased airflow. I also ordered the modern replacement "Maglev" series, which produces the original 19cfm but runs slower so should be less noisy. I will try both and see for myself. The fan is given for 60,000 hours life expectancy, so I will die long before the fan does.

Farnell has less choice than Digikey, so I could not get the sleeve type bearing that Tek specified. Sunon does offer it, on both of these models, but it's not the ones Farnell chose to pick. The old KD series will come with a ball bearing, and the modern Maglev will come with the " Vapo" type bearing, whatever that is. Seems like the latest hype in the fan bearing scene, I assume.
I will compare them and pick the one that pleases me best, the other one will still be there handy, as a back up if need be.

As for my question about the relationship between air flow and static pressure, the manufacturers give graphs that plot the latter vs the former.

Hopefully I will receive these fans this week and be able to reassemble the scope for good in the coming days... and then move on to the next "project" in the pile...

[WastelandTek]

There has been a great deal of effort spent on optimizing impeller design in recent years, yes.  It is actually pretty impressive how much noise to flow/static pressure performance they have been able to squeeze out of these constrained form factors IMO.  One would have thought that this would all have been sorted years ago, but no.

[David Hess]

Hi Dave, thanks for the comments especially the operation of the probe signal circuitry, that was interesting.

That diode disconnect method is common in Tektronix designs and particularly suitable for a operational amplifier based one because it removes amplitude variation do to operational amplifier output saturation.

- identification of the two ribbon cables : Tek did not label them, so no easy way at reassembly time, to know which is CH1 or CH2. Easy to figure out though : just swap the cables in store mode if they don't show up where they should...

- orientation of the above cables : again just enter store mode and flip the cable(s), live, 'til you get the correct polarity on the screen.. result : yes, Tek indeed did NOT match pin #1 of the cable connector, with pin #1 of the socket on the board ! How "nice" from them.... grrr...

I had to do the same thing on my 2230 on which I spent weeks tracking down a problem in the storage channel switch.  Swap the cables to get channel 1 and 2 right and then flip the cables to get the polarity correct.  See the photograph below.

The old fan was putting out 19cfm at 4600 rpm, fast type then, hence noisy.

Modern units must have improved aerodynamics of the blades a bit, because now the 19cfm class fans fall in the medium speed range. The high speed ones therefore achieve better airflow, 23cfm or so.

Watch out for the voltage specification; a 12 volt fan is specified but it is actually running on 8.6 volts.  It is not real clear to me if the 2232 fan specification of 12 volts, 1.5 watts, 4600 RPM, and 19 CFM is suppose to apply at 12 volts or 8.6 volts where they used it.  Sunon makes a fan with a suspiciously close part number which is more powerful than the Tektronix specification so I suspect this may be the case.

There has been a great deal of effort spent on optimizing impeller design in recent years, yes.  It is actually pretty impressive how much noise to flow/static pressure performance they have been able to squeeze out of these constrained form factors IMO.  One would have thought that this would all have been sorted years ago, but no.

The fan on the older 2230 has to be seen to be believed; it looks like a model of something NASA would use for a wind tunnel.  But the fans on both my 2230s and my 2232s are practically silent so I am dubious of modern improvements in aerodynamics or at least that they are commonly used.

[tautech]

Nice work Vince and well done for taking the time to write it all up.......thinkings out loud included.   

[wasyoungonce]

Well done Vince....you have that puppy nearly tamed!

[Vince]

Thanks chaps. Yes I try to think loud and document... so that other newbies like me can learn as much as possible when following this repair.

David for the fans well... when I get them, the two nes ones and the old OEM one, I think I will power them all at the same time on 12V, and get a feel for what kind of airflow they produce so I can compare them.

While I am waiting for the fans to come in, I thought I would do some measurements on the regulator and inverter, for educational purposes (and for peace of mind as well).

Below are PDF copies of the SMPS schematics and waveforms.

The waveforms are remarkably similar to those in the service manual, shape, amplitude and frequency, but was still interesting, see below. In order of appearance :

- WF #63 : ramp up voltage for the PWM oscillator.

- WF #64 : PWM output : duty cycle is less than advertised, but I guess that's to be expected since Tek based his measurements obviously on US mains, 115V. It's 230 on my side of the pond, so obviously the duty cycle has to be lower in order to compensate for that.

- WF #65 : collector voltage on one of the two paired transistors driving the primary of the main transformer.

- WF #66 : If I read the schematics about right, this waveforms is an image of the current flowing through the primary winding. The waveform in my case is a little weird as you can see : kinda "doubled"... but the trace is actually perfectly stable on the screen. Frequency is double that of the collector voltage, note. So my explanation would be : one of the two transistors is drawing a bit more current than his brother. Hence the waveform is the same shape, but slightly taller. My analog scope obviously has no means to tell one cycle from the other, so both get super-imposed, making the trace look doubled. But I guess if looked at the same signal with my digital scope, and did a single sweep, then we would see what is actually going on : a tall cycle, followed by a shorter one, then a taller one again, etc.  Yea, I might look at it with my TDS544A to confirm that. Can't make the 2232 probe itself like I would "normally" do, because obviously the SMPS ground is not referenced to chassis ground. I don't want to blow my scope again...

Now the question is : should I be worried about this current difference ? Is it caused by the primary winding(s) not having exactly the same resistance/number of turns.. that would seem strange. Is it cased by one of the transistors not being fully saturated ? That would seem unlikely too, as I expect them to be waayyy over-driven to make sure they don't overheat/cook/die by dissipating abnormal amounts of power.  I guess that's easy enough to figure out, I could just measure the voltage drop across both transistor, and also power off the scope and measure the resistance of the two halves of the primary winding. But the scope is working just fine so I am not overly worried ! ;-)

Then some more interesting stuff I did not expect !  Might be of interest to newbies, sure caught my attention !  Before I started to check waveforms, I checked DC levels where indicated on the schematic.

Main rail is fine, 42+ V as expected.
The " fun " part came when I tried to probe across R927, it says it should be 6.8V.  As soon as I put the multimeter probes on these terminals... the scope would immediately shut down !!!  And would start again just as soon !  And this was perfectly repeatable, like a Swiss watch !
So, detective cap on again ! 

That resistor is part of the voltage divider that drives the base of Q928 which itself drives Q930 which provides the power supply for the PWM chip... hence has the magical power of turning the whole scope on and off at will... so what I was witnessing was not completely unreasonable.
What seemed unreasonable though, is why Q928 would "trip" when I measure the voltage across R927 !

1) I thought maybe there is a cold solder joint on that tranny, and putting the probe on that nearby resistor just so happens to flex the PCB a little, and upset the dodgy solder joint/transistor.  that said, I was hardly pushing on the PCB with my multimeter probes, I was just "caressing" it, barely touching it. So that seemed strange. But just to be sure, I grabbed a piece of plastic and applied pressure on the PCB right on the transistor.... nothing happened. So, not a cold solder joint... joint looked good anyway, and that tranny was actually the only one that did NOT get replaced when I refurbished the regulator, so it could not have been me doing a bad soldering job...

2) Then I thought OK, if it's not mechanical, then has to be an electrical problem... but what could be happening. If it were a FET, hence enormous gate impedance, I could understand that it could be possibly trip "just like that", by picking up ambient noise,... assuming the gate was floating... but 1) it's not a  FET it's a BJT and 2) its base is hardly floating... and its solder joint is good so there is a good connection to the rest of the circuit.

3) The biasing voltage divider has quite high value resistors, so I thought maybe the 10M resistance of the multimeter, across that 100K resistor, might upset/load the biasing just a tad... just enough to trip the tranny. But... 10M is a couple orders of magnitude higher than this resistor, so no way, stop kidding me please...

4) maybe the multimeter impedance is NOT 10M... maybe somehow it's faulty/lower than this. So I used another multimeter to measure the impedance of the first meter... 2.5M instead of 10 !  Eh, what's going on ?? So cross checked.. and the other multimeter also presents a 2.5M impedance not 10 ! Suspicious... likely it's some problem with the way their ohmmeter works. No definitive answer on this subject, but I didn't want to waste time chasing red herrings, because 1) the fact that they both read 2.5 instead of 10 likely means that they are both good and that there is a valid explanation for this unexpected reading and 2) even 2.5M should still be plenty high enough not to upset the biasing of that transistor anyway !!!

So I started random experiments to try to get some clues.

6) I tried touching the resistor with only ONE probe/test lead leaving the other probe floating, resting ton the bench... this way the meter would not load the biasing circuitry anymore.  I first put that unique/lonely probe on the bottom terminal of the resistor, the one that is NOT connected to the base of the transistor... oh oh, it does NOT power off anymore, some change !

7) did the same thing but this time probing the upper terminal of the resistor, hence probing the base of the transistor : shuts down immediately !

So at this point I knew that the transistor "tripping" has nothing to do with the impedance of the meter (whatever it actually is...) loading the biasing circuit, and I also knew that it powered off only if I made direct contact with the base of the transistor.

