Still trying (and NO LONGER failing) with the LM2588... 5V -> +/- 12V

[Skimask]

Still trying to make a flyback boost / inverting converter using the LM2588.  Been at it for the past couple of days now.

The circuit I'm using...almost exactly...is described here...
http://www.ti.com/lit/ds/symlink/lm2588.pdf
Figure 13, Page 17.
The only difference is that I'm using the LM2588-ADJ with the appropriate feedback resistors, R1=8.756K, R2=1K
12v = 1.23 * ( 1 + ( 8756 / 1000 )  (ok actually 11.99988v calculated, but my parts don't exactly have .000001% tolerance).

It's a dual output flyback converter, with 4-6v DC input, +12/-12v output @ 300mA.

In my proto circuit, I have 2 82ohm power resistors on the +12v and -12v rail, giving me about 146mA of load on each rail.

At most, I can only get 6.2v on each rail.  6.2v @ 82ohm = ~76mA.
When I open circuit both rails, I can get the full +12v and -12v.  I've tried a handful of other resistors on the outputs, and of course, the voltage drops with decreasing resistance.
But, I'm not getting anywhere near the expected results.

The circuit is soldered down.  I checked for excessive resistance on the traces.
I get the same voltage at the battery and the "switch" pin of the LM2588, so no drops there.
The battery pack hasn't dropped below 4.4v while I've been testing the circuit.  When it does, I switch out battery packs for a freshly charged one...with the same results at the outputs.
Nothing on the board gets hot.  The LM2588 warms up to about 10-15 above ambient (80F) then steadies up.  No change in output voltage while it warms up.
I can see the ~100Khz waveform on the 'scope at the switch pin of the LM2588, and other relevant waveforms are similar to the waveforms noted on the datasheets.

So...what's my problem?    (besides the obvious)
If I didn't know any better (and I don't), I'd almost say that the coil itself was wired backwards.
4.4v input gives me +/- 6.2v output.  12.4v total differential / 4.4v input = 2.8, and the coil input/output ratio is 1:2.5

I'm still wokring on trying to build this curve tracer, but make it battery powered rather than using a PC/AT PSU (with +12v and -12v rails), and need/want at least +12v and -12v rails for the op-amps.  I'm using a 4 cell battery pack for now, but if the circuit can handle it without letting all the smoke out, I'm not opposed to switching over to an 8 cell NiMH battery pack.  Might end up doing that anyway...4 cell pack might not last all that long.

[PA0PBZ]

Did you use one of the recommended transformers? (page 19, transformer type T2)
I can't see anything else wrong. Maybe a picture of your setup would help.

[Skimask]

Yep, got a couple of samples of the Q4337 directly from Coilcraft...part is the recommended T2 on pg. 19.
And other than using the LM2588-ADJ and appropriate feedback resistors vs using the LM2588-12 in the datasheet, the circuit I built is identical to the one in the datasheet as noted above.

I'm fairly confident in the wiring as I've disassembled and reassembled the circuit three times now.  It's working as described in the datasheet, except that it won't support any sort of load above around 12ma (1K load) on each rail before the voltage starts to drop off from the desired set-point of the feedback resistors.  I'm wanting the 300mA described by the example circuit, but I'll settle for 200mA, possibly as low as 100mA.
All of the grounds are tied back to a single point.  I've paralleled an extra 120uF with the existing 100uF for Cin, also tried paralleling the 330 uF Cout caps with a couple of big ol' 4700uF caps.
Also tried a 2nd LM2588 as well as the 2nd coil.....just in case I got a bad part from the distributor.
No dice...

[PA0PBZ]

If you load the output with the 82R resistors, do you see the feedback (pin 3) drop to about 0.6 volt?

[Skimask]

+ and - rails loaded with 82R, output voltage on each rail is about +6.4v and -6.4v, fdbk @ pin 3 = 396mV

With both rails open-circuited, I get + and - 12.1v on the outputs, and 731mV at the feedback pin.

[PA0PBZ]

With both rails open-circuited, I get + and - 12.1v on the outputs, and 731mV at the feedback pin.

Feedback should be 1.2V I think, and I wonder how you get 731mV with the rails open, does the feedback pin draw current?

[Skimask]

Nothing measurable other than floor noise on my meter, and the datasheet confirms nothing more than nA at the FB pin.

Double-double checking the grounds now.  Even though I've checked them once, I'm wondering if I've got a high resistance wire somewhere.....

[PA0PBZ]

So how do you explain the 700mV? 8K7 and 1K, 12V, should be something like 1.2V isn't it?

[Skimask]

I can't explain squat!    
On a whim, I just hooked up my 'scope to the switch pin on the '2588.  LOTS of noise.  I don't have a fairly nice square wave like shown on the datasheet diagrams, and certainly nothing resembling a 100khz switching signal.
Gonna do some more reading of the 'sheet.  Might have to add the 'snubber' RC filter across the postive output diode...probably a dozen other things as well.  Might be that all that noise is causing the '2588 to do some stupid stuff.  The output voltage on both rails is relatively fairly clean, albeit low...but, start putting all that noise into the feedback and it ends up getting amplified and causing all sorts of crap elsewhere...

[jerry507]

Tell us some more about the parts you're using. Specifically what diode, output bulk capacitor, compensation capacitor. Pictures of your circuit would also help. Also which exact version of this chip are you using? I think you said the Adjustable version, just want to be sure.

[Bored@Work]

You don't happen to build that thing on a solderless breadboard?

[CarlG]

If you post a picture with the switch node and the trafo-anode node for the +12- it might give a clue to your problem. Of course you have short ground connection of the probe

bored@work asked a very good question. If the answer is yes, that's most certainly the problem. If you've built on a solid copper board, then I would say that you might have the answer in your first post: not that the windings are reversed, but rather that you put the diodes the wrong way, so that you get exactly what you measured: a simple 1:2.5 trafo (instead of a coupled inductor).

Another problem might be that you get switch noise into the feedback input, even with such low impedance FB-network. Minimize the FB-node; connect the output to the FB top R away from any high di/dt and du/dt.

//C

[Skimask]

@jerry - Using exactly the parts noted on the datasheet's schematic, LM2588-ADJ, Coilcraft Q4337, 1N5819's.
The big caps are genuine Nichicon's and fairly new, about a year old, 100uf on the input (tried 330uf as well), 330uf on the output (tried 4700uf in parallel too).
I don't have an ESR tester, but I've tried more than one in each position and they all act the same way.
The .1uf and .33uf are ceremic disc types, and I tried a couple of big ol' square poly () types as well.

@Bored - uh...ashamed to say, yes I did at first    and an old one at that, then got smart and moved it over to a regular proto-board.  Good solder, tinned connections, the whole bit.  Same thing.

@CarlG - trafo-anode node?  Wha?  Probe-yep, the ground connection that feeds right off the tip of the probe itself.  I double-checked the polarity markings on the transformer and diodes to make sure everything matched up.

@AcHmed99 - Thought about the battery pack too.  AA's might have too much internal resistance, crapping out when the switch is on, the V drop too fast for the 'scope to see, so switched over to a 4 pack of D cell NiMH (and tried a 4 cell NiCad pack as well).

Gonna tear it down again tonight and build it up one more time.
I gotta be doing something fundamentally wrong here.

