This page will look at the voltage regulator adaptation based on the self bias CCS that John Swenson designed as well as adding more functionality.
For testing power supplies I modified one of the amplifier modules based on the self bias CCS. By adding a second bias string it's possible to dial in the standing current then drive the unit with a test signal to see how the power supply or regulator responds.
Schematic of the power supply tester:
The DC pot is used to set the DC current loading the power supply or regulator under test. The AC pot allows the signal level to be set outside the sound card in the lab computer.
The first set of measurements is output impedance VS frequency. In this test the power supply tester was driven from the lab computer sound card. The Lab computer is equipped with an M-Audio Audiophile 192 sound card and a custom input signal processor box. The input signal processor box protects the sound card inputs as well as attenuate the incoming signal by 40dB, 20dB, 0dB, or amplify by 20dB. Audio Tester software is used for these measurements. Audio Tester was setup to sweep from 1Hz to 95Khz and use the right channel as a reference. Using the reference channel allows the sound card to measure very flat down to 1Hz.
The Swenson regulator circuit
John created a very good voltage regulator based on the self bias CCS by slightly rearranging the bias string and adding an error amplifier to lower the output impedance. The error amplifier is the third LND150 (Q5). Q5 compares the output voltage of the regulator to the reference voltage on C2.
Changing the values of R10 and R11 sets the output voltage max and range. The circuit seems to be stable with 1ma minimum load.
This screen shot shows the output impedance VS frequency for the regulator
The basic test setup has the regulator powered from a 200V unregulated supply with .5 volt of ripple. The regulator is set for 100V output. The power supply tester is set to draw 100ma DC from the regulator. The AC signal is set to 34.6ma RMS (100ma P-P). This swings the load from 50ma to 150ma.
The blue trace is the response of the regulator with John's capacitor values. The red trace shows the response with C2 increased to 1uf. The green trace is with C2 increased to 10uf. The black trace is the same setup as the green trace except that a .1 ohm resistor was added to the output.
To measure the output impedance at 1Khz the power supply tester was set to draw 100ma DC and 20ma RMS AC. The AC ripple was measured with a o'scope then Ohm's law was used to calculate the impedance. With 20ma RMS signal applied the ripple on the regulator was 5.8mv p-p. Converting the P-P measurement on the scope to RMS> 5.8mv p-p/2*.707= 1.98mv RMS. Impedance = 1.98mv(E)/20mv(I)= .099 ohms.
During the frequency sweeps there was some occasional instabilities. Increasing C1 to .47uf took care of the issue.
Updated schematic with the larger cap values
Ripple rejection. The next screen shot shows the regulators power supply rejection. Darn nice!
The setup for the ripple measurements has the input box set for 20dB of gain. The .5 volt of ripple was just over the sound card's input limit with the 20dB gain engaged so the unregulated input was captured with 0dB gain. The file from Audio Tester was massaged using Excel to add 20dB to the signal level. The main artifact from this manipulation shows up above 40Khz where the ripple met the sound card noise floor.
The power supply tester was drawing 100ma DC with no AC component for these tests.
The 3 traces overlaid are good to compare the 3 values of C2 but not good to try and see details.
Here are 3 screen shots each showing one value for C2 as well as the measurement of ripple rejection at 120Hz.
C2 .1uf
C2 1uf
C2 10uf
The regulator is performing very nicely but there is one issue left... If you short the output it blows up. I want to use several of these regulators to build a bench supply with multiple adjustable outputs. A bench supply has to be bullet proof!
The mostly bullet proof version...
First issue is gate protection for Q5. The 18V back to back zener diodes (VR3, VR4) provide protection for Q5. Another issue is the charge stored on C2. With a hundred or more volts on C2 there is a lot of energy stored. There is no way the protection zeners on Q5 would survive the surge of shorting the output. Adding a diode between Q5 and C2 removes the need to discharge C2 when the output gets shorted. If the output is shorted to ground D1 reverse biases and disconnects C2 from the circuit. R10 and R11 discharge C2.
The circuit also needs current limiting. Q6, R1, R3, and R4 are the current limit circuit. The output current causes a voltage drop across R1. When the voltage reaches around .5 volts Q6 starts to conduct removing gate drive from Q1 and shunting the reference bias current to the output causing the output voltage to drop.
To test the current limit the power supply tester was set to 100ma DC. The AC drive was increased until the output voltage dropped .1 volt. The current was measured with a DC current probe on the bench scope.
With R1 at 2 ohms the output current is hard limited to ~300ma. The first current limit circuit did not include R3, the fold back resistor. Without R3 in the circuit hard limit was 300ma but the current limit started working when the current peaks were at 170ma.
Adding R3 provides what is called fold back current limiting. When the output voltage is up a small current flows through R3 pulling the base of Q6 down about .24 volts. This allows the regulator to supply greater current when the voltage across the regulator is lower (i.e. normal operating). As the output voltage drops the current through R3 decreases allowing the current limit circuit to be more sensitive.
With the values shown in the schematic above the hard current limit is ~300ma but with the output at 100 volts the peak signal current can be as high as 420ma before the output voltage starts to drop.
To get the power supply tester to deliver the drive necessary for this test R1 had to be reduced to 5 ohms.
The only down side to the current limit circuit is the resistance of R1 reduces the tranconductance of Q1. This increases the output impedance some. The output impedance measurement with the current limit circuit in place shows the output impedance increased from .099 ohms to .142 ohms.
