high side current monitoring

[J M Fahey]

For raw screen current sensing you can try this:
stands up to 1000V , provides roughly 5V at the sense output when screen current reachs 65mA, scale R1 to suit.

[NateS]

I'm having trouble getting my head around the cascodes. I've not really looked into them before, wasn't interested (and the fact that these are upside down (pnp) doesn't help) But what I can find all assumes a gain stage with an constant current source with an active load. However in this case the input itself is a constant current source (albeit a programmable one, with Vbe on the CCS set by the drop across the current sense resistor.)

Also the 1M collector load on the bottom transistor, will itself drop 500V at full scale - regardless of whatever is happening above it, 5V out means 500u through the 10k resistor, which means 500V drop, so it's likely to bottom out at some combinations of almost 500V B+ and max current load.

[R.G.]

Cascodes are one of those odd things that sometimes are the ideal thing for the job.

Imagine a supply voltage, say 100V, and a voltage divider of two equal resistors between it and ground. To the divider middle, attach the base of an NPN. Put an emitter resistor to ground and connect the collector to the + supply. This is a dim-witted emitter follower with no input. As long as certain limits are observed, like the available current gain of the transistor and total power dissiption, etc., then the emitter will always sit at (100V/2), the divider voltage, 50V, minus one Vbe of about 0.5 to 0.7V.

This is true for all values of emitter resistor from nearly infinity down to the resistance that pulls too much current and either kills the NPN or makes HFE fall off and drags down the reference voltage through excessive base current. The collector current is Ie minus Ib, and for HFE>100, your accuracy in conversion of Ie to Ic is over 99%. You could sense the current somehow in the collector of the NPN if you wanted to.

Another way of looking at this is that the emitter stays at (V+/2) for all emitter currents between X and Y, where X is nearly 0 and Y is the death current. So we can replace the emitter resistor with another NPN transistor's collector and emitter, and drive that transistor to make a variable current flow in the emitter...

... which makes the same (to 99% or better accuracy) current flow in the collector, at every instant as long as high frequency effects don't get you.

This is a cascode. You can set the base to any voltage, and as long as the transistor *can* make it come true, the emitter is one Vbe lower than the base, and collector current follows emitter current fairly accurately, irrespective of the voltage on the collector. Cascodes are big in RF work where you can drive the emitter with a signal and get a much-amplified signal at the collector without interference of the collector-base feedback capacitance being amplified by Miller effect.

Now flip it upside down. The voltage divider goes down from V+ to some lower voltage, the base of a PNP is tied to the voltage divider, and the collector of the PNP conducts current from V+ to ground. You can measure the current at ground and the PNP more or less gracefully eats up all the voltage variation to V+ in its collector-base junction.

In J-M's case, I'd put the resistive divider for biasing the cascode over on the raw supply side instead of the screen supply, so as to not have the sensing affect the sensed current, but the current needs are probably small enough to not have this be a problem.

The PNP base-emitter conducts a current that's (the voltage across the 10 ohm resistor minus Vbe) divided by 10K. That goes out the collector side, which has a voltage compliance of BVcer for the top transistor, and the collector is held at 1/2 the supply voltage by the lower transistor being cascoded.

This is really the same scheme as floating the sensing IC you mentioned on the power supply, to do the same job as the top PNP is doing, "echoing" a sense of the through-current down into a sensed current; then using one or more P-polarity cascodes to carry that current to ground for sensing.

JM is correct, this works. I worry about the low-end inaccuracy of having zero output until the voltage on the top PNP's base emitter reaches the point where HFE comes up to let it start conducting, and on the inaccuracy of the HFE of the bottom transistor. I think those may be nit-picking. It all depends on how accurate you have to be.

Notice that you can do the same down-cascode with P-channel MOSFETs, as the cut-in threshold of most MOSFETs is 3-6V, so the voltage on the measuring device gets a bit sloppier, but the current accuracy is assured to be 100%. As I said, this is probably nit-picking unless you're doing lab instrumentation.

[NateS]

Sometime I should get you to explain women.

Looking at the application, both of my HSCM's have minimum expected loads which are a substantial fraction of max load (heaters, and screens) - I'm not really dealing with itty bitty currents (wrt the range). But it's good to make intentional choices about such things - the next application might be different.

Regarding divider tieing into the supply or load side - it doesn't seem like it would matter at all, so long as don't drag one of the transistors out of the desired operating range. You did say "You can set the base to any voltage, and as long as the transistor *can* make it come true, the emitter is one Vbe lower than the base, and collector current follows emitter current fairly accurately, irrespective of the voltage on the collector".

