6
\$\begingroup\$

First prototype

I'm working on MCU programmable bench power supply. Its range is 0-50V and 0-3A programmed in 10mV and 10mA increments. It will be published as open hardware and source code that anyone can use it if find it usable. My "expertise" is more on digital and software then analog side and I got some issues with main regulator circuit. Voltage control loop works let's say acceptable for such non-professional grade equipment. As far as I can measure PS (power supply) in CV (constant voltage) mode (1A resistive load) works predictable and stability and precision is very good. Output ripple and noise is within 2mV using only linear serial regulator but 100Hz component (for rectified European AC) is unfortunately still present and visible. I'm talking about current status which does not include planned switching preregulator.

Something that is far from ideal is current control loop. First I spent some time to make it stable since it was oscillated every time when PS enter CC (constant current) mode. So far I have workable solution with adding C10 (150pF) as negative feedback on transistor Q4. With same load (1A) in CC mode output ripple and noise is huge 50mV. I tried many things and something that I still didn’t find out how to play with poles and zeroes in practice to insure stability and good regulation. My question is how to improve circuit presents below to get better load regulation and avoid stability issues.

Measurement for same load in CV and CC mode

I removed from the schematic for the sake of simplicity digital control (ADC/DAC/MCU). Switching pre-regulator and voltage control loop are drawn as boxes. I_SET is using for set max. current (0-1.5V for 0-3A). -2V derived from -15V using ZD2 was required to go down to 0V otherwise it cannot goes below approx. +1.6V. Not nice at all but currently I have not idea how to eliminate that issue.

PS Current control loop schematic

EDIT1 (2014-10-30):

A new proposal in accordance with gsills suggestions and decision to change current shunt monitor is depicted below. The picture is exported from TINA 9 simulator. New proposal also required even lower Q4 emitter voltage to reach 0V therefore Zener diode is now for 9.1V. AS you can see I_OUT reference voltage is reversed. I'm wondering is it possible somehow to use possibility to reverse current shunt monitor inputs to get negative current readouts that with some additional modification is possible to use positive I_SET values as before.

New proposal, image exported from TINA9

EDIT2 (2014-11-07):

I got yesterday INA193 which I decided to use for further testing. Gain is now 20V/V and I'm still using R010 shunt resistor that is a little bit to small since TI recommendation is to have a 100mV drop on shunt resistor for the full scale. Therefore optimal value for 3A will be R033 but I'll probably end up with R025/3W (75mV for the full scale) that is easier to find. New schematics is depicted below. IC2C is added to invert I_REF voltage (0-1.5V for 0-3A). Please note that switching preregulator and voltage control loop are not deployed (voltage control loop need to be inverted, preregulator is still waiting to be tested).

PS Current control loop schematic r2

Thanks to Gsills recommendations few important things was achieved:

  1. C5 between Q4 C and B is not longer necessary (before without it current control loop was unstable).
  2. Feedback loop for IC2A now can works with resistor added in series with C9 (before adding any value makes current control loop unstable).
  3. Simpler control of Q1 (BF245B is no longer necessary).

\$Vout\$ ripple is still considerable, around 50mVpp for 1A and 250mVpp for 3A (full load).

New design and Vout measurement

As it's shown on the picture above the main component in ripple&noise is 100Hz which for some reason this control loop cannot reject/filter successfully. I added additional input C6 (tested with up to 10mF) but that does not improve situation. So, now remain the question how to decrease 100Hz on output? Another important question arise in the meantime: how to put under control over shot during power up and power down? Is it possible to achieve that with N-channel MOSFET or P-channel must be used (i.e. IRFP9240)?

\$\endgroup\$
  • \$\begingroup\$ Why are you using diodes on the outputs from the voltage sence and current sence circuits. I see problems here. \$\endgroup\$ – Andy aka Oct 24 '14 at 15:03
  • \$\begingroup\$ Diodes makes "wired-OR" and prohibit that output from voltage loop influence current loop and vice versa. I have same situation even if D7 is removed and whole voltage control loop 'box' is removed. \$\endgroup\$ – prasimix Oct 24 '14 at 15:16
  • 1
    \$\begingroup\$ Did you look at the output current with a current probe when regulating in CC mode? Looking at the voltage isn't entirely fair since the voltage loop is doing nothing at that point. What are you using as a load? (An active load has its own loop and can end up fighting with the PS loop, making a steady load look dynamic. Use resistors if possible) \$\endgroup\$ – Adam Lawrence Oct 24 '14 at 16:23
  • 1
    \$\begingroup\$ You can still have a positive I_Set voltage by applying a positive offset voltage to the non-inverting input of what the TINA schematic has as U2. Instead of grounding R8, connect a positive voltage, 0.5V there would give I_Set range of ~2V to ~0V. \$\endgroup\$ – gsills Oct 31 '14 at 0:35
  • 1
    \$\begingroup\$ Thanks for the accept. Mains ripple with 1A and 2mF bridge filter is probably ~ 4.5Vpp, so 50mV out is ~40dB rejection. That looks about what the current loop gain is at 100Hz, so no surprise. It is possible to improve loop. Overshoot on start could be integrator windup or power sequence. You'll probably want to open another question or two when you are ready to cover. \$\endgroup\$ – gsills Nov 10 '14 at 0:35
2
\$\begingroup\$

