Another CC source update

Well I assembled my circuit on breadboard and now added some of the controls

It works pretty well, although with some minor issues, like the driver transistors for the FET's getting a bit warm. Which is odd cos they are spec'd for the job and aren't even exceeding the power rating.

I basically set the load supply low to limit the power as at this point I don't want to overload it and cause some damage!

I've tweaked a few things, the display amp is now an LT1014, instead of an LM324, as I thought I may as well since I have a handful of them, keeps BOM count down I guess. But I'm pleased with how its coming out!

CC Source Project update

General Update

So: I have moved into the new house and all (but a few boxes of crap) have been unpacked.

I made a start on my new work space and already I am getting my hands dirty; The Mini is getting its gearbox removed again and am sending it off to a company that specialises in gearboxes! They actually quoted me a reasonable price if I dropped the box myself! Will mention them if they do a good job!

At least this time my tools are steps away rather than 100m away and don't have to lug them from the top of the garden to the end of the driveway! Also: nice smooth tarmac rather than bloody pebble driveway, bliss on my back! So given these benefits, this took me 2 hours! Plus I knew where everything went!

CC Project update

So the next logical step after simulating the circuit was to build it up and test it:

I butchered the old breadboard of its parts and plonked a new one on the side and began to build the main CC source part of my design:

This schematic has the changes I made from test findings

U5 output was connected directly to a 10k trim pot instead of all the pin headers for the SET/EN switch. The wiper of the trimmer was then connected to all of the non-inverting inputs of U3:A through D.

The black a red wires dangling off to the right are load supply connections and also the varistor is not fitted for this prototype test.

I cut a piece of 2mm aluminium and bolted the MOSFETs to them each with their own insulating pad and colet. I wanted the MOSFETs to be a thermal equilibrium with each other so one didn't run away with itself.

As such, I wasn't going to exceed a load current of 1A or a load voltage of 5V, I wanted to keep the power as low as possible. I just wanted to see if it worked first before punishing it!

 Testing

Upon initial power on I found that I had the LT1014 op-amp wired in the wrong way! Idiot! No harm done, just got a little warm. Fixed that.

The other think that was getting warm was the emitter resistors: R25 - R28. As there was no load supply attached, and therefore no voltage feedback, the transistor was being driven fully by the op-amp and thus nearly the full supply was across the 100R resistors. Although not an issue for the small load currents for this test, but definitely and issue when it comes to testing bigger currents! I worked out they need to be 2W or at least 1.5W. Another change to the list!

A steady 2.5V out of the voltage reference: Good stuff!

I then did a load test on a bench supply: 3.3V and set the current limit to 250mA. The idea was to slow wind the trimmer up till I reached the limit, least then know it worked. But it didn't: power on, instant supply trip. I had wired the FETs in wrong now! Facepalm 2.0!

Re-wired the FETs right way this time, repeated above and same result...ok so one of the FET's are possibly poorly? removed the shunt resistors one by one till the fault cleared and replaced that FET.

The next day I came back, different work station (single supply not dual supply this time) so I hooked the supply for Vcc to the load and expected it to work as expected: nice linear increase in current as I increased the voltage at the non-inverting inputs of the op-amps: it did and didn't

I decided to wind the current limit up to 1A for this test. The Vcc was taking 400mA (quite high), so I wanted to allow some head-room. This supply also had a moving needle meter. It went up to about 200mA then jumped up to about 500mA, crapping my pants thinking I had accidentally induced some runaway, I back the trimmer down and yet it started to go down linearly, yet when it reached approx. 200mA, it stayed there, even as the trimmer reached the end of its travel. OK weird! I decided to leave it and come back later.

I then went to another dual supply, MNM display again. hooked up the Vcc (12V) and hooked the other supply up to the load source and this time set the voltage to 3.5V and wound the trip all the way past 1A. I basically wanted to repeat the last test and rule out using the single supply, I suspected this was the cause.

Powered on both supplies and observed a nice steady linear control of the current on the load supply. I set the load so the display read 200mA and sure enough the shunt voltage on each was within at least 1mV of 50mV.

50mV over 1R = 50mA; 4 x 50mA = 200mA!

Bang on!

I then wound the trimmer up so that there was 250mV across one shunt resistor and sure enough the needle on the display deflected toward 1A nice a smoothly throughout!

Conclusions and Plans

So it works at a glance. I'm going to do some more accurate testing:

1. observe how linear the current increases (assuming the trimmer I have is not log! Best check that first)
2. introduce a method of observing the transient response as in the simulation and observe the transient current.
3. observe the emitter voltage as well and see what if there's anything interesting I can see.

So far then I am pleased that it works up to this point. After the testing above I'm planning to then move onto developing the over-temp protection and physically sizing the heatsink by buying it in!

