The Call of the Open Sidewalk

In a previous post I talked about a prototype 1.5V powered LED flasher capable of producing high pulse currents. The trick was to use supercapacitors and pulse transistors in a voltage tripler arrangement. Here is how I expanded the idea to produce a circuit that would flash two high power LEDs at once:

This is more or less the same thing as described in the previous post. C4 and associated components just provide another branch for the extra LED. The common branch through C2 needs twice the charging current so R10 was added. I spent some time exploring the idea of using two supercaps with low internal resistance and then using 1 ohm resistors to limit the current through the two LEDs. Unfortunately very small differences in forward voltage caused great differences in LED brightness. I then said the heck with it and more or less duplicated the circuit resulting in the use of four supercaps. That would not fit in the available board area. The three supercap solution was the compromise. I couldn't see any reason not to keep using the low internal resistance supercaps (other than the larger size) so the 1F (0.4 ohm) supercaps are now 3F (.09 ohm) supercaps. This allowed the increase of the charging resistors from 22 to 33 ohms for more or less the same pulse current. The other 22 ohm resistors were changed to 33 ohm just to eliminate the extra component value.

In the previous post I claimed that there was something wrong with the description of the timing components on the datasheet for the ZSCT1555 timer. I now withdraw my scurrilous accusation. The problem was with my understanding of the non-ideal operation of the device at 1.5V. The new values of R1, R2 and C1 result in flashes of 50ms at a rate of 1 per second. The capacitor is now a 1uF ceramic that fits on a 0805 footprint.

With a fresh battery this circuit produced pulses of about 0.4A through each LED. When the battery is at 1V I would expect the current to be about a quarter of that. The system will enter "infinite battery life" mode with the battery at 0.83V. There is no reverse battery protection here. Tests revealed that on a reverse battery the only component that takes abuse is the 555 timer. I think the package should be capable of dissipating the excess power. It did get pretty hot. Because of the larger supercap value the system takes a noticeable length of time to come up to full brightness when the capacitors are uncharged. I consider that pretty cool. Unfortunately the capacitors retain their charge quite well over time. I designed a single sided board to mount on the traditional bicycle water bottle cage mount (5mm bolts, 65mm spacing). The assembled board looked like this:

The gEDA files: Schematic Layout

The design files: Gerbers Postscript PDF

Most of the board was done with 0.5mm lines and spaces with 0.35mm lines and spaces where needed. As a result it should be realizable with the toner transfer method. Here are the mirrored copper layer images. Note that the board is 79.75mm by 19mm.

Mirrored copper: Postscript PDF

I sent my board out to get made. Here are the files I sent. The company was Olimex. You have to edit the readme.txt file before you send these files off to Olimex. This approach gets you 10 boards.

Pretty much any super/ultra cap will work here as long as the internal resistance (ESR) is reasonably low. Higher ESR means less charging current and a dimmer flash. Adjusting the value of the charging resistors (R5, R7, R9, R10) downward can be used to compensate. Keep in mind that increasing ESR is the major failure mode for supercaps over time so you normally are better off starting with a low value. Pretty much any 1 Watt LED will work as long as the forward voltage is in the neighbourhood of 2.5V . The transistors are special in that they must be capable of switching 1-2A with a low saturation voltage.

Here is a parts list:

2       ZXTN25020DFLTA          NPN low sat volt pulse transistor sot-23 (diodes inc)
2       ZXTP25020DFLTA          PNP low sat volt pulse transistor sot-23 (diodes inc)
1       ZSCT1555N8TA            low volt 555 timer so-8 (diodes inc)
3       MBRX120LF               1A Schottky Diode sod-123 (Micro Commercial)
1       C0805C105K8RACTU        ceramic cap 1uf +-10% x7r 0805 (kemet)
3       ESHSR-0003C0-002R7      3F .09ohm supercap 3.5mm lead spacing (nesscap)
1       LTPL-P011A              1W amber led (Lite-On)
1       LTPL-P011R              1W red led (Lite-On)
1       2173                    steel C cell holder (keystone)

These part numbers were findable on Digi-Key at the time of this post. Normally you do not get things like 1 Watt LEDs from an electronic distributor like Digi-Key. You can usually get better prices elsewhere for small quantities. I just wanted to show an example of the LEDs I ended up using.

