Announcement

Collapse
No announcement yet.

A proper computer-to-bench supply conversion

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

    A proper computer-to-bench supply conversion

    As promised, here's my AT-to-12V conversion.

    This started as a Hipro PSB-HP200 from the mid 90's. It came from a horribly outdated system and looked too nice to trash.

    All the old caps went, FB ckt was completely changed. Too many times, "12V" conversions are nothing but a load resistor on 5 to keep things stable or the duty cycle high enough for the '494 to stay powered. This is an AT supply; the regulator is powered off the 12V output, before the output inductor. So if the load falls too low, not only is there discontinuous current, the '494 also picks up and drops off repeatedly as the supply 'pulses'.

    The new feedback loop is directly from the now 13.8-14.2V output, completely dedicated. _Zero_ attempt was made to "make do" with the old group reg ckt. You can see where I cut traces and changed/removed components.

    Here's a list of the 'mods', not necessarily in order:

    Calculated and wound new transformer
    Rewound output inductor to match new operating conditions and current
    Replaced all caps/upsize the ones in the output and doubler-rectifier
    15A line rectifier instead of 4A unit
    MJE13009L's instead of 2SC4242's
    Reduced the value of the base drive resistors from 4.7 to 2.7 ohms, to drive transistors harder
    Two SBL3045CT secondary rectifiers instead of one UF10100
    Larger line and load wires
    Reversed fan and bent heatsink fins for better cooling, could be better
    Sealed vents no longer used, to direct airflow
    "Tinned" traces and added jumpers where high currents flow
    Cleaned all that junk glue off the components
    Resoldered everything
    Removed and reinstalled components that were too far off the board, trimmed leads, etc.
    Cut out punched grille for fan and replaced with wire fan guard

    The load test consisted of the 12-to-120 inverter, a few large light bulbs, a big audio dummy load, and some 'ignition coil' ballast resistors. 60A, as detailed below:

    Inverter- 15A
    Dummy load- 14A
    Ballast resistors- 23.3A
    Light bulbs- 8.33A
    Total- 60.66A/~850W

    Anything from that load all the way down to just the fan is stable. Short ckt protection is provided by a novel hookup of the driver transformer. By way of an extra diode and other parts from the center tap, a derived voltage is obtained. This is related by turns ratio to the primary current. It will even save the supply from a short at turn-on. The new transformer was a success- the primary snubber resistor (R11) runs cooler than with the original, smaller TX.

    Voltage setpoint was 14.2, under load it was 14.0. The mess with all the test leads would make Unique jealous... or confused I don't have pix of that, it was sort of "improvised."

    With the fan shown, 30A is much more reasonable. If I stack two fans and/or change heatsinks/airflow, it'll be able to run flat out continuously. Transformer windings were sized for 30A out, so I'd have to 'cheat' and specify a duty cycle limit on anything over 30, even with a hurricane through it...

    Now the amp. It was once a Memphis 16-ST500D, one of those PWM subwoofer things. I got it in '03, from someone after it blew up. It appears that the board shifted and the tab of one of the output MOSFETS grounded out. The cascading fault killed all the power semis in the power amp. Two of the MOSFETS in the power supply burned their leads off and split themselves. The four others were also shorted.

    While trying to fix it, a cap must have been charged somewhere and I backfed and blew out the 'special' PWM driver. Special meaning no markings on the two ICs, other than "B52" and "F18". The little driver board itself was all SMT, so playing with that was over.

    I put a TDA7240 in there for 'temporary purposes' to use it as a utility amp. Sort of awkward, it needed two 18V/2A laptop adapters to make full output. In late December of 2010, I bit the bullet and started the design of the new amp and the redesign of the PS.

    The amp is capable of 2x110W into 4 ohms and 1x220W into 8 ohms bridged. I have not tested its output at 2x8 ohms yet, but expect around 2x75W. Tone control ratios are straight out of my Sansui 881 and it sounds fantastic. With TL072/71 opamps, it's also dead quiet.