[Skimask]

Ya know, I just looked at LM2588 and Coilcraft Q4337 datasheets at the same time.
The polarity marks at on pins 1,6,7 on the LM2588 datasheet.
Polarity marks at on pins 2,6,7 on the Coilcraft Q4337 datasheet.

Hmmm....

[ftransform]

Why is this circuit inappropriate to breadboard? is it really that sensitive?
are you sure this is not a bit of paranoia?

[Skimask]

I'm talking about solderless breadboard vs soldered protoboard.
The solderless breadboard version's waveforms were practically unrecognizable, and like I said, the board was an old one.  Contacts probably ugly, dirty, worn out.
Someday I'll throw it out...( NOT! Who throws anything like that away? )

[ftransform]

I'm talking about solderless breadboard vs soldered protoboard.
The solderless breadboard version's waveforms were practically unrecognizable, and like I said, the board was an old one.  Contacts probably ugly, dirty, worn out.
Someday I'll throw it out...( NOT! Who throws anything like that away? )

I would have thought solderless board be OK for this application because its only 200khz. I also noticed that all the capacitors used in the circuit are way bigger then the parasitics present on a breadboard.. I thought they were OK up to +10 megahertz.
I guess age has something to do with it.

[Skimask]

Normally, I would agree, if the circuit was all digital.
But, there's an analog factor going on here, and I think the extra noise on the solderless breadboard gets amplified a bunch internally in the LM2588 in the feedback circuit and causes all sorts of stupidness to go on.

[CarlG]

The switching frequency has nothing to do with the breadboard type required (or fundamentally, the ground plane). It's the rise- and falltimes that matters (which naturally must be low if f_s is high). Even if you're switching something with 1kHz, there's nothing saying that you don't have t_r and t_f of 1-10ns, e.g. an ordinary logic gate. In this design you have a voltage swing of from -1.25*V_in to nominally 12V in say 10 ns or so, and of course also high di/dt that induces high frequency EM fields.

Here's also the requirement of keeping the ground connections of the regulator, the FB net, and the in- and output capacitors together in an optimal way, which IMO is not possible with a "ordinary" breadboard (i.e. those white/blue/whatever plastic ones). I'm not saying they're useless, but avoid them when you're dealling with high di/dt or du/dt.

[Skimask]

Right.  I couldn't agree more...except that you put it in a much more technically correct format that I by far.
Now the only question I've got is how you got the subscripts to work!  (e.g. the small letters behind the big letters, Vf, Tf, Tf, and so on )

Working on the 3rd iteration now...a few changes in order.  Heavier gauge jumper wires, adding the snubbers and other items noted in the datasheet, whether they'll be needed or not.  If it's noted, I'm going to add it this time around.

EDIT:  Looking thru the datasheet...  Thinking I might add another cap across the output right at the feedback resistor network as close to the LM2588 as I can get it.
Low cap value, maybe 10pf or so, high enough VDC value to keep it from frying itself...
Or would that be a really bad idea since it'll be messing with the feedback voltage...keeping it from changing as fast as it can?
If the switching freq is 100khz, the cap value plus the feedback resistor network, run it thru the math, keep the RC constant above 5x the switching speed, or 500khz.
Should help smooth things out?  Yes?  Or ruin it completely?

[grumpydoc]

Now the only question I've got is how you got the subscripts to work! 

When you edit the reply the buttons for making something _sub or ^superscript are on the 2nd row just below right justify text/font.

Highlight the text you want, then click the button.

[Skimask]

Aha!
^{I see it now.}
_{Learn something new every day}

[jerry507]

Pictures of construction will help a lot. Switching regulators are heavily dependent on layout for success.

[CarlG]

EDIT:  Looking thru the datasheet...  Thinking I might add another cap across the output right at the feedback resistor network as close to the LM2588 as I can get it.
Low cap value, maybe 10pf or so, high enough VDC value to keep it from frying itself...
Or would that be a really bad idea since it'll be messing with the feedback voltage...keeping it from changing as fast as it can?
If the switching freq is 100khz, the cap value plus the feedback resistor network, run it thru the math, keep the RC constant above 5x the switching speed, or 500khz.
Should help smooth things out?  Yes?  Or ruin it completely?

No, a feed-froward cap is quite common to reduce overshoot and ringing. The RC is not related to the switching frequency. Sometimes C_ff is mandatory for stability, but if the datasheet doesn't say so, you may check the ringing frequency and try out the cap value. If you still got that low impedance feedback voltage divider as before I don't think 10 pF will do any difference. For me, last time I think I tried with an RC =1/(2*pi*f), where f is the ringing frequency, and it worked out quite well. If C_ff gets too big, the step-response will be degraded.

[CarlG]

Ya know, I just looked at LM2588 and Coilcraft Q4337 datasheets at the same time.
The polarity marks at on pins 1,6,7 on the LM2588 datasheet.
Polarity marks at on pins 2,6,7 on the Coilcraft Q4337 datasheet.

Hmmm....

Trust the datasheet on this one. It's seems like it's the classic transformer-instead-of-coupled-inductor case...

[Skimask]

Trust the datasheet on this one. It's seems like it's the classic transformer-instead-of-coupled-inductor case...
[/quote]

Not 100% sure what you mean by that...

I think you mean transformer as in exactly that...a step up transformer....vs. two inductors sitting effectively side-by-side that just happen to be doing things together...which from my reading in TAOE, is almost the same thing, but not quite.

[Skimask]

No, a feed-froward cap is quite common to reduce overshoot and ringing. The RC is not related to the switching frequency. Sometimes C_ff is mandatory for stability, but if the datasheet doesn't say so, you may check the ringing frequency and try out the cap value. If you still got that low impedance feedback voltage divider as before I don't think 10 pF will do any difference. For me, last time I think I tried with an RC =1/(2*pi*f), where f is the ringing frequency, and it worked out quite well. If C_ff gets too big, the step-response will be degraded.

The only other cap's the datasheet talks about is the snubber RC across the positive rail output diode, before where the feedback gets picked off.
Doesn't sound like the same thing.  Maybe the LM2588 has a feed-forward cap internally???

[Skimask]

Pictures of construction will help a lot. Switching regulators are heavily dependent on layout for success.

Gonna finish building it up tonight and take some pics.
I'm even going to try and make it...pretty...rather than just slapping it together...for a change

[CarlG]

author=Skimask link=topic=13333.msg180523#msg180523 date=1357943958]

Trust the datasheet on this one. It's seems like it's the classic transformer-instead-of-coupled-inductor case...

Not 100% sure what you mean by that...

I think you mean transformer as in exactly that...a step up transformer....vs. two inductors sitting effectively side-by-side that just happen to be doing things together...which from my reading in TAOE, is almost the same thing, but not quite.

I mean that if pins 1 and 2 are swapped, the transformer/inductor will act as a transformer when the regulator switch is "on", instead of storing energy in the primary winding. For the +12V side, this means that you'll get 2.5*V_in on the secondary output, and the diode will conduct, instead of having a negative, -2.5*V_in, output voltage and the diode reverse biased.

When correctly wired, working as a coupled inductor, energy is "stored" during the switch "on" phase, which is transferred to the secondary windings when the switch is turned off.

As you may know, the transformer/coupled inductor always have some leakage inductance, meaning that the coupling between the windings is not 100%, so there will be some energy left in the primary winding that has to be dissapated somewhere, and that turns out as an overshoot at the switch node. This may kill the regulator if it's not capable of handling the voltage peak. That's why you may need a snubber on the primary side. This will also reduce ringing on the secondary side, but it will probably also be some ringing when the diodes turns of (snap).

