Lab Power Supply Design Part 5

[Bored@Work]

Finally a schematic.

Dave,

I see you share the ISP and the DAC SPI (and plan to put the ADC on the same ISP bus, too). This can in principle be done. However, a lot of AVR ISP programmers are a little bit allergic when the ISP interface is shared. I recommend you have a look at Atmel's application note AVR042, section 4.1.1 http://www.atmel.com/atmel/acrobat/doc2521.pdf for the official way to handle this.

When all else fails, then a jumper to disconnect SCK from the rest of the bus, and only leave it connected to ISP helps, too.

Second, the reset pin should be handled differently, esp. if you are in a noisy environment. Most people don't try their luck and add a resistor and cap. Similar to what is also given in AVR042. However, most people skip the diode recommended in AVR042.

Third, and this might be a real issue, you have connected a LED to the reset pin. Presumably you want to drive the LED via PC6. However, that is not possible. AVR reset pins can either be used as a reset pins, or as a GPIO pin (here PC6). Which function the pin has can be set by a fuse, RSTDISBL.

But there is a catch! If you program the RSTDISBL fuse you no longer have the reset pin function. However, the reset pin function is needed for ISP programming. In other words, programming the RSTDISBL fuse is the last thing you can do via ISP. After that you don't have ISP any more. A workaround would be to program a bootloader via ISP into the AVR, then program RSTDISBL via ISP, and from that moment on do all further firmware updates via the bootloader. A bootloader, however, can not change (program or clear) the RSTDISBL fuse.

The other programming method for this AVR, high voltage serial parallel programming, HVSPP, is just a theoretical option. It would allow to clear RSTDISBL, thus again enabling ISP. But that method requires to wire up something like 18 pins of the AVR to the programmer and use a 12V programming impulse at some pins. This is nearly impossible while the AVR is in the circuit.

I would recommend you free the reset pin, too and use it exclusively for reset.

[shebu18]

What if i need more then 4A? (beside 3-4 LT's or LM's) The whole current measurement must be changed, right?

[EEVblog]

Thanks BAW, the Reset pin had an I/O marked on it, so I assumed it could be used dual purpose without a nasty trap like that. I didn't bother to delve into the details of it at the time.

Dave.

[EEVblog]

To protect the RESET line further from noise, it is an advantage to connect a
capacitor from the RESET pin to ground. This is not directly required since the AVR
internally have a low-pass filter to eliminate spikes and noise that could cause reset.
Applying an extra capacitor is thus an additional protection.

Huh 
What kind of noise level would you need to overcome a 10K pull-up 

Dave.

[Psi]

Could use a relay instead of the fet to bypass the ucurrent section and eliminate the fet rson drop.

However i don't really like things that click all the time when moving the controls.
So i'd probably stick with the fet for now, the vdrop is so small its probably the best option.

Replacing it with a relay is simple enough if its ever needed

[shebu18]

Is it ok if we use 10k 1% instead of 10k 0.1%?

[EEVblog]

I see you share the ISP and the DAC SPI (and plan to put the ADC on the same ISP bus, too). This can in principle be done. However, a lot of AVR ISP programmers are a little bit allergic when the ISP interface is shared. I recommend you have a look at Atmel's application note AVR042, section 4.1.1 http://www.atmel.com/atmel/acrobat/doc2521.pdf for the official way to handle this.

I only have input pins on other chips (DAC-D, DAC-CK, ADC-CS) connected to the SPI lines, so I can't see how they can interfere with the ISP?
So it's only one extra CMOS gate input loading on each line, surely not an issue?

Dave.

[Psi]

Is it ok if we use 10k 1% instead of 10k 0.1%?

You can get issues with the opamps if their resistor values don't match since they control the gain.
Less accurate resistors mean less accurate gain, so you introduce errors.  (eg if it's supposed to be a gain of 5 you might end up with 5.05)
But yes, it will work with 1%, just won't be as accurate

If you have lots of 1% resistors you could go through them and find ones with matching values to use on the opamps.
That would be a good solution to your problem if you have lots of 1% resistors you want to use.

[EEVblog]

Is it ok if we use 10k 1% instead of 10k 0.1%?

Sure.
My aim with 0.1% is so that hopefully you won't have to trim any potential errors in software and there is as little absolute tolerance error as possible without fuss, i.e. their tolerance is better than the voltage reference used.
As PSI said, you can hand match them if you like, or use 1% and be happy with any potential absolute error.

Dave.

[IanB]

Could use a relay instead of the fet to bypass the ucurrent section and eliminate the fet rson drop.

However i don't really like things that click all the time when moving the controls.
So i'd probably stick with the fet for now, the vdrop is so small its probably the best option.

Replacing it with a relay is simple enough if its ever needed

What kind of resistance might you expect across a typical relay contact? I have measured a high current mechanical switch at about 1.5 milliohms and an ordinary metal face-to-face contact point at about 5 milliohms. Would the relay need to have gold plated contacts to achieve a good low contact resistance?

[Psi]

Could use a relay instead of the fet to bypass the ucurrent section and eliminate the fet rson drop.

However i don't really like things that click all the time when moving the controls.
So i'd probably stick with the fet for now, the vdrop is so small its probably the best option.