    More amp details:

    SG3525 driving '50N06s, fixed at 35kHz, 49%- mostly due to how many 'filars' I used in the TX and not wanting to tempt the parasitics in my 'point to point' wiring
    Dual positive and negative rectifiers feeding caps to provide +/-36V
    Resistor-fed zeners for +/-15V opamp supplies

    The amp itself is nothing real special, typical diff amp feeding a class A CC voltage amp. The finals are NJW21193/94s, driven by MJF15030/31s. The preamp is on its own little board, since it is mostly DIP. Not the kind of thing I wanted on the main perfboard.

    Sorry for writing a book here, I just didn't want two separate threads...
    -Paul
    Attached Files
    "pokemon go... to hell!"

    EOL it...
    Originally posted by shango066
    All style and no substance.
    Originally posted by smashstuff30
    guilty,guilty,guilty,guilty!
    guilty of being cheap-made!

    #2
    Re: A proper computer-to-bench supply conversion

    Wow! Great work!

    Comment


      #3
      Re: A proper computer-to-bench supply conversion

      Thanks!

      The transformers are Class 130, the cores and bobbins shown below.

      I wound them with Essex SDN #22. Don't remember offhand what the turns and fillar counts are. I've got them somewhere...

      Unique will be after me. Those EER35 cores aren't so bad after all! At least these TDK ones.


      -Paul
      Attached Files
      "pokemon go... to hell!"

      EOL it...
      Originally posted by shango066
      All style and no substance.
      Originally posted by smashstuff30
      guilty,guilty,guilty,guilty!
      guilty of being cheap-made!

      Comment


        #4
        Re: A proper computer-to-bench supply conversion

        Mmm, big caps. Me like. Class A is a big waste of power IMO but it's also extremely easy to make it sound good, if you got power to spare why the heck not. Good job.

        I'm building a new transformer for my power supply (again) because the current one works but you can cook eggs on it, the copper loss is way too high. Hopefully this is the last one. It's halfway done, in fact i should have finished it yesterday, but i was so tired that i slept from 17 to 23... woke up ate something, went back to sleep at around 4... and look it's 11:43 now. So if you sum it up i slept 13 hours.
        Originally posted by PeteS in CA
        Remember that by the time consequences of a short-sighted decision are experienced, the idiot who made the bad decision may have already been promoted or moved on to a better job at another company.
        A working TV? How boring!

        Comment


          #5
          Re: A proper computer-to-bench supply conversion

          Unique, do you have a thread on transformers? This seems interesting. Is it cheaper to make your own? Or is it purely for different output voltages...?

          I need a really flat laying transformer...hmmm....

          Comment


            #6
            Re: A proper computer-to-bench supply conversion

            Originally posted by Th3_uN1Qu3 View Post
            Class A is a big waste of power IMO but it's also extremely easy to make it sound good
            Haha...

            I should've been clearer. The voltage amp right after the differential pair is class A. The finals are class AB, quiescent bias is about 30mA.

            Q4 and Q5 are the diff amp; Q3 is its current source. Q10 is the voltage amp; Q6 is its current source. Q7/Q12 are the drivers and Q1/Q8 are the finals. Q11 is a bias tempcomp. R26 is a 500 ohm offset trimpot and R9/R7 are both halves of a 500 ohm bias trimpot. I'd have made it more 'fail-safe' but the two '4004s will have to suffice in the space available. This is just a quick screen grab/Paint edit from Multisim...

            I hear ya, Unique. Welcome to the balancing act of getting enough window/core area to get enough copper to keep the losses low.

            The new Hipro transformer has a slightly different turns ratio as compared to the original one, an EI28. I wanted longer 'on' times and lower secondary ripple currents. Lower peak primary currents are good too!

            -Paul
            Attached Files
            "pokemon go... to hell!"

            EOL it...
            Originally posted by shango066
            All style and no substance.
            Originally posted by smashstuff30
            guilty,guilty,guilty,guilty!
            guilty of being cheap-made!

