Specific aim 3: a dynamo wheel testing jig

In order to test my buck converter with a dynamo I need to be able to spin the wheel at different constant speeds.  This is not done practically by spinning the wheel by hand, so some kind of motorized system is required.  Fortunately, (DIY dynamo lighting legend) Martin over at pilom has already built one and he gave me some helpful advice on how to build my own. An abandoned bicycle fork recovered serendipitously from under a bridge and a washing machine motor salvaged from my in-laws' basement are the main ingredients.  Rudimentary speed control comes from a router speed controller, which I believe is TRIAC-based.  It took about 3 hours in my Dad's wood shop to put it all together:

hub dynamo testing jig with 700C wheel

1/3 hp washing machine motor. Routed slots allow motor adjustment to accommodate different sized wheels

The 6 inch drive wheel is based on achieving a maximum speed of 50 kph, or close to 400 RPM, from the motor's full speed of 1750 RPM.  Speed control is not great; the sweet spot cruising speed (25-35 kph) is very difficult to maintain.  In this region the motor's relay cycles on and off, resulting in a sinusoidal speed with a period of a few seconds.  Constant speeds in the 5 kph - 20 kph are easily achieved, then it cycles and finally takes off to a maximum speed.

So far so good. The contraption remains stable with minimum vibration at high speeds. I ordered a super cheap bicycle computer from DealExtreme to measure the wheel speed, but while I waited for it to arrive I figured how to calculate the speed based on the pulses coming out of the hub using a zero cross detector.  The cycle computer has arrived, so I'll be able to see how close my zero cross dynamo speedometer works (more on that later...).

Specific aim 2: a dynamo digital buck converter

I am developing hardware and software for a dynamo peak power tracking system.  Maximum Power Point Tracking (MPPT) is something I learned about over on a thread at CPF and have subsequently become fascinated about.  Widely applied in the solar industry to extract the maximum available power from a photovoltaic array under all conditions of illumination, MPPT is, in essence, a way to tweak the load resistance to reach a peak power point on a current/voltage curve under dynamic conditions.  Bicycle dynamos operate primarily as constant current sources. They reach a saturation current at reasonably low speeds and the only extra power to be gained from a 'passive' system is through the increase in voltage that comes at higher speeds.  Passively extracting more power is done typically by adding more LEDs in series. If you have 1 LED with a Vf of 3.0V you can get a maximum of 3V x 0.5A = 1.5W.  If you add a second 3.0V LED in series you can get 6V x 0.5A = 3.0W, a third gives you 9V x 0.5A = 4.5W, etc.  The problem here is that for each LED added there is an increase in the minium speed at which the dynamo reaches Vf of the series chain. Below that you get no light, or, at best, very blinky lights.  However, there are some clever passive low speed boost circuits out there to get more power at lower speeds.

Today's power LEDs have maximum currents in the range of 1-3A, so a dynamo that saturates at 0.5 or 0.6A is not able to run these even close to their maximum theoretical outputs.  I want to a employ a single LED design for front and rear lamps, so running the LEDs at more than the dynamo saturation current would be useful.  In order to pull maximum power out at all times, a switched mode buck converter can be employed.  I'm working on a digitally controlled synchronous buck converter that uses LED current and bicycle speed as feedback.  Thus far I've developed a peak power tracking system that works with a bench top supply.  Essentially, under fixed current conditions (like a hub dynamo), it modifies the pulse width of the switching signal to maximize the power from the available voltage.

200 mA in, 370 mA out!

LED Current vs pulse width duty cycle at different input voltages where Vf of the series LEDs is 4.1V

Putting dynamo LEDs under microprocessor control may seem excessive, but I'm not the first to do it.  In theory, it should be possible to drive LEDs up to and beyond 1A at cruising speeds, with the obvious caveat that the extra power has to come from the cyclist's legs!

Buck converter hardware

Hardware is under Arduino µC control

The software still needs some significant tweaking to work with the fluctuating DC of the rectified dynamo output.  I'll post a full description of this buck converter soon, hopefully once I have the peak power tracking working with a dynamo.  I'm using the Arduino platform for development, which is great and easy to use, but I did need to employ a few under the hood tricks to get it working. I will ultimately have to switch to another, smaller AVR microcontroller for the final design.

Specific aim 1: making an LED light bulb

I need to mount a single LED with sufficient heat sinking into the base of an E10 Edison screw thread bulb.

I've collected some lovely French lamps. Historically, these would be powered by a bottle dynamo.  My intention is to power them using one of the... um... 10 or so hub dynamos that I own.  Old incandescent bulbs don't light up much so a LED upgrade is in order.  Experience has taught me that commercially available LED E10 Edison bulb upgrades are OK, but I think I can do better, especially with the selection of high power LEDs now available.  Power LEDs get hot, so whatever retrofit I come up with will need to be able to deal sufficiently with the heat.

Late night scrawlings

Thus far, I think I have an idea involving a turned copper pillar of some sort that can be pressed/epoxied into an E10 bulb base.  Something like this:

Copper pillar LED heat sink

Pressed into something like this:

MES/E10 lamp base

This way, any cycle lamp that takes an E10 threaded bulb can be retrofitted without modification.  Getting the heat out of such a small package will be a challenge, though; something that my previously proposed solution might not be up to.  The best solution would be to mount the LED directly to the copper heat sink, which eliminates any of the thermal resistance encountered in an FR-4 board perforated with plated vias or in a metal core PCB that is then attached to a heat sink with thermal epoxy.