Arduino Uno IO Expander

Well I've moved house! Finally Have a garage to do my work in! But at the moment its full of boxes and crap! give it a weekend's work and I'll have the space freed up!

In the mean time: I've come up with a simple and effective little project on expanding the IO on an Adruino Uno.

The Uno

The Uno should be a fairly familiar board if your into electronics. Its one of the first boards released by the Arduino company: early versions had D-sub interfaces, migrating to USB and eventually to the current Rev 3 boards and some SMD editions. 

Early Arduino Uno - Wikipedia

Later Versions had USB

The Rev 2 (Version I own) notice the interface chip next to TxRx LED's is offset 45degrees

SMD Edition

The Rev 3. Notice the position of the reset button

There are also loads of Rip-off versions, nothing wrong with them, they may be cheaper, but I prefer going with the real deal.

I have the Rev 2, based around the Atmel ATMEGA328P: 28-pin DIP, 8-bit micro controller. Beauty of this version is if the chip blows, you can buy a new one either programmed or un-programmed and swap it out. If its un-programmed, you will need to program a bootloader to it through the ISCP header: Olimex Programmer Review

However, on mine, A0 is busted. So in the meantime until I get another chip ordered, I thought I'd figure out a way of how to get more out of the other analog inputs.

A simple way of doing this is to use an analogue switch IC. You can use a multiplexer, but analogue switches can usually work both ways, whereas multiplexer & demultiplexers can only usually work in one way. Also make sure you have an analogue switch or mux/demux for analogue applications and, digital ones won't work as required!

As mention above, analogue switches work both ways, so you can have one input accepting a number of sensors, for example, or have 4 inputs from 1 sensor. Also they can be good for auto-ranging circuits or gain adjustments circuits, by clicking in more or less feedback resistance.

The Setup & Code

I'm basing this around a DG409. Its an array of analog switches in various configurations. The 409 is 2 sets of Single Pole, Quad Throw (SP4T) switches. Each position is selected by the address pins A0 & A1. Its all there in the datasheet.

With this I was able to connect up a number of sensors to one bank of switches:

• S_1A: pull-down switch
• S_2A: A voltage divider (div. 10) from a higher voltage supply
• S_3A: An LDR
• S_4A: a Pot

I then connected the pin D_A to the analog input and some digital pins to the address pins A0 & A1 of the DG409. The full setup is shown here:

I'm using an interrupt switch on pin-2 so I can cycle through the different sensors at any time.

The COM or output of the DG409 (white wire) is connected to the A1 analogue input of the UNO. This now essentially has 4 inputs now on one analogue input! All i have to do is program a way to change the address each time I press the interrupt switch.

I managed to punch out some code in under 15 mins:
/* Title: IO expander Brief: Example sketch of how to use a DG409 to switch different sensor elements to one analog input Date: 24/06/2014 Author: Tron9000 */ void setup() { pinMode(8, OUTPUT); //set D8 as output pinMode(9, OUTPUT); //set D9 as output pinMode(13,OUTPUT); //set D13 as output for LED, just to show its running attachInterrupt(0, pin2ISR, FALLING); //attach Interrupt service routine to pin 2 Serial.begin(9600); //start serial terminal } int sens = 0; // set the sensor to view to be the first one void loop() { Serial.print(sens); Serial.print(": "); //print sensor being viewed to serial terminal Serial.println(analogRead(1)); //print the value from A1 to the serial terminal delay(500); //delay a little bit as not to flood the terminal digitalWrite(13, !digitalRead(13)); //toggle LED on/off } //Interrupt routine void pin2ISR(){ if(sens < 3) { sens++; //if sensor value is less than 3, look at next sensor } else { sens = 0; //else if not less than 3, reset sensor to be viewed to 0, only 4 switch positions } switch(sens) //set the output of D8 & D9 to the address to close the "switch" position on the DG409 { case 0: //select switch at address 00 digitalWrite(8, LOW); digitalWrite(9,LOW); break; case 1: //select switch at address 01 digitalWrite(8,HIGH); digitalWrite(9,LOW); break; case 2: //select switch at address 10 digitalWrite(8,LOW); digitalWrite(9,HIGH); break; case 3: //select switch at address 11 digitalWrite(8,HIGH); digitalWrite(9,HIGH); break; } }

So a breakdown of the code:

• In the setup I have declared some pins as outputs, set pin 2 as the pin to activate the ISR (Interrupt service routine - https://www.arduino.cc/en/Reference/attachInterrupt), and started the serial monitor going.
• I then declare the integer sens to be used to select the address of which switch to select on the DG409
• The loop basically reads the analogue input A1 and prints it on the serial monitor whilst also printing which sensor its looking at. it does this every 500ms using delay(500);
• The ISR is where most of the work is:
• 1st it checks if the integer sens is less than 3, if it is then it can be incremented
• if it is not less than 3, it resets it back to 0 again.
• it then checks the value of sens using a switch-case and sets the outputs 8 & 9 to the appropriate logic, and therefore address for the DG409 - truth table of addressing can be seen in datasheet.