[CarlG]

No, a feed-froward cap is quite common to reduce overshoot and ringing. The RC is not related to the switching frequency. Sometimes C_ff is mandatory for stability, but if the datasheet doesn't say so, you may check the ringing frequency and try out the cap value. If you still got that low impedance feedback voltage divider as before I don't think 10 pF will do any difference. For me, last time I think I tried with an RC =1/(2*pi*f), where f is the ringing frequency, and it worked out quite well. If C_ff gets too big, the step-response will be degraded.

The only other cap's the datasheet talks about is the snubber RC across the positive rail output diode, before where the feedback gets picked off.
Doesn't sound like the same thing.  Maybe the LM2588 has a feed-forward cap internally???

Sorry, I misinterpreted your post. I see whay you mean (I think): you want to put another cap in parallell with the electrolytic one? No problem with that either.

[Skimask]

Sorry, I misinterpreted your post. I see what you mean (I think): you want to put another cap in parallel with the electrolytic one? No problem with that either.

Well, yes and no.
No, in that it's purpose wouldn't be primarily to help filter the higher frequency noise on the output.
Yes, in that it's purpose would be to help smooth out the noise on the feedback loop and hence the caps would be placed right next to the feedback resistor network vs. near the output cap's.

[Skimask]

Ok, done with the rebuild, #3, and it works much better...but not perfect yet.  (pictures to follow, might not get there until tomorrow though)

I'm pretty sure one of the problems was the mislabeled primary side polarity marking for the coil in the LM2588 datasheet.  The primary side is reversed from the Coilcraft Q4337-BL datasheet, and as far as I can tell, the secondary sides both match up.

With no external load, I can adjust the output voltage as required.  The positive side has a small bit of load from the feedback network and it's steady enough.  The negative side doesn't have any load on it when wired up according to the figure, so I added a 10K to ground, and that took care of that.

According to the figure in the datasheet, it's rated for 300mA on each rail.  I'm not getting that without a significant voltage drop.
With 82ohms on each rail (because that's what I've got on hand), I get 8.6v on each side, equating to about 104mA per rail, 208mA total.  And a quick current measurement off the battery pack shows about 570mA average draw, which works out close enough for me when I do the math.

I'm feeding the circuit with a 4 AA cell NiMH battery pack.  I've got a 4 D cell NiMH battery pack on charge and will test that in about an hour.
The meter is showing a constant 4.5v under load from the pack.  On the 'scope, I get about 2v of noise on the V_in line.  I'm pretty sure one big issue here is the really tiny gauge wires coming from that 4 cell AA holder.

So...

Input = 4.5v @ 570mA = ~2.57 watts

Output + and - = 8.6v @ 82 ohms = 104mA = .89 watts x 2 channels = 1.78 watts

1.78 watts out / 2.57 watts in = ~70% efficient.

[Skimask]

Pics of the "completed" board...(this camera phone is worse than I thought when doing close up stuff...)

[SeanB]

A magnifying glass held in front helps with that problem.

[CarlG]

Good to hear you're making progress:) From what I can see there's no obvious component choice that explains why the circuit doesn't work properly. But you'll never get good performance using that kind of bread board. Try to get something like this bare Cu board instead.

Meanwhile, you can use copper tape to improve the ground plane.

It might be noise injection that disturbs the regulator, at the FB or COMP nodes. The current limit may set in due to improper ground/decoupling. What does the outputs look like on the scope? And winding/diode nodes? It'll be interesting to see scope pictures!

The component placement doesn't seem optimal.
- connect Cin (1uF) directly between Vin and GND pins (is Cin the one to the right of the electrolytic in front? Doesn't work placing it there...)
- connect the bottom feedback R directly between 2588 FB and GND pins
- connect the top feedback R directly to the feedback pin to minimise the feedback node, route the wire from +12V away from high current paths, tie it to output cap or just by it on the way to the load)
- get rid of the pots to reduce problem sources, and route +12V from the output cap to the top feedback R (you can add the pots later when you get the feeling of the basic circuit)
- tie Cc directly to the 2588 GND pin
- a small R (1-10 ohm) between main Vin and 2588 Vin pin/Cin may help isolate the regulator internal circuitry from the noise on the 100 uF

Use the backside of the board for above.

The output routing seems reasonable local (although I think you can improve it), but the input seems to make up a large current loop from 2588 GND to the input cap (100u). Always minimize current loops!

Good luck!
//C

[Strada916]

Firstly get rid of those plugs. Solder the wires directly.

A point to remember,

What goes in, must come out. Kirchhoff's second law,

So +12V@0.3A and -12V@0.3A = 7.2Watts, lets assume 1 or 2 watts is dissapated in the circuit, transformer loses, heat etc. therefore we have about 10 or so watts of power used by your circuit.

Now ohms laws says that 10w @ 5v is about 2amps.

Can those batteries supply 5v @ 2amps?? And for how long can they do it?

[Skimask]

But you'll never get good performance using that kind of bread board. Try to get something like this bare Cu board instead.

I've never had a problem with these boards (www.protostack.com), even when using them with PICs @ 64Mhz, then again, haven't tried using them with anything that's analog and so 'touchy' as this.
I've got a decent shop to work with out here, and even have a CNC machine to route out raw copper PCBs.  I've tried the toner-transfer method as well in the past.  Just haven't found a good method for making PCBs that works reliably and is repeatable.

Meanwhile, you can use copper tape to improve the ground plane.

I've got multiple extra wires placed here and there to help out the grounds, same thing with the higher current Vin rails.  It's not a ground plane, but it's a start.

It'll be interesting to see scope pictures!

I'll work on it.  I've got a Tek2245A, but the Holdoff knob doesn't seem to work well these days.  More troubleshooting for another time...

The component placement doesn't seem optimal.

Ya, no kidding!  I was shooting more for 'tight' than 'optimal' since I'm going to be adding a few MCUs and OpAmps to the circuit.

- connect Cin (1uF) directly between Vin and GND pins (is Cin the one to the right of the electrolytic in front? Doesn't work placing it there...)
- connect the bottom feedback R directly between 2588 FB and GND pins
- connect the top feedback R directly to the feedback pin to minimise the feedback node, route the wire from +12V away from high current paths, tie it to output cap or just by it on the way to the load)
- get rid of the pots to reduce problem sources, and route +12V from the output cap to the top feedback R (you can add the pots later when you get the feeling of the basic circuit)
- tie Cc directly to the 2588 GND pin
- a small R (1-10 ohm) between main Vin and 2588 Vin pin/Cin may help isolate the regulator internal circuitry from the noise on the 100 uF

-Cin is to the right of the electrolytic near the bottom.  The positive and negative battery inputs are tied to 7 each rows of pins to give me extra holes to work with.  The datasheet said .1uf, but I'll add another one in there anyway.
- feedback, noted...
- Cc - that was the only .47uf I could find in the whole shop, so that's why it's connected like that.
- small R between Vin-2588/Vin - that's what the datasheet said too.  Will work it...

The output routing seems reasonable local (although I think you can improve it), but the input seems to make up a large current loop from 2588 GND to the input cap (100u). Always minimize current loops!

