I want a cheap source for 60-100 white LEDs, to make some 24V DC LED modules of my own

I found this, looks pretty cheap : http://www.dealextreme.com/p/dx-white-48-led-light-bulb-110v-1475



Do any of you know :

Are the LEDs themselfes ok quality ? Do they have decent life and light output at 10-20mA ? (i'd guess that they failing in the user ratings on that page mostly due to the lots of LEDs together overheating and overdriving them to get high brightness, which won't happen in my setup)

Can this lamp be disassembled (to desolder the LEDs) without damaging the LEDs ? (ie is it supposed to be epoxy potted or something like that on the other side)

Is there any cheaper source for LEDs ? (In Ebay shipping is often more expensive than the LEDs themselfes)

I found this, looks pretty cheap : http://www.dealextreme.com/p/dx-white-48-led-light-bulb-110v-1475

Do any of you know :

Are the LEDs themselfes ok quality ? Do they have decent life and light output at 10-20mA ? (i'd guess that they failing in the user ratings on that page mostly due to the lots of LEDs together overheating and overdriving them to get high brightness, which won't happen in my setup)

These tend to fail pretty soon, the safe current would be round 5mA, no more (original is in already the 10..20mA range). And with that the light output would be to small. Moreover these 5mm LED's are too directional for use as a general light in a room.

Can this lamp be disassembled (to desolder the LEDs) without damaging the LEDs ? (ie is it supposed to be epoxy potted or something like that on the other side)

They look like only soldered to a PCB and tis snapped into the plastic shell.

Is there any cheaper source for LEDs ? (In Ebay shipping is often more expensive than the LEDs themselfes)

I would rather use these .  With a DCDC convertor on each of them (MC34063 based), that would be way more reliable then the 5mm LED setup.
To mount them use a 1.5mm PCB, drill a hole of the diameter of the LED body, make a connection pattern for complete SMD assembly (you can not use trough-holes, as on the other side would be a heatsink). Straighten LED's connection "wings" so, when soldered to the PCB, the heat slug is on the other side. Then mount the on top of the Al metal plate (~2mm thickness) or other metal object to serve as a heatsink, use thermal paste for better connection.

I am looking for the simplest solution - LEDs that dont require heatsinks, and ballasted by a resistor (on the output of LM317 regulated power supply or stack of AA batteries)

With triple AA batteries and single ~4.7Ohm resistor you could directly power that LED. The DCDC would be necessary to operate at the 24V AC/DC...

Regular 5mm LED's are very bad in reliability, I really discourage you from using them for illumination purpose (if the ON time is longer then few seconds).

A DCDC converter in any fixture is prohibitively expensive for the project

For the main room (3 fixtures), i could perhaps wire the 3 LEDs in series and kill the rest on a resistor - even a small one if i change the system from 24V to 12V so not much energy loss

For the other rooms, using just 3.4V out of 24 or 12V with a resistor is waste of precious energy when running on battery. I'd totally prefer to wire a series of 5mm LEDs (or few strings) to use up as much voltage as possible, leaving as little as possible (just enough to allow for some volt drop as the batteries are used up) for the resistor



For 5mm LED reliability, my observations untill now were :

 - At many gadgets they are running at 30mA and higher which make them de fast. In a USB powered lamp i have, the 5mm white LED was driven at 50mA and died (light output dropped, looks like only 1 corner of the chip is working under the phosphor) in 2 weeks or so of infrequent use. I replaced the LED and resistor to make it work at about 15 mA and had no problems with it, though it is used very little

 - At a flashlight powered by D cells i bought cheap, the 13 LEDs were wired in parallel (isnt it bad ??) on few ohm resistor, runing at about 250 mA total so about 20 mA / LED. LEDs were bright but warm after a while. I changed the resistor to drop it to about 10mA / LED and it is working perfectly since (and LEDs stay cool), though used infrequently (still running on the same batteries from 2008)

So if i run them at 10mA or less (compensating with more LEDs to get the same light output) so that they stay cool to touch is there really a problem ?



Also i found this, looks better than disassembling a LED lamp : http://www.dealextreme.com/p/14000mcd-100packs-white-5mm-led-1108

The user responses are clearly from users who try the LEDs for 5 seconds without resistor so pretty meaningless for me though

Using a resistor when you have used many lamps in series is not good, better is to use a constant current source, which will provide a non varying current as the battery voltage drops. Using 5 LED in series means the resistor has to drop 10V on a charging battery or 5V on a nearly flat battery. Thus you either overdrive or underdrive the LED string at some point, depending on your basis for calculating the value. If you use a simple current sink ( easiest to make with a TL431 and a few cheap resistors and a power transistor of very moderate cost, you may be able to get all from discarded electronic equipment using a little work and some soldering) the lamp string will have essentially no change until the battery is totally discharged. As a plus you can switch out LED units with a shunt switch to turn them off if desired with no effect on the other units, as the current source will compensate.

See www.fairchildsemi.com/ds/TL/TL431A.pdf page 6 for a simple application note describing the constant current source.