So in my example, sens: 0 = pulldown switch (addr: 00), 1 = divide by 10 circuit connected to a 30V variable supply (addr: 01), 2 = LDR (addr: 10), 3 = pot (addr: 11).

Results

Uploaded the sketch and opened the serial terminal:

Pressed the pull down switch for about 5 seconds

Adjusted the 30V supply from 30V down to 0V

 So in this case at 30V my full scale was 574.;
574/1023 x 5V = 2.8V;
2.8 x 10 = 28V, well I did have the fine adjustment turned down (its an old supply) so I'll take that as correct!

The LDR output. Here I was moving my hand over the LDR

POT adjustment, rotating to both ends.

It all works as expected. So that's a win in my books.

Conclusions

Well it works, its pretty simple to do and you can get a lot more I/O for very little extra cost and effort.

However there are limitations:

• Time frame & speed limits - if the sensors require fast monitoring, and you have 4 of them, then you have to switch between them all really quickly.
•  This takes time and to get an accurate reading at a given time frame (say 1uSec) and the time it takes to go through them all takes 10uSec, then your going to loose some data on that sensor whilst reading the other sensors - so bear this in mind. 
• Make sure your sensors a slow when compared to the speed of the UNO (or whatever board your using) or utilise some sort of averaging in your programming.
• Giving up some digital IO - ok its only 2 pins! But if your applications uses lots of digital signals, then you might have to work out some sort of addressing or serial interface to your analogue switch, or find a switch with some serial interface.

Those are the only 2 major ones I can think of.

Changes?

So why did I use the ISR: well because between each reading I was waiting 500ms, which is quite a long time, so I needed to interrupt right in the middle of a delay function, rather than wait 500ms to check a change in state of the sensor select button.

I could have just collected the data from all the sensors and spat them at the serial terminal in one go, that too would have been quicker.

I did originally use some port manipulation to set the address of the switch. but that meant a lot of pins weren't being used and would look somewhat intimidating to noobs. Have to be efficient yet friendly. Though if any noobs are reading this then: seriously, click on the link and go tinker, its a function worth knowing.

Anyway, this should help if you are looking to get more I/O out of your UNO, or any dev. platform. Go have fun!

Constant Current Source - Back to the drawing Board

So, I'm right in the middle of moving. New digs, bigger house for the family and, as a bonus, a Garage with an apex roof! More storage, more room to tinker, its gonna be biblical!

Ask a question...

...get a different answer! Its a typical case of asking a question and then someone points out something else, opens a massive can of worms.

http://www.eevblog.com/forum/projects/i-should-know-this-but/

So this lead to some interesting research. Parallelling MOSFET's is not as straight forward as it seems...

The Issue

So compared to my original sec of the constant current driver, I was off target to reach a max current of 7A. It was turning out to be 5A. And although Parallelling MOSFET's was the right way of going about it, I was doing it wrong: http://www.ixys.com/Documents/Articles/Article_Linear_Power_MOSFETs.pdf

Because each FET does not have exactly the same R_DS or V_GTH. This effects the Drain current, I_D. This means that each FET will not pull exactly its equal share of current. This can cause a number of problems: One is one FET gets hotter and this changes the value of R_DS. But there are also 2 modes where the R_DS has positive and negative coefficients in the Linear region and the "fully on" region. For a constant current source, these MOSFETs will be working in the linear region, this means the the R_DS has a negative tempco and as temp goes up, R_DS goes down, more current flows, FET gets hotter and before you know it your MOSFET is smoke!

Take the proposed circuit above and the Total current I_t is 5A, therefore the voltage across the single shunt/source resistor R₁would be 5V for a 1OHM shunt. The same MOSFETs aren't always cut from the same die, as it were: Q1 could have an R_DS of 0.1OHMS at a V_GS of 3V and Q2 could have an R_DS of 0.105OHMS at the same V_GS. The data sheet does have these curves of R_DS Vs V_GS 

This means I₁ does not equal I₂, I₁ could be 0.5A and I₂ could be 4.5A, yet you'd still get the same voltage drop across the shunt resistor, R₁, the current is therefore unbalanced, Q2 bears most of the load and eventually, regardless of heatsinking, it goes pop!

Even if the datasheet say its bang on, if one FET is 1^oC warmer or cooler than the other, it could mean the difference between a few milliOHM's R_DS

Have a look at page 3. for the FDP6670AL -  it shows all the graphs.