So, again the symptoms were similar to a FET with its gate floating, picking up noise, presumably from the long test leads.
But again, it's not a FET and it's not floating ! Still, that was my best shot. So I did more experiments in that vein :

I contacted/touched the base of the transistor with a plastic piece (the handle of an anti-static/ESD safe brush), thinking it's picking up noise, it should not pick anything via a plastic element !  Bingo : that did NOT power off the scope !

9) So, metal is has to be. Then assuming it's picking up noise because the test leasds are both metallic and very long... maybe using a metallic piece but that's very short, would not trip the tranny. I grabbed a tiny/precison screw driver... touched the base, nothing ! Put some more pressure... nothing !  it took quite a few attemps to get it to trip the tranny ! So it was clearly much, muuuch less susceptible with a short piece of metal.

10) I though OK, maybe it's nothing to do with the length of the metal piece, and more to do with the mass/weight of metal at play, quantity of charges at play. So, I grabbed a large screw driver... whose mass would be of the same order than that of the copper in the test lead, yet still be much shorter. Result ? It did NOt power the scope off ! 

Bingo.... so looks like that base is indeed tripped by the test lead picking-up noise. So now the problem becomes, again : it's a BJT so much lower impedance than the insulated gate of a FET ! So how could the base we THIS susceptible to noise ! GRRRR !!! 

So I now was looking at the schematics, trying to figure out how to make sense of all that..... So I turned memory lane channel on, rolled back 20+ years earlier, school days, rusty lectures... what did that electronics teacher say about BJT biasing, with regard to impedance matters.... hmmm....

1) The circuit must be designed so that the impedance of the base, as seen by the biasing circuitry, is at least 10 times/an order of magnitude, higher than the impedance, OK resistance here, talking DC, of the biasing circuitry. usual > x10 rule of thumb to limit loading of cascaded stages, fine.

So, the biasing circuitry is just a simple voltage divider, but the resistors are quite high value, 100K and 120K, so aournd 50K equivalent impedance. So... Tek must have made it so, that the base presents an impedance of at least 500K . now that's not quite a FET gate, but that's on the high side none the less... so at least the noise pickup hypothesis is not as ridiculous as it first seemed to me at the very beginning of this investigation !

2) lecture two, god only knows how I remember this 20 years later without practicing at all during all that time but : the impedance presented by the base equals the resistance  at the emitter of the transistor, multiplied by the gain / Hfe of the transistor...from memory ? The resistor at the emitter is actually not a resistor but a Zener biased in reverse, which I am not mistaken, behaves like a very high impedance ?
So, we have an unknow but high value impedance, multiplied by the Hfe of the transistor which is also unknown but also high since it's a smalll transistor, so let's say 100 minimum. So 100 times the impedances of the Zener.... now yes, that gives us a pretty high impedance at the base, doesn't it So maybe yes, it is possible that it therefore susceptible to noise pick up ?

If none of this lousy theory makes any technical sense, please forgive me.... but it was my best attempt... 

So, assuming the above is right, then I guess the transistor is just fine and should not worry about it, and I can put the scope back together with total peace of mind ?

Anyway, just when I thought the repair was finished, something as insignificant as probing the voltage across a resistor finally ended up in quite some head scratching !   I like that... when things  go wrong when you least expect it, and you have to pull you hair out... that's where you really learn the art, I find  ! ;-)

[Vince]

Back again...

I don't know if I made sense with my base impedance crap, however I think I was right about my explanation for the "doubled" trace on waveform #66, the current flowing through the primary winding(s) of the main transformer : I could not help but did what I suggested, fired up the TDS 544A so I can have a new look at the signal, and bingo it confirmed my supposition. The trace is not doubled anymore, and we can see the succession of tall and short traces, as the transistors cycle between the two primary windings.

Sorry about my ugly mug reflecting on the screen. I did try and make hard copies on a floppy disk, since the scope can do that... but then realized how useless that was, considering that I don't have a floppy drive on my desktop computer to retrieve the pics ! So, sorry, old fashioned camera pictures will have to do...
I think I will invest in one of those SD memory card readers that emulate the old floppy drive so that it is transparent to the scope. Someone showed me info about this some time ago, there are some "slim" units that can fit in these scopes IIRC.

[Vince]

Normally, rms current is under 2A and there are not much reasons to overheating the connector.
But I have found already 2 times a far much higher rms current in 2235 oscilloscope: 
Why can this happen ?
Pre-regulator works as a buck converter.
T906, even if it has several coils, is in fact a dc inductor with gap.
If inductance of this inductor is lower than required, then working of the buck converter become fully wrong, with high peak current and high rms value

In one case, inductance was very low because ferrite core of T906 was broken, in the other case, there was an internal short between layers or turns of one of the coils.

The best way to find this problem is to check the current waveform on shunt resistor R907

Quoting an early message from Oldway here. Figured I would probe one last thing before I reassemble the scope and loose accessibility to the board.
Remembered Oldway's message a month ago, so for peace of mind, not wanting to have a disaster waiting to happen again... I decided to monitor the current flowing through the FET and its coil. After all I guess that's the most vital signal to in an SMPS...  See below. The 2232 is of course powered via an isolation transformer, so I could probe the voltage right across the terminals of the shunt resistor.

Waveform seems perfect to me ?  There are big spikes of course, every time the FET switches the current in the coil on and off, but I guess you have to live with that in SMPS designs...and the LC filtering networks on every secondary winding (but the two CRT ones), are able to very effectively get rid of those spikes, I could see that for myself when I checked all the rails for ripple : they are dead flat/"clean".

So, if we forget about the large spikes, then I would say it's working just fine... You can see the current ramping up nicely/linearly as expected, during the short time that corresponds to the narrow duty cycle that the PWM chip is driving the FET at.
Then again linearly ramping down.

Amplitude wise, we are way below 2A so Oldway, that means I can sleep on both ears, the coil is fine and not agonizing, waiting to blow the scope again ?! 

Voltage amplitude averages at 275mV, spanning 250mV up to 300mV.
Shunt resistor is 0.2 ohms, so 5A per Volt. So the current in the coil averages at 1.38A and lives inside a 1.25A / 1.5A. window.


Today I received the 2 new Sunon fans. I compared them with the old / original Sunon fan. The noise difference is not that obvious at first, but it becomes much more tangible once the new fan get actually installed in the scope rather running in free air on the bench. I am happy with the result.

Now the dominant noise is the hissing noise noise from the regulator coil. I received the replacement coil from the US (to replace the one that got damaged during transport the other day). Might try to install it in the scope to see if it's less noisy than one currently installed.

One thing I did not think of/anticipate, is that the new fans did not come with a connector !  So had to cut the old connector and solder it on the new fan. Put electrical tape to insulate the solder joints, but the life expectancy of this tape, especially inside a hot PSU... is about 2 hours 3 minutes and 45 seconds, I would say !  Then the wires will short and bang... trouble again.  But that was just temporary, to check how the new fan would perform in its actual environment. Now that I am happy with this new fan, I need to get some heat shrink stuff to properly insulate the solder joints... so that will delay the re-assembly of the scope a little bit.

I would also like to replace the connector of the FET, the one that overheated. Its plastic did not get hot enough for it to melt or get deformed, and the contacts in it are still fine (no oxidization, not burn marks, metal looks healthy and shiny), and they are not lose... but they aren't very firm either. So if I can find a new connector, why not...

Anyway, the scope is basically fixed and it's now only a matter of days before I put it back together for good. Can't wait.

[David Hess]

My analog scope obviously has no means to tell one cycle from the other, so both get super-imposed, making the trace look doubled.

Sure it does; either set the holdoff control so triggering happens on alternate cycles or use a separate trigger channel.

Is it cased by one of the transistors not being fully saturated ? That would seem unlikely too, as I expect them to be waayyy over-driven to make sure they don't overheat/cook/die by dissipating abnormal amounts of power.

Differential pair Q938 and Q939 form an error amplifier which controls the bias of the output transistors through driver transistor Q944 so the output transistors are operating in their linear mode to control the output voltage; transformer T944 switches the bias between output transistors.  Primary side feedback through CR947 and CR946 controls the secondary side output voltage.

I suspect hfe mismatch in the output transistors contributes more to the imbalance than the transformer and as I recall, at some point Tektronix was grading or matching the output transistors.  You did not mark scale factors on your diagrams so I am not going to bother evaluating this.  I consider the readout provided by the 22xx series DSOs in analog mode to be a feature; it makes documentation easier.

The " fun " part came when I tried to probe across R927, it says it should be 6.8V.  As soon as I put the multimeter probes on these terminals... the scope would immediately shut down !!!

This point is high impedance, about 33 kilohms ignoring the base impedance, and your multimeter coupled enough signal into it to shut off Q928.  See below.

This circuit really should have some decoupling at pin 12 but in practice the circuit works reliably.

