der_wille_zur_macht
Team RC
I'm writing out a way too long response to your question about setting current but I need breakfast first, so stay tuned. 
DIY LED driver for reef lighting
- Thread starter der_wille_zur_macht
- Start date
der_wille_zur_macht
Team RC
Coffee consumed. Now I can think clearly. As with all my posts in this thread, if any of the more experienced EE types see error in my logic, PLEASE point it out! I'm a self-taught hobbyist and the below is based on experimentation and my interpretation of the supporting docs for this chip.
And while I'm on the subject, a big thanks to the various early LED pioneers that got me hooked on this stuff in the first place (Soudwave, stugray, liveforphysics) and kcress for answering a lot of basic EE questions in threads and via PM when I was first building these drivers.
Bear with me for a moment. I feel a quick overview of the chip's function might help those (like me) who are interested in knowing how things work. You don't need to understand all this to do this project though. I don't want to scare anyone away - if you just want the answer, skip to the bottom.
So, this chip is essentially a switching voltage regulator that operates in a constant current mode. The Rsense resistor is exposed to the current flowing through the LEDs, so the chip can determine that current by reading the voltage drop across Rsense. It switches current to the LED array on and off to keep this voltage at a preset number (.235v). That feedback voltage is preset in the chip, so to adjust the current it equates to (and hence the current through the LEDs) you just change values for Rsense. To find out what value a particular Rsense will result in, V = I/R: .235v = Iled/Rsense.
However, that's only part of the story. Since this is a switching reg, the internal switch is basically turning voltage on and off. We don't want the LEDs to see that on/off waveform - we want them to see a constant voltage at the correct value to maintain the desired current.
Regulators for LED use like this chip can generally operate in three ways. The difference in operation is in the relationship between input voltage and output voltage. The difference in implementation is where the inductor is in the circuit in relation to Vin and the switch.
1) Buck: The regulator chops down the input voltage to keep the LEDs at the desired current. In a buck regulator, the output voltage must always be LOWER than the input voltage. Think of it this way: The input voltage is trying to spike the LED array too high, so the regulator chops it down to keep it correct.
2) Boost: The regulator boosts up the input voltage. The desired output voltage must always be LOWER than the input voltage. Think of it this way: The input voltage is not high enough, so the regulator switches it higher to keep the LEDs happy.
3) SEPIC: It doesn't matter. SEPIC topology uses two inductors, one in each position, so the voltage can be anything (in range of the chips min/max) compared to the desired number.
Buck and boost are much more efficient, but SEPIC is much more flexible. In a situation where input voltage might be all over the map (think: LED lighting in automobiles, where system voltage is pretty wild) SEPIC is good. But in our builds, we generally have DC power supplies that are nearly dead-stable, so we don't need the flexibility. Hence that leaves buck and boost. Boost is a little more efficient and allows for higher LED count on a given input voltage, but buck allows for more resolution when dimming. I picked boost for this design, mostly due to the efficiency.
Ok, moving on. In either buck or boost topology, the chip's switch basically creates a square wave that we don't want to expose the LEDs to. That's where many of the other components in the design come in to play - the inductor stores and releases energy in it's magnetic field as the switch goes on and off, allowing constant conduction instead of harsh on/off. The big cap on the output side further smooths the ripple, protecting the LEDs, and the big cap on the input side smooths the ripple the power supply is exposed to.
Reviewing the rest of the components: The little tiny cap on input and output just cancel high frequency noise, and the timing cap (on TCAP pin) sets the frequency of operation for the internal switch. Out of the major components, that leaves R1, which is a current limiting resistor for the chip - to basically limit the max peak current that can flow through it.
Now that we know what the different components do, we can have a thorough discussion of setting current. There are two things we can change related to current:
1) Drive current From above, we can adjust Rsense to adjust the drive current. What we're really doing here is adjusting the average drive current, since there will always be some ripple. Luckily, we can also change. . .
2) Current ripple We know that there's going to be some ripple on the output, and we can change the intensity and frequency of that ripple. To a certain extent, average drive current (#1) is the most important adjustment, but to be sure you're not frying either the chip or the LEDs you need to understand the ripple, too. There are lots of components that affect the ripple:
1) R1, the current limiting resistor. In normal operation, this resistor doesn't come into play, but if your design has a lot of ripple and/or a very high drive current, the chip will read that through this resistor and chop the current down. Basically the ripple will be a sawtooth wave, and this resistor will hack the tops off the sawteeth. This protects the chip from overcurrent (the chip's max is 1.5A). But if your driver is ever operating with this occuring, you probably won't be reaching your desired average drive current. IMHO it's better to design with an average current and ripple that don't allow this to happen.
2) C1, the big cap on the output side. This cap smooths the output so the LEDs see a less-harsh wave.
3) The inductor. A larger inductor means lower peak current through the driver.
4) The timing capacitor. This sets the frequency of operation for the chip's switch, which basically means it sets the frequency of the ripple in the output. A higher frequency means less ripple, but the chip's switch has a pretty narrow window of operation.
5) Potentially, using an external switch. You don't HAVE to use the built in switch - you can use it as a trigger for an external switch. This gives you much more freedom in the design, because you can use a larger frequency range, and not worry about heat dissipation in the IC itself. IMHO though this isn't really required unless you want very high drive currents, which aren't typical for our reef lighting applications.
