This has some advantages. First, as will be seen below, varying the frequency is very easy. This is especially so, because varying the voltage is not too easy. Most of these flash circuits essentially have the batteries in short circuit, that is, there is no current-limiting resistor. If you try to put a voltage regulator in its place, the behavior will be completely different (if you can get it to work), because the circuit expects the battery voltage to drop as more current is drawn.

As I was learning about strobes, I started looking more deeply into the Sunpak 622 schematic, and I could see the similarities between it and the disposable strobes. After reading that the recycle times with the Sunpak AC adapter are actually slower than with batteries, I wondered if I could just put a DC supply in place of the batteries and achieve a super fast recharge. The biggest AC adapter I had was something like 6A at 9V, and it was constantly cutting out (internal protection, I guess). By the time I added enough resistance (long wires) between it and the flash, so that it could run continuously, the recharge time was not all that fascinating.

Then I finally realized I had a use for the 5V, 40A supply that was on sale at the Electronic Goldmine. I have to say those guys are a thing of beauty. Anyway, that supply can recharge two 622′s to full power simultaneously in less than 10 seconds. Not only is that fantastic, but the whine of the two flashes plus the supply working hard is very much reminiscent of the Ghost Busters’ ghost zappers.

So back to the task at hand—being able to vary the output voltage of a flash capacitor charger by not having to deal with the input voltage at all was a blessing. To try it out, I built a simple circuit with a 555 timer and a transformer off my numerous disposable cameras.

There is plenty of 555 stuff on the web. But none is as useful as this circuit simulator. With it I quickly picked the components I needed to start playing.

The resulting circuit is below. I don’t have any information on the transformer (orientation, number of turns, or anything). Suffice it to say it came from a Kodak camera and its characteristics are not important for this exercise. For simplicity, I was running the 555 off a DC power supply and the capacitor charging from a single AA cell (just like in the disposable cameras).

(Note: the schematic above and the text in this paragraph were edited for small errors on 2008-12-20. The schematic did not show a necessary connection between pins 2 and 6, and the description had reversed the roles of R2 and R4.) First, forgive me for the ugliness and confusion of the schematic. I don’t have much practice drawing them, and I’m trying to learn Eagle at the same time. Anyway, there are two adjustable quantities in the circuit: the input [to the transformer] pulse width (varied by changing R2) and the input frequency (varied by changing R4). Without getting into any calculations, I used the simulator mentioned above to test out some values and then used what I had in my bins for the circuit. I monitored the input with my oscilloscope and the capacitor voltage with a multimeter. The minimum pulse width achievable was 45 microseconds; I could vary this up to more than 1.7 milliseconds. The frequency range was from around 140Hz to 9kHz.

As expected, the pulse width does not make too much of a difference; the transformer steps up the voltage according to derivatives of voltage in time. With the really long pulse widths the voltage actually dropped (comparing to the same frequency); I suspect it was because it was draining the battery more between pulses—I don’t really know. The drop is negligible anyway. When I do this again, I will see if I can go shorter than 45 microseconds and see what the difference is; I suspect at some point inductances and capacitances will prevent the pulse from even happening and performance will drop considerably.

For the capacitor I used a 1000V 4700pF ceramic, simply because it charges very quickly, and because I didn’t have anything big that could go to 1kV or higher. In any case, here are the results:

Impressed? I am. The purpose of this circuit will be to test some capacitance/voltage/bulb combinations. I may have to fork out some cash for some more expensive capacitors, because of the ones I have, only one goes to 800V.

The first thing I did is modify the layout of the Fuji circuit I was using. Soldering and desoldering the capacitor started to rip the traces off the board, so I soldered some solid wire to have leads (to which I would in turn solder the test capacitors). I didn’t have solid wire so I used the leads off a high-voltage capacitor. The mechanical trigger switch (normally hit by a moving part from the shutter mechanism) I replaced with a pushbutton, even though it was relatively easy to actuate as it was. The flash “on” button (normally actuated in the front of the camera) takes a bit of pressure and is dangerous when the whole circuit is exposed like this, so I also replaced it with a pushbutton. I put a power switch on the battery holder (I used a metal 2-AAA holder because it wasn’t pre-wired for series or parallel) so I could stop the flash from recharging (typically they auto-recharge after firing). As always, I hooked up my trusty capacitor discharge tool to the capacitors before charging them. This time I also connected my voltmeter to that so that I could see what the capacitors were charging to (which was a surprise; see below). I also hot-glued the flash bulb assembly (which luckily in this camera comes out cleanly with the reflector and “lens”) to the board so it would be in a consistent place, and held the board in a small vice always at the same angle for the tests. With a tape measure, I put the flash facing more or less the same direction some 16.5 inches from both my Sekonic Flashmaster L-358 and my home-built strobe light output test circuit.