3) The biasing voltage divider has quite high value resistors, so I thought maybe the 10M resistance of the multimeter, across that 100K resistor, might upset/load the biasing just a tad... just enough to trip the tranny. But... 10M is a couple orders of magnitude higher than this resistor, so no way, stop kidding me please...

The multimeter's differential capacitance is 10s to 100s of picofarads and its common mode capacitance is 100s of picofarads.  Special voltmeters are used when lower input capacitance is required.

4) maybe the multimeter impedance is NOT 10M... maybe somehow it's faulty/lower than this. So I used another multimeter to measure the impedance of the first meter... 2.5M instead of 10 !  Eh, what's going on ?? So cross checked.. and the other multimeter also presents a 2.5M impedance not 10 ! Suspicious... likely it's some problem with the way their ohmmeter works. No definitive answer on this subject, but I didn't want to waste time chasing red herrings, because 1) the fact that they both read 2.5 instead of 10 likely means that they are both good and that there is a valid explanation for this unexpected reading and 2) even 2.5M should still be plenty high enough not to upset the biasing of that transistor anyway !!!

Charge pumping from the multimeter's ADC and input circuits can cause this problem which is why I have a couple of older "inferior" multimeters which have no charge pumping at all to use as a sanity check.  This is an example of how modern technology makes cheaper test instruments in the sense of quality.

So at this point I knew that the transistor "tripping" has nothing to do with the impedance of the meter (whatever it actually is...) loading the biasing circuit, and I also knew that it powered off only if I made direct contact with the base of the transistor.

The multimeter's common terminal is connected internally to the multimeter's shielding which capacitively couples to everything around it yielding 100s of picofarads to all kinds of things.  The multimeter's input terminal is capacitively coupled to the multimeter's common by 10s to 100s of picofarads.  The leads themselves add more coupling capacitance.

The impedance presented by the base equals the resistance  at the emitter of the transistor, multiplied by the gain / Hfe of the transistor...from memory ? The resistor at the emitter is actually not a resistor but a Zener biased in reverse, which I am not mistaken, behaves like a very high impedance ?

So, we have an unknown but high value impedance, multiplied by the Hfe of the transistor which is also unknown but also high since it's a small transistor, so let's say 100 minimum. So 100 times the impedances of the Zener.... now yes, that gives us a pretty high impedance at the base, doesn't it So maybe yes, it is possible that it therefore susceptible to noise pick up ?

That only applies when the transistor is operating in its linear region.  Q928 is saturated with a collector current of 7.2 volt / 6800 ohms = 1 milliamp and a base current of 66.5 microamps for a forced beta (hfe) of 15.  Once about 50 microamps is drawn out of the base circuit and Q928 starts to shut off which in turn starts to shut off Q930, the voltage at pin 12 drops further lowering the current into the base of Q928 and positive feedback latches the circuit off until the voltage across C925 rises above 23 volts turning Q928 back on.

Because of the positive feedback and fast response time of the circuit, Q928 only has to be momentarily shut off to cause shut down and the input capacitance of the multimeter is obviously sufficient to do this.  Assuming that the multimeter's input resistance is 10 megohms, put a 10 megohm resistor in series with the tip of the multimeter lead to isolate the input capacitance and then double the measured value.

And now you know why I hate cheap multimeters which have an input resistance that changes with different ranges.  Having an infinite input resistance on low voltage inputs is sometimes acceptable and makes the above technique easier if the voltage is in range but I wish they would indicate or otherwise mark it and have some way to disable this "feature".  Oddly enough my old and obsolete DM502 has a jumper to select this feature which is almost ideal.  One of Fluke's old handheld meters had the ability to select infinite input resistance mode using the controls.  I do not know of any modern multimeters which allow this but that is the price of progress.

[David Hess]

Waveform seems perfect to me ?  There are big spikes of course, every time the FET switches the current in the coil on and off, but I guess you have to live with that in SMPS designs.

There is a good chance that the spikes or a good part of them are a measurement artifact.  It is difficult to probe circuits like this and get accurate results.  For low impedance circuits, I often solder a coaxial pigtail to the point to be measured and then attach the high impedance oscilloscope probe via a coaxial BNC to probe tip adapter.  If using a differential probe, then a short length of twisted pair transmission line can be used.

Put electrical tape to insulate the solder joints, but the life expectancy of this tape, especially inside a hot PSU... is about 2 hours 3 minutes and 45 seconds, I would say!

Kapton tape works great for this but I usually use heat shrink tubing.

I would also like to replace the connector of the FET, the one that overheated. Its plastic did not get hot enough for it to melt or get deformed, and the contacts in it are still fine (no oxidization, not burn marks, metal looks healthy and shiny), and they are not lose... but they aren't very firm either. So if I can find a new connector, why not...

I would remove the connector and solder the wires directly to the transistor pins.  I thought Tektronix released a change notice about doing this but could not find it.

[Vince]

Hi Dave,

Thanks again for these explanations ! Will read them again and again...

I went to the lab to do some more experiments following your comments, to bring some more food to the table so to speak but.... first thing I did was try my Fluke 11 voltmeter in its low impedance mode (a few kohms from memory, can't remember), don't ask me why, even I would be hard pressed to give a plausible explanation... seems so stupid now. Obviously it shorted the base of the transistor and the scope turned off, no surprise here but.... problem is that it now will NOT power up again !!  The scope keeps blowing fuse after fuse ! So I plugged my dim bulb tester, and now unlike the first time where it was "blinking"/oscillating with the SMPS... now it' s a solidly lit bulb, a dead short !  Oh no...... sorry, can't experiment further with the scope, must fix it first !   

A dead short... I guess it should be easy and quick enough to trouble shoot...  crossing fingers. Hopefully I will have it figured tonight... I was really hoping to finish this scope by this week-end, the start of a 3 week vacation during which I was really hoping to start work on a more "serious" problem... the TDS544A now has POST error messages, acquisition problem mostly...

Anyway, never a dull moment ! 

[Vince]

OK, 3:30AM, fixed it !

Turned out the FET blew ! WHAT ?! "playing" with that little tranny to turn the scope on and off... how could that have killed the FET ?
And how comes that this time, the failure was completely silent ? No magic smoke escaped, no smell, no noise no big bang, no nothing... FET blew silently ?!

Regardless... this time I was able to fix the thing in record time, getting good at it !   

Removed fan assembly and pulled the painful heatsink again, to get access to the components on the board. lifted a pin of the shunt resistor, fed the inverter with the bench power supply... scope was fine, wonderful. 
So checked the regulator. Coil survived, shunt resistor as well. The FET is shorted but the gate somewhat survived : 20ohms or so. So I thought maybe the PWM chip survived too. So checked the circuitry that's between the PWM output and the gate of the FET (Q908, R909, R908, CR908), and all these components were fine. So that meant that the mains didn't pass through this section, hence the PWM chip behind it was most likely fine.
So, I replaced the FET (I had ordered two of them "just in case".. glad I did ! ), taking that opportunity to get rid of its overheated plastic connector, that turned out to be so brittle that it shattered like it were a piece of glass ! So no regrets... I soldered the wires straight onto the pins of the FET. added some heat shrink tubing I bought this morning, was cheaper than I feared, so ended up buying 9 meters of the stuff, in all, in various sizes and colours. 
Anyway, once the FET was replaced everything was fine, phew !!!   

Added some heatshrink tubes to the fan solder joints too then,  colour coded to boost, what a luxury... my 2232 will be the prettiest in town !     


Then I could resume my experiments :

- variable hold-off : sorry I looked dumb ! Last year I spent some time reading the 2232 manual about the dual time-base so I could master this magical thing and be really comfortable using it. But the variable hold-off feature I had not yet investigated ... I am glad this SMPS gave me a real world example of the usefulness of this feature ! Indeed my old Hameg could display a perfectly clear trace once I had played quickly with the variable hod-off knob, lovely ! 

As for scale in my Hameg screenshot, sorry I didn't not mention it, thought I did. The screen capture on the TDS show all the relevant information, though.

200mV per division. Shunt is 0.51ohm so about 2A per volt. Waveform shows (on the TDS) 4.2 DIVs for the shorter trace/smaller winding current, and 4.8 DIVs max for the taller trace, 4.7 average, 0.5 DIV more than the shorter trace.  So that gives us about 1.70 A and 200mA more for the other winding.

But since you said the output transistor driving the windings actually used in their linear region hence the whole is designed to cope with the trannies dissipating power, then I guess there is no need to care any more about one switching more current than the other... it's not gonna cause any harm...

Thanks again for your the circuit analysis of the inverter. Will read it 50 times...

Anyway, I think I have fixed this scope properly and probed/learned quite a lot in the process... so might be time to put the cover back on,, before I blow it again (again) ! 

Now I just need to buy 4 new little plastic knobs for the front panel, to replace those that disintegrated the other day, and I think this scope will be sorted, and I can start working on the TDS 544A now...

