So, I got to the " Order Status: Order received " page just fine (minus render preview, but the gerbers looked fine in the few viewers I tried). Now when I try to login again, I'm getting " error::Incorrect email or password. "
With "Forgot my password", I haven't received anything yet (> 24h, not spam-filtered); With "Register", I get " Email is already in use " so I'm probably halfway registered or something, but nothing I can do about it now !
Regarding eliminating the huge 4066: I don't think it's big enough an issue to start a whole thread. Of course the simplest that comes to mind, like rsdio mentioned, is just bunching all the pull-up resistors to a common, selectable V_PU like the first drawing. Very cheap : replaces the 4066 with an extra PNP+diode for enabling V_EXT.
Problem : now, if the all pullups are disabled (Q1-3 not conducting), the data/clock lines are all connected together through various resistors. If you're running a TX/RX UART with strong drivers at both ends; probably no big deal. But if RX is driving +5V while TX is driving 0V, then the other lines are at 2.5V ! That may be a problem either for the PIC pins or other circuits connected to the unused lines, I'm really not sure of the implications.
Solution: we can add diodes in series with each pull-up resistor. That's a few extra parts (2x BAT54A this time, common-anode diodes in one SOT23) but solves the inter-connection problem. The additionnal "diode drop" shouldn't be a problem : if a 2k pull-up resistor is sourcing 1mA (unlikely), it already drops 2V so the additional ~0.3V of a schottky is small.
Is it worth it ? extra parts = 1xPNP, 1x resistor, 2x bat54; but removed 1x4066. Pros: smaller footprint, nearly the same cost, cheaper to repair (but how would it break ?), expandable for >4 bus cons: extra diode drop (small con !), PCB change required, more parts
Tough call ! For my own stuff I would go with the discrete solution because I don't like, and don't stock 4066 ICs. For a production board... I think it would be cheaper to go with the IC. [EDIT : I just remembered another problem with the discrete solution : if VEXT can be larger than 5V, it's necessary to drive its transistor with an open-collector output from the PIC, but only if the PIC pins aren't protected by diodes to VCC. If so, it needs an NPN transistor as well !! So unfortunately another point in favor of the 4066]
[quote author="Colfaxmingo"] Is this project alive? It sounds pretty awesome. I haven't read anything new about it. I would be sad if it died. [/quote]
It's still just at the "idea" stage. Personnally I don't badly need one of these, and I'm a bit short on time. However, there needs to be practical experimentation. There were many cases of "this should be built to verify it works" that still remain to be tested... and last time I checked there was still a lack of concrete engineering, i.e. knowledgable design. I drew up a lot of schematics (not all good), and rsdio + another member seemed to know what they were talking about but none of us logged much serious design.
If you're handy, I would say there's enough material in the old posts to let you build something that will most probably work. Not necesarily safe / reliable / stable / performant though.
I would prefer a scheme where the input impedance remains constant, especially since our input impedance is unlikely to be so high that it doesn't matter at all (>=10Mohm). A switchable resistor divider would work too, and allow us to skip the PGA, either way would be fine for me. Whatever proves the easiest/cheapest/best (accuracy of at least 1% would probably be a requirement, even a crap DMM or panel meter does <=0.5%).
Agreed. How much impedance would be enough ? I figure 1M would be quite adequate if we can manage - this is comparable to many passive oscilloscope probes.
[quote author="tayken"] OK, I'll look up for something to limit the damage (something like blowing up a fuse or burning up the MUX only) and I'll try to design a way to simplify it as much as possible. But to measure negative voltages, we need a negative voltage supply on the board for the analog mux, otherwise we have to do the signal conditioning for all 8 channels. [/quote]
A negative rail may not be essential - one possibility (yes, op-amps again...) is to wire an inverting stage symmetrical around ground like the first ASCII schem here: http://electronics.stackexchange.com/qu ... with-a-adc Not sure about the risk of latch-up in the op-amp, however... a rail-to-rail input IC is definitely necessary.
Quote
The USB stack is in C as far as I know but aren't we going to use an FTDI IC or something else to isolate the signals? Also I think PPS is essential for connecting different devices with different protocols to an "expansion header" such as the one in Web Platform. My idea is to route at least one ADC channel, I2C pins if they are not used for some other thing on the board (or we can route them even if they are used), some other pins of the PIC and of course supply and ground voltages.