6) The ratio of input voltage to output voltage. If they are close together, the ripple is lower, because there's less switching to do. IMHO this is the single most important factor in these designs because it ALSO effects overall efficiency - the lower the difference, the higher the efficiency of the IC itself, AND of the entire design, because there will be less losses through the external components, AND they can be smaller (less ripple!) So, when doing these drivers, figure out what your voltage drop across the LEDs will be, and use a power supply close to that value (above or below depending on buck or boost.)
You want low output ripple for a few reasons. The frequency of the ripple will always be above what we can see, so luckily there's no visual impact. Mainly it's because we want to keep from frying anything with peak currents being too high. The chip itself has a max of 1.5A, but the LEDs we're using typically have a max of 1A or so. Hence, if we raise drive current close to those limits, we NEED to limit ripple to prevent our LEDs and ICs from being exposed to peak currents beyond their limits.
Generally, it might seem like you want to just max all the above variables to get the least output ripple, but they alter the design in other ways (efficiency, etc) so you need to choose carefully. As I noted above, keeping input voltage close to desired output voltage is like free money, so do that first. All the other variables (increasing frequency, increasing inductor and output cap size) drop efficiency, so use them sparingly.
Circling back to the heart of the matter - I chose the component values above to give reasonably efficient performance at 500mA drive current. OnSemi actually has a spreadsheet on their website that will let you play with different component values and see what the resulting drive current and ripple is. If you're going for a higher drive current, you'll probably want to make some changes, to keep from spiking the LEDs too high. I'd probably start by increasing size of the inductor. The size I'm using (100uH) allows for 30% ripple at 700mA, which brings you pretty close to 1000mA in the spikes. I'd probably jump up to 150uH for 700mA, which drops ripple to 20%. These values are for input voltage of 24v and output of 28v (about what 8 LEDs will want). Like I mentioned above, it's also worth keeping the input voltage close to the output voltage. Many good DC power supplies have a trimpot that let you adjust the voltage. If you had a 24v supply and turned it up to 26v, you'd get 20% ripple with the 100uH inductor, which means more efficiency (the 100uH will have less resistance) and you'll have an overall increase of efficiency in general because less switching will be required.
The only downside to having a supply voltage very close to the output voltage is that for those of us that want dimming, you lose resolution. The chip cannot cut current below the input voltage, so the input voltage defines the "floor" of your dimming capabilities. What this means is that if you're driving 8 LEDs on 24v at 500mA and they require 28v to hit that target, you'll only be able to dim them down to 24v, which probably equates to 100mA drive current or so. It's pretty dim, but definitely not "off". For most of us this is probably fine.
Another note on circuit design. The cap I'm using on the input side is very big. I did this because I plan on having several of these per DC power supply, and I didn't want to hammer the PSs too hard by all the on/off switching. I've tested several of these running on the same PS and it works totally fine, so I probably have a too-large safety margin. You could probably use a lower value cap if you really wanted to.
The cliff notes:
1) If you want 350mA or 500mA drive current, do my design above. On a 24v supply, it'll run 8 LEDs at the selected current.
2) If you want 700mA, bump the size of the inductor up to 150uH.
3) In general, keep the power supply voltage close to (BUT NEVER OVER) the desired output voltage - unless you need really fine control over low-range dimmability.
I'm a self-taught hobbyist and the below is based on experimentation and my interpretation of the supporting docs for this chip.And while I'm on the subject, a big thanks to the various early LED pioneers that got me hooked on this stuff in the first place (Soudwave, stugray, liveforphysics) and kcress for answering a lot of basic EE questions in threads and via PM when I was first building these drivers.
Bear with me for a moment. I feel a quick overview of the chip's function might help those (like me) who are interested in knowing how things work. You don't need to understand all this to do this project though. I don't want to scare anyone away - if you just want the answer, skip to the bottom.
So, this chip is essentially a switching voltage regulator that operates in a constant current mode. The Rsense resistor is exposed to the current flowing through the LEDs, so the chip can determine that current by reading the voltage drop across Rsense. It switches current to the LED array on and off to keep this voltage at a preset number (.235v). That feedback voltage is preset in the chip, so to adjust the current it equates to (and hence the current through the LEDs) you just change values for Rsense. To find out what value a particular Rsense will result in, V = I/R: .235v = Iled/Rsense.
However, that's only part of the story. Since this is a switching reg, the internal switch is basically turning voltage on and off. We don't want the LEDs to see that on/off waveform - we want them to see a constant voltage at the correct value to maintain the desired current.
Regulators for LED use like this chip can generally operate in three ways. The difference in operation is in the relationship between input voltage and output voltage. The difference in implementation is where the inductor is in the circuit in relation to Vin and the switch.
1) Buck: The regulator chops down the input voltage to keep the LEDs at the desired current. In a buck regulator, the output voltage must always be LOWER than the input voltage. Think of it this way: The input voltage is trying to spike the LED array too high, so the regulator chops it down to keep it correct.
2) Boost: The regulator boosts up the input voltage. The desired output voltage must always be LOWER than the input voltage. Think of it this way: The input voltage is not high enough, so the regulator switches it higher to keep the LEDs happy.
3) SEPIC: It doesn't matter. SEPIC topology uses two inductors, one in each position, so the voltage can be anything (in range of the chips min/max) compared to the desired number.