Here is what I did for mounting:

I suppose I could be accused of flaunting my geekiness here. I mostly do stuff like this to attract the groupies. Social aspects aside, if you live in a more hostile and/or corrosive atmosphere than I do you might want to do a bit more with the enclosure ... such as actually having one I mean. There is no switch, the battery is removed (or just moved) when the flasher is not in use. If the bike is locked up outside the battery ends up in a pocket. Any time the bike is in use the battery is in place. That is just where I keep it.

The silicone sealant seen in the picture is intended to dampen the vibrations of the capacitors to prevent lead breakage.

Here is what I did to mount the red LED on the rear:

This is attached low on the read fender. This demonstrates an advantage to centralized electronics I guess. You can put the actual light emitters pretty much anywhere. The overhang was intended to make a shade for people who are close to the back of the bike. It didn't really work. It also protects the soft silicone lens of the LED. I was told by a fairly LED adverse friend that the flashes were not that bothersome. I think that is because the LED is mounted low to the ground.

Here is what I did for the amber LED on the front:

This is mounted on the most forward extreme of the front fender. Having a bright flashing light on the front of a bike is a bit of a problem in that the rider has to be shielded from the flashes. I made a reflector out of some shiny aluminium flashing in an attempt to direct the light in line with the rider away to the front. It mostly works but the top reflector directs some light onto the tire and the road. Fortunately the tire is not all that reflective and the road is far away. If I decide later that that light bothers me I will blacken the offending part of the top reflector. Mounting the LED relatively low like this produced an unexpected benefit. It greatly reduced the flashing reflective sign effect.

These types of LEDs are often sold with a combination heat spreader and terminal board. We don't need any sort of heat sink for this application so I just soldered the bent up leads and glued the bare LED in place with some silicone sealant. This flasher system is quite insensitive to voltage drops in the wiring. Drops just make the flashes a bit dimmer and the battery last a bit longer. As a result I used 22 gauge wire to hook everything up.

Here is the best I could do in terms of photographing the working flasher system:

For those that can relate; this was taken at F2.8 and ASA 400. This shows a single flash. In terms of sending light in all possible directions we are doing quite good. Too good really, a lot of the light ends up either in space or on the ground. Better optics could help here but those optics would have to be larger than what I have now. Fortunately we have quite a lot of light available to waste.

Speaking of waste, you can adjust the brightness of the flashes by adjusting the charging resistors (R5, R7, R9, R10). Reducing them to 15 ohms would get you up to into 0.9A pulse range. I am fairly certain that the resulting system would be reliable. It would however drain the battery more than twice as fast. I don't think it would really get you any more attention (at least of the positive kind).

To speed up the flash rate you can reduce the value of R1. Halving the value of R1 would get you a pulse rate of 2 per second and a flash brightness of half at more or less the same battery consumption. The brightness could be restored by halving the value of the charging resistors (as mentioned in the previous paragraph) at the cost of twice the battery drain.

Speaking of batteries, this system would work well with a NiMH cell. In fact it would provide a more or less constant voltage/light output over the useful battery life. A NiMH cell has a high current capacity so a NiMH AA cell charged every week or so would likely be optimal.

I looked up the local laws with respect to bicycle lighting to try to determine if a front and back flasher system is legal. As far as I can determine it is not. I don't think anything someone might actually use for lighting in this century would be. It has been a long time since the laws were updated. I guess I'll just have to live with the uncertainty. Around here the only reason law enforcement does not spontaneously do something like hand out a safety award when a bicycle/rider combination is seen emitting light at night is because they are afraid they might be up to something.

The other research inspired by this project was night time bicycle safety in general. From my experience a bright flashing light on the front of the bike pretty much eliminates the common case of collisions caused by cars turning left across the path of the bike. The other common collision occurs when a car pulls out of a side street/driveway. I have often seen a driver glance in my direction between the flashes before deciding to go. This could be expected to reduce the problem to the fully stealthed bike/rider case. Unfortunately there is a complication. At some point during the manoeuvre the driver will see the next flash in their peripheral vision. Then their behaviour becomes impossible to predict. One of three things will happen:

• They will slam on the brakes.
• They will hit the accelerator.
• Those driving at the very edge of their ability will go into overload mode. This will produce a situation where there is a driverless car on the road for a time.