As I see it, using an FTDI vs PIC wouldn't make an isolated PSU more simple or complex, just different. Here are some possibilities :
I would personally go for the last of these; having a PIC to arbitrate the opto-isolated bus is better IMO. We can also program it to send a "keepalive" signal to the slave PICs : in case of a USB comm failure, disable the outputs or take other safe actions... As for PPS, well it depends what the project is meant to be. I didn't see this as another "swiss-army" device like a BP, but rather as a configurable power source... In any case all reasonably-sized PICs have many extra ADC channels (internally multiplexable), so PPS would just allow you to use one pin instead of its neighbour... As for I2C or other data busses, I don't quite see where you would want this - perhaps on the isolated side ? but then the datastream adds traffic on the opto-bus, and calls for a more elaborated protocol. Is it worth stepping up to a 24F then? I think specifying a 44 / 64-pin QFP device for this kind of project is overkill, and would actually make the PCB more complicated, even with PPS !
I don't think converting designs like this to DAC input is very risky, the pots are just voltage sources.
Hmm, that's not how I understood the E3620 circuit... It seemed rather like a resistive divider used for feedback? (The voltage control pot is between +5V(reg) and -Output) Short of using a digi-pot I don't see other easy conversions. Anyway.
[quote author="tayken"] For PWm generation, I was thinking of using a 24F series PIC and as far as I can remember they have multiple CCP modules (correct me if I'm wrong). .... I'll go for a PIC24F either way as peripheral pin select feature really helps with routing and may add "hackability" to the project in the future if we route some unused pins to a header (such as the control panel addition I have in mind). Also I'm more familiar with them and really liked them wrt 8-bit ones I used before. [/quote]
Check out arhi's previous posts, as he pointed out there are now a nice variety of 16F and 18F PICs with two or more CCP modules. I really don't think PPS is essential (PWM outputs will be PWM outputs, ADCs will stay ADCs), and the routing shouldn't be exceptionally hard. But since I probably won't be doing much coding on this project it's not my call... Is the USB stack programmed in asm or C ? 24F asm would certainly need more work to port down to 18F, but C should be pretty manageable. How people can program <16-bit CPUs in C is beyond me, it feels sooo clunky p-))
[quote author="tayken"] I think the only thing left is to finalize the linear regulator part. [/quote] Hehe I like the optimism
[quote author="alm"] I expect me and fenugrec to converge on a design soon, no major disagreements as far as I see. I would prefer to breadboard this supply (analog part, can use lab supply instead of DAC and DMM instead of ADC) in both Spice and real hardware and run some tests (eg. the National app note) before committing to PCB. I probably have most of the parts in stock, unless we decide on more exotic pass transistors or op-amps. [/quote] We'll get there yet ! To be honest I don't have any religious attachment to this or that layout... If we need bipolar supplies, so be it. I just find that a totally single-ended circuit would be really elegant, and maybe easier to float. But in any case, it's definitely not time for a PCB yet.
ADCs : We want to monitor at least output V and I, but it would be vary nice to also take a reading of the input voltage, to compute a maximum power dissipation: if the user plugs in 25VDC, the output current should be capped in relation to the max transistor dissipation... Something like Imax = Pdis / (Vin - Vset) , not counting the sense resistor. Without this, the regulator can hardly be considered "short-circuit proof" - we can count on people not paying attention and asking too much of the reg. Or just assuming it's "smart enough" to not self-destruct.
Re ADC range selection : what about a resistive divider with MOSFET-selectable legs ? Top resistor = 10k, a few 10k bottom resistors which can be switched in through software control. Add diodes between the ADC pin and VDD, VSS to clip the range to safe levels... I figure the added Rds resistance should be negligible with 1% resistors , two or three magnitudes higher in value.