I retested the capacitors from last time so that I could have a true comparison. The Flashmaster gave me a photographic exposure value, and at 16.5 inches away, the setup represented a realistic setup so the number should be quite valid.

The expensive capacitors tested were all obtained from Mouser (which had a better price over Digi-Key). The table below shows all the capacitors tested, including the ESR as measured with Smart Tweezers, and the exposure measured with the light meter set to 400 ISO (16.5 inches or so away from the flash).

Each capacitor was fired twice. Surprisingly (for me anyway), the smaller capacitors easily charged to higher voltages. In some cases I got scared, especially since the board as-is would charge the capacitor it came with to a voltage higher than that written on the label. I did not record the “decimal number” on the aperture measurements from the light meter, thus the capacitors with asterisks (*) showed both values depending on how high a voltage I let them be charged to.

The first thing to note about the expensive capacitors is that they are enormous. All of them were bigger than the entire circuit board. A custom board could elegantly be mounted to the capacitor housing, making a pretty neat little unit.

The first plot above includes both the expensive capacitors and those tested last time. The second plot shows only the expensive ones. The curves for the Epcos and Cornell-Dubilier 15μF capacitors are effectively identical. The cost for the Epcos part is nearly 4 times less—though, in all fairness, the voltage rating of the CDE part is 800V (vs. 450V for the Epcos). The low ESR of these capacitors (as compared to the cheap ones) makes the peak intensity of the pulses higher, but, more importantly, they shorten the “tail” at the end of the pulse. Compared to the cheap 22μF capacitor, the peak intensity is some 10% higher for the expensive 15μF capacitors. The intensity at 50 microseconds for the cheap capacitor is perhaps only half of the peak intensity, whereas for the expensive capacitors, it is almost a factor of 8 less—meaning we would have a much better guarantee of a truly short exposure with the expensive capacitors.

Increasing the capacitance to 25μF increases the exposure by one stop, but the pulse length increases by some 50% and the price by over 50%.

Conclusion

The $8 price tag for the Epcos B32676-G4156-K is certainly attractive. It meters at the same exposure as the 90-cent 22μF capacitor—f/11 at 400 ISO. I should reiterate that these exposure values are without any flash-focusing whatsoever—only the reflector that came with the camera was used. It may be possible to gain even one more stop with a snoot or more “active” methods of guiding the light (lenses and/or reflectors). The most recent conclusion from the lens tests is that we would use an aperture of f/11 or f/16. So, surprisingly, it seems that we don’t need a flash array at all! If we go with the Epcos, we may have enough light with just one or two flashes at an exposure length of 50 microseconds or so. To gain one stop, we must double the number of flashes. The following table counts the number of Epcos-powered flashes necessary for a given film/aperture combination.

The cost of getting [guaranteed] identical flash circuits is some $3 (expired disposable cameras). The main parts of interest are the transformers and flash bulb, regardless of whether or not a custom board is made. The capacitors, as mentioned, are some $8 a piece, though at quantities of 10 and higher there is a slight price break. This means that for some $250 it would be possible to build a 16-flash array with a short pulse length and enough light to expose 50 ISO film through an aperture of f/16, all while having logic-level triggering (with extra parts, of course).

A lower ISO film speed is important in case our shutter does not have a fast enough shutter speed, because a lower ISO will ensure less natural light blurs the image. Then again, increasing the number of flashes in the array may increase the total pulse length due to small differences in the components and triggering. There is no way to know the effect until multiple flashes are built and fired simultaneously.

Also, this may be a good time to test the lenses again. Last we left it, we still have to choose an aperture based on sharpness/depth of field balance. A test target (a dead bee maybe?) could be used to test the depth of field and actual exposure quality with these flashes, and a fast moving target (such as a rotating disc) can be used to test the actual pulse length.

I have bought a few different cameras from ECameraFilms to evaluate which would be easiest to modify for the final camera—I want to be sure all the circuits are identical, and I figured my best hope is to buy all the disposable cameras together. In the next post I’ll report what I find.

Sam recommended that the first test I should do to try to shorten the flash duration is to simply use a smaller capacitor. So I took one of the disposables I have—a Fuji, I think—and did just that. The camera had a 100μF capacitor on it; I tested along with that 10, 22, 47, and 82μF. None of them were anything special—just high voltage (at least 350V) electrolytics.

To do the tests, I resoldered the capacitor, then fired the flash (from about the same distance every time) toward a light meter much like the one on Sam’s site. The distance and angle were not always exactly the same, but it was certainly sufficient to check for basic behavior. The results are shown below (normalized to the maximum intensity of the largest capacitor).