[Vince]

David (Hess, of course), may I bother you one last time with circuit analysis, then I put this scope back together and I leave you alone for good I promise...

I am glad my scope is fixed of course, but this all turned out to be an SMPS introduction to me, so the second point of this repair thread is for me (and other silent newbies reading this maybe), to make sure I understand the basics of SMPS operation... because there are still some basic things I don't really understand, I am not happy with that ! ;-)   I would like to make sure I get the basics right.  I swear I did try to understand, I am not asking to be spoon fed... but since I failed at figuring it out by myself, I have the choice between not bothering you, but remaining ignorant and not having learned as much as I could from this repair.... or asking you, and becoming smarter... so I chose option #2 !

So here are my remaining questions on this circuit, ie, hight-voltage/mains connected SMPS in general, I guess. Some of the questions are somewhat inter-related.

INVERTER
____________

1) to finish with the inverter side of things : you said the output transistors work in their linear region, and come to think of it, the waveform #65 (collector of the upper transistor) kinda confirms this, given how slowly the voltage ramps up, hence dissipating a fair amount of power. Why can't we drive the transistors more abruptly so to speak ? Is this meant to purposefully reduce the rise time, for what purpose ? Is it to generate less harmonics, so as to induce less noise/crap  on the secondary windings, to make filtering/"cleaning"  easier downstream ? Or could it be that the transformer is more efficient when driven by a signal that's as close as possible to pure sine wave ? clueless about transformer, so just a shot in the dark always, sorry... 


REGULATOR
_____________
There are still some basic things that I don't get about the design of this thing, I am hoping maybe you can help me understand.

1) The PWM chip appears NOT to be referenced to ground. Its ground pin (pin #7) is connected to the ouput/ 40V rail... its sitting on its head, make my head spin. Why isn't it possible to simply reference is to ground ? a common ground for the input, ouput, and regulator circuitry ? Sure, the PWM doesn't care/know where its referenced... all it sees is a suitable potential difference between its power pins, its all it needs to operate I guess. But still... why not make it simple, with everything referenced to ground ?


2) Generating the power supply for the PWM chip : basically it looks simple to me : we start from a mains/rectified/filtered, a nice DC voltage... quite a high-voltage sure, 300V in my neck of the woods, but still a nice DC votlage. Why can't we just derive a lower voltage from that, to supply the chip... in a simple way ? There is already this fancy high-voltage DALE 150K resistor that feeds the regulator circuitry, and we have already a Zener as well in there there. Why can't we just (silly me I guess) just connect the Zener to that 150K resistor and supply the chip with that ? Why do we need the extra complexity of Q930 switching power to the chip, itself switched by Q928 and 5 resistors around it ? Something to do with impedance ? A simple resistor + Zener would not provide a suitably low impedance to supply the chip and make it operate properly ? It can't be drawing that much current can it ? Can't find much info about it in its datasheet...

3) This all "start-up" thing, where you kinda have to prime the pump to get it going, so to speak... why is it needed ? I mean, a low voltage/"local" DC-DC regulator doesn't appear to need this : you just give it it's DC input voltage, and as soo as you power it up, it gets going.  Why can't the same be done with a high-voltage/mains SMPS ? The input voltage is much higher for sure, but the principle of operation is the same no ?

4) Granted, we have tis start-up sequence, so let's try to see how it operates exactly...  what happens exactly, at power up, step by step ? I can see on the schematic what components/paths could be involved, but saying I fully understand what happens at power up, would be a big lie....
I can see the upper secondary winding of the transformer (pins 6 and 7), which provides a direct path that can feed the output voltage back to the power pin of the chip, via CR920 which presumably keeps the rectified mains from being dumped to the output in the process...
The best I can come up with is this : at power up, the filter caps of the 40V rail, and C925, are discharged, so the rail is sitting at zero Volt. These caps are progressively chargin up, hence voltage increasig, as the rectified mains trickles through this  speicla purpose DALE 150K resistor, R926.  The ouput voltage rises, rises... and is fed back to the power circuitry of the PWM chip, via the upper winding of the transformer and the diode CR920.  Once the voltage is high enough, Q928 turns on, the Zener conducts, and the voltage is then stabilized, and green flag is given Q930 "supply is good and stable, you can wake up/power the chip now ! ".  At this point, the voltage would still increase until the the 40V rail actually  reaches the rectified mains voltage, which would be catastrophic (but crow bar is there to protect the inverter of course)... if it were not for the PWM chip now starting up, and it regulating the voltage at 40V, keeping it from increasing to dangerous levels.

Please tell me that I got at least some that kinda right (and correct the rest !), other wise I would be kinda depressed...

5) On the role of the secondary windings. That is one of the main things that always made real/practical SMPS seem like black magic to me, compared to the very simple theory we learn at school, where only a simple coil is used.  We have no less thant 3 windings. I guess after working on this SMPS I am a lot less inthe dark, btu still. 

- Upper winding : OK, so as seen above, I guess it's used as part of the start-up circuit.

- Middle winding (pins 9 and 10). It's connected to nothing ?! I seem to remember someone, on this thread not sure, that is was some clever trick to help with EMI emissions/compliance ? That would explain why it's connected to no other components in the circuit. I suppose it picks up (some) of the noise from the coil, and "dissipates" it to the earth connection, and ground plane (according to the symbols I can see) . Capacitor C908 I guess is used to capacitively couple that noise, because it there was a DC path, it would "load" the coil/primary too much ? I am really out of my comfort zone again, so I am doing the best I can but probably not good enough I know... 


- Lower winding (pins 4 and 5). This one I can not make heads or tail about it at all. On side of the winding is connected to ground, Then CR907 keeps current from flowing from flowing from the output rail back upstream.. but why not simply make no connection at all in this case ?!  So that means you would want current to flow TO the rail... but where would that current come from ? It can  not come from the secondary winding, as it's connected to ground so not much happening there. So it would have to come from the main coil/primary/mains ? But there is C907 in the way ?  And well, why would you want the voltage from the coil, to leak to the output, hence interfering with FET that's trying hard to regulate the voltage ?  That does not look like team playing...
So I could really do with some explanation about the role and operating of this winding and C907 / CR907 that seem to go with it.



MISCELLANEOUS
________________

1) Where the hell does the hissing noise from the regulator coil come from ? I mean OK the FET is switching current abruptly in it, so generates "noise"... but the switching frequency, 60+ KHz, is well above the audible range... so if that caused/induced a corresponding mechanical vibration/sound via the pot core, then we would not be able to hear it either, right ? OK, maybe the pot core resonates with some harmonics... but these would be at even higher frequencies  by definition (?) so even less audible, so again not a problem. So... is there some physical phenomenon by which the 60Khz of the current flowing the coil, could somehow induce a mechanical vibration at a LOWER frequency ?!  Looks stupid maybe, but I would like to know !

The humming/low frequency  noise of a transformer connected to the mains, makes sense I guess, since it's running at 50 or 60Hz, so in the audible range... though not by much. But a coil switching at 60 KHz ?!

I guess that goes for any coil... they also often make noise in old  CRT computer monitors for example...


2) In more modern SMPS, I notice they almost always use an optocoupler, so as, I hear "provide feedback to the primary side while maintaining the galvanic isolation the main transformer provides". OK but... feedback for what/where to ? so that the inverter or/regulator can adjust their voltage in a closed-loop ? The Tek from what I can see is open loop, the inverter is just tweaked with a trimmer and the tongue at the right angle, then you just hope that it stays there... and it seems to work just fine. So feedback would allow to stop having to pray, and bring peace of mind that the inverter really is guaranteed never to drift ? Isn't it a bit overkill, given that in practice, open loop operation proved to work just fine ?
I can see maybe a use case where it could be useful, though ? In the case of an adjustable lab power supply, where you could have a pre-regulator that would track the (adjustable/variable at will) output voltage so that the ballast transistor always drops the minimum amount of voltage hence minimum heat dissipation. Just enough voltage to be able to regulate without risking to become unstable, and not too much voltage so as not to waste uselessly power. this way you can increase the life expectancy of the power transistor or I guess, as well as make for a more compact power supply/smaller heat sink requirement.
So in this case, yes I can see that feedback would be very useful indeed. But for a PSU producing fixed/known voltages levels ??


If you (or who ever volunteers ! ;-) can shed some light on any of the above, I will have learned something...


Oh, and I just found one last little job to do on the scope before I can put it back together ! I almost forgot... need to rewind the thread back to the beginning : There is a filter cap  (-8.6V rail from memory) on the digital board, that got caught/rubbed on to the cabinet when I slid it off the chassis. Cap stil measures fine somehow, but it got bent quite a lot, and the bodty of the cap slid off on of it's terminal... I though that would kill it, shorting all the layers inside.. but no, still works fine somehow ! So I need to replace that one, have loads of caps salvaged from....25 years ago. don't want to put a bad cap in the scope, so I first need to get an ESR meter to find a good one... but I guess they will allbe bad after such a long time, since I understand they go bad just by sitting on a shelf. However I alos understand that they can be healed/brought back to life, so an ESR would still be useful, to ensure that the healing process performed as required... or I can just play it safe and buy a brand new cap... but placing an order just for a little cap, argh.... :-/   Will check my local store to see if they have something suitable, but not holding my breath.