Oh ya...it's not optimal by any stretch of my, or anybody else's, imagination!  The current loop you refer too is 'shortened' a bit by the two extra wires on the left, but like you say, is not optimal at all.

Rework time!

[Skimask]

Firstly get rid of those plugs. Solder the wires directly.
A point to remember,
What goes in, must come out. Kirchhoff's second law,
So +12V@0.3A and -12V@0.3A = 7.2Watts, lets assume 1 or 2 watts is dissapated in the circuit, transformer loses, heat etc. therefore we have about 10 or so watts of power used by your circuit.
Now ohms laws says that 10w @ 5v is about 2amps.
Can those batteries supply 5v @ 2amps?? And for how long can they do it?

Plugs - not for awhile yet.  Those plugs give me a way to pull the battery pack out in a hurry in case I see smoke!  But I understand where you're coming from.

Can those batteries supply 5v @ 2 amps?  Well, probably not for long, which is why I've got a handful of 4 AA cell NiMH packs on charge waiting to go.
On the upside, the 4 D cell NiMH pack came off charge a little while ago, so that's my next thing to try now that I know I won't be letting the smoke out of various components...

[Skimask]

Can't find any low ohm resistors with decent wattage to put between Vin and the LM2588 Vin...left it out for now...
Got rid of the pots.  Using 2x 4.7K resistors for the 'top' of the feedback resistor chain, and a 1K resistor for the 'bottom'.  In a perfect world, that should come out to 12.792 volts at the output.

No load results (practically no load...feedback network on the positive rail provides about ~10.4K, added 10K to ground on negative rail, about ~2.5mA idling load)
Vin = 5.35v measured at the LM2588 and coil primary side
Vout+ = 12.62v
Vout-  = 12.62v
V @ feedback pin = 1.217v

1.5K load on each rail (3 470 ohm resistors in series, in this case ~19.2mA total load post-regulator)
Vin = 5.34v measured at the LM2588 and coil primary side
Vout+ = 12.51v
Vout-  = 12.52v
V @ feedback pin = 1.207v

82 ohm load on each rail (in this case, ~210mA total load post-regulator)
Vin = 5.17v
Vout+ = 8.48v
Vout-  = 8.48v
V @ feedback pin = .829v

  Put the meter on the ground rails, positive rails, looking for any obvious resistance...
Didn't find anything measurable either in resistance or in voltage differences.
'scope pictures up next if I can get the Holdoff knob to work right and clear up the picture.

(EDITs vs. new posts each time I make 'progress' should keep this a bit cleaner...)

EDIT:  I can't get the 'scope pictures to clear up at all.  Secondhand Tek 2246A I've had for about 10+ years.  Never had it "tuned up" or calibrated other than using the built-in menus.

EDIT#2:  Added the "snubber" R & C to the positive output diode.  Made it worse.  Output voltage dropped down to + and - 8.1v.  Noise appeared to stay the same.  Removed previously mentioned snubber since all it was snubbing was my power!

EDIT#3:  Added a 1uf across LM2588-Vin to ground.  Output voltage started at 8.4v, but rose to about 9.1v after the circuit had "warmed up" for about 2 minutes (power resistor load and LM2588 w/heatsinks rose to about 30F above ambient, 105F).

EDIT#4:  What if the polarity marking for the coil was actually on pin #1 according to the LM2588 datasheet  vs. on pin#2 according to the Coilcraft datasheet?   Hmmm....
Edit#4a:  Nope that's not it.

EDIT#5:  Added a smattering of .01uf caps, one each at the input and output of each positive and negative rail diode, one at the Vin/Pin7 to go along with the 100uf & 1uf, and another at 'top' of the feedback resistor chain.  Also put a fat 4700uf cap across the battery input wires on the PCB itself.
Now...I'm getting 11.6v at the output with the 82ohm load...I think.  It's seeming like the regulator wants to be "warmed up" before it'll start giving me a decent output.
Using 2 4.7K resistors for the 'top' of the feedback, 1K resistor for the bottom of the feedback.
Vout     = 1.23 * ( 1 +   R1  /  R2   )
12.792 = 1.23 * ( 1 + 9400/1000 )
At any rate, 11.6v is within the tolerances from the resistors in the feedback chain as well as the tolerance of Vref inside the LM2588.  The calculated worst case each way gives me a range of 11.453v - 14.294v (5% resistors in the feedback, along with 2% feedback voltage tolerance according to the datasheet).

EDIT#6:  No dice...
Loaded each rail down with 44ohms (82ohm + 2x47ohm in parallel).  Got 7.4v on each output.  That's 168mA @ 7.4v, where I should be getting ~12v @ ~270mA. 
But, after the regulator/heatsink warmed up a bit, 130F (~60F above ambient), I got 9.58v output, with 4.84v on Vin.
If the 70% efficiency I calculated earlier holds up, that's 438mA load at the outputs, 626mA draw from the battery pack.

Rechecked it after the regulator had cooled down to 60F (at the heatsink) and the output was only 7.4v.  (Vbatt = 4.82v)
Re-recheck it after the regulator had heated up to about 100F (again at the heatsink), output up to 8.42v.  (Vbatt = 4.81v)
Re-Re-rechecked it with the regulator heated up to about 130F, output up to 9.58v.  (Vbatt = 4.81v)

That a repeatable response.  Did it 3 times, same temps, same voltages, a few millivolts lower on the battery pack each time though.

What the heck is going on here?

[Skimask]

Zzzzz....pissed off am I....unable to make it work properly I am....backwards I will talk.

Added 4,700uf across Vin and ground right where the battery pack is connected to the PCB.
Added 2 extra 220uF across the plus and minus outputs right before the loading power resistors.
Added 2 extra 1uf and .01uf cap's across those same 220uF caps.
Added another .047uF across the feedback resistor network right at the LM2588.

My waveforms still looking nothing like those in the datasheet, not nearly as clean, and almost unrecognizable if I didn't know what I was looking for in the first place.
Still getting bunches of the same noise on Vin, Vout+, Vout-, Vfeedback.
The voltage on the switch spikes (very VERY narrow spike) up to about 50+ volts, but hangs out between ground and 2xVin.  What gives here???
On every pin i checked, the noise seems to have a decent frequency component of about 5Mhz, other than the main switching frequency of ~100Khz.  Mine seems to hover around 109Khz.

Charging the battery packs for more tries again later.

[jahonen]

Believe it or not, but lack of stiff ground (i.e. plane) is #1 reason for strange and difficult-to-solve problems with switch mode regulators. Having few wires here and there for ground simply doesn't cut it. Do it "Jim Williams" style.

Regards,
Janne

[Skimask]

I couldn't agree more...well I could, but I'd sound stupider than I already am 
I don't have the facilities to make a really good ground plane...so trying to make due.  Ain't workin' out so good is it?   I'll get 'er figured out sooner or later...

[jahonen]

But certainly you could get a copper clad board (unetched single/double sided PCB)? That makes a very good ground plane. Then solder all your components there using "3D construction".

What I (and others have meant by copper clad construction) mean, look at Linear Technology AN-47 http://cds.linear.com/docs/Application%20Note/an47fa.pdf, somewhere around page 102 or so. Of course, the result will not be pretty or even mechanically very robust but I suspect that will work fine electrically.

BTW, I just remembered that have tried to do something similar in the past with LM2585, and I had also horrible problems (the PCB is now in the box of "unsolved problems projects"), so I think we have done same mistakes, no need to repeat them

Regards,
Janne

[Skimask]

That AN-47 is quite the read, but it covers a lot of stuff not in "the good book" (aka The Art Of Electronics).