[quote author="arhi"] [quote author="fenugrec"] Only trouble is that we would need 2 PWM channels, very few PICs have two independant PWM modules. [/quote]
I'm not sure I understand this, sorry for slowing discussion down but you would need 2 PWM channels that have different duty cycle, they can have same frequency right? any pic with few pwm channels can do this, they can't have different frequency but they can have different dc .. [/quote]
I take that back, I see microchip has some new devices that have two CCP modules and more. I was thinking of older/smaller devices like 16F628, 16F88, 12F683, etc. We're fine, then p-)
[quote author="alm"] About the R2, I think a transistor/JFET constant current source would be better, since HP used a resistor to a constant voltage above Ve(Q1). No need for great accuracy, it just should be better than between 14V/R2 (almost zero output voltage) and 2V/R2 or so (max output voltage). Something like -50%/+100% should be fine. [/quote] That part definitely needs improvement... I like the JFET idea. As it turns out, all transistor models I've seen specify ludicrous beta factors, and I'm not aware of any re-adjustment in function of Ic... What I drew was definitely more along the lines of a proof of concept - I wanted to make sure it could theoretically work.
Quote
I have some doubts about your design, even though you appear to know more about this: 1. How's the stability of Q3 without emitter resistor? Isn't it very sensitive to tiny Vbe and temperature variations, making it hard to keep stable? 2. The HP design of Q1/Q6 appears be temperature compensated, since the Vb(Q6) is at about the same level as Ve(Q1). Your design isn't, which may make the job of the op-amps harder. 3. Why isn't this topology used in other lab supplies? All commercial designs I'm aware of use a negative and positive voltage relative to the output voltage. Even in a commercial design, this has to add some extra costs, why do they use it if they could easily omit it? There has to be some downside, it's not like an LM324 is a recent invention. 4. In general, the major changes compared to the original design makes me somewhat uncomfortable without doing extensive analysis and testing. Hence my preference for something closer to the HP design if we can (doesn't add much costs with the symmetrical design).
All excellent points, sorry I can't answer most of them. One thing I might say is that the E3620 was designed for potentiometer panel controls. I saw no other simple way to convert it to DC control... Last time I checked HP service manuals, I didn't find a linear lab supply that could be digitally controlled with i.e. GPIB. The digitally-controlled units I found were SMPS designs.
Quote
If you're convinced that it will work fine, I'm willing to believe that. But if I would have to build it, I would go for a more conservative design.
Oh I'm far from convinced it'll work well, and I certainly don't want to sound over-confident. I do feel it's worth actually prototyping one of these, but I really can't say if it's going to turn out to be a waste of time. On a side note: I had built a cheezy current regulator similar to this layout, with an LM358 driving a transistor in a single-supply circuit. It was far from perfect, and initially I had some oscillation problems, but after a bit of tweaking it became quite usable. I actually charge some battery packs with it...
I would ask what you would consider to be a more conservative design ? Some problems I see to using i.e. the E3620 (or any other) circuit is : component availability, unkown design decisions and different specifications.
Quote
[...] Designing/manufacturing an AC wall wart is much harder to screw up.
(Sorry, I trimmed the part mentioning using two wallwarts to get symmetrical / floating outputs) They do make center-tapped AC wall warts. I'm not sure how internet-available they would be however (I bought a bunch at a local electronics dollarstore)
Quote
Quote
whether to float the output needs to be decided.
I don't think extra parts will be too bad, some opto's and an extra PIC/FT232. Supply for the grounded part can be USB.
Ok, as I said I'm all for it - but the guy making the boards gets to decide, in the end.
Quote
The reason for considering an FT232 was that we don't really need more than an USB interface on the grounded side. We could isolate the UART connection and connect the PIC to lab supply ground. I've used the small MCU per output on the isolated side in the past, because an SPI/I2C ADC is not much cheaper, and the MCU can also control the DAC's and switch some other stuff (I had relays for tap switching). With multiple channels, we would need more intelligence on the grounded side, since we would need to control two outputs with one UART. Using an FTDI device with dual UART is ugly in my opinion, and probably even more expensive.
Indeed. FT232 is ok, but I'm partial for a USB PIC speaking to the output channel(s) through an opto-isolated bus.
Quote
[quote author="tayken"] - For the initial release we might go for PWM + caps instead of a seperate ADC IC. Lets see its performance, if it is really poor, then we can go for the ADC IC in the real design.