Anyway, the reassembly is again delayed a little bit, but not by much luckily 

[tautech]

I've got a few thoughts but to start with I too would be worried about audible noise and wonder if the switching bipolars unequal current waveforms could be the cause. Time to get a matched pair ?

The cap on the -8.6 rail is an electrolytic, they don't self heal. Replace it.

More later..........

[Vince]

I too would be worried about audible noise and wonder if the switching bipolars unequal current waveforms could be the cause. Time to get a matched pair ?

Seems that most SMPS make hissing noise so can't be that dangerous... but is sure is driving my ears mad....
The transistor pair is not at fault I think, because they drive the main transformer and this one is quiet, and is the transformer in the inverter. When I fed the scope/inverter with the bench power supply, there was absolutely no noise whatsoever, absolutely zero.
Nope, the hissing noise only comes from the coil/transformer of the regulator, driven by the FET.
Maybe because the FET is switching it so hard/fast rise times, that it generates lots of energy at frequencies that happen not to suite our ears. Still, how can a coil driven at 60KHz, end up producing mechanical waves in the audible range. I am truly interested in learning a well grounded technical explanation. I want to understaaaand what's at play here ! 
I also noticed something "interesting" : from COLD, the hissing noise is very high pitched and very loud, but over the course of a 2 or 3 minutes, the noise progressively decrease in amplitude to a much, much more bearable level. The sound frequency also get a little less high-pitched.
So I take it that the SMPS warming up has some effect on this coil, direclty (the coil itself warming up internally), or indirectly (warm up affecting the surounding circuitry which in turn would affect the coil's operating conditions) though the exact mechanism is still to be determined...

The cap on the -8.6 rail is an electrolytic, they don't self heal. Replace it.

I might be mistaken in some way then, but I gathered that fans of ancient Tektronix tube scopes from the '60's routinely "revive" their 50 year old electrolytics.

I also recently came across this video I found very interesting :



It's up to date, ( a year old), an engineer from from "Kemet", a major capacitor manufacturer, giving a talk about the most common types of capacitors, to an audience of H/W design engineers. It's 50 minutes long but I found every second of it very interesting. Worth taking the time.  However you can get straight to the part where he talks about electrolytics, if you seek directly to 16minutes and 35 seconds or thereabout. Lasts for 5 or 6 minutes.
The guy clearly explains the mechanism by which elecgtrolytics self destruct "by design", teh second they leave the factory, hence why the have a shelve life. He also explains how/why they also do self heal. Later in the talk he also talk about self healing in the case of tantalum capacitors.

So from what I understand, the wear out mechanism that invariably destroys all electrolytics (the commonly used/wide spread ""wet" electrolyte type at least), is as follow : the dielectric is a thin oxide layer grown on the aluminum anode "plate". Physics make it so that the electrolyte actually "eats" this oxide layer, the moment the cap is manufactured, even when in storage. When in use, this effects is also present and even amplified, and the higher the applied voltage, the faster it east the oxide layer/dielectric. So, that's why "derating" is common place when using electrolytics : it slows down the process, increasing the life expectancy of the caps. It's a way to buy some time, short of being able to actually avoid the problem completely. Derating is commonly 50%, meaning it is advised to use a say 16V cap when dealing with an 8Volt signal. In the case of the filter caps on the digital board, Tek used 35V caps for the 8.6V rail, so quite an enormous derating ! Helps to make them last.

He also explains how chemical reactions in the cap, make it so that the capacitor actually indeed heals itself.  The thing to do is simply to slowly ramp up the voltage, to let the reaction take place progressively. If you apply the operating voltage all of a sudden, as when in power up the instrument, then of course it can be damaged because if it's too far gone, the dielectric/oxide layer is so thing, or completely eaten, that the cap is effectively shorted, so a current surge will damage it at power up. But if power it with a small voltage, and limit the current, then progressively increasing voltage, then you avoid current rush piercing the anode, and you allow enough time for the chemical reaction inside, to progressively re-grow the oxide layer to it's original thickness, and the cap is good to go again.

Aside from that, I found it particularly interesting, among many other things he talked about, to understand how electrolytics managed to achieve such high capacitance. it actually kinda relates to why the caps "die" : the high capacitance is achieved by increasing the effective contact area by etching the oxide layer, which created hills and valleys in it, so to speak. This means that when in a "valley", the oxide layer is actually, locally, much thiner than in other places, meaning the electrolyte will be able to eat through it even faster.

 This talk is really packed with interesting and practical information, I find.  Well at least to a newbie like me, I gleaned many interesting pieces of information throughout the video.

Anyway, I checked the website of my local shop, and they have a broader choice of electrolytics than I imagined. I was able to find something suitable. Same package size, capacitance and voltage rating.  Actually the package is slightly smaller, 5mm in diameter (same height). If I take the same diameter as the original part (6+ mm ), then I corresponds to a 47uF cap rather than the original 33uF. Can't be a bad thing for a filter cap, to be a big bigger, eh ? So I bought that... but also the small package so I can put the exact same value, should there be a reason not to use a larger value.

The replacement will also be improved with regard to temperature rating : the Tek parts were rated only at 85° (despite being close to the power hungry acquisition chip, and zero airflow in the area.). Replacements parts will be rated at 105°.  Tihng is, the shop ha da much borader choice of caps in the 105 rating. The appropriate cap was not available in their 85°C range.

Obviously, I anticipate I will get some unknown/cheap brand, not the original Chemicon, but I guess a brand new "noname" 105°C cap is better than a 30 year old Chemicon.

Since they are dirt cheap (10 cents a piece) I ordered 10 of each (33uF and 47uF), so I can replace all the electrolytics on the digital board while I am at it : only 6 of them, all filter caps, all the exact same specs.
I should get them in the post in 2 or 3 days max, I would imagine. When I ordered Zener diodes from them the other day for the regulator, they were very quick.

So hopefully in a few days the scope will be re-assembled.

[tautech]

Yep, seen the Kermet vid, it's very good.
You and I have just a different terminology, for E caps most call it reforming where you call it healing, a term IMO moreso applies to Tants and Polyester caps.
No biggie. 

For the prereg then some temp dependency could be attributed to old E caps............I've seen LV stuff warm considerably and then not trip overcurrent protection. You'll ESR tests to confirm this.

[Vince]

You and I have just a different terminology, for E caps most call it reforming where you call it healing, a term IMO moreso applies to Tants and Polyester caps.
No biggie. 

Oops yes, you might be right... I came across both terms and thought they were used interchangeably by people, but now you mention it yes "reforming" refers more to a human being actually putting the cap through a process, out of the board, whereas "healing" might refer more to the chemical reaction that happens naturally in the cap, during its normal operation, in circuit, without any particular intervention from outside.  Thanks for clearing that up, I will be more accurate next time ;-)

For the prereg then some temp dependency could be attributed to old E caps............I've seen LV stuff warm considerably and then not trip overcurrent protection. You'll ESR tests to confirm this.

Hey good idea !  Yes, I definitely need to get an ESR meter quit soon... I am working more and more on fixing old test gear so it's quite hard to do a good/serious job of it, without an ESR meter to check all the caps. I admit...
I did have a quick look at the options on Ebay fairly recently, but not much I fancied. There were either things in kit form but I prefer a finished product. The cheap oens look ugly. There is a little one that looks real cute but costs 150 buck from memory, not gonna pay that for a simple ESR meter... come on.
Of course, long term, I plan to get a proper bench RLC meter, but even old stuff costs 500 buck or more, so can't afford it at the moment.
So I am thinking of a modern/new handheld RLC meter, as a compromise. I need to see what's available on the market at the moment, price/performance/feature wise... I am hoping I can get something decent for 100 to 150 bucks, will see.

[tautech]

The $20 LRC one is pretty popular, there's a thread on them in the TE board.

This one:
https://www.eevblog.com/forum/testgear/$20-lcr-esr-transistor-checker-project/

There's a few variants of this on eBay and Aliexpress.

[David Hess]

1) to finish with the inverter side of things : you said the output transistors work in their linear region, and come to think of it, the waveform #65 (collector of the upper transistor) kinda confirms this, given how slowly the voltage ramps up, hence dissipating a fair amount of power. Why can't we drive the transistors more abruptly so to speak ? Is this meant to purposefully reduce the rise time, for what purpose ? Is it to generate less harmonics, so as to induce less noise/crap  on the secondary windings, to make filtering/"cleaning"  easier downstream ? Or could it be that the transformer is more efficient when driven by a signal that's as close as possible to pure sine wave ? clueless about transformer, so just a shot in the dark always, sorry... 