A constantly illuminated headlight would help with this so I have to acknowledge that such a thing is still a good idea no matter how bright my front flasher is. Having said that, I think that I will still often find myself with just the flashers. History has shown that it is too much to ask that I maintain much in the way of lighting over the long term. I can however manage a single battery. That simplicity is the greatest advantage of a system like this in my opinion.

posted at: 14:57 | path: /ledlight | permanent link to this entry

Once upon a time, someone invented a solid state light source called a light emitting diode (LED). Sometime after that a company called National Semiconductor created an integrated circuit (IC) called the LM3909 that flashed a LED.

The LM3909 was considered a very exciting thing in spite of the fact that flashing a LED was pretty much all the LM3909 did. It did not produce particularly bright flashes. It did not even flash the LED all that quickly. It would however flash a LED for a very long time (years) from a single 1.5V cell. That was enough.

This was something new in the world. The LM3909 showed how profoundly different a light source the LED was compared to everything that had come before. People that did electronics for fun absolutely loved the LM3909 for providing this insight. At one time you could buy a LM3909 in pretty much any large shopping mall in North America (the ones containing a store called Radio Shack). Combining a LM3909, a capacitor and a battery just to watch a LED flash is something that pretty much everyone of my generation with an interest in electronics has done at one time or another.

The LM3909 did something sort of clever. A voltage of 1.5V was not enough to cause one of the red LEDs of the day to emit light. The LM3909 would charge up a several hundred uF capacitor to a significant fraction of the 1.5V battery voltage. It would then force the capacitor to be in series with the 1.5V power source causing the LED to light until the capacitor discharged. The capacitor was charged through a couple of resistors that drew relatively large currents when the capacitor was forced into series mode. To get reasonable efficiency the resistor values had to be on the larger side which limited the flash rate to something like once a second.

In the late 90's, I was enduring an entire team of people who were attempting to teach me undergraduate engineering math. This caused me to yearn for the life I had left behind (electronics). Due to the nature of the Electrical Engineering program I got to take my bicycle home in the dark fairly often. Once a woman in a van pulled over and told me a story with a very unhappy ending. It was about her friend who shared with me a love of stealthy bicycle travel. My thoughts naturally turned to lighting. Lighting involving electronics pretty much leaves some sort of LED flasher.

At various points during the ensuing project people would make the helpful suggestion that I could in fact just buy a bike flasher. I suppose those same people would suggest that a Jedi Master could hop into the land speeder and go pick up a light sabre. I didn't want just a bike flasher. I wanted the bike flasher. ... and well, the LM3909 was not it. It was power efficient at the cost of a very weak flash. I was sort of disappointed.

I still wanted to be able to use a single 1.5V cell so I used the idea of using a capacitor as a voltage multiplier as seen with the LM3909. I found that a 1000uf capacitor made for a reasonable flash. I wanted to use 3 LEDs with good lenses (10mm) adjusted to produce a fan of light directed down the road and somewhat to the sides. I just did not have enough light available that I could waste it. I eventually ended up with a single large power transistor switching individual 1000 uF capacitors for each LED. That was 3 amber LEDs and 3 capacitors on the front of the bike and 3 red LEDs and 3 capacitors on the back. I had to look to find some sort of convenient oscillator that would run on 1.5V (really 1V). Turned out that the LM3909 was ideal for the job and was used as the timing source. Sometimes you just can't avoid using the LM3909.

The flasher system was quite bright and was visible even on cloudy days. It served me well up to the point that the bicycle it was attached to was stolen from outside a post-university workplace. There was no on-off switch, I simply put a C cell in a battery holder on the bike frame. The battery life was pretty much perpetual. I didn't really care if I forgot and left the battery installed. I could have used a smaller/lighter power source but I do not like having to think about changing batteries.

We now draw a bit nearer to the actual point of this post... While investigating the present state of high power LEDs for a circadian light fixture project I spent some time wondering what a flasher using a high power LED would be like. I had felt limited by the amount of current I could blast the LEDs of the late 90's with. I was pulsing the LEDs with something like 100 mA. A 1W LED is rated for 350mA and can be pulsed to .5A-1A. They are also more efficient at converting electrical power to light. In particular I wanted a flasher bright enough to be seen in a urban streetlight environment from all directions.

