Wednesday 20 May 2015

Lab Power Supply - Current sensing/limiting

I've made some progress on the current sensing and current limiting circuit design but I'm still not quite there yet. This turned out to be a lot harder than I thought. I split the design into the part that measures the current and the circuit that controls the voltage regulator to limit the output current (current error amplifier).

My goal was to be able to measure and limit the output current to an accuracy of 1mA across the whole current range (0-5A) and the whole voltage range (0-30V). I wanted the current limiting circuit to be as quick as the voltage regulator but I found this isn't possible.

Basics of Current Sensing.

The idea is that we want to be able to set a maximum output current so that should something go wrong in the circuit powered by the supply, the circuit won't be damaged to much. The first part of the problem is actually measuring the current flowing.

While I will try and explain the basics here, I suggest looking at Linear Technologies Application note 105 as I think they do a better job.

The current sensing circuit measures the current flowing out of the power supply by inserting a small resistor in the power supply circuit and then measuring the voltage across this resistor. The resistor can be connected either between the positive side and the load (high-side) or between the load and the negative negative size (low-side).

High-side sensing is harder to implement as the inputs are likely close to the power supply rail (more on this later). The problem with low-side sensing is that if there is another path to ground that doesn't go via your sense resistor, you won't see any of it. I don't think this is a problem for my power supply as even if you tied the negative terminal to ground (and then grounded the negative side of the load) it would still complete the circuit via the ground part of the supply.

Current Sense Resistor

Choosing a current sense resistor and the amplifier to measure the voltage is quite tricky as you need to trade off the cost of heat dissipation in the sense resistor with the accuracy.

Working with the worst case settings of 5A, if I choose a 1 ohm resistor as the sense resistor then each mA of current generates 1mV across the resistor (which is easy to measure). However this will dissipate  P = I^2 * R = 25W of power at full load. That's enough to require a heatsink and the increased temperature will cause the resistance value to vary (due to the resistors tempco). Also, this will drop 5V from the maximum voltage the power supply can deliver at full output.

On the other hand if I use a 1mOhm resistor then this will only burn 25mW and there is only 5mV drop. The problem then is that 1mA only generates 1uV on the resistor which is well below the offset voltage of many op amps. 

The solution is to use 10-50mOhms which is still only 250mW - 1.25W, and 50mV - 250mV drop. Then 1mA will measure between 10uV and 50uV which is still pretty small but much better.

Current Sense Amplifiers

There are a few different ways of implement the current sensing circuit. Apart from high-side vs low-side you also have the choice of using a dedicated current sense module or to build one from op amps. Also, some of the high-side options are powered from the circuit they are monitoring and some can be run off an independent supply, There are modules that support digital output but this is no good unless they also provide an analog output as I need this for the current limiting circuit. Some have a fixed gain (or a set of fixed gains) where others can be configured with resistors.

A couple of the critical parameters for me were the ability to sense the current when the voltage is close to zero, having 1mA resolution (so therefore 5uV-25uV offset voltage) and an input range (common mode range) of 0 - 30V.

So I started by looking at dedicated, high-side current sense amplifiers and went through the manufacturers.

  1. From TI I found the INA-225, INA-250A2 to be the only ones that came close. Both had offeset voltages in the vecinity of 100uV or more which would mean I need at least a 200mOhm sense resistor which equates to 1V drop and 5W of power lost. The bandwidth is pretty bad too at between 5kHz and 250kHz.
  2. From Linear Technology I found the LT-1999 and LT-6105 that support he common mode voltage range I am after but both have silly offset voltages (300uV and 1.5mV). 
  3. Maxim Integrated had two parts that fit my common mode voltage range - the MAX44284 and the MAX9643 although the MAX9643 had an offset voltage of 130uV. The MAX44284 looked perfect in that it supported separate supply voltage from the sense voltage, supports the full voltage range and has 10uV offset. The only downside is the thing is quite slow (between 3kHz and 400Hz and have settling time of 1.5ms).
All in all I was a bit disappointed with what was on offer. I'm not completely convinced that high-side monitoring is essentially and I was wondering if I could get away with using an op amp and monitoring the low side.

One of the issues with using an op amp to measure current on the high-side is that the quoted Common Mode Rejection Ratio (CMRR) apparently doesn't cover the case where the device is close to the rails. I don't have this issue as I am running most of the circuit at 40V so I can drive the MOSFET gate - so long as the op amp can handle this it won't be close to its rails.