Buck and boost are much more efficient, but SEPIC is much more flexible. In a situation where input voltage might be all over the map (think: LED lighting in automobiles, where system voltage is pretty wild) SEPIC is good. But in our builds, we generally have DC power supplies that are nearly dead-stable, so we don't need the flexibility. Hence that leaves buck and boost. Boost is a little more efficient and allows for higher LED count on a given input voltage, but buck allows for more resolution when dimming. I picked boost for this design, mostly due to the efficiency.
Ok, moving on. In either buck or boost topology, the chip's switch basically creates a square wave that we don't want to expose the LEDs to. That's where many of the other components in the design come in to play - the inductor stores and releases energy in it's magnetic field as the switch goes on and off, allowing constant conduction instead of harsh on/off. The big cap on the output side further smooths the ripple, protecting the LEDs, and the big cap on the input side smooths the ripple the power supply is exposed to.
Reviewing the rest of the components: The little tiny cap on input and output just cancel high frequency noise, and the timing cap (on TCAP pin) sets the frequency of operation for the internal switch. Out of the major components, that leaves R1, which is a current limiting resistor for the chip - to basically limit the max peak current that can flow through it.
Now that we know what the different components do, we can have a thorough discussion of setting current. There are two things we can change related to current:
1) Drive current From above, we can adjust Rsense to adjust the drive current. What we're really doing here is adjusting the average drive current, since there will always be some ripple. Luckily, we can also change. . .
2) Current ripple We know that there's going to be some ripple on the output, and we can change the intensity and frequency of that ripple. To a certain extent, average drive current (#1) is the most important adjustment, but to be sure you're not frying either the chip or the LEDs you need to understand the ripple, too. There are lots of components that affect the ripple:
1) R1, the current limiting resistor. In normal operation, this resistor doesn't come into play, but if your design has a lot of ripple and/or a very high drive current, the chip will read that through this resistor and chop the current down. Basically the ripple will be a sawtooth wave, and this resistor will hack the tops off the sawteeth. This protects the chip from overcurrent (the chip's max is 1.5A). But if your driver is ever operating with this occuring, you probably won't be reaching your desired average drive current. IMHO it's better to design with an average current and ripple that don't allow this to happen.
2) C1, the big cap on the output side. This cap smooths the output so the LEDs see a less-harsh wave.
3) The inductor. A larger inductor means lower peak current through the driver.
4) The timing capacitor. This sets the frequency of operation for the chip's switch, which basically means it sets the frequency of the ripple in the output. A higher frequency means less ripple, but the chip's switch has a pretty narrow window of operation.
5) Potentially, using an external switch. You don't HAVE to use the built in switch - you can use it as a trigger for an external switch. This gives you much more freedom in the design, because you can use a larger frequency range, and not worry about heat dissipation in the IC itself. IMHO though this isn't really required unless you want very high drive currents, which aren't typical for our reef lighting applications.
6) The ratio of input voltage to output voltage. If they are close together, the ripple is lower, because there's less switching to do. IMHO this is the single most important factor in these designs because it ALSO effects overall efficiency - the lower the difference, the higher the efficiency of the IC itself, AND of the entire design, because there will be less losses through the external components, AND they can be smaller (less ripple!) So, when doing these drivers, figure out what your voltage drop across the LEDs will be, and use a power supply close to that value (above or below depending on buck or boost.)
You want low output ripple for a few reasons. The frequency of the ripple will always be above what we can see, so luckily there's no visual impact. Mainly it's because we want to keep from frying anything with peak currents being too high. The chip itself has a max of 1.5A, but the LEDs we're using typically have a max of 1A or so. Hence, if we raise drive current close to those limits, we NEED to limit ripple to prevent our LEDs and ICs from being exposed to peak currents beyond their limits.
Generally, it might seem like you want to just max all the above variables to get the least output ripple, but they alter the design in other ways (efficiency, etc) so you need to choose carefully. As I noted above, keeping input voltage close to desired output voltage is like free money, so do that first. All the other variables (increasing frequency, increasing inductor and output cap size) drop efficiency, so use them sparingly.
Circling back to the heart of the matter - I chose the component values above to give reasonably efficient performance at 500mA drive current. OnSemi actually has a spreadsheet on their website that will let you play with different component values and see what the resulting drive current and ripple is. If you're going for a higher drive current, you'll probably want to make some changes, to keep from spiking the LEDs too high. I'd probably start by increasing size of the inductor. The size I'm using (100uH) allows for 30% ripple at 700mA, which brings you pretty close to 1000mA in the spikes. I'd probably jump up to 150uH for 700mA, which drops ripple to 20%. These values are for input voltage of 24v and output of 28v (about what 8 LEDs will want). Like I mentioned above, it's also worth keeping the input voltage close to the output voltage. Many good DC power supplies have a trimpot that let you adjust the voltage. If you had a 24v supply and turned it up to 26v, you'd get 20% ripple with the 100uH inductor, which means more efficiency (the 100uH will have less resistance) and you'll have an overall increase of efficiency in general because less switching will be required.
The only downside to having a supply voltage very close to the output voltage is that for those of us that want dimming, you lose resolution. The chip cannot cut current below the input voltage, so the input voltage defines the "floor" of your dimming capabilities. What this means is that if you're driving 8 LEDs on 24v at 500mA and they require 28v to hit that target, you'll only be able to dim them down to 24v, which probably equates to 100mA drive current or so. It's pretty dim, but definitely not "off". For most of us this is probably fine.