I wanted to be able to retain the single C cell power source even if I ended up with a lot less battery life. Now, it turns out that .04 seconds is a kind of sweet spot for flash duration when dealing with human vision. For a given available optical energy, putting that energy into a .04 second flash produces the greatest perceived brightness. Doing the capacitor trick as before got me a capacitor value of around 0.22 Farad for a reasonable .04 second flash. That is a big capacitor. Fortunately there has been a recent technology breakthrough in the form of what are often called supercapacitors. I really needed .022 Farad with a low internal resistance (ESR). Contemporary supercaps do not roll that way. I could get a much larger value than I needed with a low ESR but that would make for a really large pulse. In the end I was still able to use the supercap to multiply the voltage of the battery but I had to do it a different way than the venerable LM3909 did.

The LEDs I was pulsing way back when had a forward voltage drop of around 2V. Typical high power LEDs are 2.5V for red and amber and 3.5V for green and blue/white. OK, goodbye to green and blue/white then. There is still a problem with red and amber. If we are only doubling the voltage the battery can only discharge to 1.25V (2*1.25V=2.5V). Dry cells are considered used up at about 1V. We are losing half the battery life. OK then, hello second supercap. Now we can triple the voltage. That does cause a pretty severe hit in terms of efficiency. Since there are no inductors here all extra voltage gets lost someplace in a resistance. The doubler has a maximum possible efficiency of 2.5V/(2*1.5V)=0.83 and the tripler comes out at 2.5V/(3*1.5)=0.56. Still, 56% efficiency with twice the battery life available is significantly better than 83%. We also need enough extra room to charge the capacitors with resistors. So the tripler is the way to go here. I will leave it as an exercise to come up with something better using an inductor.

I still needed a timing source. The LM3909 is now discontinued (sniff) but you can get 555 timers designed to run off of a single 1.5V cell. I used the one from Zetex. Here is the schematic:

Here is a list of important parts used for the prototype (these were the ones in stock at Digi-Key at the time of the order):

1       ZXTN25020DFLTA          NPN low sat transistor sot-23
2       ZXTP25020DFLTA          PNP low sat transistor sot-23
1       ZSCT1555N8TA            lo volt 555 timer so-8
2       MBRX120LF               1A Schottky Diode sod-123
2       B0810-2R5105-R          1F supercap
1       LTPL-P011A              1W amber led
1       BHCL                    plastic C cell holder
1       9081CA                  so-8 proto board

U1 is just a 555 timer so I will not bother explaining how it works here. The values of R1, R2 and C1 gave a pulse length of 0.06 seconds and a period of 1.1 seconds with the prototype. Close enough I guess. By the way, the Zetex data sheet for the part seemed actively misleading. I had to figure out the timing (which is different from a normal 555 timer) on my own. The circuit formed by D1,C2 and R5 is the same thing as the circuit formed by R7, C3 and D2 (just turned over) so I will mostly just discuss one of them. C3 is charged during the non-pulse time through Schottky diode D2 and R7. The Schottky diode is quite over specified here in terms of maximum current. I was looking for a diode with low forward voltage at only 30mA. This one managed about 0.2V. Note that capacitor C3 never gets charged up to 1.5V. It only charges as much as is needed to fire the LED. That is because the RC time constant of C3 and R7 is relatively long at 22ohm*1F=22 seconds. This time constant means that the flasher actually needs to warm up over the course of 10 seconds or so before it reaches full brightness.

