Anything wrong with this linear PSU design? (now selecting parts)

[David Hess]

Unless you are going for a design which requires no trimming, the absolute accuracy of the reference will not matter and even if it does, the absolute tolerance of the current shunt and output divider will create separate gain errors which will need to be trimmed anyway.  That implies at least two separate trims whether you use an ADC to read back the output voltage and output current or not.

An LM317 is not all that good as a reference with its 100ppm/C temperature coefficient.  A TL431 is better at 50ppm/C.  An LM336, LM385, or LM4040 is the cheapest common reference with a 20ppm/C typical temperature coefficient which is a good match if you use inexpensive 25ppm/C metal film resistors.

You might be able to make a better but inexpensive reference by spending time to grade 6.2 volt zener diodes and diode pairs for low temperature coefficient.  One of these days I want to try configuring a 723 regulator to operate with a constant elevated die temperature so it can be used as an ovenized reference.

Dithering or PWMing a DAC improves resolution but does not improve accuracy in the form of INL.

[Powermax]

Does the INL rating depend on the resolution? Like suppose I have 2 identical DACS, but one has 2x resolution. For them to have the same relitive acuracy (INL) would the higher resolution one have a proportanally higher INL?

Does high INL mean nonlinearlity?

When it comes to the R2R ladder, what is the maximum unbuffered voltage? It cant be exactly VCC because its still acting like a resistor divider.

[blackdog]

Hi,

Just some info...

Multiple users of this forum, saying that a compound transistor often oscillate...
That's not true if you use it right, directly mounted to the base resistor to the first transistor.
So do not install the resistor on the board and then go with a piece of wire to the base, Wrong!
And mount a small resistor also in the emittor wire of the first transistor, see the schematic .

Here you can see a earlier version of my design, with two compound power sections.
Extreem low noise reference section and good dynamic behavior, this schematic is not compleet! just to let you see how you aslso can do it.
DC Ri, how good are you building it?  in my test setup it was smaler than 0.0001 Ohm.
Noise onder full load? about 2uV RMS 80Khz bandwith.




This schematic is tested with the ADA4077-2 and the NE5532A, but know i have better Opamps, to get the Ri @ high frequenties, of lower,  better fase margin.
This is the Ri  @ high frequency, "F" is the dual compound transistor,  100Khz 14m Ohm!


Dynamic behavior
Just a sample of one of the measurements, NO! the the design is not ringing, This is one of my torture test for power supply's
The dynamic load is connected via two cables of 50 cm and in the dynamic load is mounted a MKP capacitor 6,8uF.
The ringing what is visible, is caused by the residual induction of the twisted connecting cables and the 6,8uF.
It is verry hard for the power supply to get the error puls low as in this picture.

Test with 2 compound transistors, NE5532A and a 9.5-Amp load puls.
If i remove the 6,8uF capacitor then there is no ringing, just about 5 a 8mV drop and in 15uSec flat line. look at the line in the middle of the picture,
no droop 10mV/Dic is not enough to sense the Ri of this measurement.



Opamps
A OPA140/2140, TLE2071 and my compound power transistor wil give you an nice phase margin and a low Ri of your power supply.
This design is from the Harrison Division of Hewlett Packart @ the end of 1950's... still going strong with modern components and some atention

My comments are intended only to show what is possible with an old design and new components.
Not to demonstrate a complete power supply design.
There are many ways to make a good linear power supply, this is just one of them.

Kind regarts
Blackdog

[Kleinstein]

INL is the integral nonlinearity. This is the maximum deviation from a straight line (sometimes with 0 fixed).
INL is often noted relative to resolution (e.g. in LSB) - so this number depends on the resolution.
For those cheap microchip DACs there are 8 / 10 and 12 Bit versions. The overall accuracy of the 12 Bit version tends to be slightly better than for the 10 Bit version, but the INL expressed in LSB is nearly 4 times as high.


For an R2R chain the maximum you can get is between  Vcc * (1-2^N) and the full Vcc. It depends on were you put the last end. So you get quite close to the full voltage. However with just normal digital outputs and resistors the accuracy is limited. More than 8 Bits is already hard.