Another note on circuit design. The cap I'm using on the input side is very big. I did this because I plan on having several of these per DC power supply, and I didn't want to hammer the PSs too hard by all the on/off switching. I've tested several of these running on the same PS and it works totally fine, so I probably have a too-large safety margin. You could probably use a lower value cap if you really wanted to.
The cliff notes:
1) If you want 350mA or 500mA drive current, do my design above. On a 24v supply, it'll run 8 LEDs at the selected current.
2) If you want 700mA, bump the size of the inductor up to 150uH.
3) In general, keep the power supply voltage close to (BUT NEVER OVER) the desired output voltage - unless you need really fine control over low-range dimmability.
Right on Wille!
Thanks for the explanation....that's exactly what I was looking for. You explained things perfectly. I think I'll end up buying both the 100uH and the 150uH and then if I find I'm not getting the light deep enough I can put the larger one in..That's the beauty of doing the drivers yourself...one can play.
So would it also not be an idea to increase the size of the filter cap on the output to smooth the ripple a little more if one bumps up the current...or possibly add another 100mF? Most powersupplies have multiple filter caps on the output to provide a cleaner DC. Would doing this just be overkill given the amount of extra smoothing that would be gained?
Sorry if my questions seem basic! I've been out of electronics for many years! My understanding of the principals is pretty sound but my experience is a distant memory!!
Don
Thanks for the explanation....that's exactly what I was looking for. You explained things perfectly. I think I'll end up buying both the 100uH and the 150uH and then if I find I'm not getting the light deep enough I can put the larger one in..That's the beauty of doing the drivers yourself...one can play.
So would it also not be an idea to increase the size of the filter cap on the output to smooth the ripple a little more if one bumps up the current...or possibly add another 100mF? Most powersupplies have multiple filter caps on the output to provide a cleaner DC. Would doing this just be overkill given the amount of extra smoothing that would be gained?
Sorry if my questions seem basic! I've been out of electronics for many years! My understanding of the principals is pretty sound but my experience is a distant memory!!
Don
cutting cost
cutting cost
Seeing how DIY drivers will cut costs, I was wondering if anyone has found a different material to mount these lights to. Ive got a 150g tank and i would need 3 8.5x22 aluminum heat sinks costing around 175 dollars. I plan on using 72 cree lamp 32 white and 32 royal blue with some version of a DIY driver. Any ideas on cutting costs on these large builds would be great.
cutting cost
Seeing how DIY drivers will cut costs, I was wondering if anyone has found a different material to mount these lights to. Ive got a 150g tank and i would need 3 8.5x22 aluminum heat sinks costing around 175 dollars. I plan on using 72 cree lamp 32 white and 32 royal blue with some version of a DIY driver. Any ideas on cutting costs on these large builds would be great.
der_wille_zur_macht
Team RC
Right on Wille!
Thanks for the explanation....that's exactly what I was looking for. You explained things perfectly. I think I'll end up buying both the 100uH and the 150uH and then if I find I'm not getting the light deep enough I can put the larger one in..That's the beauty of doing the drivers yourself...one can play.
So would it also not be an idea to increase the size of the filter cap on the output to smooth the ripple a little more if one bumps up the current...or possibly add another 100mF? Most powersupplies have multiple filter caps on the output to provide a cleaner DC. Would doing this just be overkill given the amount of extra smoothing that would be gained?
Sorry if my questions seem basic! I've been out of electronics for many years! My understanding of the principals is pretty sound but my experience is a distant memory!!
Don
Keep in mind that the size of the inductor just makes the ripple a different shape - it's not what actually sets the current (Rsense does). In theory you could run a high current on a small inductor, but you'd have more ripple. If you think you might switch, I'd just build a proto of the driver both ways and see if there's any perceptible difference, and/or just go with the bigger one.
Regarding your cap question, you're starting to get beyond my range of knowledge - I don't think the cap size matters TOO much as long as it's about in the right range for the frequency of the noise you're trying to cancel. Too big and it'll react too slowly for the frequency of the ripple. I'm sure there are calculators out there that can help size caps for certain applications, I basically followed guidelines in the docs for this chip, staying a bit on the big size. You can technically run this without caps on the output side according to the app notes, but that would strike me as a noisy solution.
Also, bigger caps = more resistance = less efficiency!
Look around metalworking supply shops in your area. I bet you could find some thick angle aluminum (U channel maybe?) that would work fine. IMHO most people are fairly overkill on heatsinking these things, especially if you stick with reasonable drive currents and use fans. Though overkill is better than not enough cooling!
der_wille_zur_macht
Team RC
Steve,
I have some photos of the drivers "in action" but am trying to sort through them before posting. I've lent a few LED fixtures based on these drivers out locally and I'm waiting for some photos of them in action on those tanks, too. Soon though hopefully.
As far as replacing MH, as you can imagine, it can be tough to judge. Different MH will have different output, different LEDs will have different output. Based on harvesting data from the 10 - 20 people who have reported results in the various LED megathreads linked at the beginning of my post, IMHO it's fair to assume that an LED fixture will use about 40% of the energy of a comparable traditional fixture, if done right. Like I said though, there are a TON of variables - which LED you buy, drive current, spacing, color choice, brightness bin, etc. The best LEDs commonly available (Cree XR-E Q5 cool white or Luxeon Rebel 100 bin cool white) are around 90 - 100 lumens/watt at their best. An "average" HP LED might only be 40 or 50 - so right there you've got 100% variability! And unlike MH or T5, LED is very easy to dim or drive at whatever wattage you want (within limits) so even among fixtures built from the same components, output can be extremely variable.