I would offer both options in that case, so we can compare (populate one board with DAC and one with PWM, or switch with a jumper). [/quote] Only trouble is that we would need 2 PWM channels, very few PICs have two independant PWM modules. Of course we can generate the PWM in software, but if we need to manage ADC conversions & communications at the same time, it might not be worth the trouble.
[quote author="arhi"] I would like to see 3-5A more then 1-2A .. but .. maybe it would be cool to make the project out of 2 segments where one segment does the read/write and the other segment does the "power" stuff so one can easily change the power segment to get 5A or 10A if needed [/quote] I'm against this for a v1 board. Modularity isn't useful to everyone. Plus, having 2 PCBs make things more complicated for fab'ing. Also: 12V @ 5A makes a big wall wart, and a hefty linear regulator + heatsink !
[quote author="ian"] My preference would probably be an FT232 chip on the PC side, and whatever PIC on the power supply side.
Floating outputs are great, but if it adds a lot of trickery to isolate it could be a further drag on getting a real v1 out.
Quote
2- op-amps U1, U2, and their compensation network (absent on the SPICE schem I drew), are critical.
Could anyone following this thread please contribute a little more about this. A discussion or napkin sketch could be enough to help me work it into the schematic. [/quote]
Re FT232 : do we have a usable PIC18-compatible USB stack ? I.e. can the BP4 stack be adapted to this project ? The same command-response structure could be used, or we could implement our own vendor requests/replies , with interrupt transfers. As I mentioned devices like the 18F13K* are quite nice, and a lot cheaper than an FT232. Plus they have an internal 1.024 / 2.048 / 4.096V reference for the ADC !
I think we should really keep floating outputs for a v2... Besides v1 will still be quite usable for a lot of people in a lot of typical situations.
As for the compensation networks, this is really at the limit of my knowledge, and will ultimately depend on the pass transistor, the drive trans, the output capacitor, and the op-amp type. If you recall the HP schematic, the op-amps all had small capacitors around them. In my SPICE schem I left them out except for a ~100p cap near the current-limiter. It slowed the response of the limiter, but consequently also added a bit of overshoot (the limiter set to 1A would still allow a short period of >1A output before cutting in).
Hi guys, sorry I've been away from this thread for so long... I'm still interested though.
alm said:
Quote
I think I found an error in Fen's schematic (he didn't present it as a finished product), Q1 was a Darlington in the Agilent design (not indicated in the schematic), and I don't see how a regular MJE3055 will have enough gain for the base current for R2 to be enough. I'm also not a fan of the 'constant current source' provided by (Vin-Vb(Q1))/R2, which will vary with output voltage.
Thanks for looking at the design in detail. I was afraid nobody would go so far! Indeed, I didn't "select" the pass transistor, I just took one for which I had a SPICE model (I forgot HP used a darlington). You're right, it probably only has a hFE of 15-20 at Ic=1A. You're also right that the bias resistor is probably inadequate in a large number of conditions.
Just to clarify some of the features of the linear stage I posted some time ago ( http://dangerousprototypes.com/forum/in ... 87#msg9187 ) : 1- it will regulate down to 0 volts, without requiring a negative supply, if the op-amps are well chosen (LM358 should work, I can't say how well though) 2- op-amps U1, U2, and their compensation network (absent on the SPICE schem I drew), are critical. To repeat myself once more : it's farily easy to mess up the component choice, and have a supply that sorta-works but will be unstable. Unstable to the point that you can get substantial oscillation on the output. Or just terrible transient load change response. It's a great topology, and very flexible, but more delicate than a single LM317.
Misc thoughts: - having two symmetrical regulators is nice but a bit restrictive for those times when you would really need a 12V supply and a 5V supply. Sure you can set one supply to 7V and the other to -5V, but it's still not great. The best is two floating regulator outputs.
- for an initial release, I think it should be most basic. Straight DC-input with polarity protection is most important, and the .3 or 0.4V drop caused by a series diode is insignificant. Or use a "crowbar" type circuit with an input fuse (useful in any case) with a diode to ground.
