Electronics > Beginners
DC dummy load circuit calibration
<< < (12/20) > >>
Ian.M:
A LM317 is quite predictable and you don't need a 10K pot to set it as the exact value of the 9V output voltage is non-critical.    Assuming you put the power LED + 1K resistor on the 9V rail, that will draw at least 6mA, assuming the LED Vf is under 3V.   Therefore you only need 4mA more to make up the 10mA minimum load requirement for the  LM317.   The top resistor in its feedback divider always has 1.25V across it so 270R would draw 4.6mA.   A 1.5K lower resistor would give an output voltage of 8.2V - near enough to 9V for this circuit.  However the reduced voltage means the LED Vf must be under 2.2V or we must decrease its series resistor or we must recalulate for more current in the LM317 divider to meet the 10mA minimum load requirement.   Try 220R for the upper resistor which will draw 5.7mA and a lower resistor of 1.2K for 8V out.

Feeding 8V to the 7660 will give a negative rail of about -6V to -7V depending on the load on it.    That's great for the OPAMP Vee (V-) supply, but its uncertainty is less good for the zero adjustment.    If you put a 2K2 resistor from Vee to a small silicon diode (e.g. 1N4148), anode to the Gnd rail, you'll get an approximately -0.6V rail that's ten times more stable than the Vee supply to the OPAMP from the 7660 (neglecting the temperature coefficient of Vf for the diode).  Tap down on that with a 10K preset feeding a 10R resistor and its easy to get the few mV of negative bias required to get the MOSFET to zero current cut-off, and avoid fluctuations in the 7660 output significantly affecting the load current.

Your latest circuit (#8) has a fatal bug - you've put the panel meter on the output of the 'regulator'.  Unfortunately your unrealistic regulator sim as a voltage source is fooling you because it doesn't draw its load current from +12V, so the return current of the panel meter isn't going through the diode that provides the negative Vee rail.   Connect the panel meter so it draws from +12V and see how horrible it really is.
VEGETA:

--- Quote from: Ian.M on May 19, 2018, 10:49:13 pm ---A LM317 is quite predictable and you don't need a 10K pot to set it as the exact value of the 9V output voltage is non-critical.    Assuming you put the power LED + 1K resistor on the 9V rail, that will draw at least 6mA, assuming the LED Vf is under 3V.   Therefore you only need 4mA more to make up the 10mA minimum load requirement for the  LM317.   The top resistor in its feedback divider always has 1.25V across it so 270R would draw 4.6mA.   A 1.5K lower resistor would give an output voltage of 8.2V - near enough to 9V for this circuit.  However the reduced voltage means the LED Vf must be under 2.2V or we must decrease its series resistor or we must recalulate for more current in the LM317 divider to meet the 10mA minimum load requirement.   Try 220R for the upper resistor which will draw 5.7mA and a lower resistor of 1.2K for 8V out.

Feeding 8V to the 7660 will give a negative rail of about -6V to -7V depending on the load on it.    That's great for the OPAMP Vee (V-) supply, but its uncertainty is less good for the zero adjustment.    If you put a 2K2 resistor from Vee to a small silicon diode (e.g. 1N4148), anode to the Gnd rail, you'll get an approximately -0.6V rail that's ten times more stable than the Vee supply to the OPAMP from the 7660 (neglecting the temperature coefficient of Vf for the diode).  Tap down on that with a 10K preset feeding a 10R resistor and its easy to get the few mV of negative bias required to get the MOSFET to zero current cut-off, and avoid fluctuations in the 7660 output significantly affecting the load current.

Your latest circuit (#8) has a fatal bug - you've put the panel meter on the output of the 'regulator'.  Unfortunately your unrealistic regulator sim as a voltage source is fooling you because it doesn't draw its load current from +12V, so the return current of the panel meter isn't going through the diode that provides the negative Vee rail.   Connect the panel meter so it draws from +12V and see how horrible it really is.

--- End quote ---

So what do you suggest to overcome such issues?

So far we knew all the resistor values necessary, so how can we solve the panel meter issue? I thought the regulator will solve it but you seem to back down from this idea due to simulation.

Yes, simulation shows a bad thing as you said but how can we be sure?

If we don't want the current to pass through the diode or the negative rail, then using 7660 is a must. However, you don't seem to like that idea and you are right since I would need to keep re-adjusting the zero point due to the switching of 7660.


--- Quote --- If you put a 2K2 resistor from Vee to a small silicon diode (e.g. 1N4148), anode to the Gnd rail, you'll get an approximately -0.6V rail that's ten times more stable than the Vee supply to the OPAMP from the 7660 (neglecting the temperature coefficient of Vf for the diode).  Tap down on that with a 10K preset feeding a 10R resistor and its easy to get the few mV of negative bias required to get the MOSFET to zero current cut-off, and avoid fluctuations in the 7660 output significantly affecting the load current.
--- End quote ---

Isn't that what I did in #8? or I am misunderstanding you?


--- Quote ---Your latest circuit (#8) has a fatal bug - you've put the panel meter on the output of the 'regulator'.  Unfortunately your unrealistic regulator sim as a voltage source is fooling you because it doesn't draw its load current from +12V, so the return current of the panel meter isn't going through the diode that provides the negative Vee rail.   Connect the panel meter so it draws from +12V and see how horrible it really is.
--- End quote ---

Ok, so what to do now? I really don't understand why the voltage peaks like this? it is a very low current and we have a high power supply.

If we cannot find a solution, then using 7660 is the only way to go as far as I can see. So I am waiting your input since I tried to go around the diode with no use  :palm:

_____________________

EDIT:

Do you mean like the image?