I think it's going to take some time for our hobby to really get comfortable with the fact that LEDs are like opening Pandora's box - you get total control over every aspect - color, intensity, spectrum, angle, etc. down to a pretty fine resolution. It would be really easy to have different combinations of those variables in different areas of the same tank. And with a microcontroller, you can alter those variables automatically, on the fly, throughout the day. The possibilities are really endless. That's what's attracted me to this, at least as much as the energy savings.
And to get back on track with your question. I'll be using LEDs on my 360g tank. If I didn't use LEDs, I'd probably have 1200 - 1500w of conventional lighting (probably 4 250w MH and some supplemental.) Let's say 1200 to be fair. The LED fixture I'm planning on will PEAK at about 500w, and average 400w throughout the photoperiod. (that's around 200 LEDs driven at a max of ~600mA, most likely). With a 10 hour photoperiod, and 15 cents per kWh, the annual price of the traditional fixture would be $657 in electricity and with $75 x 4 MH lamps replaced annually and $150 of supplemental lamps annualy; $450 in lamp costs annually. That's $1107 annually to operate. If the fixture cost me $1000 to DIY, the 10 year total cost of ownership is $12,070. (Everyone who's thinking about a big tank - put that in your pipe and smoke it a bit!)
If we assume the LED fixture is run flat-out at 500w for the same 10 hour photoperiod, the annual cost is $273.75 - let's call it $275. If the LED fixture costs me $2k to build ($10 per LED, which is easy to do if you're careful), that's a 10 year total cost of ownership of $4750.
So, in 10 years, I'll save $7,320 in operating costs by going with the LEDs. Of course, no one's run LEDs for 10 years on a marine tank. :lol: It's still fairly experimental, but I'm either confident enough - or enough of a nut - that I'm willing to try it. Even if it only lasts 5 years and falls to bits, I will have still saved money. But based on failure data from the LED manufacturers, 10 years shouldn't be a problem - technically, the LEDs should be at 70% output after around 50,000 hours at 700mA and junction temps of 125c. (lifted from my memory, someone can check the datasheets if they want to verify.) I'll be running below that current and temp, so I should get more than that. 50k hours at 10 hrs/day is almost 14 years. And even at that point, they'll "only" be at 70% output, so I can just run 70% more current through them. OR, replace them all (some will inevitably fail along the way) with the more efficient LEDs sure to be out at that point in time. Even if you factor in replacement cost for 100% of the emitters in the cost estimates above over 10 years, the LEDs are still several thousand dollars cheaper to run. And, already, the Cree XP-G has been released, which has about 30% more efficiency than the LEDs I mentioned above, and the industry isn't slowing down, so they should only get better in the future.
Once I have the material gathered on the prototypes and nano fixtures I've been playing with over the last few months I'll post that too, but in the meantime there's probably a few weeks' worth of reading in the big monster threads I posted at the beginning. And once I build the big rig for my 360g, it'll be fully documented in my build thread for that tank.
I have some photos of the drivers "in action" but am trying to sort through them before posting. I've lent a few LED fixtures based on these drivers out locally and I'm waiting for some photos of them in action on those tanks, too. Soon though hopefully.
As far as replacing MH, as you can imagine, it can be tough to judge. Different MH will have different output, different LEDs will have different output. Based on harvesting data from the 10 - 20 people who have reported results in the various LED megathreads linked at the beginning of my post, IMHO it's fair to assume that an LED fixture will use about 40% of the energy of a comparable traditional fixture, if done right. Like I said though, there are a TON of variables - which LED you buy, drive current, spacing, color choice, brightness bin, etc. The best LEDs commonly available (Cree XR-E Q5 cool white or Luxeon Rebel 100 bin cool white) are around 90 - 100 lumens/watt at their best. An "average" HP LED might only be 40 or 50 - so right there you've got 100% variability! And unlike MH or T5, LED is very easy to dim or drive at whatever wattage you want (within limits) so even among fixtures built from the same components, output can be extremely variable.
I think it's going to take some time for our hobby to really get comfortable with the fact that LEDs are like opening Pandora's box - you get total control over every aspect - color, intensity, spectrum, angle, etc. down to a pretty fine resolution. It would be really easy to have different combinations of those variables in different areas of the same tank. And with a microcontroller, you can alter those variables automatically, on the fly, throughout the day. The possibilities are really endless. That's what's attracted me to this, at least as much as the energy savings.
And to get back on track with your question. I'll be using LEDs on my 360g tank. If I didn't use LEDs, I'd probably have 1200 - 1500w of conventional lighting (probably 4 250w MH and some supplemental.) Let's say 1200 to be fair. The LED fixture I'm planning on will PEAK at about 500w, and average 400w throughout the photoperiod. (that's around 200 LEDs driven at a max of ~600mA, most likely). With a 10 hour photoperiod, and 15 cents per kWh, the annual price of the traditional fixture would be $657 in electricity and with $75 x 4 MH lamps replaced annually and $150 of supplemental lamps annualy; $450 in lamp costs annually. That's $1107 annually to operate. If the fixture cost me $1000 to DIY, the 10 year total cost of ownership is $12,070. (Everyone who's thinking about a big tank - put that in your pipe and smoke it a bit!)