As you can see, the minimum usable flash duration I got was with the 22μF capacitor, and is somewhere between 50 and 75 μs (this would obviously depend on the optical system and the exposure curve of the film). With the 10μF capacitor, the flash does not reach its peak output. At 22μF the peak output is not as high as with the stock 100μF capacitor, but I figured I would be making a large array of these anyway (and focusing them as best as I could). If I cared, some pulse-shaping electronics (inductors) could make the pulse much sharper, but I doubt its “tail” would influence the image all that much.

These preliminary tests were easy to do. But is this pulse duration short enough? According to “The Aerodynamics of Hovering Insect Flight. III. Kinematics” by C. P. Ellington (Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences, Vol. 305, No. 1122, Feb. 24 1984, pp.41-78), the honey bee Apis mellifera flaps its wings through a 131-degree arc at 197 Hz. Ellington’s measurements show that during the stroke, the velocity of the wing tip is nearly constant. For the sake of calculation, if we assume a 120-degree stroke at 200 Hz, then over 1/400th of a second the wing swings 120 degrees. Since the velocity is constant, we can say the wing tip is like the edge of a solid disk rotating at 400 x 60 / 3 = 8,000 RPM. According to this site, the length of the bee wing is 9mm or greater (we’ll say 10). So the velocity of the wing tip through this constant-velocity stroke is 10 / 1000 x 2 x π x 400 / 3 = 8.4 meters per second (19 miles per hour).

So if the exposure time is 1 millisecond, the wing will “move” 8.4 mm—an awful amount of blur (one down or up stroke takes 2.5 milliseconds, if the beat fequency is 200 Hz). At 50μs, this is reduced to 0.42 mm. Not bad!

But these pictures will be big. If the magnification is 1:1, then the wing’s image is also moving 420 microns on the film plane, but if it’s 2:1 (as is likely it may end up being) then the wing moves 840 microns—almost 1 mm. A high-end film scanner these days can achieve 4800 DPI. Typically, the pictures will be printed at 300 DPI (for 6×6 film, this comes out to a 38×38 inch print) which is a magnification factor of 16. That means for a 1:1 camera the wing blur will be almost 8 mm and for a 2:1 camera almost 16 mm. In both cases, this will be about 4% of the size of the bee’s wing on the print. In a worst-case scenario, having multiple flashes firing at once (and not being identical), I’d say we can expect an exposure time of some 75μs, which brings the blur up to 6 to 8% of the bee’s wing.

The nice thing about all this is that bees flap their wings unusually fast for their size. Other insects of that size will flap much slower. Bumble bees flap around 150 Hz, but, for example, the crane fly Tipula obsoleta flaps at around 60 Hz.

It’s hard to judge how sharp the image through the lens is in these terms—but you can bet with a high-end lens the image of a point source will be well below 400 microns. In other words, the motion blur will be discernible in the image. Still, these results are encouraging, because with the 2:1 camera (and the 16x printing enlargement), there’s 16mm (some 5/8 inch) of blur on a bee that is probably around 60 centimeters (almost 2 feet) long on the image.

I think next I’ll try a few other tricks to shorten the pulses, then maybe go on and try to build the array and start measuring (with a photographic light meter) how much usable light I can get out of them.

Free Email Dating Service
Single Night Club
Online Dating Service Uk
Top Online Dating Service
Valentines Day Dating Game
Speed Dating Clapham
Single Web Site
Dating Sexy Girl
Asian Female Single
Sea Of Love Dating Service

The Kodak MAX-derived supply was charged until it shut off, and the voice coil was “fired” in one direction with a one-millisecond pulse. To know how charged it was, I had a multimeter carefully clipped to the capacitor – and the high-side collectors. Whatever leaks in the circuit (could it be just the capacitor?) makes this thing seep voltage pretty fast, so I tried to fire as soon as I could after the supply stopped charging the capacitor.

Using the “afterthought” flash trigger I set up, I took photos at 1.0, 1.5, and 2.0 milliseconds after the leading edge of the input trigger. Thus in the animations you see below, each frame is a separate firing action, so the voltage is probably not exactly the same for all of them (and neither may the relative position of the drive in the picture—though I aligned them as best as I could). I used my Sunpak 622 for illumination (set to 1/128th power) and measured its illumination pulse width to be about 100 microseconds. I tuned the timing so that the flash delays mentioned above were measured to the half-power point of the flash pulse.

The room was pretty dark, and I used a digital camera set to a 2 second exposure so that I could manually fire the voice coil so that it would “land” on a picture. The aperture on the lens was tuned to provide enough light given the flash’s output. Since the room was not absolutely dark, you may notice a red “ghost” image of the hard drive head; this is its position at “neutral” (which, for one of the drives, was not always the same). I didn’t use a spring or anything else to hold the head off to one side; I just let it sit where it does (which is governed by the flex-circuit going to the voice coil and the heads themselves for all drives but one).