[Dave]

Why are you using JFET schematic symbols for MOSFETs?

[blackdog]

Hi Dave,


Its a old schematic, i have found the MOSFET picture in a library know 

Kind regarts,
Bram

[Powermax]

Hi,

Just some info...

Multiple users of this forum, saying that a compound transistor often oscillate...
That's not true if you use it right, directly mounted to the base resistor to the first transistor.
So do not install the resistor on the board and then go with a piece of wire to the base, Wrong!
And mount a small resistor also in the emittor wire of the first transistor, see the schematic .

Here you can see a earlier version of my design, with two compound power sections.
Extreem low noise reference section and good dynamic behavior, this schematic is not compleet! just to let you see how you aslso can do it.
DC Ri, how good are you building it?  in my test setup it was smaler than 0.0001 Ohm.
Noise onder full load? about 2uV RMS 80Khz bandwith.




This schematic is tested with the ADA4077-2 and the NE5532A, but know i have better Opamps, to get the Ri @ high frequenties, of lower,  better fase margin.
This is the Ri  @ high frequency, "F" is the dual compound transistor,  100Khz 14m Ohm!


Dynamic behavior
Just a sample of one of the measurements, NO! the the design is not ringing, This is one of my torture test for power supply's
The dynamic load is connected via two cables of 50 cm and in the dynamic load is mounted a MKP capacitor 6,8uF.
The ringing what is visible, is caused by the residual induction of the twisted connecting cables and the 6,8uF.
It is verry hard for the power supply to get the error puls low as in this picture.

Test with 2 compound transistors, NE5532A and a 9.5-Amp load puls.
If i remove the 6,8uF capacitor then there is no ringing, just about 5 a 8mV drop and in 15uSec flat line. look at the line in the middle of the picture,
no droop 10mV/Dic is not enough to sense the Ri of this measurement.



Opamps
A OPA140/2140, TLE2071 and my compound power transistor wil give you an nice phase margin and a low Ri of your power supply.
This design is from the Harrison Division of Hewlett Packart @ the end of 1950's... still going strong with modern components and some atention

My comments are intended only to show what is possible with an old design and new components.
Not to demonstrate a complete power supply design.
There are many ways to make a good linear power supply, this is just one of them.

Kind regarts
Blackdog

Nice design! You really went all out on some parts of it! I am particularly interested in the switching preregulator section and how you implemented current limiting. It appears my original circuit had a basic version of what you are doing for current limiting, (more specificially, the method I asked about on in this instructables question: http://www.instructables.com/answers/Can-I-use-the-same-resistors-for-current-shunt-AND/ )

To me it looks like Q7 serves as a fast current limit while the op amps can take time to slew down and take control of the pass elements very simalar to my original design (without Q7 however.) Because you used a potentiometer to set the current limit there is no requirement to figure out how make it settable from a grounded source as with my design. My solution was to use a variable current DAC to replace the constant current sink and pot in my original design, although it turned out this would not have been a cost-effective solution.

Also what process did you use to calculate or otherwise figure out the compensation components around the op amps? I just did a guessing game with the LT1007's I choose.

Would you consider your design low noise? I read that potentiometers are noisy so I tried to avoid them when possible.

[blackdog]

Hi Powermax, 

Look at these pictures how good Q7 works.

No Q7, +18 Amps for 1 transistor (no current loop, switch off)



Q7 mounted, +7 Amps for 1 transistor (no current loop, switch off)



This will keep the power section better protected.

I realy love the LT1007 and the LT1037, nice low offset and low noise opamps, doe not use the LT1037 in a power supply, minimal gain is 5x.

Low Noise
Is 2 to 5uV RMS  80KHz bandwith to much for you!!!   

Kind regarts,
Bram

[Powermax]

Hi Powermax, 

Look at these pictures how good Q7 works.

No Q7, +18 Amps for 1 transistor (no current loop, switch off)



Q7 mounted, +7 Amps for 1 transistor (no current loop, switch off)



This will keep the power section better protected.