If we assume the LED fixture is run flat-out at 500w for the same 10 hour photoperiod, the annual cost is $273.75 - let's call it $275. If the LED fixture costs me $2k to build ($10 per LED, which is easy to do if you're careful), that's a 10 year total cost of ownership of $4750.
So, in 10 years, I'll save $7,320 in operating costs by going with the LEDs. Of course, no one's run LEDs for 10 years on a marine tank. :lol: It's still fairly experimental, but I'm either confident enough - or enough of a nut - that I'm willing to try it. Even if it only lasts 5 years and falls to bits, I will have still saved money. But based on failure data from the LED manufacturers, 10 years shouldn't be a problem - technically, the LEDs should be at 70% output after around 50,000 hours at 700mA and junction temps of 125c. (lifted from my memory, someone can check the datasheets if they want to verify.) I'll be running below that current and temp, so I should get more than that. 50k hours at 10 hrs/day is almost 14 years. And even at that point, they'll "only" be at 70% output, so I can just run 70% more current through them. OR, replace them all (some will inevitably fail along the way) with the more efficient LEDs sure to be out at that point in time. Even if you factor in replacement cost for 100% of the emitters in the cost estimates above over 10 years, the LEDs are still several thousand dollars cheaper to run. And, already, the Cree XP-G has been released, which has about 30% more efficiency than the LEDs I mentioned above, and the industry isn't slowing down, so they should only get better in the future.
Once I have the material gathered on the prototypes and nano fixtures I've been playing with over the last few months I'll post that too, but in the meantime there's probably a few weeks' worth of reading in the big monster threads I posted at the beginning. And once I build the big rig for my 360g, it'll be fully documented in my build thread for that tank.
AcroSteve
Make my Funk a P-Funk
Very good explanation.
I have not read the threads you link to, but do any of these LED's experience a phase shift at all?
The way you describe them, I like the advantages of LED's. And, i am definitely a DIY type so it gives me plenty to think about.
Keep up the good work.
I have not read the threads you link to, but do any of these LED's experience a phase shift at all?
The way you describe them, I like the advantages of LED's. And, i am definitely a DIY type so it gives me plenty to think about.
Keep up the good work.
der_wille_zur_macht
Team RC
Steve, the LEDs should stay very accurate to color over time, if that's what you're asking. In theory, the spectrum shifts slightly as they're dimmed below the design currents, but we're not typically dimming them that low - or if you were, it would only be for sunrise/sunset. And even if you decided it was a big deal, you could dim different colors of LED separately to cancel it out. Typically, people are using a mix of "cool white" LEDs (roughly equivalent to a slightly yellow 10k) and blue or royal blue LEDs (think: 20kk and actinic, respectively). Since you can dim the different colors independently, you can get pretty much any shade of white or blue-white you want. And of course the sky's really the limit, since you can also get red, green, yellow, amber, UV, etc. LEDs are typically VERY accurate to a specific wavelength of light - if you look at a spectral plot, it's typically a nice big narrow clean-looking spike right on the wavelength specified. In a build like mine, I'm going to have roughly 200 LEDs, so that gives a very high resolution in terms of control.
And just to be clear, since there are people like you reading this thread who aren't totally versed in LEDs for reefs yet - you can buy off-the-shelf drivers roughly equivalent to what I'm building if you don't want to spend a week soldering. Of course there's a cost difference. A typical buckpuck can drive 6 LEDs and costs $15 - $20 depending on options. My driver probably costs $6 - 7 and can drive 8 LEDs.
Above I quoted a number of $10/LED to DIY a rig. That's based on a "typical" build as detailed in the other threads on here. I'm willing to bet you could bring it down around $6 - $7/LED if you DIY'd the cheapest driver design you could come up with (probably not this one, though it would be close) and used cheap heatsink material. At that point, it's probably only marginally more expensive than an MH/T5 rig, if at all - which really removes the last big argument point (upfront cost) against doing LEDs.
Though, again, a disclaimer - it's still relatively untested longterm. You could probably count the number of people using a DIY fixture longer than 1 year on one hand and have digits left over - and I'm not yet in that group. So many of the projections I'm presenting are based on a combination of short term experience (yes, corals can grow under these things for a few months) and interpretation of manufacturer's data based on lifespan and performance.
And just to be clear, since there are people like you reading this thread who aren't totally versed in LEDs for reefs yet - you can buy off-the-shelf drivers roughly equivalent to what I'm building if you don't want to spend a week soldering. Of course there's a cost difference. A typical buckpuck can drive 6 LEDs and costs $15 - $20 depending on options. My driver probably costs $6 - 7 and can drive 8 LEDs.
Above I quoted a number of $10/LED to DIY a rig. That's based on a "typical" build as detailed in the other threads on here. I'm willing to bet you could bring it down around $6 - $7/LED if you DIY'd the cheapest driver design you could come up with (probably not this one, though it would be close) and used cheap heatsink material. At that point, it's probably only marginally more expensive than an MH/T5 rig, if at all - which really removes the last big argument point (upfront cost) against doing LEDs.
Though, again, a disclaimer - it's still relatively untested longterm. You could probably count the number of people using a DIY fixture longer than 1 year on one hand and have digits left over - and I'm not yet in that group. So many of the projections I'm presenting are based on a combination of short term experience (yes, corals can grow under these things for a few months) and interpretation of manufacturer's data based on lifespan and performance.
kcress
New member
der_wille_zur_macht; You are a typin' phool!! Wow.