Thanks for explaining the scaling - I saw that 10/10k = 1:1000, but I didn't see a divider - instead I saw big current through R1, R2 passes only itty bitty base current, so Vbq1 ~ Vscreen therefore Vbe = i*R1, and the thing would saturate at roughly iR1 = Vbe = 0.65v. I could see how R1 set the drop at Vbe, but that alone only sets the max current sensing range, not the scaling of the current output.

[defaced]

LEM Coil? The servo drives that are in the welding power supplies I use at work use them. That specific part is used in a circuit that has a 4 amp limit, so they probably will have the resolution you need to monitor the plate supply, not sure about the screens though. I have used the bigger ones to monitor welding current (thanks Steve for the suggestion!) and they work great.

[NateS]

"The collector current is Ie minus Ib, and for HFE>100, your accuracy in conversion of Ie to Ic is over 99%. You could sense the current somehow in the collector of the NPN if you wanted to."

Something else to consider - High Voltage PNP's hFE starts to drop at 400V. (availability that is). The highest gain readily available 500V PNP's I see is 82. If you're already stacking, you can get hFE up to 100 or 120 by dropping to 400V, and shave a dime or two at the same time.

[J M Fahey]

Well, thanks for the detailed analysis.
I'm a dyed in the wool minimalist and in this case went for the simplest+cheapest solution to comply with the stated goal.
This is not meant to *measure* screen current (nobody asked for that afaik) but is an "alarm system" which applies 0V to an Arduino (or any other Logic device, well beyond my comfortable area) unless current surpasses a given trigger value (user defined) where 5V appear at the sense terminal.
This is not a linear circuit but a switch.
Don't worry about high frequency response or linearity.
Top transistor turns on when getting 650mV BE ; a PNP is used because it fits there (it's working with *negative * voltage, referred to where it sits.
The 1M/10K divider was chosen for 3 reasons:
1) it minimizes current and dissipation
2) supplies roughly 5V when ON; 0V when OFF.
Zero knowledge of Logic circuits but faintly remember that in 5V circuits, the rule of thumb was *roughly* "3 to 5V is Logic 1 ; 0 to 2V is Logic 0". Or something similar.
If 10K source impedance is too high, or transition is not clean, I bet a Gate or Schmitt or whatever is used nowadays will clean and condition it.
3) being such a high impedance load, practically any base current into the top PNP will saturate it and the full 500V will be applied to the voltage divider.
4) yes, the bottom PNP can be seen as part of a cascode, but in this case, linear amplification is not needed or expected, it simply works like a common base amplifier (fixed base voltage, current enters the emmiter, comes out the collector) and current gain is 1 , by definition (ok, 0.99 , close enough).
The net effect is that it will stand 1/2 the rail voltage, sharing the work with the other PNP.
5) yes, its base divider could be better connected to top PNP emitter rail, I left it on the left because I was lazy. Oh well.
I leave that homework to the end user

PS: since now this became overspec'd (who needs 1000V capability?) , "better" transistors can be used.
As was mentioned above, perhaps 350 or 400V transistors will have much better Hfe or dissipation , or at least be easier to find.
Good luck.

[NateS]

And thanks JM for the explanation as well. Perhaps it wasn't explicit, but I am calculating screen dissipation, not just monitoring current, so I wanted an analog value. The same principles still apply though - create a scaled copy of the current passing through Rsense and get it down to ground. I don't think I need 800V+ protection - but I'm glad it came up, and I do plan to implement the no-load throttle + clamp.

[R.G.]

Well, thanks for the detailed analysis.

Sorry if that came off as pompous. I realize it might read that way because...

This is not meant to *measure* screen current (nobody asked for that afaik) but is an "alarm system"

... I knew from some out-of-band communications that Nate is trying to do measurements.

I've used similar for trip-wires before, as well.

By the way, I hope I was clear that your suggested circuit will, in fact, serve as a measurement solution once it gets over the turn-on minimum.

I'm a dyed in the wool minimalist and in this case went for the simplest+cheapest solution to comply with the stated goal.

Nothing wrong with that at all.

[woodyc]

...ZXCT1009 appears to be a clever little device.

Thanks, thats a cool chip that wasn't on my radar.
I've done similar measurements using chips like ISO124 and AD202 which aren't cheap but are easy to make into a HV diff probe.

[J M Fahey]

Sorry if that came off as pompous.