Amongst other things Wong with the design: the LM324 was pointed out to be, well; a bit shit! Really bad offset voltage.

So how did I overcome these problems...?

Some Changes

First off; a better Op-amp: LT1014 - low voltage offset, same package and pinout as the LM324, and I have some in stock, perfect.

So that was easy, but now had to put thinking cap on about my FET array.

Instead of using 1 series shunt resistor and one Op-amp to do the control, I decided that 4 MOSFETS, each with a series shunt resistor and control op-amp.

As mentioned before though, someone will pick your design apart, and usually not in the area you asked your original question, you know: "you do know that...?" or "I'm surprised that works!"

Well that's exactly what happened, not that I'm ungrateful for it, far from it, ignoring criticism is a fatal mistake I find, that's if its constructive!

Seems my feedback network was somewhat thrown together (see Current Source Update for schematic): C8 & R18 on the output of U1:A are there to suppress some ringing on that output which showed itself on the load source, not desirable. So I just threw in a LP-RC filter and it got rid of it.

However I didn't measure the transient response, didn't even cross my mind to be honest. I referred back to the design I was taking my inspiration from and he had saw he had an RC in parallel in his feedback (R7 & C5). Possibly to compensate for the slow reaction time of the system to stabilise.

Someone else on the forum suggested that I reduce the output impedance of U1:A by using an emitter follower and that this would improve transient response too.

Only one thing for it: TO THE SIMULATOR!

Testing

3 circuits to test

1. Standard feedback constant current driver - no compensation nor common emitter output
2. Compensated feedback - as above but with an RC in the feedback network
3. CE output stage - standard feedback with a CE output stage

The shunt resistance will be the same in each design: 1OHM.

Each circuit will be test 10uS rise time stepped input first for 2.5A load current and then 0.1A. So V_REF in each will be set to a max of 100mV and then 2.5V. So lets have a look at the results

Circuit 1: standard circuit

RED = Gate voltage signal
GOLD = Load Current
BLUE = inverting input / feedback
GREEN = input signal/ref

As it can be seen there is some minor overshoot on the gate voltage and this also shows on the load current. So it'll overshoot by about 100mA and takes about 15uS to settle. Now lets look at 100mA:

Dear god! Well that's serious overshoot if ever I saw it! Clears the set point by 60mA (160%)! not good!

Lets see if circuit 2 is any better?

Circuit 2: compensated feedback

RED = Gate voltage signal
GOLD = Load Current
BLUE = inverting input / feedback
GREEN = input signal/ref

Really not that much difference at 2.5A, least not that I can notice.
100mA?

 Same as circuit 1 again. it doesn't seem to do much for the transient response anyway - might be doing something wrong here

Circuit 3: CE output of op-amp

 S when compared to the other 2 you can see the Gate voltage lags a bit behind the reference signal, however the overshoot seems to be lower and the settling time seems faster.

I didn't do a 100mA transient response cos I did some playing about and reduced the value of  R4 down to about 100R and repeated the above response.

The overshoot is much lower than R4=10k and settles much quicker. The gate signal still lags by a couple of uS but its not worth complaining! Here's the best bit at 100mA transient:

Wow! Compared to the other 2 that is pretty good! Very little overshoot and fast settling! I think we have a winner!

Conclusions

So why does this work - well I haven't done much research, but it looks like larger current gain by employing Q3, essentially making Q1 & Q3 a hybrid BJT-FET Darlington pair.

If I remember my control systems correctly, larger gain = faster response/sensitivity (https://en.wikipedia.org/wiki/Control_system#Proportional_control - paragraph 2) - which rings true with the vastly improved overshoot we see.

Edit: I have recently been corrected above: the output impedance from the CE stage coupled with the gate capacitance, forms a 1st order low pass filter that has better phase response than that of just the op-amp output connected to the gate - lower resistance - lower phase shift - better loop response. I'd like to simulate this effect and post it up later.

So I guess then I'll be doing a new schematic for this project then. But on top of this improvement, some more additions:

1. use 4 of these circuits in parallel to sink more current. Each shunt resistor will be a 1OHM 7W ceramic resistor. This means I can sink 2.5A per channel, meaning a total current of 10A! Mega!
2. Cos each channel is now 2.5A I can change my reference voltage to 2.5V, and guess what, I have some AD680 voltage references in DIP-8. Really good tempco on them!

So I now have a more precise, and better performing design, that exceeds spec! the max power I worked out to be about 68W, which is still ok, but not an improvement over the first design!

So I have a max rating for my design: 30V, 10A, 68W - so long as any of these are not exceeded, it should work.

The Down side: the only thing I can see is I have to employ another op-amp to do the current control, but meh, its not too difficult!