I realy love the LT1007 and the LT1037, nice low offset and low noise opamps, doe not use the LT1037 in a power supply, minimal gain is 5x.

Low Noise
Is 2 to 5uV RMS  80KHz bandwith to much for you!!!   

Kind regarts,
Bram

I plan to use the LT1007 because, well I have those lying in my junk bin. Strangely enough.And because they are in my LTspice component lists. I've never figured out or really bothered to add other component models. I am in need of adding the OP27, I have a bunch of those and would like to at least try them. I would also like to simulate how much worse a LM741 would perform.

[Powermax]

I went ahead and got my arduino to output 14 bits of resolution natively, and used a 4 stages of RC networks (22K and 0.47uF) for filtering, but my UT61E claims 1.92mV of AC voltage still present, my Fluke 117 is OL. (but I think the fluke is wrong, I suspect the AC mv range is broke, as both AC and DC mV range have identical readings. I suspect maybe a blown DC blocking capacitor. Maybe time for a teardown? Got it for like $25 paired with a cheap $2 garbage multimeter so no clue of it's origins.)

So after messing with this, I think I might be able to achieve 4 sig fig accuracy, although it might be hard. My filtering takes almost a second to settle, which is not too huge of a deal. At the half point (output set to 7.5V to allow the PWM to have the strongest primary frequency content) those 3 millivolts will be multiplied by 3 in the supply design, which is 10 mV, which will make a significant error on the LSB particularly at lower voltages. That's of course not even mentioning issues regarding powering the arduino with precisely 5V (or maybe 5.461333...V precisely so that 15000/2^14 outputs 5.000V), and whatever the effective INL would be, (my guess is that it would be related to jitter in the clock)

So a cheap DAC may be better, what do you all think? Whats a good DAC under $5 12 or preferably 14 bit?

I would not mind that inaccuracy too much if it was not periodic, audible, visible on a scope, etc. It just bothers me! So a proper 12 bit to 14 bit DAC may be a better option.

[blackdog]

Hi powermax,

Just one thing before i go to work, LT Spice is OK, but...
You learn more by building circuits!!!

All your attention should go to the loop controle, to get it stable and fast as posible.
You have to learn about wiring in a power supply, a good power supply is wired with the same attention as a HF  Amplifier 
Digital control... one of the last parts of de power supply!

My schematic is easy to convert to digital control.

Use your soldering iron more and les Spice, just a tip 

Kind regarts,
Bram

[VEGETA]

This DAC is 16-bit and cheap (3$): http://www.digikey.com/product-detail/en/maxim-integrated/MAX5217BGUA-/MAX5217BGUA--ND/3758774


Here is the Digikey search: http://www.digikey.com/products/en/integrated-circuits-ics/data-acquisition-digital-to-analog-converters-dac/701?k=&pkeyword=&pv153=5&pv153=11&pv153=22&pv153=19&FV=2640005%2C264000a%2C264000b%2C2640013%2C2640016%2C1f140000%2Cffe002bd&mnonly=0&newproducts=0&ColumnSort=1000011&page=1&stock=1&quantity=1&ptm=0&fid=0&pageSize=25

and this is an ADC with 16-bit: http://www.digikey.com/product-detail/en/microchip-technology/MCP3428-E-SL/MCP3428-E-SL-ND/2179226  << costs around 3.6$.

I guess they are fairly cheap really with very high precision of 16-bit. Is there any problem with these 2 parts?

[Kleinstein]

The DAC looks good for that price. Well good enough for a power supply. One might want a little filtering at the output, if low noise is important. But this would also need a low noise reference.

The ADC is good, especially for the low voltage at the shunt.

The LT1007/LT1037 are essentially the same as the OP27/OP37, just a different manufacturer. It is more like too good for a power supply - but chips are not that expensive any more. For the voltage loop one could still get away with the 741. The current loop might have some use for a low noise precision OP and the extra bandwidth.

With a voltage regulator tuned extremely fast (like the one shown by blackdog), parasitic inductance and coupling could get important. So at that level stability depends on the layout and inductance of the parts. Spice usually does not include the extra inductance in the models. At such a low impedance even a few 10 nH could make a difference. One might at least include the inductance of the shunt and maybe the power transistors in the simulations.