- I don't like AC-input with voltage-doublers: how do you accomodate the unknown range of voltages that a user might use ? Supplying 1A (or 1A + 1A) will put a lot of strain on capacitors (high ripple current). Also, if you don't want huge 60Hz ripple you want large capacitorvalues, which are expensive and physically large.
-whether to float the output needs to be decided. One thing is for sure : the very desirable flexibility automatically means it will be more complex & more expensive. Is it worth it ? In my experience, it's a nice feature, certainly worth a few extra $. I'll also vaguely quote someone on this thread (I think it was rsdio) concerning the basic utility of a lab supply :
Quote
If I need to use a lab supply, it's either because A) I don't have a supply built yet (unknown requirements) or B) I'm troubleshooting and suspecting the regulator
-I really see no need for a dsPIC. I think almost everyone now agrees that we're not going to kludge up a software feedback loop, therefore the requirements for the regulating uC would be: *a bunch of ADC channels for metering the output (10-bit is plenty enough, this is not a 5-digit calibrated multimeter); *either a couple DAC channels or PWM outputs that are capacitor-filtered to give variable DC. I would tend to prefer a DAC solution here since filtered PWM inherently has ripple, and possibly a slow response time. Now here's the catch. The most basic USB-enabled PIC is about 3$. Mchip lists the new 18F13K50 @ 1.30$ in large quantities. But at that price you don't get a DAC! If you want USB + DAC, I actually haven't found any PIC that had more than 2 DAC channels. In any case it would be expensive, like 6-10$ , and wouldn't easily be opto-isolated. That leads me to the next point :
-if a DAC chip is used in addition to the USB-speaking chip (PIC or FT232), it becomes easy(er) to add opto-isolation and have fully floatable outputs. But then you want to meter the output, so you also need ADCs on the isolated side ! At this point it would then sound justified to add a small, very basic PIC on the isolated side, which would serve only to monitor and steer the regulator. I like this idea, but one thing to consider is how much cost overhead it adds if we need to program 2 or 3 MCUs. Ian ?
To summarize : floating outputs are nice to have, but significantly more complex and expensive. I would opt for a single-channel, single-output version 01 with modest, but solid features. I'm attaching a symbolic diagram of what I have in mind.
[quote author="alm"] Using a bridge rectifier would certainly make it more fool-proof, although it will drop 2V or so. If it would be possible to provide pads for wiring in an electrolytic cap (not on the board), it would be easy to adapt for transformers.
Quote
Using a dsPIC is the best way in my opinion. These guys have application notes for creating SMPS with 8-bit microcontrollers, surely a dsPIC will give better and acceptable results. We might even go for getting rid of the DC/DC converter IC and using PWM to drive MOSFETs, we might need some MOSFET drivers (or a transistor) but this might give us waaaaaay better flexibility. If these tiny guys can crunch data for audio stuff, I am sure that we will be able to control a DC power supply.
I'm sure it can work, the question is if it works better than tried-and-true analog. What kind of flexibility would the dsPIC give? What would it do better than a $0.30 LM324 in this application? To me this seems like a solution in search of a problem. It's not about number crunching, but about latency, and it's hard to beat analog electronics on that. No point in responding to a transient that's already over. Are there any good commercial lab supplies (who also try to save money and reduce parts count) using this? [/quote]
Alm you make some good points in this and your previous post. To restate one of them : if anyone is being put-off by the complexity of an analog-based supply (whether linear or switching), don't think that bringing it into software control will make things any easier. In fact, it's a pretty safe bet that it can become software hell. Tayken, I'd like to know what extra flexibility you're expecting by using a dsPIC ? Just remember that in some cases, application notes should also be seen as "proof of concepts", not recommended designs. Also remember that we'll need a customized but yet-inexistant USB stack !
As for the linear regulator I posted a few pages back, I just want to point out again that it's an unfinished design and while it should work, some thought needs to be given to loop stability. The op-amps doesn't magically make the thing work, and in fact minor component changes can easily make the regulator unusable ! A power supply like the HP3620 wasn't just slapped together in an afternoon...
alm you say a lab PSU should have an irreproachable output, I would tend to agree to that but I would opt for a linear post-regulator. The ability to source large currents at low voltages, without requiring a huge heatsink, can be useful in a number of cases.