Don't worry, didn't see it that way, at all.
In fact, I was truly amazed that so much could be "read" from such a simple circuit
And if the idea was to measure screen current and "email" that value to a point 500V away, well, that can be easily designed too.
Of course the ZXCT1009 *is* a clever dedicated device which does exactly that.
Personally, living in Argentina means for one side I was always up to date with the latest developments (well, 60 days behind), because I regularly got Popular Electronics, Radio Electronics, Electronics World, plus all the great British magazines (ETI, Practical Electronics, Elektor) *but* on the other side, could not get locally all the yummy new stuff I read about (special ICs, Laser Diodes, etc.) , so I had to study what they did and try to approximate equivalent functionality with general purpose components, specially Op Amps, the Universal building block.
And as they say, "what does'nt kill you, makes you stronger".
Oh well.
PS: I think Australians also faced a similar challenge, although in a somewhat lesser degree.

[Steve Conner]

My favourite circuit so far is the LT6101. If you had trouble sourcing the chip, you could easily reproduce it with an op-amp and a P-channel MOSFET. (You'll be needing a bunch of them for the cascode string as described by RG anyway. )

Look for an op-amp whose input common mode range includes the positive rail, like the TL0xx series. I forget which one is the low-power one, but you'll want that one, as all the power for it is dropped from the B+ supply.

The Hall effect current sensors ("LEM coils") are great for welding machines and 50hp inverter drives, but they're not much use for low currents.

[NateS]

Arrrgh. Wifi here is so terrible that i hree times i thought I repsonded but xyzzy.

The LTC6101 solution comes out to $6.18 @ qty 1. That's fine by Nathan's Cost Minimization Rule for Hobby Projects: Shortcuts on a one off hobby project can be justified if tbe price is less than you might spend on a hamburger or other temporal splurge." But I think i can do better, for the challenge.

Consider this: A current source which is driven from iMeasured * Rsense, biased one diode drop so that it starts measuring from zero. Current ratio is set by the emitter resistor. Cascode takes it to ground. I'd consider FET's if it would let me do it with fewer stages, or I couldn't get the hFE i needed. With Vmax =500V, clamped. One cascode stage allows me to use MSB92S (Vce =300, hFE=120). The current source at top needs to be BJT anyway for predictability. Using it both places lets me get to ground in two stages, and the cascode also holds Vce for the current source constant, and provides backing for the bias diode.

Total price as drawn $1.31 qty 1

[NateS]

Arrrgh. Wifi here is so terrible that i hree times i thought I repsonded but xyzzy.

The LTC6101 solution comes out to $6.18 @ qty 1. That's fine by Nathan's Cost Minimization Rule for Hobby Projects: Shortcuts on a one off hobby project can be justified if tbe price is less than you might spend on a hamburger or other temporal splurge." But I think i can do better, for the challenge.

Consider this: A current source which is driven from iMeasured * Rsense, biased one diode drop so that it starts measuring from zero. Current ratio is set by the emitter resistor. Cascode takes it to ground. I'd consider FET's if it would let me do it with fewer stages, or I couldn't get the hFE i needed. With Vmax =500V, clamped. One cascode stage allows me to use MSB92S (Vce =300, hFE=120). The current source at top needs to be BJT anyway for predictability. Using it both places lets me get to ground in two stages, and the cascode also holds Vce for the current source constant, and provides backing for the bias diode.

Total price as drawn $1.31 qty 1

[R.G.]

As an eternal circuits kibitzer:
- you want matching of the first transistor Vbe and the diode. I'd use another same-kind transistor diode connected for the diode.

- you want to think about current gains; the currents divide in the ratio of the 10R and the 1K, I think. So the current in the 1K is 0-1ma. That means the base current should be 0-1ma/HFE. Transistor current gain changes with collector/emitter current, going down at low currents. This is the basis of most VCAs. I think you'll get linearity errors at low current. I don't know how bad that will be, but the concept has bitten me before.

- you want to think about divider-string currents. You'd like the base-emitter of the top transistor to have a similar forward voltage to the diode. Using a transistor diode connected for the diode helps, as the base-emitter of the diode-connected transistor/diode only conducts a similar current to Q1 and the collector-emitter conducts the rest. If you could make the collector current of the diode/transistor be the same as Q1, the Vbes would track as well as they could, and you'd only have the differences between them as offset/linearity errors.

That says that the divider string needs to conduct currents equal to the Q1/Q2 string. That's not possible with resistor loads, but some thought about the resistor values is in order to keep them similar and not starve or overload the resistor string.

Hmm. I wonder if the best thing to do is to use a double transistor ladder, mirroring Q1/Q1 on the R3/R4 side. That might even get you tracking and make HFE funnies go away.