[ZeTeX]

This DAC is 16-bit and cheap (3$): http://www.digikey.com/product-detail/en/maxim-integrated/MAX5217BGUA-/MAX5217BGUA--ND/3758774


Here is the Digikey search: http://www.digikey.com/products/en/integrated-circuits-ics/data-acquisition-digital-to-analog-converters-dac/701?k=&pkeyword=&pv153=5&pv153=11&pv153=22&pv153=19&FV=2640005%2C264000a%2C264000b%2C2640013%2C2640016%2C1f140000%2Cffe002bd&mnonly=0&newproducts=0&ColumnSort=1000011&page=1&stock=1&quantity=1&ptm=0&fid=0&pageSize=25

and this is an ADC with 16-bit: http://www.digikey.com/product-detail/en/microchip-technology/MCP3428-E-SL/MCP3428-E-SL-ND/2179226  << costs around 3.6$.

I guess they are fairly cheap really with very high precision of 16-bit. Is there any problem with these 2 parts?

It has 0.65mm pitch, might be a little annoying IMO to solder.
MCP4922 is 12 bit and duel, and it is a little bit cheaper but more common:
http://www.digikey.com/products/en/integrated-circuits-ics/data-acquisition-digital-to-analog-converters-dac/701?mnonly=0&newproducts=0&ColumnSort=1000011&page=1&stock=0&pbfree=0&rohs=0&k=MCP4922&quantity=&ptm=0&fid=0&pageSize=25

[David Hess]

I plan to use the LT1007 because, well I have those lying in my junk bin. Strangely enough.And because they are in my LTspice component lists. I've never figured out or really bothered to add other component models. I am in need of adding the OP27, I have a bunch of those and would like to at least try them. I would also like to simulate how much worse a LM741 would perform.

The LT1007 is an improved OP27 so it should perform identically.

It would be fun to test a low bandwidth but high slew rate part for the error amplifiers like the LT1371 or various low bandwidth JFET input operational amplifiers.

[VEGETA]

zetex, 12 bit is not gonna help him achieve 10mV and 1mA minimum set value as I remember, so he needs 14 bit or more... So go for 16 bit. 0.5$ more and much more precision. Maybe I am wrong here and 12 bit is enough, I don't remember. Is there a device that has both dac and adc? that would be nice xD.

the usage of external dac and adc with this accuracy will make him use a very cheap MCU since it won't be doing much, so maybe 1$ PIC MCU is going to be enough.

[ZeTeX]

zetex, 12 bit is not gonna help him achieve 10mV and 1mA minimum set value as I remember, so he needs 14 bit or more... So go for 16 bit. 0.5$ more and much more precision. Maybe I am wrong here and 12 bit is enough, I don't remember. Is there a device that has both dac and adc? that would be nice xD.

the usage of external dac and adc with this accuracy will make him use a very cheap MCU since it won't be doing much, so maybe 1$ PIC MCU is going to be enough.

3000mA / 4096 is 0.7mA per 1 value so it will work, right?

Sent from my Nexus 5X using Tapatalk

[Kleinstein]

The MCP4722 and similar are 12 Bit resolution, but with limited accuracy. So the 12 Bit with not so good accuracy is not really giving you 0.7 mA steps. It is more like 0.2 - 1.5 mA steps depending on the code. So it is more like a 10 Bit ADC with 2 more bits than are nearly lost in the noise.
Similar the 16 Bit version of the max5217 is not a very accurate 16 bit DAC, but still good for true 14 bits plus two more bits of limited use.

If one accepts some extra time for adjustment, one could use the ADC to check the set point too and this way use it to correct a low accuracy but high resolution DAC (e.g. 2 10 Bit DACs combined, or 16 bits of R2R). This is quite some effort. For the current it might be a very real option to have two ranges and use two ranges for the shunt. The accuracy is limited at the very low end anyway, not just by the DAC but also the OP used and thermal EMFs: having a burden voltage of 200 mV at the upper end (more causes trouble with heat), means you are at 200 µV at 1/1000 of the range and thus in the range of thermal EMFs.