            Comment


              #7
              Re: A proper computer-to-bench supply conversion

              Originally posted by TheLaw View Post
              Unique, do you have a thread on transformers? This seems interesting. Is it cheaper to make your own? Or is it purely for different output voltages...?
              For me it's about different output voltage, different topology, different purpose. Cores are quite cheap, but for the price of a decent size core + bobbin (around $7) i can get three dead ATX power supplies that contain other goodies besides the transformer. More bang for the buck.

              Originally posted by TheLaw View Post
              I need a really flat laying transformer...hmmm....
              You could try doing a planar transformer. You'll have to buy the cores though, and because of the small space you will need to work at high frequency. Not for the inexperienced, layout becomes critical in such a situation.

              Originally posted by kaboom View Post
              Haha...

              I should've been clearer. The voltage amp right after the differential pair is class A. The finals are class AB, quiescent bias is about 30mA.
              Looks quite a bit like mine, but i'm using more current sources. I still haven't gotten around to building this, but after finishing this power supply and the class D subwoofer amp it is made for, i'll be upgrading the amp for my main speakers as well.

              Mine will be built in an old Romanian-made amp, i used to have the black version that worked, but i really really wanted the silver edition so i traded my working (but not without its flaws, it was on its last legs), for a blown silver unit. One of the channels can't be fixed because the PCB is old and brittle and it's damaged everywhere, i was able to fix the other, however it's getting some very weird waveform distortion. Over a certain level, the top side of the input signal "snaps" to the rail. You can imagine how bad that sounds.

              The strangest thing is that i have the original schematic and i have simmed it in LTSpice. I get the exact same distortion in the simulation, while the physical circuit worked, and still works fine in hundreds of units... just not in mine. Right now i'm using an old home theater system for the amplifier, but the background noise is higher than i'd like it.

              Originally posted by kaboom View Post
              I hear ya, Unique. Welcome to the balancing act of getting enough window/core area to get enough copper to keep the losses low.
              Welcome to the balancing act of shoving crazy amounts of wire in a tiny core... Well, as long as the bobbin doesn't fall to pieces, it's good. Vise ftw.
              Last edited by Th3_uN1Qu3; 04-02-2011, 02:43 AM.
              Originally posted by PeteS in CA
              Remember that by the time consequences of a short-sighted decision are experienced, the idiot who made the bad decision may have already been promoted or moved on to a better job at another company.
              A working TV? How boring!

              Comment


                #8
                Re: A proper computer-to-bench supply conversion

                Originally posted by Th3_uN1Qu3 View Post
                For me it's about different output voltage, different topology, different purpose. Cores are quite cheap, but for the price of a decent size core + bobbin (around $7) i can get three dead ATX power supplies that contain other goodies besides the transformer. More bang for the buck.



                You could try doing a planar transformer. You'll have to buy the cores though, and because of the small space you will need to work at high frequency. Not for the inexperienced, layout becomes critical in such a situation.
                Shoot...I think I'll just stick to buying them or not trying anything crazy with them. Do you know any good sources that can teach you how to make your own?

                Comment


                  #9
                  Re: A proper computer-to-bench supply conversion

                  Depends on what transformer you need. If you need a mains transformer, it isn't all that hard since the mains frequency is the same everywhere and you can't change it, so there are tables for it. But you have to wind lots of turns in the primary (and since you say you need a small one, with very tiny wire). Unless you have a coil winding machine i wouldn't bother.

                  For switching transformers it depends on topology chosen, frequency and flux density. If you are simply modifying an AT/ATX power supply, remove the first half primary, remove the secondary, wind another secondary to your specs, then put the old half primary back on. Easiest way to do it. The cheap PSUs with EI33 core are usually 20+20 turns primary.
                  Originally posted by PeteS in CA
                  Remember that by the time consequences of a short-sighted decision are experienced, the idiot who made the bad decision may have already been promoted or moved on to a better job at another company.
                  A working TV? How boring!