Nonetheless, I don't have the capability at the moment to do anything close to what that AN47 shows.
I think the best I'm going to be able to do in the near future with the "Protostack" boards is to tie all of the unused buses together in and around the entire circuit back to ground.
As well, if I move a few parts around, it should shorten up any ground loops that might exist.
I've got a handful of single sided copper clad board.  You think if I drilled a handful of holes in it and connected it up as a ground plane, I might make progress?

[CarlG]

That AN-47 is quite the read, but it covers a lot of stuff not in "the good book" (aka The Art Of Electronics).

Nonetheless, I don't have the capability at the moment to do anything close to what that AN47 shows.
I think the best I'm going to be able to do in the near future with the "Protostack" boards is to tie all of the unused buses together in and around the entire circuit back to ground.
As well, if I move a few parts around, it should shorten up any ground loops that might exist.
I've got a handful of single sided copper clad board.  You think if I drilled a handful of holes in it and connected it up as a ground plane, I might make progress?

So you got it? Well, then there's no problem. You don't need to drill anything, solder ground pins to the copper and "hardwire" the rest in the air! It will look messy, but that's part of the beauty!

[jerry507]

Bad physical construction is the downfall of every switching regulator. No amount of added capacitors anywhere will fix the problem. Somewhere Jim Williams kept a tally of the causes of problems that he got calls about. Well over half of them were caused by layout problems, another big chunk by crappy parts.

Every time I see one of these posts and the person just wants to keep slapping parts of to fix the problem, I sigh. They always have some kind of non-ideal layout that causes a huge amount of inefficiency.

[Skimask]

Can't disagree with you there jerry507, especially when I pulled all those extra cap's out of there and found the overall noise went DOWN!, then again so did the 'regulate-ability' of the output voltage (is that even a word? )

I'll get it figured out eventually.
That is unless somebody else has a more easily workable idea for generating a split rail supply for opamps using (preferably) a 4 cell (or 8 cell, got both, either would be nice vs using a PC/AT PSU) battery pack.

[grumpydoc]

If you just want to split the supply then a "virtual ground" buffer should do the trick - in fact wasn't the circuit in the "perfboard prototyping" thread just such a beast?

[Skimask]

Ya, I saw that.  But I'm not sure if a "virtual ground" buffer is the right tool for me in this application, e.g. the advantages and disadvantages of using a virtual ground.  I'm going to do a bunch of reading tonight on the "virtual ground" subject, see what I can come up with...

My project is the battery powered curve tracer using a PIC to do the A/D work and an LCD for the display.  Been working on it for awhile now.  Guess a little bit of a circuit description is in order.
The basics are done up already on the PIC, LCD, A/D, D/A, etc. side of things.  That's all down to tightening up the programming, making it more user friendly, etc.

It's the power supply that's tripping me up.

I want +15v / -15v to drive the opamps (along with another 3.3v for the PICs/LCD/etc).  Could make due with +/-12v, heck even as low as +5/-5 volts would probably do for my purposes.  Using TL084 opamps for now since I happen to have a tube of 20 sitting in the parts bins.
After I get the PSU issues sorted, I'll switch over to something with more current source capability and can drive closer to the rails.

I want to run it off a 4 cell NiMH pack, whether it's a 4 cell pack of AA's or D's doesn't matter (AA's are preferable, but D's will do).
An 8 cell AA NiMH pack would be ok too but would increase the size/weight of the whole thing.
What I don't want is to have to run the thing off a PC/AT PSU just to get a decent +/-12v and have to plug it in everywhere I go.

Maybe, in my situation (with the lack of proper facilities, PCB building materials, etc.) the way to go would be to boost that input battery volts to ~30, split that with the virtual ground, then post regulate that back down with a couple of linear regulators leaving plenty of headroom while under load.  I have successfully boosted a 6 cell pack up to 12v using an LM2576 (?partnumber?) awhile back.  Didn't have any problems with that.  But all that was doing was pulsing a solenoid.

Amateur hour for me... 

EDIT:  It was the LM2577 that I used successfully for the 7.2v -> 12v step-up converter that I made back in the day.  Worked good enough to run a hard drive for the MP3 player I built and that was on a protoboard board very similar to what I'm using now.

[Skimask]

I give up on the LM2588 in the flyback configuration.  I don't have the proper PCB making equipment to make a 'clean' layout to make it work correctly and cleanly.

New plan...In flux...still working out the details...

Use the LM2588 in the boost configuration, not a flyback config.  5v in, 12v out, Page 22 of the LM2588 datasheet.
http://www.ti.com/lit/ds/symlink/lm2588.pdf

Use an LM675 and make a virtual ground with that.  Page 2 of the LM675 datasheet.
http://www.ti.com/lit/ds/symlink/lm675.pdf

The ground of the LM2588 section becomes V-, the LM675 provides a Vgnd, the LM2588 booster gives me V+.
Keep the grounds separated as required between the PSU and the actual logic doing the work.

Whether or not this will work in this configuration using the same layout methods, I don't know.  I do know, as I said in the last post, I've used the LM2577 as a booster practically identical to this and it worked, using the same layout methods I'm going to use now, but a fair bit cleaner (I didn't know what I doing back then anyways).

We'll see what happens.  I gotta order a few LM675's.  Closest thing I've got to that is an LM386 and that ain't gonna cut it.

[jerry507]

Look, it's not that you lack the tools. You said you've got copper clad board laying around and you've got a soldering iron. Maybe a little wire and that's all you need. When we say build it Jim Williams style, or dead bug style or whatever phrase, that means:

http://cds.linear.com/docs/Application%20Note/an47fa.pdf

Look at page 18 (of the PDF).

Keep the leads as short as you can, everything as tight as you can. Your circuit will work. That's all you're missing. You're thinking to yourself "Man, it can't be the circuit board I'm using, right?" Yes, it is. Switching supplies are easy to build assuming you have an acceptable construction technique and you choose good components. You've got good components, just a poor layout/construction technique.

The flyback is a great way of doing what you're doing, and dropping it for a different switcher won't solve your problem and you certainly won't learn anything more. Stick with it, and listen to what we're saying.

[Skimask]

Ok, I will disassemble the board and give it a try, ugly as it may end up, I'll try it.  Not really a lot of connections to be made, and a lot of them are directly to the "ground plane" at that.
But, I think I'm going to 'mount' (where's that hot glue gun?) the components on the one side, drill holes in the board and solder the ground connections to the copper side.
Lessen the chances of shorting anything out.

[Skimask]

Well...here it is...
Haven't tried it out yet.  Just had to get some pictures up here.  Gonna spend awhile with the circuit under the microscope looking for anything miswired.
Obviously not the tightest layout.  Now that I look at it, I likely could've squeezed it together a bit better, but I think the wiring is about as short as I can get it.

Changes from the example schematic in the LM2588 datasheet:

Ran out of anything near 330uF cap's for the output.  Used 4,700uf in parallel w/ 4.7uf.
In the other version, the negative rail wouldn't regulate since it didn't have any load, so added a 10K on the negative rail to ground.
Added the extra Cin and 3.9ohm across the battery Vin and LM2588 Vin (as suggested by the datasheet).
82 ohm sand resistors are there for load testing only.
Feedback, R1=1K, R2=a pair of 4.7K.  Should give me about 12.792v at the output, assuming all of the values are dead on.
The twisted wires are a pair of 22ga jumper wires feeding the coil primary side.