[Powermax]

Hi powermax,

Just one thing before i go to work, LT Spice is OK, but...
You learn more by building circuits!!!

All your attention should go to the loop controle, to get it stable and fast as posible.
You have to learn about wiring in a power supply, a good power supply is wired with the same attention as a HF  Amplifier 
Digital control... one of the last parts of de power supply!

My schematic is easy to convert to digital control.

Use your soldering iron more and les Spice, just a tip 

Kind regarts,
Bram

Thanks! Don't worry, I 100% agree! I have been doing that as well with every idea and revision. As I am afraid to kill my better op amps, I have prototyped my last design below. It has problems with current regulation at the very low end (it only goes down to a few mA's) most likely due to the really bad offset voltage of the UA741, and does not appear to have good line regulation at 100KHz, the frequency my 4A 24V switching supply operates at.

Check out the rats-nest attached below

[ZeTeX]

The MCP4722 and similar are 12 Bit resolution, but with limited accuracy. So the 12 Bit with not so good accuracy is not really giving you 0.7 mA steps. It is more like 0.2 - 1.5 mA steps depending on the code. So it is more like a 10 Bit ADC with 2 more bits than are nearly lost in the noise.
Similar the 16 Bit version of the max5217 is not a very accurate 16 bit DAC, but still good for true 14 bits plus two more bits of limited use.

Correct me if I'm wrong because I have never used DAC's before but lets say that I have a 12bit DAC with +-4INL max, and lets say that the +-4INL is at all code ranges (e.g from 0 to 4096)
so if my max current is 3000mA, it means about 0.723mA per 1 value, so if I set the value to 1000 I will get in theory without noise and such:
0INL = 732mA
-4LSB = 729mA
+4LSB = 734mA

Is it correct?

[Powermax]

I don't know how to make attachments appear/load at full size automatically (I'm still pretty new on this forum) so any advice is appreciated!

So I got a rough idea of what filter capacitors I'll be using, 10 of these for 666 cents.  :
They have damn good ESR, good ripple current, 105^oC rated, panasonic, and best of all 66 cents a piece!
http://www.mouser.com/Search/ProductDetail.aspx?R=EEU-FM1V122Lvirtualkey66720000virtualkey667-EEU-FM1V122L


For the voltage reference, I like the LM336, looks like a jellybean part, I think someone recommended it, and it's one of the cheapest! (yes, I am a penny pincher, as strangely all the parts together cost like $40! I think mouser is rigged  )
http://www.mouser.com/Search/ProductDetail.aspx?R=LM336BDG4-2-5virtualkey59500000virtualkey595-LM336BDG4-2-5
But I don't like the package. What a waste of 6 out of 8 pins!  I found TO92 versions of the LM336 but why would the package affect the temp-co? The one above is rated 10PPM/^oC while the TO-92 version rated 34PPC/^oC (?)

I did come across this bad-ass super-duper voltage reference, 0.05% initial accuracy, 3PPM/^oC 10V reference but the 10V part is kind of a deal-breaker. I would need a dodgy resistor divider to bring that down to a more manageable level and then I basically throw those ratings out the window. It's too good and unsuitable. But I am certainly keeping an eye on it for future designs.
http://www.mouser.com/ProductDetail/Maxim-Integrated/MAX6043CAUT10TG16/?qs=sGAEpiMZZMuBck1X%252b7j9fADMRbLaMGMSpKa%2f1OuhCQ8%3d


I am still going for this rotary encoder, as it has 30 indents, more than the 20 my encoder has at the moment as well as a pushable switch. Unlike many of the other offerings, this one has half the resolution of the indents, so you get 2 pulses per turn. Why is this? Is the purpose of a full cycle per indent to allow programming the trigger to be easier? Or is the full cycle per indent common so that there is plenty of redundancy in case a pulse or 2 is missed?
http://www.mouser.com/Search/ProductDetail.aspx?R=EC12D1524403virtualkey68800000virtualkey688-EC12D1524403