So then I went looking for op amps
  1. Linear has the LT6015/6016/6017 which has 50uV offset and 3,2MHz Gain bandwidth product. This is close but I still need to use a 100mOhm sense resistor which means 2.5W power consumption. 
  2. There is an LTC2057 which has over-the-top input (so rail-0.3 to rail+0.3), has 4uV offset and is quite low noise. Again it is slow (GBP of 1.5MHz) but not too bad.
  3. I looked at Max Integrated, searched for op amps with  the required supply range, high CMRR and low offset voltage and one that caught my eye was MAX44241 which has an offset voltage of 5uV max! Also it has a bandwidth of 5MHz. Only down side is that it has a slightly lower voltage range (36V). The one issue with this part is that based on some quick googling it appears to be currently unavailable anywhere (certainly not for a while anyway)
  4. The TI OP27A looks perfect - fast, very low offset voltage, common mode range of -10 to 10V (which is fine for low-side sensing). Just one problem - they cost $55 each!

This looked pretty grim in that there are few choices and they are expensive or unavailable. Then it occurred to me that if I am sensing from the low side and say I configure the circuit to output 1V/1A then the maximum output will be 5V. I could use an op amp running at say 5V and this would work (well 5.5 would be better but anyway...).

Conveniently this post on the EEVBlog forum has a similar circuit and includes a parametric search of op amps. I looked through of those but the one that really stood out was the LTC2050HV that can run at 10V and has a 3uV offset and gain bandwidth product of 3MHz. Overall this looks pretty damn good and they cost about $3AUD in small quantities.

Current Sense Circuit

The current sense circuit is quite simple in that it measures the voltage across the sense resistor and the amplifier feedback is setup to multiply the voltage to get 1A per volt.


The output matches pretty closely with what I want  - it is pretty close to 1V/1A


However the response is a bit slow (as expected):


Current Control Circuit

The usual current control circuit you see is like this where you do a diode-or between the output of a voltage error amplifier and the current error amplifier. Whichever of the two has a lower output will control the gate of the MOSFET. I found I had to buffer the monitor voltage as otherwise the capacitance used to compensate the current error amplifier messes with the current monitoring signal.


This did work and limits the current. As expected, the circuit takes some time between the output current rising and it clamping it to the limit. I don't think I can improve on this as I am limited by the speed of the current sense circuit. In the trace below I set the output voltage at 25V, I configured the load to switch between 1.3A and 2.5A and I set the current limit to 1,8A. You can see the load shoot up to 2.5A and then, about 100uS later, get clamped back to the limit. The current settles in around 200uS. 


This is pretty much perfect I think. There is no way to make it react any faster and there will always be a short overshoot in current.

When the load falls away you can see the current undershoot slightly and then the output voltage and current climbs back up (slowly). I'm not bothered about the current undershoot but I noticed there is a significant voltage overshoot in the output which I don't like at all.


I thought about this for quite a while and figured out that the reason is that while the current  error amplifier has control, the voltage error amplifier output shoots off to the positive rail. When the current error amplifier returns the voltage error amplifier has to come back down and this is dominated by the op amps slew rate.

I had an idea - what about if we use the current error amplifier to control the set voltage on the voltage error amplifier. If I do a diode-or between the set voltage and the current error output then when the current drops off it should not climb higher than the set voltage. Essentially something like this;

The one downside of this approach is that the output voltage will always be one diode drop higher than the set voltage. This means I can't take the supply down to below 0.6V unless I use a negative set voltage which isn't ideal. It does fix the problem very nicely however - here is the voltage/current trace when the current goes down.


The current/voltage recovers within about 8uS and there is no overshoot. Perfect!

I tried using another diode to get rid of the voltage offset but this doesn't work at low voltages. Maybe 0.6V might be a sensible minimum output voltage? I have to think about this.

What does the Jaycar supply do?

As before I thought it would be interesting to see how the Jaycar supply handles over current. As before the short answer is woefully!

In the scope trace below the yellow trace shows the current flowing (1V/1A). The dummy load was set for 2A and you can see the current peaks at around that. The supply was set for a current limit of 500mA and you can see the current rise to the full 2A and then the limit circuit kick in and bring it back down. The green trace shows the output voltage while this is happening and it goes from near zero to just under 6V

It is just so slow though. It takes 3-4ms for the current to be limited. Checkout the slow rise time on the voltage too. Crazy!


Next I hope to get some parts and test the current limiter on the breadboard.


Tuesday 12 May 2015

Lab Power Supply - Progress!

Since the last post I added a new winding to my transformer, sorted out my ground problems and figured out why my pre-regulator was messing up the output. The voltage output now actually looks really good!

Transformer Winding

So I needed an additional winding so that I could get a voltage a few volts above the bulk capacitor in order to drive the MOSFET gate. I tried a voltage doubler but this messed with the AC waveform enough that it meant the pre-regulator zero crossing detection didn't work.

The plan was simple - I wrapped a few turns of insulated wire around the transformer, attached the ends to a multimeter and turned the transformer on (being careful not to short the outputs as that would be bad!). I calculated I needed to add 40 turns to get 10V. In the end I only got 8V but it doesn't matter.