The LED fires when pin 3 of U1 goes low. Q3 is turned on through Q2 and drags the positive end of C3 to the V- rail. D2 is reverse biased and turns off. Let us assume that the V- rail is 0V. Then the negative end of C3 (and the negative end of the LED) is now as little as -1.5V. Q1 does the same sort of thing to C2 with respect to the V+ rail at +1.5V to drive the positive end of the LED as high as +3V. Thus potentially the LED could have as much as 4.5V across it. This would be too much voltage for the LED. The RC time constant mentioned previously prevents this excessive voltage at the LED. There simply isn't enough time for R7 to change the voltage of the capacitor any significant amount before the next pulse. Eventually an equilibrium is reached with the amount of charge going into the capacitor during charging time coming out during the LED pulse time. In practise the voltage across the capacitors is exactly enough to add together with the battery voltage to make up the LED voltage. It tends to be higher than expected because of the internal resistance of the capacitors when the led fires. The capacitors are rated at .4 ohms internal resistance and showed a drop of 0.2V with a 0.5A pulse. Lower internal resistance is better but makes for a more expensive supercapacitor.

And speaking of things that drop voltage, the transistors used here are special in that they drop very little voltage when turned on. They need to be driven hard to do this. For the old flasher I ended up using a transistor in a huge TO-220 package. As these transistors come in SOT-23 packages this ends up being a sort of technology improvement as well.

The LED current here is limited quite accurately by the combination of the value of R7 and the duty cycle of the pulse. This is possible due to the large value of the supercapacitor and is a very nice result. Unfortunately as the battery voltage drops the voltage across R7 drops. That means that the LED gets dimmer over time. With the prototype the combination of extra voltage available due to tripling combined with the internal resistance of the capacitors produced a situation where the LED brightness was still a bit more than a third of the full voltage brightness when the battery hit 1V ("dead"). That is also a nice result. This could be improved by finding a less voltage sensitive way to charge the capacitors.

Here is what the prototype looked like:

With a new battery it produced pulses of about 0.5A. The light output is really intense ... which is good. I think that battery life should be something like 100 hours with a C cell and 150 hours with a D cell. The front and back dual flasher I now hope to build should thus get about 50 hours on a C cell.

This flasher suffers from a problem I refer to as the infinite battery life syndrome. You often see bicycle flashers in the wild producing barely perceptible light output. I think that is caused by a combination of human perception and the characteristics of LEDs. As the battery voltage gets closer and closer to the LED forward voltage the current decreases and as a result the battery life increases. Eventually the current gets down into the microamp range but the LEDs are still showing light. Since people are really bad at judging brightness there never comes a time where it is obvious that the battery needs to be changed. Since this is safely critical equipment I do not think this is good design. Presently the prototype is happily producing 63 mA (0.07V across R5 while charging) pulses at a battery voltage of 0.83V after running steady for a week and a half and now seems fairly stable. I guess one could add a voltage detector and do something on low battery voltage. I have no idea what this something should be.

I know that 1 second period flashers are considered with some disdain in some quarters. The issue is that it makes it difficult for observers to infer the speed and direction of the bicycle. I can rebut this from experience as I have used such a flasher for some years. On a totally dark street there might be an issue but in an urban environment if you can draw attention to yourself there is almost always enough light to follow the path of the bicycle/rider afterwards. This is especially true if the bicycle is liberally plastered with reflective tape as mine is. Since attracting attention is important it seems reasonable to throw out as much light energy as possible during the pulse. The best way to do this is to pulse as infrequently as is practical.

The stuff about the pulse energy brings up another objection. Some people feel that very bright flashers are obnoxious. The amount of light intensity produced by my prototype during the pulse is more or less the same as the light intensity produced for a much longer time by the flash of a automotive signal light. The reason it is annoying is because the light comes from a very small area. The problem is that it is impossible to separate annoying from attention attracting. One implies the other. From my interest in circadian light I know that even very bright red LEDs can only zap one photoreceptor (amber is pretty much one and a half). There are three types left so I do not think it is possible to night blind someone with red/amber LEDs. So given that, the concise response is:

Annoying? Good! I wanna live...

posted at: 21:39 | path: /ledlight | permanent link to this entry

The small area of LED light emitters makes it somewhat challenging to use them effectively in low level lighting applications. The light is hard to shade and diffuse efficiently. In the past I have used light coloured ceilings and walls as diffuse reflectors. This works fairly well but it can be a problem avoiding a bright area on the wall/ceiling where the light is mounted. Here is a description of a light I built that physically moves the light emitters away from the wall to deal with this issue. This works but the resulting fixture is quite large.