I had an oscilloscope monitoring the signals: the TTL input to the optoisolator, whose leading edge was considered the “fire” time, the optoisolator output into the IRS2183 input (to check for inductive spikes, etc.), the flash pulse (via the very simple photodetector circuit from Sam’s site), and the Kodak’s capacitor’s voltage.

The expended energy is calculated from the measured capacitor voltages before and after firing:

The average power is simply this amount divided by 1 millisecond (the pulse length of the discharge). So if the energy is in Joules, then the average power is the same number in kilowatts. The speed of the head (in RPMs) is simply the angle traversed divided by the time between the leading edge of the power pulse and the photo being taken, so it is an estimation of the average speed for that period.

The results are really fascinating. There is actually motion blur in the heads—with a 100 microsecond exposure!

Drive 1

This one I had ripped the flex circuit off, so after much patience, I managed to peel the coil wire (without destroying it) and soldering some wire to it. The thickness of the big wires sure didn’t help performance. Also, since AC current flows on the surface of conductors, it is probably best to use solid wires, but I didn’t have any on hand (and it would have gotten in the way even more). You can also see in the 1.5 millisecond frame that the head had started in a different place than the others. Really, this wasn’t a reliable test. Still, I include it for completeness—since it was one of the two drives that survived (if you don’t count the last one).

• Energy expended: 4.74 J (4.74kWs)
• Angle traversed at 1.0ms: 6.3 deg (1050 RPM)
• Angle traversed at 1.5ms: 16.2 deg (1800 RPM)

Although it’s hard to tell since the drives were not guaranteed to stay put after a firing, it seems like this one’s head is on the verge of breaking off.

Drive 2

• Energy expended: 3.26 J (3.26kWs)
• Angle traversed at 1.0ms: 12.8 deg (2133 RPM)

This drive was fast. But it didn’t last. The head bearing was held to the chassis by a hollow shaft that sheared off after the first firing, presumably when the head hit its physical stop.

Drive 3

• Energy expended: 4.88 J (4.88kWs)
• Angle traversed at 1.0ms: 14.6 deg (2433 RPM)
• Angle traversed at 1.5ms: 27.4 deg (3044 RPM)
• Angle traversed at 2.0ms: 36.1 deg (3008 RPM)

This was a fast, solid drive. At 2 milliseconds, it had hit its stop, and since this one has a “resting” position near one end of its swing, this represents the maximum stroke. Recall from previous analysisthat we need over 100 degrees of travel on each side of the shutter for the wedge to have minimum inertia. This is not possible with a stock hard drive head—the two options are to use different magnets that allow for longer travel, or make the wedge long enough so that these 30-odd degrees are enough to clear the lens opening needed. According to the last lens tests, the lens opening necessary will be at most around 30mm; at 36 degrees this means the shutter arm length (wedge radius) must be around 120mm (almost 5 inches). The performance is bound to suffer with such a long arm, and because these accelerations are pretty violent so the wedge’s structural integrity will be a limitation. With enough identical drives, voice coils could be mounted to work together, especially since it would probably be possible to machine them down and fit two on the same bearing (magnets and all). This would effectively double the force if the power supplies are independent. Note also that, depending on how much abuse the coils can take, it may be possible to use a bigger capacitor and longer pulse widths to pump more current through them.

Drive 4

• Energy expended: 4.94 J (4.94kWs)
• Angle traversed at 1.0ms: 12.3 deg (2050 RPM)

This drive didn’t do so well. The coil’s frame (or “cradle”—that’s my word for it) is made of plastic (all the other drives had aluminum ones). The minute it slammed into its physical stop half of it broke off.

Drive 5

• Energy expended: 5.61 J (5.61kWs)
• Angle traversed at 1.0ms: 4.4 deg (733 RPM)

This was the ancient giant. Six or eight platters (can’t remember), and the only one with a significantly different voice coil design. First, its energy consumption was huge, with little performance (probably because the head is so bulky). Instead of the coil sitting flat in an air space between two magnets, this one is actually wound into two slots in a stack of magnets—it is impossible to separate the coil and the magnets. This drive was strong, but slow, and after the first firing there was a pretty strong spike in voltage in the storage capacitor and the input of the IRS2183 (and probably everywhere else in the drive circuit) right at the shutoff of the coil current. This actually disabled the drive circuit, and I thought it had fried one of the IGBTs, but by the time I had taken the setup apart to check the parts, I couldn’t find a problem, and a subsequent test showed the circuit seems to be fine. Still, I didn’t want to risk it, so I didn’t fire it again.

Conclusion

Smaller drives are better. I don’t mean smaller capacity (although it may still correlate to some extent), but generally it seems that the lower the number of platters the better, because the head will have less stuff hanging off it. Of course that may not matter much in the long run because the shutter wedge will replace all that.