For the current shunt, I still plan to use this glorified piece of wire, to further improve temperature drift, I will series 2 of them together (thus spreading the heat) That should also allow slightly improved regulation, as a larger voltage drop increases sensitivity of the error amp.http://www.mouser.com/Search/ProductDetail.aspx?R=OAR3R100JLFvirtualkey66200000virtualkey66-OAR3R100JLF

Or do you think it would be better to go with one of these instead? Same one, but a flattened-out version. It should have slightly better heat-dissipation.
http://www.mouser.com/ProductDetail/Bourns/PWR4412-2SCR1000F/?qs=sGAEpiMZZMtlleCFQhR%2fzUMu1%2fA3jOoDI9fAYn4YTO8%3d

I'm not entirely sure if I want to bother with selectable shunts, I'll throw together a perf-board semi-permanent circuit using only the parts I have to judge performance at the low end, using cheap pots instead of arduino for this. (also because I am in desprete need of any sort of decent lab supply at the moment.)


Also, what about connectors? What kind of connector should I add to the PCB layout? I don't really know what to look for in Mouser in this regard.

[Kleinstein]

For the voltage reference, it depends a little on the circuit / DAC, which voltage you need. With just a pot to set the voltage a higher voltage might be a better option.
Other relative good cheap ones are TL431 (cheap and adjustable) and LM329 (relatively low noise, 7,x V).
With just a pot to set the voltage, there is no need for a really low TC.

The filter caps don't need to be low ESR ones. I am not sure you really need 10000 µF.

It is a good idea to start with a simple version, without digital part.

[Powermax]

For the voltage reference, it depends a little on the circuit / DAC, which voltage you need. With just a pot to set the voltage a higher voltage might be a better option.
Other relative good cheap ones are TL431 (cheap and adjustable) and LM329 (relatively low noise, 7,x V).
With just a pot to set the voltage, there is no need for a really low TC.

The filter caps don't need to be low ESR ones. I am not sure you really need 10000 µF.

It is a good idea to start with a simple version, without digital part.

It's for the input, I am getting 10 mostly because of economies of scale. (10x quantity costs less per unit) and I know I don't need low ESR ones, just seems to be a really good value. I simulated the amount of ripple that 5A will see from a 80Hz transformer and I am seeing spikes of current reaching close to 20A. 10 of these can safely handle 36A.

[David Hess]

For the voltage reference, I like the LM336, looks like a jellybean part, I think someone recommended it, and it's one of the cheapest! (yes, I am a penny pincher, as strangely all the parts together cost like $40! I think mouser is rigged  )
http://www.mouser.com/Search/ProductDetail.aspx?R=LM336BDG4-2-5virtualkey59500000virtualkey595-LM336BDG4-2-5
But I don't like the package. What a waste of 6 out of 8 pins!  I found TO92 versions of the LM336 but why would the package affect the temp-co? The one above is rated 10PPM/^oC while the TO-92 version rated 34PPC/^oC (?)

Check the datasheets very carefully; Mouser tends to be inconsistent when listing some specifications like temperature coefficient.  There is no difference between the same grade of SOIC and TO-92 part in this case.

[Powermax]

For the voltage reference, I like the LM336, looks like a jellybean part, I think someone recommended it, and it's one of the cheapest! (yes, I am a penny pincher, as strangely all the parts together cost like $40! I think mouser is rigged  )
http://www.mouser.com/Search/ProductDetail.aspx?R=LM336BDG4-2-5virtualkey59500000virtualkey595-LM336BDG4-2-5
But I don't like the package. What a waste of 6 out of 8 pins!  I found TO92 versions of the LM336 but why would the package affect the temp-co? The one above is rated 10PPM/^oC while the TO-92 version rated 34PPC/^oC (?)

Check the datasheets very carefully; Mouser tends to be inconsistent when listing some specifications like temperature coefficient.  There is no difference between the same grade of SOIC and TO-92 part in this case.

ERRR 😠. Mouser, why!? Whatevs, thanks for letting me know.