The ideal angle of the wedge of light from a wall mounted fixture is something a bit under 90 degrees. We want to avoid illuminating the wall behind the fixture while preventing people from seeing the small bright emitters directly. We also do not want a lot of light on the ceiling closest to the fixture as this tends to make a bright spot, particularly if the fixture is close to the ceiling. The raw LED chip itself emits light from a flat surface. This means that the light distribution tends to be proportional to the cosine of the angle from the normal. Cutting this distribution in half would produce a fairly good light distribution for the application. That could be done by adding a horizontal reflector. Since we have no need for optics other than the reflector we might as well use surface mount LEDs and make the fixture small.

I wanted to experiment with monochrome light as we are targeting night vision here. This plot:

from here suggests that the best wavelength for night (scotopic) vision would be 505 nm. That by coincidence is the wavelength that is perceived as the North American traffic signal "green" colour. As a result LEDs that emit this odd wavelength are relatively easy to find. The colour ends up being sort of a cyanish green. Light with a wavelength of 505 nm is fairly active as circadian light but the light levels achieved here are much too low to be disruptive.

My thinking up to this point inspired me to go out and buy these from here. I ended up with the larger 1206 packages because the supplier I ordered from would only provide the wavelength I wanted in that size. The extra width (1.6mm) ended up not being a problem and they were easier to handle than the 0805 packages I normally prefer for hand made surface mount based prototypes.

I spent some time planning to use a thin sheet of stainless steel as a reflector until I stumbled on this at a local home improvement store:

It was originally intended as an edging for a counter top. It was made out of a shiny alloy and was close to the desired shape. It comes with a prefabricated wall shade in the form of the overhang.

The use of the surface mount devices pretty much mandated the use of the PCB. Rather than measure anything I made a drawing at 5X scale:

and then printed it at 20% to use as a guide. Much to my surprise the Xfig program on Linux got it exactly right with no fiddling. That's not all that expected for a mere figure creation program. Here is the drawing in various formats: fig, postscript, pdf, svg, png.

A sharp tool was used to transfer the dimensions from the 1X scale drawing to the copper:

followed by a layout:

followed by some use of a cutting disk:

Note that the edges of the copper have been ground off to prevent shorts through the metal reflector. I just dragged the cutting disk along the sides. A more appropriate tool would make for a straighter edge.

I first applied liquid flux to the PCB. This is probably optional for such a simple assembly but I had it available and it did make things easier. A popular method of hand soldering surface mount devices starts by adding solder to just one pad. In my case this meant I would first add solder to each alternate area of copper like this:

Then the device is moved into position and held in place with something like a round toothpick. Heat from a soldering iron is applied to the solder (but not directly to the device) to sweat solder the device in place. Then it is fairly easy to solder the other pins. Once the device is well secured the original pin can have solder added if needed. This was the scene after soldering was complete:

I have ended up with 3 LEDs and 2 resistors. When I made the drawing I was enough of an engineer to want to cover the case where 4 LEDs were required to make up the voltage. The idea was to just skip one cut for the 3 LED case. When I made the board I was apparently enough of a technician to be compelled to make things as shown on the drawing. Thus I managed to make a classic error involving an ambiguous drawing in a way that normally requires two people. Using 2 resistors solved the problem in the assembly phase which meant I had also achieved pointless symmetry which is another sort of classic error. There is probably some insightful observation on the human condition available to me here but it escapes me.

I picked a LED current of half the normal specification based on the hope that it might increase the lifetime of the fixture. I have no idea if that makes sense with LEDs. Heat might not be an issue. LED manufacturers like to quote a lifetime of 100,000 hours (11 years) which really just means they think their product might last a long time. With a target current of 10 mA we end up with (12V-(3.2V*3))/0.01A = 240 ohms of resistance. I ended up using two 110 ohm resistors.

After soldering the wire connections and cleaning off the rosin based flux with pure isopropyl alcohol I was done with the electrical assembly:

I used a small amount of RTV silicone sealant to glue the board in place in the reflector. The board would not stay where it was supposed to go. I ended up using the weight of the wires to keep it in place:

The preceding image shows that the reflector is quite a lot wider than would be required just to reflect the light from the LEDs. The idea was to prevent light from escaping from the ends of the channel without any extra fabrication. This worked but the channel itself was still a bit bright where it could be seen on the ends.