Since the transistors are operating in their linear region to regulate the output voltage, they are already "inefficient" and a fast switching time is not required.  The slow switching time also contributes to a low noise output which is important in this application.

Note that this is a saturable core inverter.  Saturation of T944 is the only thing controlling the switching frequency.  Normally the base control winding would come from output transformer T948 but by using a separate and smaller saturating transformer for controlling the transistors, the peak flux and losses are lower and flux leakage is minimized.  Among other changes, later versions of this power supply did away with T944 and clocked the inverter synchronously with the preregulator.

1) The PWM chip appears NOT to be referenced to ground. Its ground pin (pin #7) is connected to the ouput/ 40V rail... its sitting on its head, make my head spin. Why isn't it possible to simply reference is to ground ? a common ground for the input, ouput, and regulator circuitry ? Sure, the PWM doesn't care/know where its referenced... all it sees is a suitable potential difference between its power pins, its all it needs to operate I guess. But still... why not make it simple, with everything referenced to ground ?

The regulator controller is referenced to the output voltage so that no level shifting is necessary to control n-channel MOSFET Q9070 and the output current can be measured directly via R907.  The output voltage however *is* referenced to ground via voltage divider R915 and R914 which connects to ground.  The +5 volt reference creates a current of 5/10k=500uA through R915.  This current is balanced by a current through R914 when 500uA*84.5k=42.25 volts exists between ground and the output.

2) Generating the power supply for the PWM chip : basically it looks simple to me : we start from a mains/rectified/filtered, a nice DC voltage... quite a high-voltage sure, 300V in my neck of the woods, but still a nice DC votlage. Why can't we just derive a lower voltage from that, to supply the chip... in a simple way ? There is already this fancy high-voltage DALE 150K resistor that feeds the regulator circuitry, and we have already a Zener as well in there there. Why can't we just (silly me I guess) just connect the Zener to that 150K resistor and supply the chip with that ? Why do we need the extra complexity of Q930 switching power to the chip, itself switched by Q928 and 5 resistors around it ? Something to do with impedance ? A simple resistor + Zener would not provide a suitably low impedance to supply the chip and make it operate properly ? It can't be drawing that much current can it ? Can't find much info about it in its datasheet...

The drive to Q9070 also requires considerable current.

3) This all "start-up" thing, where you kinda have to prime the pump to get it going, so to speak... why is it needed ? I mean, a low voltage/"local" DC-DC regulator doesn't appear to need this : you just give it it's DC input voltage, and as soo as you power it up, it gets going.  Why can't the same be done with a high-voltage/mains SMPS ? The input voltage is much higher for sure, but the principle of operation is the same no ?

It is much more efficient to power the regulator from a separate low voltage winding on the inductor and this also provides for safer operation by inherently shutting the regulator down if the output voltage cannot be maintained by for instance excessive output current.  Practically all modern off-line switching power supplies work this way also for the same reasons and the really high efficiency ones disconnect R926 once they have started.

4) Granted, we have tis start-up sequence, so let's try to see how it operates exactly...  what happens exactly, at power up, step by step ? I can see on the schematic what components/paths could be involved, but saying I fully understand what happens at power up, would be a big lie....
I can see the upper secondary winding of the transformer (pins 6 and 7), which provides a direct path that can feed the output voltage back to the power pin of the chip, via CR920 which presumably keeps the rectified mains from being dumped to the output in the process...

Notice the phasing marks on the inductor windings.  Tektronix calls it a transformer to to my mind, it is really an inductor.  Pin 6 goes positive with respect to pin 7 only when pin 4 and 8 go positive with respect to pin 5.  So CR920 only conducts when the transistor switch is off and the inductor is discharging into the output via CR907.  Further, the voltage across pin 4 an 8 to pin 5 is a constant 42.25 volts (actually 42.25 volts plus the voltage drop of CR907) so the voltage across pin 6 to pin 7 is also constant and depends only on the winding ratio between the two windings.  So when the output voltage is in regulation, the regulator's supply voltage is also in regulation.  At startup, the output voltage has to reach 42.25 volts (or close enough) before C925 discharges.

The best I can come up with is this : at power up, the filter caps of the 40V rail, and C925, are discharged, so the rail is sitting at zero Volt. These caps are progressively chargin up, hence voltage increasig, as the rectified mains trickles through this  speicla purpose DALE 150K resistor, R926.  The ouput voltage rises, rises... and is fed back to the power circuitry of the PWM chip, via the upper winding of the transformer and the diode CR920.

That is right but CR920 is reverse biased the whole time so the winding is isolated while C925 charges.

Once the voltage is high enough, Q928 turns on, the Zener conducts, and the voltage is then stabilized, and green flag is given Q930 "supply is good and stable, you can wake up/power the chip now ! ".

Q930 is actually just powering the entire regulator controller.  Pin 12 is the power pin for U930.

At this point, the voltage would still increase until the the 40V rail actually  reaches the rectified mains voltage, which would be catastrophic (but crow bar is there to protect the inverter of course)... if it were not for the PWM chip now starting up, and it regulating the voltage at 40V, keeping it from increasing to dangerous levels.

U930 draws more power supply current through pin 12 and pin 11 to drive Q9070 than R926 can supply so the voltage across C925 drops until the output voltage rises high enough for CR920 supply power to the regulator circuits.

5) On the role of the secondary windings. That is one of the main things that always made real/practical SMPS seem like black magic to me, compared to the very simple theory we learn at school, where only a simple coil is used.  We have no less thant 3 windings. I guess after working on this SMPS I am a lot less inthe dark, btu still. 

- Upper winding : OK, so as seen above, I guess it's used as part of the start-up circuit.

Exactly, modern regulators have this winding also.

- Middle winding (pins 9 and 10). It's connected to nothing ?! I seem to remember someone, on this thread not sure, that is was some clever trick to help with EMI emissions/compliance ? That would explain why it's connected to no other components in the circuit. I suppose it picks up (some) of the noise from the coil, and "dissipates" it to the earth connection, and ground plane (according to the symbols I can see) . Capacitor C908 I guess is used to capacitively couple that noise, because it there was a DC path, it would "load" the coil/primary too much ? I am really out of my comfort zone again, so I am doing the best I can but probably not good enough I know... 

That is my understanding; this was a clever way to meet EMI requirements by driving an opposite current into the chassis for cancellation.  I suspect this has to do with C903 and C902.  Later versions of this power supply did not require this.

- Lower winding (pins 4 and 5). This one I can not make heads or tail about it at all. On side of the winding is connected to ground, Then CR907 keeps current from flowing from flowing from the output rail back upstream.. but why not simply make no connection at all in this case ?!  So that means you would want current to flow TO the rail... but where would that current come from ? It can  not come from the secondary winding, as it's connected to ground so not much happening there. So it would have to come from the main coil/primary/mains ? But there is C907 in the way ?  And well, why would you want the voltage from the coil, to leak to the output, hence interfering with FET that's trying hard to regulate the voltage ?  That does not look like team playing...
So I could really do with some explanation about the role and operating of this winding and C907 / CR907 that seem to go with it.

This power supply design is unusual and I have trouble visualizing it as well.

If the output of Q9070 went to ground and the controller was referenced to ground, then this would be a flyback regulator and C907 would not exist.  By referencing Q9070 to the output however, both phases of the switching cycle contribute to the output.  Further, now windings 1-2 and 5-4,8 can be 1:1 and C907 added so that the leakage inductance spike from winding 1-2 get redirected by C907 to the output.  This lowers the peak voltage seen by Q9070 so a lower voltage part may be used.

It could be considered a floating CUK converter with the output shorted.  I wonder if Tektronix might have done this to avoid the patents on the CUK converter which existed at the time.

1) Where the hell does the hissing noise from the regulator coil come from ? I mean OK the FET is switching current abruptly in it, so generates "noise"... but the switching frequency, 60+ KHz, is well above the audible range... so if that caused/induced a corresponding mechanical vibration/sound via the pot core, then we would not be able to hear it either, right ? OK, maybe the pot core resonates with some harmonics... but these would be at even higher frequencies  by definition (?) so even less audible, so again not a problem. So... is there some physical phenomenon by which the 60Khz of the current flowing the coil, could somehow induce a mechanical vibration at a LOWER frequency ?!  Looks stupid maybe, but I would like to know !

I think you ruled out damage to T906 with your current measurements across R907.

How stable is the switching frequency?  The lack of decoupling at pin 12 of U930 bothers me.

Hissing is common with constant off time switching regulators but this one is not that and I suspect it could indicate a problem with unstable operation of U930.  It will be a difficult measurement to make but I would try to take a close look at pin 3 of U930 to see what the error amplifiers are doing.  Pin 3 is high impedance so it will need to be buffered and of course it is about 50 volts above ground further complicating things.  Pins 5 and 10 might also reveal what is going on.

Magnetostriction in inductors and transformers can make them noisy but they are often designed with dampening materials to make them quiet.