The most important conclusion didn’t require driving the voice coils at all—they do not swing far enough to implement the wedge with a minimum inertia. Getting around this will require custom magnets (as in magnets not from the hard drive) or coping with a long shutter arm.

I’m going to see if I can get my hands on a laptop drive.

Edit: as far as extending the swing by using different magnets, this is only possible in the construction like that of drive 5. More on this later.

As I mentioned, I had no idea what I was doing at the beginning, so the driver circuit went through several iterations:

Version 1.2: This was a complete failure, because all switches were N-channel IRF740′s driven from 0 to 5V signals. The high side switches cannot be N-channel without special arrangements. Also, 5V is not enough to switch them, and, most likely, the IRF740′s couldn’t have put up with the abuse. I had put in provisions to hold the voice coil to one side or another—a low voltage supply to replace the high voltage flash circuit—but this didn’t work because of the flyback diodes built into the IRF740. Still, the circuit worked at low supply voltages.

Version 2.0: This was the first version with the IRS2183 and the STGW30NC60VD. I abandoned, for now, the idea of a “hold” circuit. At some point, one of the driver chips died. I was also getting a lot of spikes, so I needed to add capacitors to suppress these.

Version 2.1: This had bypass capacitors for VCC on the driver chips, and also capacitors between the gate and emitter of the IGBT’s. After some successful tests, the circuit began behaving erratically and draining the batteries. Turned out one of the IGBT’s had failed—the gate was conducting straight to the emitter.

Version 2.2: After replacing the dying IGBT, everything seemed fine. Still, there were a lot of ripples in the gate signals and in the capacitor voltage right when the voice coil turns off, so I tried to add some small capacitors right at the output of the H-bridge. These have to withstand the full 300V, and the smallest I had was 22nF, which wasn’t enough. Two of them in series gave 11nF, which almost completely killed the spikes, but not quite. The schematic of this version is here.

In this last version I also added a quick-and-dirty provision to trigger a flash from the signal generator so I could take precisely timed pictures and see just how fast the hard drives were moving. At first I reached for an IRF740 (that’s the first thing my hand lands on most of the time) and, no matter how I connected the two poles of the trigger cord from the Sunpak 622, it would fire the minute I connected it to the MOSFET. Tried another one (TIP29C or something) and the same thing. I have no idea why, but I could only properly trigger it with an NPN transistor, connecting the positive trigger to the collector and the negative to the emitter. I’m feeding this straight off the optoisolator (see below) and I didn’t even bother looking at what the specs are on the transistor, so my guess is it will probably die one of these days. I should hook something up a bit more carefully.

The circuit is powered by two 9V batteries in series to provide 18V. I definitely need to get around this, because these batteries are stupidly expensive.

The Power Supply

The power supply for the voice coil actuator comes from a Kodak MAX disposable camera. There are several circuits for these cameras depending on when they were made—see Goldwasser’s page. I drew a schematic of the circuit, which requires removing the transformer and checking the winding direction (if you care enough). The schematic is here. I also have some high-resolution pictures of the board: the back and the front. I’ve also labeled some of the parts according to the schematic on the back and on the front.

To use it as the power supply, it’s necessary to remove everything to the right of C1 in the schematic. I replaced the very rudimentary charge button with a better one (so that I could put the whole thing in an enclosure), and added a switch to turn off the battery power—this is how I limit the charge in the capacitor.

The power supply is powered by a single 1.5V battery, just as the disposable cameras are. My guess is that caution should be taken if powering it from a power supply that can source a lot of current (or a big battery) because the circuit basically shorts the battery when it’s running—batteries respond to this by lowering the voltage and thus the current gets limited to some value, but something that can put out more current (including NiCd batteries) may fry everything; I don’t know.

The Input Signals

Timing is controlled by a pulse generator I borrow from work. (Now that the circuit seems to be working, I want to put it all in a small microcontroller to make it all portable.) To protect this [expensive] piece of equipment, I built a really simple optoisolator circuit using two Lumex OCP-PCT4116/E optocouplers (for a total of 8 channels). These are not particularly fast, but at this point, I don’t really care. Using them is simple; the LED side only needs a current-limiting resistor, which can be shared between all of them (all the cathodes tied in parallel to it). I used 220 ohms. On the output side, a resistor must be placed to make sure the output is something when the transistor is not conducting. I think if the collector of all the transistors are connected to the supply voltage via some resistor, then this can be achieved, but the output will be inverted. I didn’t want to deal with that, so I used one resistor for each emitter (actually, a resistor network) connected to ground, with the output between the emitter and the resistor. Thus when the transistor is off, the resistor pulls the output to ground, and when the transistor is on, the current through the resistor provides a voltage drop such that the output is the supply voltage minus the diode drop of the transistor.