Thanks for the cudo BTW. Your controller description was very good. I liked it.
You asked for any corrections. I have only two.
You seemed concerned about ripple or current fluctuations thru LEDs. That is actually of NO concern at all within limits. The only limit is too much current.
Let me refine that. LEDs really have two numbers with respect to their currents, "average" and "peak". You can do anything to them you want with regards to average, anything at all! as long as you never exceed their peak. The peak is the current at which either the actual wire bonds going to the semiconductor bulk material explodes or the current density in the bar semiconductor does instant physical damage. Often in Crees case they lamely do not state this peak current directly, they only state some pulse current limit at some duty cycle. In a Cree XR-E case they say 1kHz 10% duty cycle at 1.8A.
LEDs do not care a twit about ripple. Indeed most LED drivers do everything you described above but when you go to dimming anything less than the full set current they cycle the LED drive OFF and ON. This means that the normal regulation all occurs - but only while the LED is actually on!! The rest of the time the dimming is a function of how long the LED is completely off as apposed to completely on.
Indeed it has been shown that humans perceive brightness more towards peak brightness than average brightness. This is used in two applications. One is to make LEDs as bright as possible. Running the highest peak PWM allowed by the maximum peak current will provide the human perception of the brightest lighting. This would be used to make LEDs that must indicate something in bright ambients do their best.
The second application is in extremely low power consumption applications. But using less average current but with higher peak currents a designer can get the most bang per watt-second.
The key to all this is human visual perception. Most humans cannot see flickering if it happens above about 110Hz. I'm a rare phreak who can see it up to about 135Hz. Our eyes being chemical receptors can only respond so fast. For us not to see flickering we need any LEDs to be driven at least 120Hz. Better would be a couple of hundred hertz. At that frequency we cannot detect any flicker what-so-ever.
LEDs, they don't care as long as you stay below their peak current. With those Crees that means 10% at 1kHz OR FASTER. Slower than 1kHz and they will fail almost instantly.
Drivers need the inductor etc to regulate the current. If you just used a switch to turn on/off the current to a LED it would not limit the actual current to anything below the LED's peak allowable current.
Summary:
Dimming is usually accomplished by full on/off PWM of the regulated current.
The PWM must be at a higher frequency than 150Hz or we will visually detect it.
The PWM must be at a higher frequency than 20kHz or we will detect it audibly.
No LED current waveform does any harm or causes any damage to LEDs unless it exceeds the allowed peak, or pulse duty.
I haven't run the numbers but you wouldn't need to increase the current %70 to get back to the same brightness. Probably more like couple of 10's of percent.
Again, thanks for the excellent write up. :dance:
Thanks for the cudo BTW. Your controller description was very good. I liked it.
You asked for any corrections. I have only two.
You seemed concerned about ripple or current fluctuations thru LEDs. That is actually of NO concern at all within limits. The only limit is too much current.
Let me refine that. LEDs really have two numbers with respect to their currents, "average" and "peak". You can do anything to them you want with regards to average, anything at all! as long as you never exceed their peak. The peak is the current at which either the actual wire bonds going to the semiconductor bulk material explodes or the current density in the bar semiconductor does instant physical damage. Often in Crees case they lamely do not state this peak current directly, they only state some pulse current limit at some duty cycle. In a Cree XR-E case they say 1kHz 10% duty cycle at 1.8A.
LEDs do not care a twit about ripple. Indeed most LED drivers do everything you described above but when you go to dimming anything less than the full set current they cycle the LED drive OFF and ON. This means that the normal regulation all occurs - but only while the LED is actually on!! The rest of the time the dimming is a function of how long the LED is completely off as apposed to completely on.
Indeed it has been shown that humans perceive brightness more towards peak brightness than average brightness. This is used in two applications. One is to make LEDs as bright as possible. Running the highest peak PWM allowed by the maximum peak current will provide the human perception of the brightest lighting. This would be used to make LEDs that must indicate something in bright ambients do their best.
The second application is in extremely low power consumption applications. But using less average current but with higher peak currents a designer can get the most bang per watt-second.
The key to all this is human visual perception. Most humans cannot see flickering if it happens above about 110Hz. I'm a rare phreak who can see it up to about 135Hz. Our eyes being chemical receptors can only respond so fast. For us not to see flickering we need any LEDs to be driven at least 120Hz. Better would be a couple of hundred hertz. At that frequency we cannot detect any flicker what-so-ever.
LEDs, they don't care as long as you stay below their peak current. With those Crees that means 10% at 1kHz OR FASTER. Slower than 1kHz and they will fail almost instantly.
Drivers need the inductor etc to regulate the current. If you just used a switch to turn on/off the current to a LED it would not limit the actual current to anything below the LED's peak allowable current.
Summary:
Dimming is usually accomplished by full on/off PWM of the regulated current.
The PWM must be at a higher frequency than 150Hz or we will visually detect it.
The PWM must be at a higher frequency than 20kHz or we will detect it audibly.
No LED current waveform does any harm or causes any damage to LEDs unless it exceeds the allowed peak, or pulse duty.
I haven't run the numbers but you wouldn't need to increase the current %70 to get back to the same brightness. Probably more like couple of 10's of percent.