2) In more modern SMPS, I notice they almost always use an optocoupler, so as, I hear "provide feedback to the primary side while maintaining the galvanic isolation the main transformer provides". OK but... feedback for what/where to ? so that the inverter or/regulator can adjust their voltage in a closed-loop?

Exactly and the last evolution of this power supply design dropped primary side regulation for optocoupler feedback.  It is worth noting that optocouplers are notoriously unreliable.  Maybe newer ones are better but old ones have an appalling failure rate and they are one of the old parts I tend to replace spontaneously.

The Tek from what I can see is open loop, the inverter is just tweaked with a trimmer and the tongue at the right angle, then you just hope that it stays there... and it seems to work just fine. So feedback would allow to stop having to pray, and bring peace of mind that the inverter really is guaranteed never to drift ? Isn't it a bit overkill, given that in practice, open loop operation proved to work just fine ?

The regulation is still closed loop but feedback it is from the primary side of the inverter.  CR946 and CR947 rectify the primary side waveform and that is compared to the 6.2 volt zener reference via differential pair Q938 and Q939.  The forward voltage drop and temperature coefficient of CR946 and CR947 roughly compensate for the forward voltage drop and temperature coefficient of the rectifier diodes on the secondary side.  Since the diodes roughly match and the peak voltage on the primary side of the transformer is proportional to the peak voltage on the secondary side, the secondary voltages can be regulated by sampling the primary side voltage.

I can see maybe a use case where it could be useful, though ? In the case of an adjustable lab power supply, where you could have a pre-regulator that would track the (adjustable/variable at will) output voltage so that the ballast transistor always drops the minimum amount of voltage hence minimum heat dissipation. Just enough voltage to be able to regulate without risking to become unstable, and not too much voltage so as not to waste uselessly power. this way you can increase the life expectancy of the power transistor or I guess, as well as make for a more compact power supply/smaller heat sink requirement.
So in this case, yes I can see that feedback would be very useful indeed. But for a PSU producing fixed/known voltages levels ??

Making a variable output voltage and current switching regulator adds a lot of complication.

This power supply is so unusual because it requires lower output noise than normal and had to be made with the low performance switching transistors available at the time.  There is probably something to be learned on this account by studying the designs it led to.

Oh, and I just found one last little job to do on the scope before I can put it back together ! I almost forgot... need to rewind the thread back to the beginning : There is a filter cap  (-8.6V rail from memory) on the digital board, that got caught/rubbed on to the cabinet when I slid it off the chassis. Cap stil measures fine somehow, but it got bent quite a lot, and the bodty of the cap slid off on of it's terminal... I though that would kill it, shorting all the layers inside.. but no, still works fine somehow ! So I need to replace that one, have loads of caps salvaged from....25 years ago. don't want to put a bad cap in the scope, so I first need to get an ESR meter to find a good one... but I guess they will allbe bad after such a long time, since I understand they go bad just by sitting on a shelf. However I alos understand that they can be healed/brought back to life, so an ESR would still be useful, to ensure that the healing process performed as required... or I can just play it safe and buy a brand new cap... but placing an order just for a little cap, argh.... :-/   Will check my local store to see if they have something suitable, but not holding my breath.

I suspect you are talking about 33uF C9006.  I would probably replace it with a 10uF 16V solid tantalum capacitor but if I did that, I would probably replace the other bulk decoupling capacitors on the storage board as well.  If I had to order parts, I would definitely replace them all.

ESR meters are very useful although I get by with an old style impedance bridge which makes measurements of capacitance and dissipation at 1000 Hz.  It is great for evaluating capacitors but time consuming to use.  I am inclined to believe that there is a market for a dedicated handheld capacitance meter which measures capacitance, ESR, D, and leakage at the operating voltage but I have never seen such a thing.  Dielectric absorption might be fun also but is probably too specialized.

[David Hess]

I might be mistaken in some way then, but I gathered that fans of ancient Tektronix tube scopes from the '60's routinely "revive" their 50 year old electrolytics.

There are no good reasons to reform or attempt to repair a dinky inexpensive 20 cent 33uF low voltage aluminum electrolytic capacitor.  For immediate replacement if I did not have one, I would just a part from my collection of parts that have been used for prototyping or pull one off of a parts donor.  If I was refurbishing a 2232, I would replace them with solid tantalum capacitors.

[Vince]

David, thanks a lot, again, for taking the time to write all this, was most interesting as always.

Seems to me that SMPS design is really a very complex subject if one really needs/wants to take it seriously and really master it.

For a "general purpose" design engineer, or a serious hobbyist, I think the best we can expect is to get a general knowledge of the various types and configurations that exist, principle of operation, most common tips and tricks, taking some basic measurements like I did, so that we are able to trouble-shoot/repair a defective unit, or maybe slightly modify an existing/well documented design, to be able to adapt it to slightly different requirements.
But as far as mastering the thing in such fine detail as to be able to actually design a complete SMPS from scratch, make it reliable, make it meet all applicable regulations, take legal responsibility for it, and actually charge some other company for your design/product... I think that's an entire world altogether !

I mean. Picture the average modern engineer, fresh from school. His first job will likely be to work on some modern digital stuff. He will spend all day long putting big digital chips together, like Lego bricks : FPGA, 32 bits MCU, RAM and ROM, and fight his compiler to try and get Linux to run on his embedded product.... doing this for 10 years. The guy is therefore doing more "computing" than real electronics. Then 10 years later his boss suddenly tells him : "hey kid, I decided I should not put all the eggs in the same basket. I decided the company should diversify and get into analog design as well. I have just secured a contract with Tektronix to design the SMPS for their next generation 40,000 USD fancy Spectrum analyser, how cool is that. So you better get reading fast on SMPS stuff because Tek gave us just 3 weeks to show them a working prototype. Here are the specs they want, get started kid !!! ".

Nooooo... this just can't happen... I hope so anyway !
I mean, sure, whacking tiny/simplistic/minimalistic local low voltage DC-DC converters on a digital board, with just a tiny well crafted 8 pins chip with its built-in high performance FET, a small coil and a couple filter caps at the input and output...  no worries. Digital boards have these spread all over the place
But designing the mains SMPS for the entire gear, from scratch, with zero prior experience ? No matter how clever or talented the guy might be, I don't see that as a reasonable expectation from his boss...   

So... I guess I understand why, starting with the early '90's by the looks of it, (when gear started to enter the highly digital age) just about any test gear manufacturer moved to third party power supplies. And it is my understanding that the company supplying these SMPSes, do nothing but SMPS design. It's become a highly specialized job, so requires highly specialized engineers, that do this and only this, alllll dayyyy loooong...

I think you ruled out damage to T906 with your current measurements across R907.
How stable is the switching frequency? 

Hard to give numbers, but from what I recall when I looked at it on the scope, it was perfectly stable, signal was not jerky at all.

The lack of decoupling at pin 12 of U930 bothers me.

I guess it wouldn't be too much work whacking some cap in there just to see if it makes any difference...

Hissing is common with constant off time switching regulators but this one is not that and I suspect it could indicate a problem with unstable operation of U930.  It will be a difficult measurement to make but I would try to take a close look at pin 3 of U930 to see what the error amplifiers are doing.  Pin 3 is high impedance so it will need to be buffered and of course it is about 50 volts above ground further complicating things.  Pins 5 and 10 might also reveal what is going on.

Might try to probe for that, while waiting for the new filter caps to arrive, for the digital board.

I suspect you are talking about 33uF C9006.

 
Almost, I am talking about his brother, C9007 ! 

I would probably replace it with a 10uF 16V solid tantalum capacitor

I sticked to old style electrolytics because it was the original part and I don't know enough to feel confident replacing it with a different type of cap. But since you are confident, then yes I am all for a change, if it means better performance and/or long term reliability : no leaky electrolyte corroding the PCB traces (what ruined my TDS 544A !) , to start with...
Come to think of it, yes it should be alright : just about every modern digital board I see now, professional/expensive stuff at least, now have boards populated with numerous yellow tantalum caps (SMD type of course), the industry must have moved to this new "standard". I still do see bunch of electrolytics in modern stuff but only in cheap consumer gear, for example in that little 'DVB' TV tuner I opened up the other day. It was full of them.
OK, so let's go with tantalums ! Will have to be the "dipped" /leaded type so that I can re-use the existing through-hole pads. I can't solder SMD tantalums on that board. Not without it looking ugly, at least.

Luckily my local shop has some dipped tantalum available, that was not a given... 
However they cost an absolute fortune !  The same votlage and capacitance rating, 33uF / 35V then, costs a whooping 3,50 Euros ! The electrolytics I have ordered from them yesterday cost only 10 cents ! That's 35 times more, oh !

However if we lower the voltage rating to 16 volts as you suggest, which is half what Tek did, but which is still x2 the operating voltage (it filtering the 8V rail) so still comfortable... then prices drop significantly ! A 33uF 16V is only 70 cents. Still 7 times more expensive than the electrolytic mind you, but at least it's back to the tens of cents range, phew... 