Now, electrical isolation. This increases circuit complexity, especially since we have an USB interface. Fixed-output, offline PSUs often use voltage references on the output side coupled to the input side through optocouplers, but we can't really do that because our voltage reference from the PIC is necessarily on the non-isolated side. We can try to use optocouplers with calibrated networks on both side to get a usable voltage feedback. Another way would be to try to separate the PIC from the DACs. This could be either by using an isolated DAC chip, or maybe route a PWM signal through an optocoupler, which would then get filtered on the Power side. Or, use a dedicated USB chip (either FT232 or a generic PIC) which communicates to a very simple PIC to do the actual controlling. That is to say, one PIC would manage the USB protocol, and send data to the power section through an opto-isolated bus. This could be as simple as sending a 10-bit voltage word, a 10-bit current word, and receiving a 10-bit voltage reading and a 10-bit current reading. This means an extra PIC is needed, obviously, but this power-PIC can be pretty basic : it just needs a handful of ADC + DAC channels, plus 1-2 pins for communications. As a bonus it could also handle the dummy-load, which then becomes easy to isolate.
I was working on a programmable battery charger - an independant project of mine - which is quite similar to what we're building here - and reading a few datasheets I put things back in perspective: First, I'm assuming the main desirable feature of this project is a digitally-controlled CV/CC supply which should ideally be efficient, i.e. switching. We're going to have an SMPS stage anyway, which means an IC, inductor, diode and probably a MOSFET, which also means doing some calculations to size things properly. An all-in-one chip like MC34063 and tons of others is certainly nice, but not as customizable as we need. I had suggested the idea of adding an op-amp stage in the feedback path, which is theoretically sound but not necessarily stable. As I drew it, no current-limiting was provided either. *** Enter the TL494 and its compatible friends, KA7500, etc. They're dirt-cheap (like the MC34063), but include every thing needed for a custom regulator. They have two internal error amps with all inputs accessible, which means we can use one to control voltage (drive the reference pin with a DAC, smoothed PWM or whatever), and use the other EA to current-limit. TI's datasheet for TL494 is "arid", at best, so the key is really the application note "designing switching regulators with TL494", which you can grab from http://focus.ti.com/lit/an/slva001d/slva001d.pdf
Pros: *only one IC *efficient *no software feedback loop! *high power *passive soft-start is just 1-2 extra components *full control: 0-100% output voltage @ 0-100% current.
Cons: *Reference voltage from DACs may need to be scaled, which could mean an extra op-amp or two. A few were probably going to be necessary anyway to monitor voltage + current. *Does not have an internal switching element, so an external MOSFET is required *Output ripple. Some calculations will be needed, but <50mV is probably realistic. The question is whether we need the extra-low ripple that a linear post-regulator provides. It may be possible to add a tracking post-regulator which just regulates the output down to, for example, 1V lower than the switcher's output. This could be as simple as an LM317 and a bunch of resistors. *Current sense resistor on low side of the load is going to require a bit of thought with regards to signal/power grounds. Nothing that can't be managed. *Minimum input voltage 7V. I guess we were aiming for a more typical 12-24V anyway. The TL494 will take up to 40V ! Universal input, anyone ?
So yeah, I'm sold to this idea, and I'm probably going to build one for my charger.
[quote author="ian"] Here's a schematic from the simulation. [/quote]
Just note that the 100 ohms at the output was just there for testing, I would make it 2-10k just to make sure the filter capacitor doesn't stay charged for too long p-) Remember the pass transistor and op-amp were very preliminary choices, you can probably use different / more available parts. It's an unfinished design ! Probably good enough for testing though.
Quote
Can you also include LTSpice files so that I can check them out?
Sorry, you'll have to draw it yourself - I didn't save the file. Should be pretty easy ...
I thought of another simple way of controlling the output voltage: DAC outputs offset voltage, which shifts the voltage feedback routed to whatever SMPS regulator IC, MC34063 for instance. However, without a negative supply, it will be impossible to regulate to a lower voltage than the IC's fb comparator (1.25V in this case). I'm rather busy these days, so I don't have time to further simulate, let alone build this, but I think it would be worth a try. Just wire up a potentiometer followed by an op amp buffer to replace V2.