There are several elaborate methods described for doing this. Lambda Lights seem to have mounting LEDs directly to a large copper heat sink down to a science, so I might contact them before attempting to reinvent the wheel.  From what I can tell, they turn a copper pillar and then machine a square nub on the top where a PCB is mounted with a slot milled in it so the nub can poke through and be soldered directly to the thermal pad of the LED.  Something like this:

While ideal, this solution may be cost prohibitive...

Sturmey Archer lamps: second attempt

Back in the summer of 2008 I was quite proud of myself when I managed to get a hold of a set of NOS Sturmey Archer lamps for my wife's Raleigh Superbe from a Canadian seller for about half of the going rate at the time ($67, to be exact).  Two years later my faith (what little was left) in humanity was shattered when a very bad man (I've called him worse) very craftily stole the rear light.  This faith was somewhat restored when a very nice reader of this blog offered to send me a replacement rear lamp at no charge (thanks Jeff!).  A last minute discovery of an auction with less than an hour to go has turned up another NOS Sturmey light set, also going for well below market price (and also, coincidentally, sold by a Canadian seller).  I bid, finished my dinner and was delighted (but not surprised) to find that I had won!

Sturmey Archer NOS light set

Just a week ago I won an auction for a NOS GH-6 Dynohub, a 1978, from the same seller:

1978 NOS Sturmey Archer GH-6 Dynohub. Not quite as elegant as earlier models.

Setting goals and things in the works

Being housebound taking care of two small children leaves the mind with enough excess capacity to come up with an increasingly complex and ambitious list of aims.  To try to keep track of all the stuff bumping around up there, I'm going to commit some of these goals to this here blog page.

Specific aims:


1.  Mount a single LED with sufficient heatsinking into the base of an E10 Edison screw thread bulb.


2. Develop hardware and software for a dynamo peak power tracking system. 


3. Build a speed-controlled dynamo wheel testing jig. 


4. Upgrade the Sturmey Archer GH6 magnet to modern Neodymium magnets.


A flurry of posts will address each of these aims.

Luxor dynamo lamps

These really are my favorite dynamo lights of all time. I have one set of the hammered aluminum Luxor Martele lamps. I find them to be achingly beautiful.  Just stumbled across a NOS set on offer through Classic Rendezvous back in February of this year.  Have a look.

getting the heat out of LEDs (as cheaply as possible)

Well, it turns out designing a PCB for power LEDs isn't as simple as I'd previously thought.  Now, to be fair, with the intention I had in mind, I was only going to run my power LEDs at about half of their rated maximum current. With that in mind, I'd put a single via through the thermal pad to connect it to a large plane on the bottom side. It was also connected to a plane on the component side. Like so:

Cree XP-G SMT pad with via through thermal pad and connection to component side thermal plane

I figured that was a lot of copper to dissipate the heat.  After some more reading, however, I realized that while my design was probably adequate for my current requirements, there was no way I was going to get away with running the LEDs at their maximum current.  At about 300-400 mA the thermal planes are quite warm, but not alarming.  At 500mA the LEDs on this design do get pretty darn hot to the touch and that isn't good as it reduces their efficiency which ultimately results in more heat, less light and reliability issues.  I do have a planned application where I want to run these guys up to 1A or more, so my current PCB design needs revision.

Many power LEDs come mounted on the familiar metal-core PCB star:

Cree LED on MCPCB star

MCPCBs have a metal core that conducts heat much more efficiently than the standard FR-4 PCB laminates from the top layer to the bottom layer (where a heat sink can be attached).  In other words, the metal core has a much lower thermal resistance than FR-4; 5-10 times lower.  MCPCBs seem like a great solution for mounting LEDs, but they are very expensive to have custom made.  Fortunately, there are several application notes out there that describe how to design an FR-4-based PCB using thermal vias to get a top to bottom layer thermal resistance closer to metal core.  I used this one from Cree to guide my design.  The application note is summarized thusly:  instead of using a metal core to get heat away from the LED's thermal pad, use as many thermal vias as you can fit in the thermal pad to transfer the heat away from it through to the bottom layer, where a heat sink can be attached.

First, I had to rejig the XP-G solder pad, as the orignal layout created a real bottleneck for heat transfer:

The big 'I' can now be filled with vias to conduct heat away from the thermal pad.  Riddling the board with thermal vias isn't necessary; just getting the heat way from the immediate vicinity of the thermal pad has the greatest effect.  Here's what the PCB design looks like:

Lots of thermal vias.  On the back there is a 14x14mm square solder pad to attach a copper heat sink:

These are little RAM sinks.  You can get 8 for about $15.  They come with their own thermal adhesive already attached but the general consensus among user reviews to to scrape that off and use something like Arctic Silver thermal adhesive instead.

Hopefully the vias will provide sufficient transfer to keep the LED cool enough to remain bright and not overheat at full current (1A+).  They are the bottleneck between the LED's thermal pad and the heat sink. The application note recommends more vias in the actual thermal pad itself but my cheap PCB fabricator has a minimum drill size that doesn't allow this.  Following their recommended design they claim an FR-4 PCB can have a thermal resistance close to that of a metal core PCB.  In my case, I doubt I can achieve that with only 3 vias in the pad proper (they recommend something like 14!).  If these aren't sufficient I might have to find another fabricator that can do smaller drill hits, although this could ultimately defeat the purpose of doing this on the cheap!

Another potential pitfall here is that BatchPCB, who I'm using for fabrication, only fabricate 1 ounce copper PCBs.  Thicker copper would provide a larger conduit to take the heat away, but for the sake of not spending a fortune on prototyping, I'm kind of stuck with their 1 oz limit.  I placed the order last night, so I'll know if a few weeks if this design will work.