So OK, let's got tantalum, and let's derate less than Tek did, but what I don't understand is why you want to replace their 33uF caps with 10uF ones ? Surely they must  have determined that they needed 33uF not 10. OK maybe just like the voltage derating, they played it extremely safe, too much maybe, so we can afford to lower the value and still have a perfectly reliable scope... but the point is : we don't KNOW that, it's a gamble, for what benefit ? The 10uF tantalum is only 30 cents cheaper (40cents a piece ) than than 33uF one (70 cents)... and shipping is 3,50 Euros ! So why gamble ?

but if I did that, I would probably replace the other bulk decoupling capacitors on the storage board as well.  If I had to order parts, I would definitely replace them all.

Yep, I later posted (when chatting with Tautech last night) that I would just as well replace all the filters caps on that digital board, especially since there are only 6 of them and all the same specs, so money is no concern here.

I am inclined to believe that there is a market for a dedicated handheld capacitance meter which measures capacitance, ESR, D, and leakage at the operating voltage but I have never seen such a thing.

Yep would love to be able to buy such a thing indeed ! 
Failing that, I will get a cheap ESR meter to get me going short term, then invest some money later on when I can afford it, for a good/proper bench RLC meter. Will most likely be again an old '90s unit so might need some repairing of its own too !  LOL

There are no good reasons to reform or attempt to repair a dinky inexpensive 20 cent 33uF low voltage aluminum electrolytic capacitor.

Yes sure ! LOL

As I said (I think ?) the people who do that, I gather (might be wring again) are generally owners of vintage/classic tube/valve Tektronix oscilloscope, who are trying to revive the existing caps so as to be able to preserve the "originality" of the scope as much as possible. So in this case it's not guided by pure/cold technical reasoning as much as it is driven by a will to preserve a historical piece of equipment for posterity. Well, I can understand that, no worries.

But in the case of my "modern" 2232, for which replacement caps are still available and for cheap, then yes, no reason whatsoever to be dicking around, it's a no brainer.

[David Hess]

Seems to me that SMPS design is really a very complex subject if one really needs/wants to take it seriously and really master it.

As if a switching regulator was not complex enough already, there is definitely a step up in difficulty when a switching regulator has to operate directly off of the power line which typically means either 150 to 370 volts for a universal input or 305 to 370 volts for a voltage doubler input.  The 2225 series of oscilloscopes which followed the 2235/2232 series use a 60 Hz transformer to provide isolation followed by a switching regulator.  The 2335 series of ruggedized oscilloscopes which preceded the 2235/2232 series used a linear power supply.

The 2235/2232 switching power supply design accumulated various improvements when it moved to the 4 channel 22xx series and the 24xx series of analog and digital oscilloscopes and these later designs were more reliable.  Based on various changes over the 2235/2232 series production time and the alternate design in the 2225 series power supply, I get the feeling that Tektronix was not completely happy with the 2235/2232 series power supply.

For a "general purpose" design engineer, or a serious hobbyist, I think the best we can expect is to get a general knowledge of the various types and configurations that exist, principle of operation, most common tips and tricks, taking some basic measurements like I did, so that we are able to trouble-shoot/repair a defective unit, or maybe slightly modify an existing/well documented design, to be able to adapt it to slightly different requirements.
But as far as mastering the thing in such fine detail as to be able to actually design a complete SMPS from scratch, make it reliable, make it meet all applicable regulations, take legal responsibility for it, and actually charge some other company for your design/product... I think that's an entire world altogether!

Off-line switching regulator design really is a specialized field but most products these days use a third party power supply brick or module.

I mean. Picture the average modern engineer, fresh from school. His first job will likely be to work on some modern digital stuff. He will spend all day long putting big digital chips together, like Lego bricks : FPGA, 32 bits MCU, RAM and ROM, and fight his compiler to try and get Linux to run on his embedded product.... doing this for 10 years. The guy is therefore doing more "computing" than real electronics. Then 10 years later his boss suddenly tells him : "hey kid, I decided I should not put all the eggs in the same basket. I decided the company should diversify and get into analog design as well. I have just secured a contract with Tektronix to design the SMPS for their next generation 40,000 USD fancy Spectrum analyser, how cool is that. So you better get reading fast on SMPS stuff because Tek gave us just 3 weeks to show them a working prototype. Here are the specs they want, get started kid !!! ".

Nooooo... this just can't happen... I hope so anyway !
I mean, sure, whacking tiny/simplistic/minimalistic local low voltage DC-DC converters on a digital board, with just a tiny well crafted 8 pins chip with its built-in high performance FET, a small coil and a couple filter caps at the input and output...  no worries. Digital boards have these spread all over the place
But designing the mains SMPS for the entire gear, from scratch, with zero prior experience ? No matter how clever or talented the guy might be, I don't see that as a reasonable expectation from his boss...   

This is common enough to be a known joke.  After the digital guys finish, they throw the design over the wall to the guy responsible for the power supply and regulators with a tiny bare area of the board left over for him.  One of the lessons is *not* to hand the power supply design over to the new engineer assuming that it will be the easiest part.

So... I guess I understand why, starting with the early '90's by the looks of it, (when gear started to enter the highly digital age) just about any test gear manufacturer moved to third party power supplies. And it is my understanding that the company supplying these SMPSes, do nothing but SMPS design. It's become a highly specialized job, so requires highly specialized engineers, that do this and only this, alllll dayyyy loooong...

We also saw a lot of "cookbook" designs coming out of the semiconductor parts industry from companies like National and Linear Technology including some cookbook off-line designs using standardized magnetics.

The lack of decoupling at pin 12 of U930 bothers me.

I guess it wouldn't be too much work whacking some cap in there just to see if it makes any difference...

Maybe but on the other hand, messing with it risks causing damage and you have it apparently working at this point except for the hissing.

I would probably replace it with a 10uF 16V solid tantalum capacitor

I sticked to old style electrolytics because it was the original part and I don't know enough to feel confident replacing it with a different type of cap. But since you are confident, then yes I am all for a change, if it means better performance and/or long term reliability

The general rule that I use is that a good aluminum electrolytic capacitor needs to have 2 to 4 times the capacitance of a solid tantalum capacitor that it replaces.  This is to keep the ESRs roughly equal.

Luckily my local shop has some dipped tantalum available, that was not a given... 
However they cost an absolute fortune !  The same votlage and capacitance rating, 33uF / 35V then, costs a whooping 3,50 Euros ! The electrolytics I have ordered from them yesterday cost only 10 cents ! That's 35 times more, oh !

I have a local shop with inexpensive NOS (new old stock) dipped tantalum capacitors so if they are that much more expensive for you, just use aluminum electrolytic replacements.

Surely they must  have determined that they needed 33uF not 10. OK maybe just like the voltage derating, they played it extremely safe, too much maybe, so we can afford to lower the value and still have a perfectly reliable scope... but the point is : we don't KNOW that, it's a gamble, for what benefit ? The 10uF tantalum is only 30 cents cheaper (40cents a piece ) than than 33uF one (70 cents)... and shipping is 3,50 Euros ! So why gamble ?

See above about aluminum electrolytic versus solid tantalum ESR.  These capacitors were sized for ESR and not capacitance.

[Vince]

See above about aluminum electrolytic versus solid tantalum ESR.  These capacitors were sized for ESR and not capacitance.

Well that's quite interesting ! So the sizing of filter caps, as I learned it at school, is mostly relevant when used behind the main bridge rectifier in a linear power supply.. but for "local" decoupling on individual boards, ESR is the most important factor. I thought that's why there was this common practice of associating a big electrolytic cap with a small 100nF "plastic" cap, for better high-frequency/transient response ?
I guess using a tantalum kinda make for an "all in one" compromise : ESR low enough to be able to get rid of the extra 100nF cap, along with a capacitance high enough to be able replace the old electrolytics. OK, so basically in modern designs they use tantalum to kill two birds with one stone, as well as increasing long term reliability I guess.

OK so no worries, I will order some dipped tantalum right away !  Might even e-mail the shop to see if they can modify my previous order (which they have not yet processed, of course, as I placed it during the week-end...) to add the tantalum  caps to it, so that I don't have to pay for shipping twice.

Well, come to think of it. Since you said the ESR decreases as the capacitance increases, does that also hold true for tantalum ? In this case wouldn't it be worth using 22uF or 33uF caps instead of 10uF ? I mean, the price difference is not that great, it's not a factor... so no need to be cheap here. The 35V ones are expensive, yes, but the 16V ones, that we are going to use here, are much cheaper :

http://composants.e44.com/composants-passifs/condensateurs/condensateurs-tantale/16-v/#topliste

So if we can make that precious ESR that little lower still, why not ! 

[Vince]

I think I may have (?) unintentionally mixed two different things above, ES-R and ES-inductance, maybe. Was 5AM when I posted that so I hope I will be excused...