I bought some 1.2mm (18AWG) magnet wire from Jaycar which is way more than I need as the current required from this winding should be minimal. I got a bolt from the shed and put this into the chuck of a battery drill. I wrapped a couple of turns on the bolt (particularly the square head bit) to get the wire attached. I wrapped a cloth around the wire so I could hold it and turned the drill on slowly. Every so often I squashed the wire together. I put the whole real onto the bolt (as I didn't want to run out!)

I fed the bolt through the centre of the transformer and wound 40 turns roughly spacing them on the torroid. I cut a short length of green wire, stripped the end and used pliers to pull the wires out of the centre. I fed this over the copper wire so I could colour tag the winding wires. I used a file to clean the lacquer off the ends of the new windings I added. I plan to add a layer of tape to protect the winding but I don't want to do anything too permanent until I am sure it all works.



With the transformer disconnected from AC I connected a signal generator to the primary and started a 50Hz sine signal. I attached one scope probe to the primary, another to one of the original secondaries and one to the new secondary. Using this I could see the phase of the winding relative to the input and the other windings.

I powered the transformer and measure 8V RMS on the new winding. Ok not the 10 I was aiming for but it is near enough.

Using the new Winding

I was talking with Peter Oakes about how to integrate this winding and he was suggesting I attach it so that the ground reference of the new winding and circuit is the voltage on the capacitor. This way I could use a small (say 9 or 12V regulator) to regulate the output. I didn't like this for a few reasons:
  • I'm using this to power the regulator op amps and they have a maximum voltage of 40V
  • The output voltage will have a ripple that matches the bulk capacitor and I don't want this sneaking into the output.
So my plan was to attach a bridge rectifier to the new winding and then attach the negative output of the rectifier to the positive output of the main bridge, Then I will use a high voltage regulator to keep this at 40V (when the supply is running in 30V output mode) and 25V when the supply is running in 15V mode. An LM317HVT will do this. I only have the regular LM317 so for now I am running it at 33V which isn't quite enough but close.

Here is the circuit with the new winding

And here is what LTSpice thinks it will do

LTSpice

At this point I integrated the extra winding plus the pre-regulator circuit into a single circuit so I could simulate this. I found that even though the extra winding works it would take forever to simulate and would sometimes fail. I ended up going back to using voltage sources.

I also found that the small overshoot that occurs at turn on would send the down-regulator nuts. It would try to soak up current to handle the overshoot and would go into oscillation as a result. I haven't figured out how to fix it and for now I have just removed the whole thing. It seems like a good idea but I am not absolutely sure I need it.

Pre Regulator

Then I put it all back together and added the pre-regulator circuit that I designed in the previous post. The problem before was that I couldn't use this with the voltage doubler as the doubler circuit messed with the AC waveforms too much. By using the extra winding I don't have any of these issues and the saw-tooth waveform looks ok.

This looked really promising. The pre-regulator will fire when the capacitor voltage dips and charge it back up. Under no-load this is very infrequently and oddly I could actually hear a tick noise when it fired! The ticking would get much quicker when I enabled my dummy load!

I decided I didn't need the transistor to pass current to the SCR gate and it works fine without this.

I've been using a 25V zener diode to set the output voltage so I just added another one in series to set the pre-regulator output a few volts above. It took some experimentation to choose the right size zener diode so the capacitor voltage didn't fall below the desired output. I also found the firing is more consistent if I lower the compensation capacitor from the op amp to 0.1uF.

I also found that there was a fast, high-frequency ringing transient at both the output and the bulk capacitor when the SCR fires. I found a 470uH inductor at Jaycar that was rated for 5A and placed this between the SCR and the bulk capacitor. This cleaned up the transient completely.

It also occurred to me that if I use a zener to subtract a few volts from the capacitor voltage before I feed it into the pre-regulator I only need one input and it will automatically manage the voltage.

Grounding

The pre-regulator voltage seemed to be working quite well but still the signals area all over the place. I figured out that the signals varied greatly depending on where I placed my scope ground lead (I only hook one up at a time for now because of this issue).

I thought that much of my problems were because the large currents going from the bridge to the bulk capacitor and from the capacitor out to the MOSFET and onto the load were going through the breadboard. As a result it would really depend where on the breadboard you connect up ground for other parts of the circuit.

I decided I would rebuild the high-current parts of the circuit on vero-board as I thought this would have much higher current capabilities. I ran flying leads off this board for the SCR gate, the MOSFET gate, the bulk capacitor voltage sense line, the gate bias regulator and ground. I connected all the grounds to a point close to the bridge rectifier (between it and the capacitor).