Now for the good stuff...
EDIT:  Now that I see the pictures in full screen, I'm going back to clean off that flux.  Jeeze that makes it look ugly---er.
EDIT#2:  And now there is a ground on pin#4 of the LM2588.  How could I possibly forget that?
EDIT#3:  Also reconnected the switch to Vin vs. ground...

[Skimask]

Ok...works!!!
Still got a bit of high frequency noise on the outputs, not a lot, surely not NEARLY as much as before.
The trace coming off the switch @ pin 5 is sort of recognizable, but doesn't resemble the trace in the datasheet.  I can see the switch points, but I can also see a sold 3 waves of rings after the switch turns on.

Nonetheless, it works.  I get 13.02 volts at both outputs, well within the 5% resistor tolerance, easily trimmed out, but no need since I'll be post-regulating this down to +12/-12 after I raise the output voltage a bit more to cover the headroom for the regulator.

Output load on both rails is 82ohms, ~160mA per rail.
13.02 / 82 * 2 = 317mA load + 2.5mA resistive load for feedback + about 100mA for the LM2588 itself = ~419 mA
13.02 * .419 = 4.45 watts

Input voltage is a steady 4.63v @ 1.81amps = 8.38 watts input

8.38 / 4.45 = 53%

Not great, but not terrible and doesn't bug me since all I want is the voltage and I'm driving this with a 4 cell pk of D's rated at 10000mAh (whether I can actually get the rated capacity, I'd doubt it, but I'll take what I can get).

Top of the head (e.g. bottom of the barrel) math says, assuming full load, max out the output rails at 300mA each (double what it is now), 50% efficient, double the 1.81 amp input, call it 3.7amps, add in another 500mA from the PICs, LCD, other circuitry, 4.2 amps, drawing off a 10AH battery pack, gives me about 2.4 hours with the numbers, maybe 1/2 that, a little over an hour semi-realistically (.4 C draw from the batt pack, inefficiencies, other loads, etc.), on a charge.
I'm good with that...especially since I don't anticipate the end unit to run at full load maybe 10% of the time at the very most...who knows...
Full load would entail the least amount of series resistance on the part I'm testing, all of the opamps putting out maximum current, LCD at full brightness with all pixels lit up, the PICs cranking away.

Now then...
MAN WAS I WRONG!  Or rather massively ignorant!  I now see, after reading thru the aforementioned documents, what a decent ground plane can do for you.
Even though my circuit is far from optimal (need compensation values tweaked, etc), it works, and I think I can assume it will work well once placed on a proper PCB with a generous amount of copper dedicated for a ground plane only.

Next up...wait for the batteries to fully charge up, add in more resistors to get it to a full 300mA per rail for a load, and see what it does.
(And small booster, low current, <100mA, for a USB charger to kinda trickle the battery pack up a bit while I'm connected to the PC.
Ya, 100mA max going into a dead 10,000 mAh batt pack.  That'll take awhile. Like ~7 days if the batt pack was completely dead, and the circuit was off, and I left it on the USB connector continuously.  A lot of IF's there.)

On another side note...on that last try with the circuit, the regulator heated up faster than the resistors and I generally shut everything off once it hit about 160F (100F above ambient).  Took about 3 minutes under light load to get to that point.
This time around, with the 82 ohm loads, it barely got warm.  Last reading I took after about 15 minutes of running was 80F...20F above ambient, and both resistors were up to about 150F.
It also draws 250uA (assuming that's anywhere near correct with my crappy meter) with the battery connected and the LM2588 "switched" off.

[Colfaxmingo]

Looks good!  Glad you could get it going.

[mariush]

Why would you need 4700uF/35v on the outputs? I don't think you would need that much.

Did you test how the esr of the capacitors affects the circuit? Have you tried using one or a couple of 820-1000uF 6.3v polymers for the 5v with this setup? And maybe a 1500-2200uF 16v low esr cap for the 12v?

[Skimask]

Just using stuff I had in the spare parts bins.
After I play with it a bit, I'll put parts in it closer to what they actually should be (e.g. 330uf on the outputs vs. 4700uf).

[Bored@Work]

You still haven't understood the prototyping technique.

You take the copper clad board, copper side up! Up as in that you look at it. And thats where it stays.

You solder some terminals or sockets to ground to have some standoffs for a start. For the supply rails you of course need to have them isolated from ground, but fixed to have some standoffs.

You directly solder parts to these connectors. When you have part that needs to have a pin connected to ground you  rotate it, bend the pin etc until you can solder that pin directly to the copper. No wire, no hole. Put the damned pin on the copper, solder it.

You take a part that needs to be connected to a non ground pin of that part. You directly solder the pins together, avoiding wires. If you need additional stand ups to hold non grounded parts in the air you solder a resistor in the gigaohm range to ground.

[Skimask]

Yes, I get the prototype technique which has been eluded to, both by the comments and the linked text......something solderable but insulated for standoffs as needed, direct connections between parts, yes, I get it.

But, with that being said, doing what I did worked, therefore, how much change do I actually need to make?
I've only got 6 extra interconnect wires as it is.  2 of them going directly to the ground, 2 of them running from the output caps to my load resistors (which if this were to be used in an actual circuit would like go to multiple other parts anyway), and 2 are actually interconnect wires that were needed because the pins on the coil itself are too short to do anything with.  4 of those wires are really needed because the coil's pins are only .1" long in the first place.

Yes, I could've done it with the copper side up.  I didn't want to do it that way.  I didn't feel comfortable with the copper side up as such.

Where did I go "wrong" again?

[Shuggsy]

Very good that you got it going! Good job. Layouts and schematics go hand-in-hand to make circuits work, and the more high frequency components (fast switching, even at "slow" rates for example) you have going around the more attention needs to be paid to good layout practice.

As for why you should do builds like this with the copper side up, well:

Ok...works!!!
Still got a bit of high frequency noise on the outputs, not a lot, surely not NEARLY as much as before.
The trace coming off the switch @ pin 5 is sort of recognizable, but doesn't resemble the trace in the datasheet.  I can see the switch points, but I can also see a sold 3 waves of rings after the switch turns on.

That speaks for itself. YES, your circuit works! And really, for this application that may be all you're after. If, however, find yourself working on another sensitive circuit with high frequency components later on and build it the same way then you may not be so lucky. In this case, it may not be "wrong" so much as just bad practice. Maybe best to do it the right way when it doesn't matter so much so you have experience when a circuit really needs to be built carefully.

One of Linear Technology's latest app notes deals specifically with layout considerations for power supply designs: Application Note 139 - Power Supply Layout and EMI. In their case, it's written for PCB manufacture instead of the dead-bug/copper-clad/Jim Williams style which is more for prototyping and one-offs. Rather than thin traces like a wire would essentially be, nearly all the layout is done using big, low-impedance copper pours. Attention is paid to where the major current paths are and how to best separate things.

Anyway, good job again on getting the switcher going! Very good troubleshooting experience.