                  Comment


                    #10
                    Re: A proper computer-to-bench supply conversion

                    I find power supplies with ERL 35 transformers all the time. How many watts you think they can do?

                    Comment


                      #11
                      Re: A proper computer-to-bench supply conversion

                      ERL-35 or ERI-35 = Ferroxcube part number ER-35 (which is already small), but we have a problem - the ER-35 is spec'd for two E halves while the ones found in very cheap PSUs are always an E and an I (a bar). Therefore the effective area is even lower, requiring a higher number of turns => higher copper loss, more heat.

                      I'd say 250W, maybe 300 if you're lucky. Not more than 200 in the configuration they have in the power supply.
                      Originally posted by PeteS in CA
                      Remember that by the time consequences of a short-sighted decision are experienced, the idiot who made the bad decision may have already been promoted or moved on to a better job at another company.
                      A working TV? How boring!

                      Comment


                        #12
                        Re: A proper computer-to-bench supply conversion

                        Thanks Unique. I don't NEED a small one. It was just a fleeting thought. I usually don't let form factor get in the way of a better product.

                        What is the "best" core? And what kind of wire do you use? Is there a special type you recommend?

                        Comment


                          #13
                          Re: A proper computer-to-bench supply conversion

                          Core? The largest you can find that fits in the case you want to house your project. Always a good design philosophy.

                          As for wire, i use whatever i can find for cheap. Current wire is 0.35mm dia (~27AWG) enameled wire, got 300g for $12.30. Good to 130kHz, for higher frequency you'll have to go thinner. Here's a table. Need more current? Parallel several strands and twist them together. Tie one end to something, grab the other end and move back till it's stretched, and use your imagination. Or a drill set to low speed. Important note: ALWAYS cut the wire when you're done twisting, while it's still stretched. Otherwise... well, see for yourself.

                          Of course, the best wire is litz wire (lots of fine strands, pre-twisted and in the diameter you need for your particular application), but it tends to be expensive and it sells only in 1kg bobbins, which makes it even more expensive.
                          Last edited by Th3_uN1Qu3; 04-03-2011, 01:34 PM.
                          Originally posted by PeteS in CA
                          Remember that by the time consequences of a short-sighted decision are experienced, the idiot who made the bad decision may have already been promoted or moved on to a better job at another company.
                          A working TV? How boring!

                          Comment


                            #14
                            Re: A proper computer-to-bench supply conversion

                            I know you live in Romania, but where do you usually source these cores from?

                            Comment


                              #15
                              Re: A proper computer-to-bench supply conversion

                              For now mostly old PSUs and monitors, but there are local suppliers at decent prices.
                              Originally posted by PeteS in CA
                              Remember that by the time consequences of a short-sighted decision are experienced, the idiot who made the bad decision may have already been promoted or moved on to a better job at another company.
                              A working TV? How boring!

                              Comment


                                #16
                                Re: A proper computer-to-bench supply conversion

                                Originally posted by Th3_uN1Qu3 View Post
                                Core? The largest you can find that fits in the case you want to house your project. Always a good design philosophy.

                                As for wire, i use whatever i can find for cheap. Current wire is 0.35mm dia (~27AWG) enameled wire, got 300g for $12.30. Good to 130kHz, for higher frequency you'll have to go thinner. Here's a table. Need more current? Parallel several strands and twist them together. Tie one end to something, grab the other end and move back till it's stretched, and use your imagination. Or a drill set to low speed. Important note: ALWAYS cut the wire when you're done twisting, while it's still stretched. Otherwise... well, see for yourself.

                                Of course, the best wire is litz wire (lots of fine strands, pre-twisted and in the diameter you need for your particular application), but it tends to be expensive and it sells only in 1kg bobbins, which makes it even more expensive.
                                I mean what "type" of core? E-Core, C-Core? Is there one that's "better than the other"?

                                Comment


                                  #17
                                  Re: A proper computer-to-bench supply conversion

                                  Originally posted by Th3_uN1Qu3 View Post
                                  For now mostly old PSUs and monitors, but there are local suppliers at decent prices.
                                  Oh alright, thanks.