Still this didn't really work - the output was a mess. I experimented with how it looks when I place my scope ground on the breadboard ground (which is all pretty much the same voltage now that all the high-current stuff is on vero-board) and at the output terminals. With the ground on the output terminals, there is definite dip in the gate drive when the SCR fires and the capacitor charges.

This didn't make any sense. I experimented with adding resistors between components in the LT SPCIE model and found that if I ground at the same point in my circuit I get similar output results. If I move the ground for the rest of the circuit to a place closer to the output then a lot of the problems go away.

I modified my vero-board to move the ground to the front terminals. This actually made a significant difference but still the gate drive voltage dipped. I realized the ground for the gate drive regulator was also still near the bridge rectifier so I moved this also.

Output Mess

Still the output is a mess. Now the ground is consistent between the output terminals and the bread-board electronics. I found that now the gate bias was very steady (mVs of ripple). So why is the output so bad?

I looked at the output of the differential amp measuring the voltage on the terminals. The output now matched the output terminal voltage waveform exactly.

I looked at the drive going to the base of the transistor driving the MOSFET and weirdly this seemed to match the output to some extent. I would have expected it to be moving in opposition to the output to try and keep it under control.

So I looked at the voltage on the zener I am using as a reference and this was also moving a *lot*. I noticed this before but because the gate drive voltage was also moving a lot I put it down to variations in current. Now it isn't though.

It took a while but I figured out that the op amp that drives the pre-regulator was pulling big currents when the SCR fires (mAs), This caused the zener voltage to wander. I used the last op amp in my quad package to buffer the zener voltage for the pre-regulator circuit and it worked! All of a sudden the output is *very* clean.

Pictures

Here the pink/blue traces are the AC signals. The yellow signal is the SCR firing pulse and the green signal is the zero detect circuit.


This one shows the line-sync ramp generated from the zero detect.

Here is the output from the PSU with 1A load. The noise (with 20MHz bandwidth limit enabled) is around 1.3mV RMS. Pretty damn good considering it is running on breadboards!


Here is the output circuit responding to a change in load from 1.5A to 0. Recovers to within 15mV in 15us!! The overshoot is a bit bad though at nearly 170mV.


This is the undershoot when the load goes from zero to 1.5A. Again it reacts very fast but the undershoot is pretty big (400mV).


And here is the current circuit with the pre-regulator and the extra winding


So time do do some current measurements and current limiting!

Sunday 3 May 2015

Configuring Centos with a DHCP Reservation

Sigh.. I'm sure I struggled with this the last time I had to do it. Hence I am writing a little note for next time!

The scenario:

  • Windows 2003 Server running a DHCP server
  • Centos 6.4 VM I want to automatically allocate a static address to. Essentially I don't want to hard-code the IP address/DNS server addresses in the VM's config. I want it to get these automatically from DHCP but as the VM will act as a server I need the IP to be static and for the machine to be locatable via DNS.
So initially Centos is installed and the machine boots with whatever default address got allocated to it. Right click the connection icon at the top left of the screen and choose Connection Information. You will get this screen:

Select and copy the hardware address. Logon to the windows 2003 machine where the DHCP server is running and run the DHCP configuration tool for Administrative Tools.

Open the scope, right click reservations and add a reservation like this (sorry I blotted out the domain name).


Paste in the hardware address and delete the colons (:).

The description doesn't matter. Leave the DNS settings as default.

Now go back to centos, right click the connection icon, choose Edit Connections, select the  System Eth0 connection and click edit.

For some reason when  I first installed the system, the network interface was configured NOT to start at boot time. This is pretty annoying as you have to connect to the console to get the machine running. Set it to connect automatically

Initially I thought I had to set the DHCP client ID. After many reboots I figured out if I set this it wouldn't work! If I just set it to blank the DHCP server gives the machine the right address and we are all good. Apply the changes, enter your root password and let it save.

In the Windows DHCP config tool, find the current reservation for the machine, right click and select delete. If you really want to be sure restart the DHCP server (right click the node below DHCP, choose All Tasks and restart). Then in Centos click the network icon, click System eth0, wait for the spinning to stop and then check the connection settings again. It now should have the configured IP.

Then I found that for whatever reason the DNS still doesn't automatically get this new entry even though the DHCP server is supposed to update it.

On Windows 2003 server, open the DNS management tool from Administrative tools, open the forward lookup zone and open your zone. Right click and choose New Host (A). Enter the name of the host (just the short part so centosdev in my case). Enter the IP address. I left the 'Create associated pointer (PTR) record checked but I got an error when I hit add host saying it couldn't. It still works so I'm not worried. This is probably because I am running on a non-routable address range.

My client computer (i.e. the computer I am accessing the centos server with) is a Windows 7 machine. I did a ipconfig /renew and then I was able to ping the new computer by name and I was able to access it via VNC.