Again, thanks for the excellent write up. :dance:
Altpers0na
ARS Caesar
if your LED project included dozens of LEDs would it be easier to 'dim' the fixture by just turning off some them off?
kcress
New member
Of course that would work! But since each LED contributes a lot to the light footprint you would probably end up with a mottled result which, I personally, think would be cool.
der_wille_zur_macht
Team RC
Thanks for the notes on that. It definitely helps clarify that the ONLY time we really need to be concerned about ripple is when the peaks of the ripple are dangerously close to the LED's max rating - either because the ripple is very large for a small average current, or if the drive current is very high you'll need to be concerned pretty much no matter what. It's interesting to note that many of the reference designs in OnSemi's documentation that use low drive currents have NO big cap on the output.
I wonder about frequency related to photosynthesis. In other words, if X amount of intensity is enough to bleach a coral, and our dimming method "simulates" .5*X by just switching at a high frequency with a 50% duty cycle, is that actually safe, or will the coral still bleach because of the peaks of brightness? Results seem to say that my concern is not valid, luckily.
And FWIW, the design I posted uses a 150khz switching frequency, so vastly too high to see OR hear for humans. The dimming I'm doing with my Arduino uses a 400hz (or so, I'd have to look it up) PWM signal for dimming, and it isn't visible but IS audible - though barely, and only if your head is very close to the driver in a quiet room. Put it in a project box or have a fan or pump running nearby and you definitely won't hear it.
Typo on my part! Though given that I'm planning on running them below 700mA and 125c; and that 50k hours at 10 hrs/day is like 14 years, I really doubt I'll ever be adjusting them anyways.
The problem with that is that you'll quickly get spotlighting - if you use standard densities, you'll have an LED every 10 - 20 square inches of tank surface area. If you want 50% brightness and try to get it by turning every other LED off, you'll get alternating patches of brightness and darkness. It's really pretty easy to control the drivers via PWM from a variety of sources anyways - an Arduino like I'm doing, or just build a 555 circuit with a pot to adjust duty cycle.
der_wille_zur_macht
Team RC
I agree, though people should do that because they want the mottled look, not because they're trying to dim by 50%.
When I get the thread together about my nano-sized prototypes I'm planning on talking about some of the interesting results I've been able to get doing things like that - which are all pretty much impossible with conventional lighting.
terahz
1x10^12 Hz
Getting a bit off topic but there it is:
I mounted mine on individual PowerPC heatsinks (Wakefield 658-35AB, mouser part 567-658-35AB). It will cost you less than $100 for the heatsinks but then you hove to suspend them somehow. I used angle aluminum so that will probably run you another 10-15 bucks from HD.
PM me if have any questions.
der_wille_zur_macht: Strange that you can hear your PWM. I can hear mine from the fans, but nothing from the LEDs. Also the arduino PWM can be adjusted a bit over 1KHz. It is still audible but might be better. Here is a nice read about it: http://arcfn.com/2009/07/secrets-of-arduino-pwm.html
There seems to be a general preference towards dimming with PWM vs 0-10V analog. Is this just because of the convenience of using it with the Arduino? It seems a lot easier to build a 0-10V dimmer with a pot and an LM317, and then just dim it with the pot. Of course this means no automation, unless you use a ReefKeeper or Apex controller, or Profilux, etc..
der_wille_zur_macht
Team RC
I've done CPU heatsinks, too. Depending on how beefy they are, you can get away with several LEDs on each. I have some nano fixtures with eight LEDs on one CPU heatsink that run just fine.
I can hear it IF I put my head close to the driver and the room is relatively silent. In a built fixture (i.e. in a project box with fans around) you really can't hear it at all. I've played with adjusting the PWM frequency for other purposes but don't really think it's required here.
Yes, I picked a device compatible with 5v PWM from the Arduino. PWM is easy to come by. Just about any microprocessor will have compatible PWM pins and/or be capable of PWM with a little software. If you wanted to control these drivers with a pot, you could build a simple 555 circuit running on 5v that would barely be more complex than using an LM317 and pot to get a 0-10v analog signal.
That said, an Arduino is so easy, cheap, and powerful I'd rather see someone go that route than just put a knob on it! Maybe I need another thread on "arduino basics for LED applications."
+1 on the Arduino thread. I did a little research on it and was quickly overwhelmed and quit. It seems like there's a lot of programming to get it automated. And from what I could tell, it doesn't have an internal clock, so you need to experiment until you get a 24 hour cycle from it.
But sorry to digress. Back to the drivers!
But sorry to digress. Back to the drivers!
der_wille_zur_macht
Team RC
The arduino's internal clock is simply a milliseconds counter, so it's not really THAT bad - you just need to convert or do math to get things right, and be prepared for when it resets (it overflows the variable it's stored in every 9 days or so). Also, since it resets when the chip is power cycled, you need to program it with some flexibility. And it's probably a few seconds a month inaccurate.
I used an RTC (real time clock) to avoid some of these problems - it's a little more accurate, and can store a full date and time, plus if you provide it with a coin cell backup it won't lose it's memory when the power goes out.
I'm going to go ahead and do an arduino thread, but probably not till next week or after New Year's, so stay tuned.
I used an RTC (real time clock) to avoid some of these problems - it's a little more accurate, and can store a full date and time, plus if you provide it with a coin cell backup it won't lose it's memory when the power goes out.
I'm going to go ahead and do an arduino thread, but probably not till next week or after New Year's, so stay tuned.