[Skimask]

Can't disagree with any of that.
Changed the feedback resistors to get ~18-19v at the outputs, added a 7815 and a 7915 to each rail and loaded it to about ~225mA on each rail.
The post-regulated outputs look clean on the 'scope (Tek 2246A, A/C coupled, 20mv/div).  Shut the circuit down and keep the probes connected, and I can see the ambient noise here in the shop (concentrated around 58Khz and 60Hz) on the 'scope is pretty close to the noise that I'm seeing with the circuit powered up.  Hooked up an identical set of probes to Channel 2, grounded it to the PCB, left the tip 'hanging' a couple feet away from the PCB, selected add and invert, and the noise all but disappears, even down to 2mv/div.  Same thing with the circuit powered off.  So, I think I can conclude that the bulk of the post-regulated noise on the rails is due to the shop's ambient electrical noise.

For grins, bumped up the output to +/- 24.4v, still worked, but killed that battery pack fast!
Playing around...I'm having fun...

EDIT: Slightly confused on the efficiency calculations...
Input = 4.66v @ 1.81amps = 8.435 watts
Output#1 = 13v @ 82 ohms = 159mA = 2.067 watts
Output#2 = -13v @ 82 ohms = 159mA = 2.067 watts
Total output = 4.134 watts
Output / Input = 49%, which seems awful low...considering the noise, probably dead on.
Been running for an hour, heatsink on the LM2588 is steady at 62C

[Skimask]

Another note...
I had trouble believing myself, that I did something wrong in the first place (other than the obvious ground plane issues that I got schooled on!)...
So, I built up another one of these circuits last night, on the same type of protoboard I used 'back in the day' when I built up that 5v->12v boost converter roughly 10+ years ago.

It worked.

So now I've got three identical versions of the exact same circuit here...

1 - built up on a "protostack" pcb (http://www.protostack.com/boards/prototyping/prototyping-board-full-size-style-1).  Doesn't work, lots of noise, etc.

2 - built up on a piece of single sided copper clad PCB, dead bug style.  Works well, a little bit of noise, taken care of easily, and will be the type of construction I'll use in the "final" version.

3 - built up on a piece of protoboard similar to the protostack pcb, same layout as far as pins and traces go, but much older, maybe 10+ years old.  Put it together in the same manner as version #1 as far as parts placement and wires go.
This version works as well.  Just a tad bit more noise than version #2, but the input vs output numbers (volts, current draw, etc) match up between #2 and #3, at least as far as the resolution of my BK2709 can get.  And they both run about 3 1/2 hours loaded down on a charge from the battery pack before the LM2588 won't regulate any more (down to ~3.7v input voltage).

Therefore, all other things considered, I can only conclude that the 'protostack' boards are the issue.  Whether it's thinner copper, different substrate, I don't know since I don't have the spec's on the older protoboard.
I do know that the 'protostack' protoboards work well for 'purely' digital circuits...e.g. PICs running at 64Mhz (internal of course), with SD cards driving them at max SPI rates, LCDs, GPS, WIFI, Camera modules, fram chips, etc.

If it was just as easy to dead bug 40 pin PICs, I'd do the whole thing on the single sided board.  It's not, and I've got a stack of 7 protostack boards to use up.

[jerry507]

Eh, if you're having trouble believing all this layout stuff means anything then all you need to do is look around the datasheet more. They're getting inefficiencies of 75-90%. Look at their diagrams. If you're not getting waveforms that look like that, you're doing it wrong.

Yes, some of it will be ESR in your output caps (read the linked Linear app note and look at the current loops to see why these matter). The xformer you're using is from the datasheet, and will be what they used. All that ringing and stuff is layout related. Extra inductance in the leads, mostly.

You don't need a PCB to get optimal performance, and the dead bug style can do just as good if not better. But if you don't "believe" it, not much we can do.

[Skimask]

Yes, I know I'm doing it "wrong"...in a way.  That's what I'm saying.  But, what I'm also saying is that I know I had a similar project working using my "old methods" on those prototyping PCBs.  I just had to prove it to myself, mainly because back then I thought I knew a lot, and even if it worked, maybe I still had a LOAD of noise that I wasn't aware of.
I get the "goodness" of a good ground plane, component placement, cap ESRs, flux leakage, internal resistance, skin effect, current loops....All (ok, most) of that good stuff.
I'm eventually wanting this thing on a PCB so it can be fairly durable, more repeatable, something I can get more than PCB made up to play with in the future.
Like the Monkees said...I'm a believer.

[jerry507]

Well it "worked" on your old PCB too. It regulated and did stuff, it just died under load and was very inefficient. That's exactly what you'd expect to see from all the parasitic effects. You see the same stuff happen with poor quality transformers. All that ringing really messes with your circuit.

[Skimask]

One thing I can't find in the LM2588 datasheets is how to calculate the correct values for R(comp) and C(comp) coming off pin 2.
well, that and the need for a proper PCB design

[jerry507]

Assuming you're not changing the frequency (100khz default) by putting a resistor from pin 1 to ground, you don't need to change the compensation network.

[Skimask]

Nope, just the isolation circuit as described in the datasheet for the power on/off function, a diode and a switched resistor to Vin.

Did a few more runs of the circuit with the battery packs.  Output voltage = 13v, load = 82 ohms per rail = ~317mA total output load, a bit over 4.1 watts.
Using 4 cell D NiMH batteries, got a tad over 3 1/2 hours before the output dropped off @ 3.7v battery pack voltage.
Tried a 4 cell AA NiMH pack for the heck of it and it ran about 1/2 hour, close enough to what I had calculated.

I'm satisfied enough, except for the efficiency.  ~8.3 watts in , 4.1 watts out, a bit less than 50% efficient, and I can only account for about 2 watts of that using the formulas in the datasheet.  The rest is ringing, noise, etc.  Build it, rebuild it...  Eventually I'll get a configuration that takes care of business.

[ptricks]

One thing often overlooked on switching supplies is the input side, just using a battery on the input doesn't solve the problem.
A good pdf on switching designs and layout :
https://roboturk.googlecode.com/files/Boost%20Converter%20Design%20Tips.pdf

[Skimask]

More good reading material...

One note that hit me in the face in that document was the one about "always" putting a 1uf in parallel with any aluminum electrolytic to bypass the cap's inductance.
Google'd it, read up on it a bit, and of course the intarwebs agrees.
So, I added 1uf across the input and output caps.
Sure enough...  I've only got one battery run on it, so it's not by any means a statistical analysis, but after adding the caps, I got 4 hours on the charge before it went out of regulation.  Input current dropped to ~1.6 amps (from 1.81amps).  Calculated efficiency is up to about 55%.

Hey...every little bit counts.

[Psi]

75% efficiency should be possible without much trouble, so there's something causing your low 55% figure

I've not read this entire thread, so ignore me if this has been said before.

I would put a pot on the frequency input and graph the efficiency across the frequency range.
You'll get better efficiency if you tune the frequency to match your transformer.

[Skimask]

Possibly, but if you look at the circuit board I whipped up a couple of pages ago, I think you'll see why the efficiency is so low.  Ugly construction to say the least.
I've got really good inputs from here, and a bunch of tech doc's that I've read thru with good design tips and such.
For now, this circuit will work just fine for my application.  When I get around to making it properly, I'm sure the efficiency numbers will go up markedly.