Replacing it with a relay is simple enough if its ever needed

What kind of resistance might you expect across a typical relay contact? I have measured a high current mechanical switch at about 1.5 milliohms and an ordinary metal face-to-face contact point at about 5 milliohms. Would the relay need to have gold plated contacts to achieve a good low contact resistance?

True

[firewalker]

Really, really good blog Dave.

Is there a link for the schematics. Like a pdf?

Alexander.

[EEVblog]

Is there a link for the schematics. Like a pdf?

In the Youtube description:
http://eevblog.com/uploads/uSupplyBenchRevA.pdf

Dave.

[firewalker]

Oh... 

Alexander.

[EEVblog]

Oh... 

No one ever reads those! 

Dave.

[Armin_Balija]

And just when I went ahead and ordered the IC's for your older videos! I guess I've got my work cut out for me.. I hope I can put it all together myself.. I guess everyone else is comfortable with this level of detail but it's a bit intimidating to me. I'll try it anyway though.

Again, thanks Dave, we all appreciate the hard work you put into this.

[shebu18]

Dave, can you put a link to a rotary encoder you have used? I find only explications and examples of them.

Thanks.

LE: Thank you!

[EEVblog]

Dave, can you put a link to a rotary encoder you have used? I find only explications and examples of them.

http://search.digikey.com/us/en/products/EN12-VN20AF20/987-1197-ND/2408775

Dave.

[firewalker]

Dave just noticed you was going to use PB6,PB7 for SPI communication with th ADC. Did you implemented a soft SPI?

Alexander.

[metalphreak]

I was considering making my own uCurrent after the latest batch sold out so fast, but integrating it into a power supply is a brilliant idea

I too came across all the microchip ADC/DACs due to their low prices, but I was disappointed to see no quad input ADCs from them with I2C Using I2C is great from a modularity point of view. You can very easily expand to a dual/triple output supply without needing more microcontroller I/Os, and the code change is trivial (simple address change - no overlapping pins).

The Analog Devices AD7998 seems perfect for a dual output supply. Once setup you can read all 8 channels with one address/write byte + command byte, followed by one address/read byte and 16 read bytes (it just samples ch1, ch2, etc in order). For each output, I would sample current, uCurrent, output voltage BEFORE an output enable mosfet, and Vsense (either seperate terminal, or on the actual output connector, or switchable...). The AD7994 is a 4ch version.

Input voltage could simply be measured using the microcontrollers ADC as high accuracy isn't necessary.

For a i2c DAC I was going to go with the microchip MCP4728. Has an internal 2.048v Vref with a selectable x1 or x2 gain (handy for extended voltage ranges) but its accuracy is only +/- 2% with 45ppm/C...

For a micro I was going to go with a PIC18F4550 for a few reasons.

1) I've used them in the past
2) Hardware i2c, serial, and USB2.0 (all usable at once I believe)
3) Microchip USB stack and bootloader make it incredibly easy to do firmware updates over usb, or emulate a COM/Serial port over USB. Their framework and examples make it incredibly easy to get USB comms working.
4) Plenty of I/O leftover for "stuff"

Also comes in a DIP40 package although some of the other parts i've chose are SMD parts only. I'll be going for TQFP package myself since I've got a few of them leftover from a previous project.

I figured with I2C, you could build modular boards that have all the DAC/ADC/Vreg stuff, and a single control board. Thanks to Dave making his design open source other's can do things a little bit different

Also, for the main power supply, would it be reasonable to use something like an 15-18V switchmode supply? I want something I can mount internally instead of using an external jack, and AC transformers are so expensive!

Any thoughts, comments, glaringly obvious silly things? 

[Nick Gammon]

Dave, one quick suggestion as you are running out of pins, if you have to use the reset pin to drive an LED ...

Your four switches (PB1 to PB4) ... one way of doing that without four pins is to set them up, via resistors as voltage dividers, to a single analog pin (you are using an analog pin anyway). The simple method (same value resistors) could give ambiguous results if multiples switches are pressed at once (I don't know if this would matter to you). With different value resistors (eg. powers of 2) you could deduce from the resulting analog level which combinations must have been pressed. So that frees up three pins.

[firewalker]

Charlieplexing is also a potential.

Alexander.

[Rutger]

Hi, this might be a wacky idea, but how difficult would it to add a function generator option that turns the dc output specs into an ac output with the same specs (like voltage 5 VDC to 5 V peak-to-peak at a set current).

Rutger

[alm]

( Using I2C is great from a modularity point of view. You can very easily expand to a dual/triple output supply without needing more microcontroller I/Os, and the code change is trivial (simple address change - no overlapping pins).

The Analog Devices AD7998 seems perfect for a dual output supply. Once setup you can read all 8 channels with one address/write byte + command byte, followed by one address/read byte and 16 read bytes (it just samples ch1, ch2, etc in order). For each output, I would sample current, uCurrent, output voltage BEFORE an output enable mosfet, and Vsense (either seperate terminal, or on the actual output connector, or switchable...). The AD7994 is a 4ch version.

One issue with this is that all outputs will share a common ground. In typical lab supplies, all outputs are floating. Isolating an I²C interface is somewhat complex. NXP has an appnote which basically involves two I² transceivers with some optos in between. There are also some specialized I²C RF isolators by AD and others.

[JimmyM]

Heck just go with a mega324! Scads of pins! Just bring out the unused pins to PTH pads. Voila!