Here I've got 7660 giving -7 and the diode + resistor resulted in around -0.7v which is enough for zero biasing. Then this -0.7 is fed to the 10 + 10k pot as usual for zero adjustment.

However, I feed opamp negative rail with 7660 directly.

How about that? Now, panel meter goes from 9v directly to ground or better yet from 12v to ground. No diode and no 2n2222.

I guess now this issue is resolved, is there anything else?
Ian.M:
Without a 7660 or some other way of getting a negative rail that doesn't have to carry the return current for the LED meter, the problem is the drastically varying current through the diode or Vbe multiplier that's used to offset the negative side of the 12V supply to get the negative rail.   To be reasonably stable, the variation must be no more than 10% of the average.   

I suspect that most of the meter's supply current goes to light the LEDs.  Least current will be with both displays reading 1.1 and most, 88.8.   1.1 has four segments lit and 88.8 has 21 segments lit (ignoring the d.p.)  so, assuming 88.8 on both displays is the max 60mA consumption, 1.1 on both probably draws about 12mA.    However those will be average currents as its highly likely that it uses multiplexing so it can use a cheap MCU with fewer pins.  Rather than needing 42 pins for direct drive it only needs 13 to multiplex 6 digits x 7 segments.   That alters the situation considerably - peak current will be when the multiplexing is outputting an 8 digit, and least when it is displaying a blanked leading zero.   If there's enough decoupling built into the display it will average out, but it would need about 1000uF to do that, which is a physically large capacitor, and there isn't much room for it so its consumption may well swing between near zero and up to 60mA at the multiplexing refresh rate.   To know for sure, power up the display with a 1R resistor in series with its negative supply, and feed its voltage measurement wire from a pot or variable supply so you can easily change the reading.  Scope across the 1R resistor and calculate the current fluctuations from the peak to peak amplitude of the scope trace in mV.

Yes, your edit above is what I've been suggesting.   If you want to continue experimenting while you wait for your parts/board order, you could build a charge pump with a NE555 squarewave oscillator driving a charge pump (2 diodes, 2 capacitors) to do the same job as the 7660.   Ideally use a CMOS 555 clone not a real bipolar NE555 as the CMOS ones have rail-to-rail output and will more accurately emulate a 7660.  Sim attached.

N.B a practical 555 charge pump will have 100uF electrolytic + 0.1uF ceramic decoupling between pins 1 (Gnd) and 8 (Vcc) to handle the very large current spikes when it switches, and should have 0.1uF to ground on pin 5 (CV).
VEGETA:
It is done then, we'll consider this the final version and I will start modifying the KiCAD project accordingly. Getting 7660 is easy from China so people wouldn't face a problem. I will use 1N4007/1 in the bias negative voltage since it is what I have and what is available to most people.

Looks like current fluctuations of the panel meter are accurately assumed to be 10-60mA and yes the panel meter doesn't have a big capacitor but lots of tiny circuitry. We were correct to assume the problem in the diode or Vbe in the first place but looks like it won't be solved in this design as long as panel meter exists. However, I guess if it is not in the design and replaced by LCD character display then it would be better. However, that is for another project.


--- Quote ---Yes, your edit above is what I've been suggesting.   If you want to continue experimenting while you wait for your parts/board order, you could build a charge pump with a NE555 squarewave oscillator driving a charge pump (2 diodes, 2 capacitors) to do the same job as the 7660.   Ideally use a CMOS 555 clone not a real bipolar NE555 as the CMOS ones have rail-to-rail output and will more accurately emulate a 7660.  Sim attached.
--- End quote ---

Yes I saw such circuit of using 555 timer but it is just another IC so we are better yet get the 7660 if we want to get another extra IC. Originally I thought we can use the charge pump without switching but that was when I didn't understand the concept behind it.

One last issue is the 10-turn POT. I will mount it on front panel, and will put wires from it to the board. I will make it very short so I will put its footprints very close to the front panel. I hope this won't ruin everything, I don't mind few errors that could be removed by zeroing POT though.

I will build the project again in KiCAD and will post it here. Thanks for your awesome collaboration so far, gonna be credited in the videos and documentation wherever possible.

Ian.M:
I suggested the 555 circuit for testing because I expect you to either have a few already or be able to find them locally.

I'd still recommend prototyping it before doing the final PCB.   Use that 555 circuit to stand in for the 7660.  We've had enough nasty surprises that I'm reluctant for you to commit to the PCB design till you've actually had it working.   It could be simplified a bit for testing though - only two MOSFETs and a 1A full scale limit.  That will still let you see how well the heatsinks cope.

Due to the very low sense resistance of only 1/5 ohm per MOSFET, you'll need to put those parts + the OPAMP and its whole feedback loop on protoboard as solderless breadboard will have too much contact resistance.   The rest of the circuit should be fine on solderless breadboard.

As you are using a metal case, short unscreened wiring to the front panel pot should be OK.   Screened wiring would be essential in a plastic case as otherwise any nearby RF sources would be far more likely to cause it to malfunction.  The case should be connected to circuit ground.

There are lots of improvements that could be introduced but they all add complexity.   An Atmega328P controlled version, programmed using the Arduino toolchain, with a LCD, and an opto-isolated serial <> USB PC interface for control and data logging would be interesting, but I agree that the best way forward would be to build this version and get experience using it before designing anything more complex.
Navigation
Message Index
Next page
Previous page
There was an error while thanking
Thanking...

Go to full version
Powered by SMFPacks Advanced Attachments Uploader Mod