Originally (before the IGBT driver) I used a 1MOhm resistor for the transistors to save power but this is a horrible idea. MOSFETs’ and IGBTs’ gates are basically capacitors. So if you have a huge resistor, they turn on and off at some unpredicatble delay because of the slow charge/discharge rate. At the time, this wasn’t actually the reason I changed the resistor; I actually calculated it based on the Ic = 1mA quoted in one of the transistor specifications for the OCP-PCT4116/E. Since I was using 5V for the output at the time, I ended up with 4.7kOhm as the standard value.

Calculating it based on transistor characteristics I think is bogus. Obviously one should take care not to fry the transistor by pumping 20A through it accidentally, but eventually, once I had the IRS2183 circuit, I noticed that this load resistor on the transistor had some interesting effects.

From then on I was powering the output side of the optoisolator with the same 18V I was using for the driver circuit. I was still using the 4.7kOhm resistors on the transistors. I was using a DC power supply for the voice coil rather than dealing with the high voltages of the disposable camera, and at above around 20V, the driver circuit would behave normally at powerup, but after firing the driver once, some oscillations would build up on the input pins of th IRS2183 which were turning the IGBTs on and off at extremely high frequencies. By accident, I noticed that if I just shorted the inputs to ground, the problem would go away. I remembered then reading somewhere in the many IR documents that these chips have a pretty weak (meaning large) pull-down resistor on the inputs. I changed the 4.7kOhm resistors on the optoisolator output to 270 ohms and the problem went away. This means while an input signal is high the resistors are actually draining a lot of power, but since the pulses are in the millisecond range, I don’t really care. Another option is to use a transistor or FET to short the inputs to ground when the input signal is supposed to be low, but there is no need for that.

As mentioned, the optoisolator output is powered by the same 18V battery pack as the driver circuit.

The Circuits in Action

Just so you won’t get discouraged, the developments covered in this post took me about 2 months of working pretty much every weekend on it—and obviously learning a lot. In my next post, I’ll show some results with actual hard drives that I’ve collected off junk piles.

So in the interest of getting something done, I started working on an actuator suitable for the impulsive shutters.

Continuing from my first tests, I first considered solenoids. But it seemed to me that if a solenoid works because of the ferromagnetism of the plunger, then if the plunger were a permanent magnet, the actuator would be more efficient. As it turns out, that’s what a voice coil is. And the best place to get a rotary voice coil is a hard drive head.

So the idea was simple. Take an old hard drive apart, and wire the head coil to a 300 Volt supply that could be precisely pulsed. Such a supply would need a storage capacitor to source high currents, and the best place to get such a supply is a camera flash. After reading through Sam Goldwasser’s site, it became obvious that the best source for that is a disposable flash camera. I went to two local drug stores; the first refused to give me any of the used ones, but the second one had a kid behind the counter and he gave me my choice of about 10 used disposables with flash.

To enable the head to be actuated in either direction, I decided an H-bridge was the answer. I spent about a month trying to build the circuit, because I had no idea what I was doing. At first, I tried to build it simply with four N-channel MOSFET’s. Although the circuit fired, it behaved very erratically, and, well, it should have never worked in the first place. The problem is that in N-channel MOSFET’s, or IGBT’s, the gate is referenced from the source. For the switches at the bottom of the H-bridge, the source is the ground, or the power supply minus, so driving them is easy. For the ones on top, the source is the voice coil itself, so the gate drives have to have a voltage that is relative to the power supply voltage minus whatever voltage drop there is across the MOSFET or IGBT (or whatever else is doing the switching).

There are several ways of providing this high-side gate voltage. One of the easiest is to use an optoisolator to translate a typical pulse, for example, of 0 to 5 V, to the appropriate level—say, 300 to 320 V. But I figured there should be a chip to do this. Even though I was already using an optoisolator to electrically isolate my pulse generator from any high voltage mistakes I may make, I decided to look for a chip—at the time I didn’t know how to implement the direct-from-optoisolator circuit (and I’m still not sure I do). There are several IGBT drivers, and, not surprisingly, many meant to drive half an H-bridge (2 of the four switches, either on the same side of the H bridge or corresponding high and low sides), or full H-bridges. There is a large variety of these, but the ones from International Rectifier are rated for 600 V.