Main conclusions about this current loop, based on a drive-by analysis. The loop has way too much gain, and that gain is too variable over the span of output voltage. Also, it is doubtful that the bandwidth constraint forced by the INA282 is comprehended.

Let's go through each stage of the loop, in order of importance, to see how they look.

FET Driving Amp Stage

This is the most important stage in the loop since both the voltage loop and the current loop depend on it. Mess this stage up and there will be two problem loops for the price of one mistake.

The IRFP240 (Q1) is driven by a discrete amplifier made of the common emitter and common base stack-up of Q4 and Q2. It is almost a cascode amp, but because of R5 existence and modulation of Q2-B by Q1-S, isn't quite. Approximate gain for the stage is the ratio of impedance at Q2-C to the impedance at Q4-E. Dominant impedance at Q2-C is the JFET Q3 acting as current source with \$V_{\text{gs}}\$ = 0, while the dominant impedance at Q4-E ends up being \$r_e\$ of Q4.

JFETs like Q3 typically have variation in \$I_{\text{dss}}\$ of several mA unit to unit, but more important, channel impedance can vary by orders of magnitude as a function of \$V_{\text{ds}}\$. If \$V_{\text{ds}}\$ is ~2V \$R_{\text{ds}}\$ will likely be about 1kOhm. As \$V_{\text{ds}}\$ increases to ~10V \$R_{\text{ds}}\$ will increase too to something like 20kOhms. These parts don't really start to behave like current sources until \$V_{\text{ds}}\$ > 10V. For high output voltage (\$V_o\$) amplifier gain will be at the lowest and at low \$V_o\$ amplifier gain will be greatest.

BJTs like Q4 will have \$r_e\$ values from about 1 Ohm to 10 Ohms typically. It is not a directly specified parameter and will be a function of current and temperature. It may seem like R13 would swamp the wildness of \$r_e\$. Not so, since R13 is bypassed by C14, what is left is only \$r_e\$. For the calculations here, choose \$r_e\$ = 7 Ohms. Take the ratio of 1kOhm to 20kOhm by \$r_e\$ and find that over the range of \$V_o\$ amplifier gain could vary between about 40dB and 70dB at 1kHz. Too much gain, but mainly too much variation to be useable.

Here are some things to improve performance of this stage:

  • Get rid of C10, it's putting a pole at a frequency that either nothing or a zero would be needed.
  • Connect C12 and C14 to ZD2 cathode, instead of Q4-E, so as not to bypass R13.
  • Increase value of R13 to reduce gain. Maybe as much as 300 Ohms.
  • Get rid of R5, it's not doing anything for you either.
  • Get rid of C3, it makes ripple rejection worse.
  • Get rid of C6, it's not necessary.
  • Replace Q3 with R4 and increase to 10kOhms to eliminate gain variation with changing voltage. Maximum dissipation in a 10kOhm should be ~ 0.25W, so use 0.5W part.

All this will allow the amplifier to produce a lower and more stable gain with a response flat out past 10kHz.

Current Sense Amp

IC1 is a diff amp stage that, along with R1, does an inter-domain conversion of current to voltage. A look at the title description at TI website shows the INA282 to be in the Zero-Drift family of amplifiers, which means that it is a switched capacitor part. That makes this a sampled data loop. So, in this linear regulator, the current loop will resemble that of a switching power supply because, Nyquist.

Schematic shows amplifier gain to be 50V/V which is 0.5V/A or -6dB. Gain will be flat out to about 10kHz or so and then, at around 100kHz, the gain and phase will crash like a load of bricks over a cliff because of the sampling. It will be undesirable to have the loop bandwidth be greater than 10kHz because of the rapid loss of phase at higher frequencies.

Since this stage has -6dB of gain, the rest of the stages combined can have a maximum of about 20kHz bandwidth. For example, at 1kHz the remaining stages combined could have a maximum gain of 26dB with a -20dB/decade rolloff for good loop performance.