                                  Comment


                                    #18
                                    Re: A proper computer-to-bench supply conversion

                                    Depends what you want to do with it really. The regular general use cores are either EE (two Es) or EI (an E and a bar). The EIs are easier to disassemble if they're coming from some scrap equipment... the EE cores tend to crack when you pry them apart. But that's never an issue, as you simply superglue them together and they work just fine. There's other namings like ER, EC... but they will always fall into one of the two versions mentioned above. Differences between them are in physical geometry mostly, the basic properties are the same. Effective area still determines minimum number of turns, maximum operating flux is still the same, and so on. Frequency limits depend on the material that the core is made of not the geometry.

                                    The E type cores are the best all-round cores. RM type cores are the best at keeping stray flux IN (good for high sensitivity equipment) but since the entire bobbin is enclosed temperature is a concern. U cores are good for high voltage stuff (since you can wind the primary and secondary on separate legs) and they're very easy to cool effectively since most of the winding is exposed, but leakage inductance and stray fields are high. You see U cores in CRT flyback transformers mostly.
                                    Last edited by Th3_uN1Qu3; 04-03-2011, 04:29 PM.
                                    Originally posted by PeteS in CA
                                    Remember that by the time consequences of a short-sighted decision are experienced, the idiot who made the bad decision may have already been promoted or moved on to a better job at another company.
                                    A working TV? How boring!

                                    Comment


                                      #19
                                      Re: A proper computer-to-bench supply conversion

                                      Originally posted by Th3_uN1Qu3 View Post
                                      Depends what you want to do with it really. The regular general use cores are either EE (two Es) or EI (an E and a bar). The EIs are easier to disassemble if they're coming from some scrap equipment... the EE cores tend to crack when you pry them apart. But that's never an issue, as you simply superglue them together and they work just fine. There's other namings like ER, EC... but they will always fall into one of the two versions mentioned above. Differences between them are in physical geometry mostly, the basic properties are the same. Effective area still determines minimum number of turns, maximum operating flux is still the same, and so on. Frequency limits depend on the material that the core is made of not the geometry.

                                      The E type cores are the best all-round cores. RM type cores are the best at keeping stray flux IN (good for high sensitivity equipment) but since the entire bobbin is enclosed temperature is a concern. U cores are good for high voltage stuff (since you can wind the primary and secondary on separate legs) and they're very easy to cool effectively since most of the winding is exposed, but leakage inductance and stray fields are high. You see U cores in CRT flyback transformers mostly.
                                      Thanks Unique.

                                      Comment


                                        #20
                                        Re: A proper computer-to-bench supply conversion

                                        Originally posted by Th3_uN1Qu3 View Post
                                        One of the channels can't be fixed because the PCB is old and brittle and it's damaged everywhere, i was able to fix the other, however it's getting some very weird waveform distortion. Over a certain level, the top side of the input signal "snaps" to the rail. You can imagine how bad that sounds.

                                        The strangest thing is that i have the original schematic and i have simmed it in LTSpice. I get the exact same distortion in the simulation, while the physical circuit worked, and still works fine in hundreds of units... just not in mine.
                                        Do you think it's moisture or carbon in the cracks and the board going leaky after it heats up? Can you tell if the 'snapping' is from the diff amp going unbalanced or do you think the current sink/source for the voltage amp is losing bias? R12 in the partial.

                                        It works in the sim and a few hundred 'real' units so the board is the only thing left that's most obvious. Another thing to check is if any almost-broken traces separate completely when it's up to temp. I ASSume you've poked or tapped the board to see what happens. If not the board, go for the 'small' transistors.


                                        Good luck
                                        -Paul
                                        "pokemon go... to hell!"

                                        EOL it...
                                        Originally posted by shango066
                                        All style and no substance.
                                        Originally posted by smashstuff30
                                        guilty,guilty,guilty,guilty!
                                        guilty of being cheap-made!

                                        Comment

                                        Working...