Testing the light distribution against a white surface showed some problems:

To prevent light from projecting below the horizontal plane the assembly had to be tilted back towards the LED side. This is inconvenient for the eventual installation of the fixture and caused two other problems. The largest amount of light was going more or less directly up to the ceiling above the fixture creating a bright spot. The other problem was that a significant amount of light was finding its way to the wall directly behind the fixture which created another bright area. Blackening the top of the channel helped but I eventually took a different approach and bent the end of the reflector up like this:

The edge of the reflector is bent up enough to be in the same plane as the top of the channel. This means that there is no light projected below the horizontal plane when the fixture is sitting flat against a horizontal surface. The result was:

This is much better. I consider the design complete.

Next I'll go into the various thoughts and observations that came out of this activity...

When using monochromatic light even illumination can create a situation that destroys visual information. The light from the designed fixture is very uniform. Evenly illuminating large areas of the ceiling results in a situation related to an outdoors whiteout condition. Objects are still visible but with little detail. I suspect that this would only be an issue in rooms small enough to allow a single fixture to evenly illuminate the entire ceiling. A fix is to move the fixture closer to the ceiling to produce sharper shadows at the cost of a more intense hot spot.

The Homo sapiens sapiens set depend on visual transitions in their peripheral vision to help with things like walking around. The sharp vertical light transition produced by the reflector means that a strong cue should exist for orientation and balance. This artificial horizon effect might end up being the more important visual information. This system is after all intended to improve navigation. I am not sure how well this actually works at low light levels. There is probably a research opportunity here.

An important performance feature of a low level lighting system is the length of time it takes for vision to adapt enough to navigate after the high level lighting is turned off. Testing this produced a result that initially seemed odd. It seemed that the nighttime vision system (rod based) was adapting significantly faster than the daytime vision system (cone based). I would see a monochrome image before I could perceive the colour of the light. This lag was something like 5 seconds. The oddness here came from the impression that nighttime vision takes a much longer time to adapt to darkness than daytime vision. Some research revealed that this isn't really true in general. It turns out that the traditional way to check the time it takes people to adapt to the dark is to expose them to really bright light for a long time and then plunge them into complete darkness. This presumably is to simulate the common situation where one is walking along in a sunlit meadow before falling into a fast moving underground river. In more reasonable transitions from light to dark it is the nighttime vision that kicks in first (Ref, See Figure 2). At the light levels I am using the effect would be close to the maximum. The daytime vision system has to adapt to a level close to the daytime minimum while the nighttime vision system only has to adapt to a light level 100 times higher than the nighttime minimum. The transition is driven harder for the nighttime system. This I think strengthens the contention that 505 nm is the magic wavelength for low level lighting.

Since the edge of the reflector is to be bent up it might make sense to cut the reflector in such a way as to allow the bent up part to shade the bright ends of the channel. In other words, the reflector could be cut off at an angle to make the LED side shorter than the bent side. This could also hide associated wiring.

Light fixtures that point upward accumulate dirt. This creates a maintenance issue. For some applications it might be better to put the light sources below eye level pointing downward. Hallways and walkways are sometimes lit this way.

Blue LEDs at 470 nm are fairly effective for nighttime vision but are a lot less visible to daytime vision than 505 nm LEDs. This might make blue LEDs preferable where it is important for the night lighting to be invisible during the day.

The next step is to make some of these and deploy them around the house to try out the low level lighting lifestyle for myself. I should be able to hide the fixtures on the top of door and window frames. This will make it possible to hook the wires up into the attic for an easy installation. I will report on the results when and if there are any.

EDIT: proceeding -> preceding

posted at: 15:49 | path: /ledlight | permanent link to this entry

I seem to be stuck on the low level lighting thing. An advantage to not working is not having anyone to tell me to get back to work...

Here is an approachable introduction to modern colour vision theory. It helped me a lot. The rest of the site (watercolours) is a bit too technical for me but I think I now know what to expect when I mix single colour LEDs.

For my bathroom night light I now think that just green and red LEDs would of worked fairly well. I could of balanced them to give some sort of yellow. A problem would occur if the colour was exactly the same as the perceived light colour. It would not be visible against white. The 3rd colour in between removes that ambiguity. That is assuming that the green and red LEDs do not balance out to the in between colour. The apparently excessively bright green LED is thus a feature as it moves the balance point well away from the yellow LED wavelength.