Frequency VS impedance with current limit circuit in place
You can see that the output impedance (Z) of the circuit with current limit (red trace) (.142 ohms output Z) fits between the green trace where the 2 ohm resistor is shorted out (.099 ohms output Z) and the black trace where a .1 ohm resistor is added (~.2 ohms output Z).
The regulator circuit with current limit is quite robust. I've repeatedly shorted the output to ground and it survives. Just keep in mind the power ratings of the output MOSFET. With 180 volts across Q2 and 300ma the power dissipation during the shorted condition in this test was 54 watts, way over the 33 watt limit. To set the hard current limit lower you can increase the value of R1.
In the development of the current limit and protection circuitry many, many parts let out their smoke before the regulator was close to "bullet proof".
Over all, this looks like a darn nice regulator. The output impedance is quite low without using large capacitors and the ripple rejection is suburb.
The high voltage version:
One of the things I wanted to use this regulator for is a new bench power supply. Wanting to have regulated outputs in the 500V+ range meant the input to the regulator was over the 500V limit of both Q2 and Q4. Replacing Q2 with any of the many 900V MOSFETs is easy. Coming up with a high voltage capable bias CCS was more interesting. The way I did this was to stack a third LND150 (Q7) on top of Q4 configured as a CCS set to deliver ~600ua. A string of zener diodes adding up to 500V is placed between the drain of Q4 and the output terminal of the regulator. When the input-output voltage is less than 500V the circuit works pretty much the same as the lower voltage version. The 3rd LND150 is "ON" and represents only a couple of K ohms in series with Q4. When the voltage goes over 500V the string of zeners turns on and clamps the voltage across the bias string to 500V above the regulator output voltage. Q7 now acts as a CCS and delivers ~600ua to the parallel combination of the Q4 bias string and the zener stack. This allows the input to Q7 to increase another 500V above the output voltage. In the test circuit Q4 was drawing 390ua with the remaining 210ua flowing through the zener stack.
Here is a screen shot from the Tek 576 curve tracer testing the high voltage bias regulator idea.
Below 500V the current drawn is the 390ua set by Q3. Above 500V the current increases to 620ua when Q7 becomes active. Just before 1100V you can see the current start to increase as Q7 starts to break down. During the whole sequence the current output from Q3 into the bias string remains constant.
Here is the schematic of the high voltage version.
The bias string has been tested. The circuit has not been built and tested with the FQAF11N90 so the value of R15 may need to be changed. This will get updated when I get some FQAF11N90. I have FQAF5N90's on hand but have found out that they are now considered obsolete so will configure the circuit for the non obsolete part.
I've been working on a board layout for the regulator. As far as I can tell the layout is finished but wont know for sure until the first board is populated and tested. The compensation value of R15 may need to be changed a small amount for the different stray capacitance of the new layout, and may need different values for different MOSFET types. Only testing will tell. It takes up 1/2 of a mini board so there will be 6 boards per order. After the board has been populated and tested and checks out good, I'll post the board layout so you can order your own boards from ExpressPCB.
To fit my desire to use of this regulator as a bench supply I've chosen film caps for C1 and C2. Reasons for using film caps are good high frequency performance and high voltage ratings are available. The caps chosen for this project are the Cornell Dubilier DME series miniature metallized polyester capacitors. They are relatively small for film caps. Values are available from Mouser up to 10uf at 250V and 2.2uf at 630V. Even though they are small for film caps they still take up a lot of real estate on the board. The board is setup so you can use multiple parallel caps for C2. C5, the electrolytic capacitor in the bias string is in a non critical location. It only needs to have a rating as it only sees the voltage across R14 + the turn on voltage of Q1.
The desire is to have a minimum of 16 volts or so across Q1. To do this we need 16 volts across R14. /.ma=53.ok. I think 5K is the next standard value. Have to go look at the test board to see why the resistor is giving 17.2 volts. Q3 may be leaking some extra current into the bias string. Have to check to see if Q3 has survived all destruction that has been going on...
Selectable resistor values:
R2 should be chosen to draw ~1ma at your chosen output voltage. This sets the minimum current the supply delivers. If you have lots of power available you can increase the current by using a shunt resistor on the output to ground. The output impedance of the regulator decreases as the current increases.
The bias current set by Q3 should be ~.ma. The bias current will vary with each batch of LND150 MOSFETs. You can change the value of R1 to compensate for the variations in MOSFET lots. I've used from 2G to 2.ok to get the bias current in the ball park.
R10 and R11 set the output voltage. The total resistance of R10 + R11 sets the output voltage. If you want only a small amount of adjustment make R10 be most of the resistance needed and R11 be a small portion. Just the opposite if you want a wide adjustment range. There are a couple of larger pads at R11 to facilitate connecting wires to an off board pot for panel mounting.
For high power or high voltage operation the board is setup to take either TO220 or T03P packages for Q2. I highly recommend the fully en capsulated parts for safety. The full encapsulation reduces the power disapation ratings but is made up by the ease of attaching to the heat sink without worry of arc-over and corona issues.
If you don't need to be able to handle over 500 volts on the input you can leave out Q7, R5, VR5, VR6, and VR7. Put a jumper wire across the drain and gate terminals of Q7 to connect the input voltage to Q4.
After the board tests out good I'll post the board layout file and a BOM.
Here's the layout occupying half of an ExpressPCB mini board. The board size is 2.5" high, 1.9"wide.
Ugly abused test mule...