The whole thing turned out to be extremely complicated, and it took me about 2 weeks to finalize the design. The first problem is that IR makes dozens of different drivers, most of which, at first glance, seem to have the exact same specifications. My choices were limited to half-bridge drivers, of which there are apparently two types. Some are meant to drive an “I” of a bridge—that is, the high-side switch and the low-side switch that should not be conducting at the same time—and some are meant to drive a “direction” of the circuit (the two switches that are supposed to conduct at the same time). The circuit examples at the bottom of the first page of their data sheets is not reliable, and they like to explain the input/output logic by minimalistic timing diagrams that are confusing. The terminology is not clear: “half-bridge drivers” are evidently no the same as “high and low drivers”. In the end, the only way to figure it out is to look at several different ones and examine the timing diagram. The difference is in input/output logic. A half-bridge driver is mean to drive an “I” of the “H” and thus the high-side output is equal to the input pulse while the low-side output is equal to “not” of the input (which is why they designated as “HIN/LIN(bar)”, thus preventing a short circuit condition automatically. A high and low side driver has both the outputs equal to the inputs (“HIN/LIN”) so if you hook it up to an “I” you get a short circuit. It took me a week just to narrow this down, but in the end I chose the IRS2183 (and I don’t even remember why). I didn’t dare venture into the realm of models with advanced protection features.

After that came the adventure of choosing the component values, which their “typical connection” diagrams don’t detail at all. Thankfully, they have a sort of forum where dozens of people ask the same question. Whoever answers the question gives a different response, but the answer is in their design tips and application notes. They give more or less the same answers, with a bit of variation from one to the other. I ended up using DT04-4, which took me about a week to interpret, though at least they tell you where they are pulling the values from so you can copy their example. For the most part, it is poorly written, and some of the more advanced concepts, such as different switch-on and switch-off gate resistors, are completely lost because they don’t explain how to wire it. So I recommend you read some others, too, including AN-978, AN-1123, AN-944 (seems interesting, though I haven’t read it), DT99-7 (haven’t read it), and DT98-2 (all about component values). For the purpose of occasionally pulsing current through the voice coil, most of the issues reviewed in these documents don’t come up—for example, the bootstrap capacitor will always be ready and charged, especially with the IRS2183′s IO logic which keeps the low-side IGBT conducting (thus providing a path to ground for the capacitor to charge).

So then came the time to choose an IGBT. The voice coils apparently tend to be on the order of 10 ohms, and we are pumping 300V into them for 1 millisecond (actually less, since the flash capacitor will discharge quite a bit). Still, IGBT’s are just about the only option here. Although IR offers quite a few, I opted for an ST Micro because it was much cheaper.

Figuring out the SOA (Safe Operating Area) curves was another surprise, and as always, I must thank Sam Goldwasser for getting me through it. I’ll go over the SOA for the STGW30NC60VD, which is the IGBT I chose. First off, I have no idea why in the data sheet it says “Turn-off SOA”.

The first thing to keep in mind is that this SOA is for the full operational range of the transistor—which includes operation as an amplifier (rather than a switch). The other key is that there are two distinct specifications we shouldn’t exceed. First, our flash supply is approximately 300V, so we cannot exceed the voltage rating of the IGBT when it’s not conducting (in this case, 600V). Second, we cannot exceed a maximum current that is specified by the SOA when the IGBT is conducting. When used as a switch, VCE will be the saturation voltage, so the current is defined by the left side of the SOA curve. In this case, the maximum is 100A. The IR SOA’s typically show different limits for pulsed currents, but the same rule applies—only the left side matters. This IGBT is plenty for our purposes, because not only are we pulsing the current, but the capacitor will discharge relatively quickly so the 30A we roughly predict is only instantaneous.

But we have minimized the inertia of the wedge by making it a disk, and, assuming a constant torque output from our actuator, this is the way to minimize lag and exposure time. Beyond this, one idea is to add a second wedge rotating in the opposite direction. The sequence below shows the shutter opening.

Whereas before the angle from closed to full open was 120 degrees, now it is less than 60. This should save us some time. The diagram below shows us how to calculate this exact angle.

Each disc much turn so that the center of its aperture hole is on the edge of the lens aperture for the shutter to be fully closed. The triangle shown can be used to find this angle; it is an isosceles triangle with base and sides . Thus we can write

and taking from our previous derivations,

And thus

Furthermore, our wedges need no longer be disks, since they are not rotating as far anymore. With the figure below, we can find what angle of “disk margin” we need for the two wedges to completely cover the aperture when the shutter is closed.

Using the figure, we see that

So exactly. This means that now the wedge has to subtend a total of if we want to the shutter to be bidirectional. The angle could be reduced by a bit less than 60 degrees if we decide that we can reset the shutter between shots (in other words, if we allow it to have a margin only on one side of the aperture hole).

So we have reduced the angle from closed to open by more than half, and reduced the inertia of each shutter blade by quite a bit, too. What effect does this all have on performance? With the smaller wedge angle,

which is 54% of what it was before. Our new expression for the lag time is

which is 48% of what it was before. For 1 millisecond lag of the full aperture of a #5 Ilex, the torque required is now “only” 115 Nm (84 foot-pounds), an amazing 23% of what it was for the single wedge shutter (remember we are assuming we have two independent actuators here; with a single actuator we get 46% as much as before). For 300mm at f/11, the required torque is 3.8 Nm (2.8 foot-pounds) for 1 millisecond and 0.24 Nm (0.18 foot-pounds) for 4 milliseconds. The speeds are still insane, but less so—for 1 millisecond lag the shutter is going at 17,100 RPM when full open, and at 4 milliseconds it is going at 4,280 RPM.