Another possible problem with just using green and red (or blue and red) is that you would end up with more green light. Green (and blue) can blind night vision when not used in moderation. I now think of my bathroom night light as primarily yellow with a bit of some other wavelengths thown in to allow some colour discrimination.

posted at: 17:13 | path: /ledlight | permanent link to this entry

Some wisdom came out of my night light projects. I'll share that here.

Are light emitting diodes ready for prime time residential lighting? That question is surprisingly controversial. I am not really sure myself about the answer to the general question. I am sure that for low level lighting LEDs entirely rule.

Up to this point in the history of the world there has not been a practical and efficient way to make a small amount of light for a long time. Incandescents are particularly bad for efficiency and life. Electroluminescent were the closest thing pre-LED era but they are an area emitter and are thus hard to shade effectively. They also need high voltage wiring.

The result of this is that the system of vision intended to allow people to function at night is not really used by most people. People can see at levels down to 0.01 lux. Typical residential living spaces are lit to levels between 50 to 100 lux. That is as much as 4 orders of magnitude more than the absolute minimum required.

It may be that we have hit a technological threshold. If low level lighting is more practical then perhaps it will become more common. It is interesting to consider how a hypothetical low level lighting enthusiast would arrange the lighting in a house.

Let us first invent a specification. All living areas should be lit to the level of moonlight on a clear night. One source claims this is .3-1 lux. Brighter would interfere with things like sleep. Darker would mean a longish time to adapt after turning off a light. Tropical moonlight would be a bit bright I think so let's aim for the lower end of .3 lux.

For a 3m X 3m room we would have around 10m2 to illuminate. Lux is lumens per square metre so to hit .3 lux we would need 3 lumens. LEDs with an efficacy of 30 lumens/watt are available so our power budget could assume around 0.1W per lamp. Since this is mostly best case I'll double the assumption to 0.2W. It would likely take 7 of these lamps to light my house. So the total power draw is 1.4W giving us a yearly power consumption of 12 kW/h. That would cost me $0.72CAN a year.

The fixtures could be built in at time of construction. Retrofit would be easy as you would only have to snake in something like alarm wire to power the lights. The lights could be designed to run off the door bell transformer. If instead you ran them off a small sealed lead acid battery charged from a wall wart you could claim that you had established a sort of residential emergency egress lighting system.

The proposed level of lighting would of course mostly serve for navigating through the house. This would still simplify things. Most rooms would not need any ceiling lights and associated switches/wiring. Hallways and stairways would not need three way switched lights.

In general safety would be greater at night. I used to fall over the stool in my kitchen on a semi-regular basis before I installed my always-on counter light. I did get quite good at it but it would be unrealistic to think I would not eventually do it badly. The compact fluorescent that lights my basement stairs sometimes likes to wait just long enough for me to miss the top step before producing light after I turn it on. A little bit of light would help there too.

Our house of the future would probably need some built in higher level lighting. A ceiling mounted light for the dining room table is nice. Vanity lighting for the bathroom mirror is essential to allow the residents to maximize their apparent reproductive fitness. Kitchen counters and the stove would need bright lighting. Pretty much the rest of the lighting inside the house would be provided with lamps.

When one deals with lighting one is dealing with some fairly primal stuff. I always need to turn on the main light in the bathroom before having a bath. I think I am afraid to go in the dark water. Perhaps at some level I am afraid of alligators. When fluorescent lighting become popular people immediately started illuminating work spaces to some almost insane levels. Legend has it that photographers of that era used to have trouble believing the result of their light meter readings. They were seeing outside daylight levels of lighting inside buildings. A decade or so later "de-lamping" became popular when it became fashionable to care about power consumption. My point is that people like it when it is bright indoors and the brighter the better. Perhaps low level lighting as a way of life will never become popular simply because of that.

Here are some interesting/related links;

• Don Klipstein's Lighting Info Site
• The LED Museum
• Color Vision
• Scotopic Vision

posted at: 22:37 | path: /ledlight | permanent link to this entry