Thats a good option and i have enough stuff to get 431's from (blown up comuter PSU's, they are there in the 5VS optocoupler and 3.3V sense circuits). Perhaps i can even find a LM317 in something else which needs just the resistor to be a current source

However, i am unsure whether it is even so important : Driving the LEDs briefly on a new battery (untill it drops a bit) won't harm them immediately, and as it drops the inhabitants get used to the dark (vs the lots of light from the 240V incandescents) so they see better with time despite the actual amount of light going down

My biggest concern is the LEDs. I know how those LEDs fail (and i found enough horror stories about the high power LEDs decaying too, written by DIYers across the net). I want to use them anyway as they are so cheap, and just drive them at lower current to make them last long and be reliable (and just use more LEDs to get the wanted amount of light, still cheaper than the 1W LEDs). But i dont know which current i want if i want them to last X hours (like, 40K hours for continuous 5 years)

The MC34063 is cheap as dirt (about $0.6 here, coil for $0.5, other components are per few cents), the DCDC would be below $2 on material cost. But that make sense for higher wattage (e.g. 3x"1W", run it on ~200mA, so 2W total, giving about 180lm)
Or I just find this driver, it would work on both 12V AC/DC...

For the 5mm LED's (10mA) I would use two transistors (the same circuit as the TL431 on fig 14 in the datasheet, but use an NPN instead of the IC, it suffice with lower voltage, so there may be about a LED extra in the chain). It would have a negative temperature coefficient.

But to get the same light with 5mm LED's would be more expensive then with the "1W" ones, even when using the "expensive" DCDC regulator. To get the same power with 10mA drive, you would need about 30 pieces, what is for about $3. And the light output would be way less...

Problems with high power LED's is, then people frequently do not use heatsinks at all...

The issue with the 1W LED is that it is too much for some places, where a few 5mm's are sufficient

I can underpower the 1W LED (perhaps to the point where it won't even need heatsink) but then i waste too much power for the resistor / linear regulator (could us up more volts with a series string of 5mm's)

A cost of 2$ / fixture for the regulators is about 2 times more than what the LEDs themselfes cost, so it would triple the cost of the system - WAY too much. Linear regulators are ok as i have a source for them (blown computer power supplies in this case)



A side thought : I could make a multivibrator with 2 small NPN's (like 2SC945) and resistors / capacitors, so get a primitive PWM driver for the LEDs (high frequency, high current, short duty cycle). How about that ?

Some of the LED drivers on DX can be dimmed with some simple modifications (without external PWM generator, altough that is also possible). Those in Medved's link probably use the same or similiar driver IC as the one described in This thread.. These drivers are very efficient, but in my experience (and others) they do generate a bit of radio interference.

A side thought : I could make a multivibrator with 2 small NPN's (like 2SC945) and resistors / capacitors, so get a primitive PWM driver for the LEDs (high frequency, high current, short duty cycle). How about that ?

It have to have a means to stabilize the current. With a fixed duty ratio you create only a "transformer" with fixed voltage conversion ratio.
The chips in these converters use the ON time terminated not by fixed time, but by the current reaching the fixed threshold. This ensure, then the duty ratio adopt so, the current stay constant.
It is indeed possible to make the same with few bipolars, but you would have to deal with more components. But if you have plenty of them, you may indeed end up cheaper.

This circuit is a switching current regulator, it uses a few common parts but also an inductor, a fast diode (not a standard rectifier) and a medium power PNP. I have not built this, but a very similar constant voltage version is also on that website (not the same page) which I tested, and it worked quite well before I accidentally burned it out.

This circuit is a switching current regulator, it uses a few common parts but also an inductor, a fast diode (not a standard rectifier) and a medium power PNP. I have not built this, but a very similar constant voltage version is also on that website (not the same page) which I tested, and it worked quite well before I accidentally burned it out.

One thing I do not like about that circuit (and many similar):
The operation of that circuit is strongly linked to the equivalent series resistance of the C2, what being parasitic, it is everything, but under control. Under some conditions (lower temperature, C2 aging) the oscillation may stop even for higher voltages and the circuit enter the linear regulation mode, what mean the power dissipation on the Q3 become huge (compare to normal, switching operation). At the same tme the frequency may drop below reasonable control, what would cause higher peak current, eventually discontinuous current mode of operation, what stress mainly the inductor.

Way better is to use the true hysteretic current control, what mean one resistor and one extra transistor extra and remove the C2 (or reconnect it just plain parallel to LED's), while all the circuit become reasonably under control.
The example I put here , added are Q3 and R5. The R2 is reconnected. Another improvement is the well controlled base drive of the Q3, so way lower losses in resistors, mainly on higher input voltages.

If they are available in Israel, Velleman make a high power LED driver kit (K8071) that can drive up to 4 1W LEDs.  Over here they are around $15 to $16 each.

i get my solar leds from junk solar powered yard lights. i run my leds 3 in series on a resistor and 10 volt zener diode. to it take what it draw and nothing goes up in smoke. even when my batteries EQing.
junk box parts on hand so it cost me nothing.