Gain and Error Amp Stage

This stage starts with IC2D (TL074) as a diff amp, followed by IC2A as a non-inverting integrator. Since there is already a nice balanced diff amp stage (INA282 - IC1) this second diff amp is not needed. There will be a better way to do any gain and level shifting without using another diff amp, a way that would not require a bunch of tight tolerance resistors.

Non-inverting integrator for error amp. So many problems with non-inverting integrator use, stated categorically, loss of flexibility and options. The minimum attainable gain is 0dB, but usually, and this case is no exception, gain less than 0dB will be needed for some of the loop bandwidth.

Here's an idea. Turn IC2A and IC2D into inverting stages. Make IC2D a unity gain inverter and feed +2.5Vref into its non-inverting input through a 10kOhm resistor to take care of the offset. It will have better precision using 1% resistors than the diff amp using 0.1% resistors. Make IC2A an inverting integrator and feed I_Set into its non-inverting input through a 30kOhm resistor. You'll have to add a resistor in series with C9 to place a zero, but you'll have total control over where it is, plus the gain can go less than 0dB and maintain -20dB/decade rolloff until you need that zero.


Edit: About mapping I_Set to Io using inverting amps. Since IC2D would now invert the current signal, I_Set would need to be inverted too. This shouldn't be a problem since a micro-controller is being used to determine I_Set. But since most micros don't have negative outputs, an offset to the non-inverting input of IC2D will be needed. For the case IC2D has a gain of alpha, an equation for I_Set would be:

I_Set = offset - alpha(CSgain Io - offset + Vref)

where CSgain is the gain of the current sense amp (including R1), offset is the offset voltage applied to the non-inverting input of IC2D, alpha is the gain of IC2D, and Vref is any reference voltage applied to the current sense amp IC1.

For example if CSgain=0.5V/A and Vref=0V and offset=0.75V, I_Set would decrease from 1.5V to 0V as Io increased from 0A to 3A.


A Word About Power

\$V_o\$ of 0V to 50V with \$I_o\$ up to 3A is a lot of range for a linear. Let's say the Prereg voltage is 58Vand \$V_o\$ is set at 3V with a load current of 3A. Q1 \$V_{\text{ds}}\$ will be 55Vand its power will be 165W. Maintaining a junction temperature of 150C with an ambient temperature of 25C would require a total thermal resistance, junction to ambient, of 0.76C/W. Unfortunately 0.76C/W is lower thermal resistance than the junction to case thermal resistance of an IRFP240, so nothing short of refrigeration would help.

If you really want to supply that range of voltage and current, the preregulator output will have to track \$V_o\$ allowing Q1 \$V_{\text{ds}}\$ of 8V to 10V as head room. That would end up with 24W to 30W in Q1.

\$\endgroup\$
  • \$\begingroup\$ Many thanks gsills for your analysis and suggestions. Before updating schematic I'd like to say that using INA282 in current control loop was mistake. You are right, mode of operation and bandwidth is not appropriate. I found on TI site some better candidate such as INA194 (50V/V, 300KHz) or INA183 (20V/V, 500KHz) so my intention is to proceed with new current shunt monitor. \$\endgroup\$ – prasimix Oct 30 '14 at 9:23
  • \$\begingroup\$ INA19x do not require 2.5V offset and it can be now removed from the picture. So, IC2D could be used for Vout from current monitor but I don't understand how you'd like to convert whole stage from non-inverting to inverting without making any correction in "output" stage (Q4). \$\endgroup\$ – prasimix Oct 30 '14 at 9:28
  • \$\begingroup\$ I'm aware about possible power dissipation which I want to limit to 20-30W. Heatsink (on right side of the picture) will be mounted outside the box and can easily handle such dissipation. I'm also thinking about IRFP250 which has even better SOA (safe operating area). \$\endgroup\$ – prasimix Oct 30 '14 at 9:31
  • 1
    \$\begingroup\$ @prasimix Nothing really wrong with INA282, just has that unusual characteristic out around 100kHz. As to making IC2A and IC2D inverting, for loop and dynamics there is no difference between 2 non-inverting and 2 inverting sub-stages, but the inverting case is more flexible and easier to compensate. Only difference would be how I_o gets mapped to I_set, since would be inverted with inverting amps. I'll edit to make that clear. \$\endgroup\$ – gsills Oct 30 '14 at 20:06
  • \$\begingroup\$ @prasimix As to power. 30W can be done. For realistic temps like Tamb=35C, Tj=110C total thermal resistance is ~2.5C/W. \$\endgroup\$ – gsills Oct 30 '14 at 20:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.