[Shuggsy]

Before this topic gets buried with time, I'd like to add one more bit of reading material. I skimmed through the other pages and didn't see this one linked. Again, this is an app note from Linear Technology, but delves into the big world of switching regulators in a very nice, explanatory, and fun to read way. It was written by the late, great Jim Williams who has been mentioned previously in this thread. I highly recommend this app note as well as the entirety of app note 47, High Speed Amplifier Techniques, which was previously linked (much more than copper-clad build construction info inside it at almost 100 pages!). This app note is app note 27, Switching Regulators for Poets -- A Gentile Guide for the Trepidatious.

It doesn't go much into layout, but does make some mention of it (quote: "Layout is vital.") and some measures to ease the layout constraints. If you want good layout practice guides, app note 47 provides many examples as it consistently deals with high frequency techniques. App note 27 also goes over frequency compensation, an overall checklist for switching regulator designs, and common design issues. Being a Linear Technology app note, it does exclusively use Linear Tech parts, but the lessons can be applied to general switching regulator design.

Skimask, any chance we'll see some of the results from the circuit using this regulator's outputs? Also, perhaps you should edit the topic line to read something like"Still trying (...finally SUCCEEDING!)  with the LM2588... 5V -> +/- 12V" 

[Skimask]

Resulting circuit/project - that's gonna even more fun...or not...
This whole battery powered supply plus a couple of small linear regulators, charging circuit, etc., 2 DIP28 PICs, 1 DIP14 PIC, 3 DIP8 PICs, 3 TL084s, 2 6 channel digital pots, LCD w/ touch panel, 4x4 matrix keypad, SD card, USB, and 2 banks of 8 reed relays (unless somebody has a good part number for a digital pot that can handle + and - across the pot terminals, as well as up to 1 watt per digital resistor).

[Skimask]

Next issue...finding an opamp that'll do close to +/- 12v and 300mA at the outputs in a DIP8 package...
Literally hundreds to choose from.  Initial look says a TCA0372 looks like it'll fit the bill...

[Skimask]

I would put a pot on the frequency input and graph the efficiency across the frequency range.
You'll get better efficiency if you tune the frequency to match your transformer.

I'm gonna give this a try.  I'm using the recommended transformer, not the exact recommended output cap's or input cap's.
Couldn't hurt...

[Skimask]

Ok, first quick look at changing the frequency of the LM2588...

100k pot on pin #1 (freq adj / shutdown), 'scope hooked up to pin #5, the switch pin.  Alligator clip wire set is M.I.A., so couldn't get reliable current measurements.  Either that or power it off the PSU and read the current from the display.  Give me something to do tomorrow.

With the pot dialed around to 100K, sloppy waveforms on the 'scope.  Barely recognizable 100khz signal coming thru.
Rotate the pot (decreased resistance) and I can see the frequency increasing,  until I get about 3/4 the way around, and SNAP...the switch waveform gets clean, some bouncing, not much.
Not sure if that's a function of the iffy hold-knob on the 'scope though.  Could be that's the perfect frequency and timing for the hold-off itself.  And if I rotate the pot any farther, it goes into shutdown, just like it's supposed to.
Also, the frequency is about 208Khz at that "perfect point", a bit above the maximum stated in the datasheet, with "Rset" at about 8Kohm.

Even though I couldn't get a current measurement off the battery (due to missing alligator clip leads), while rotating the pot, the voltage on the output varies a few tenths of a volt, not much at all (cheap meter doing the reading for now), but more importantly, the A/C component of the output power starts out at about .6v at the lower frequencies and drops to about .1v when the waveform "snaps" into clear view at the higher frequency.  Spikes disappear, everything looks a lot smoother.  That might also be a function of the bandwidth at the input to the 'scope itself.  (Tek 2246A)
And, the regulator heat sink didn't get above 65C during this run, without a fan.  Before this, I had a 40mm fan blowing air across the heatsink to keep it from going into thermal overload shutdown @ >120C.

I'll take a video of the 'scope tomorrow while I'm messing with it.....if I get to it.

EDIT:  On another note, regarding current measurement on the meter.  Would I be better off measuring the current off the battery using the AC or DC mode?  Maybe put the biggest cap I can find across the battery pack leads?  It's not alternating current, but it's surely a varying current.  Maybe, since I'm using a cheap meter, in this case I'd be better off rather than trying to directly measure the current, I'd be better off measuring the delta of the current measured at the various frequencies.  For instance, I know I'm reading ~1.81amps using this cheap meter at 100Khz.  If I run the LM2588 up to 200Khz and the current drops to, say 1 amp, throwing out the effects of frequency on the measurement, I'll at least know I'm pulling ~.81amps less than before (and even that .81 amps probably isn't correct either, but less is less, a lot less is a lot less, and so on)...  I know I'll have the same problem using my bench PSU, current readout will vary, will the readout keep up?  will it average?  will it display garbage?  Won't know 'till I try it.

[Skimask]

Now I really feel like a dumbass...

C(comp) is supposed to be .47uf, e.g. markings should be 474, not .047uf (marking of 473)!

'scope traces on the switch, pre- and post- catch diodes, top of the load, Vin at the LM2588, are clear(er), not perfect, but resemble the waveforms in the datasheet now.
And more bonus points, the regulator didn't get above 60C during the 1/2 hour I ran it.

Used my power supply for juice.
4.6v @ 1.21 amps = 5.566 watts input
two rails of 12.8v @ 312mA amps = 3.9936 watts output
3.9936 / 5.556 = 71.9%
72% isn't the 90% the datasheet calls for, but wayyy good enough for me.

Also, the input amperage doesn't change by varying the frequency at pin#1 (freq adj), at least not as far as my psu shows (only shows down to 10mA).

So, dumbass award of the day goes to ME!

[Skimask]

More success -- ish...
Still not getting above 71% efficiency, but the circuit works.

Same circuit as before, same battery input, same 82ohm resistive load on each side positive and negative.

10000mah 4 cell D size NiMH battery pack, cutoff at 4v (1v/cell), started out at 5.1v under load, 1.1amps, ended at 4v under load, ~1.4 amps, 5.57 watts average input.
Continuous 12.8v at the output.  I could've ran the pack down to 3.7v before it cut itself off due to low input voltage, but I don't want to beat up on the battery pack too badly.
Ran 5 1/4 hours.  12.8v @ .312mA = 4 watts * 5.25 hours = 21 watt/hours.  Battery pack = ~4.4v @ ~1.21amps = 27.9 watt/hours...about 75% efficient.
Doing the math on the battery pack itself shows my run times just about match up to what I'm seeing.  Not too bad.

In the "real world", the maximum output load will be a DAC driven AC sine wave, so output current will average about 70% of the total, meaning that this 5.25 hours of run time should translate to about 7 hours of run time, assuming it's used for those 7 hours continuously.  Add in another 500mA draw from the rest of the circuit, drops the total run time down to ~6 2/3 hours.

I'm happy.  Now onto a switching regulator to drop the 4 cell pack down to 3.3v...

Thanks to all those that pointed me to good information/datasheets/white-papers/etc for tips, calculations, formulas, ideas (good and bad!), on how to properly design and set up switch mode power supplies in general.  I learned a crap load.

[Skimask]

Added a couple of PIC driven MOSFETs to the load to switch it on/off @ 100Hz, 70% duty cycle...
Added a 5W 8.1ohm resistor across the battery pack input for a "constant"-ish 590mA load...(simulating the rest of the circuitry that'll be in there)

Charged the battery pack..and BLAM...
6 hours & 45 minutes before the battery pack voltage dropped enough to pull the LM2588 out of regulation.
Not too shabby...