Perhaps adding a third or fourth blade will help even more. The geometry is bound to get complicated, and the law of diminishing returns should strike soon. But first, we should discuss other ideas that don’t rely on force for acceleration.

The lag time is important because if an insect triggers the camera and the shutter takes too long to open, the insect may be out of the field of view or out of focus by the time the picture is taken. The exposure time must be minimized to reduce motion blur under natural lighting conditions as much as possible—in other words, we want as close to 100% of the exposure lighting to come from the flash.

The size of the shutter is the limiting factor in its speed. Since according to the lens tests the aperture will be f/11 or f/16 (if depth of field doesn’t require an even smaller aperture), the size will not be too large. I wouldn’t mind having a versatile shutter rather than limiting myself to a small aperture size—who knows if I may not want to have some shots with extremely short depth of field.

Traditional iris shutters (like the Ilex Oscillo B I tested before, except with built-in timing mechanisms) have a maximum diameter covered by the shutter leaves regardless of what the aperture diaphragm is set to. In other words, time is wasted with the aperture leaves opening into the area which is covered by the aperture diaphragm. This is not an issue with traditional shutters since under natural lighting the exposure time will increase as the aperture decreases, but since we want it to go as fast as possible regardless of lighting, this timing mechanism is not optimal. For large apertures, this is especially unacceptable—the fastest size 5 shutters have a 1/50 second exposure at the fastest settings.

Although I still plan on repeating the tests with the size 3 Oscillo B, but with higher voltages, I did some calculations regarding various shutter concepts to get an idea of what is possible.

The Swinging Wedge Concept

To minimize lag and exposure, the accelerations must be as high as possible. An iris shutter requires an acceleration to start opening, then (assuming it just “slams” open) an acceleration to start closing. Conceivably then the exposure time could be cut in half by eliminating the second acceleration.

Such a shutter could be made by having a plate shaped like a slice of pizza with a hole the size of the aperture in the center at the edge, with just enough space on either side to cover the lens aperture. In the diagram below, the wedge-shaped plate, which is made to be thin, is shown with an optional wedge-shaped cut (the dotted line filled in white) to save weight. The aperture hole is shown in white, and the dotted circles show how on either side there is enough plate to cover the aperture. In the shutter “closed” position, the lens aperture would sit entirely within one of the dotted circles (A or C). The shutter is opened by pivoting it about the dark grey circle on top, and as the circular hole (B) in the wedge passes through the lens shutter, the exposure is made (the flash would fire exactly when the hole coincides with the aperture—that is, full open position). The wedge then continues to rotate until the lens aperture is completely covered by the second dotted circle region.

If we leave the radius of the wedge as a variable, then the minimum angle it must subtend can be estimated by setting the length of the dotted arc to be three times the diameter of the necessary aperture. The length of a circular arc is simply the angle it subtends times its radius (which is why the circumference of a circle is ):

Note that this equation is an approximation, because we are setting an arc length equal to the length of three straight lines, when in fact we should calculate the angle between the three straight segments and make the wedge fit that dimension.

To maximize acceleration, the inertia of the shutter wedge about its pivot point should be as small as possible. The inertia we are looking for is called the “polar moment of inertia” about the pivot point (in this case, the center of the circle the wedge came from), and is available here (note that that expression is in terms of the half angle). The expression using the full angle of the wedge is

The multiplication by is to account for the real shutter wedge having thickness and being made of a material with density.

To account for holes, we simply have to subtract the inertia of the corresponding shape. Ignoring the weight-saving wedge-shaped cut out, we only need to consider the aperture hole. The moment of inertia of a circle about its center is (omitting the mass and thickness parameters)

(Note that like area and circumference, it is related to that of the wedge by the subtended angle.)

But in our case, this hole is pivoting about a point which is not the one for which the above moment of inertia is calculated. We can use the parallel axis theorem to find the new moment of inertia:

and thus the inertia of the shutter is

If we wanted to include the weight-saving wedge cut, and assume its tip is an offset from the pivot point and its circumference is offset the same amount from the aperture hole, then its inertia contribution would be

Ignoring this for now, we write the expression for the inertia of the shutter by replacing with its relationship to [text]r[/tex] and :

In our case, is dictated by the f-number of the lens, and thus is our only variable. Since the only positive term has an in it, it would seem the best way to minimize inertia is to minimize the radius of the wedge. To minimize the radius, we have to find the shape that most tightly packs the three circles, which should be an equilateral triangle. The “wedge” which then circumscribes this arrangement is the minimum inertia version of this shutter. The shape is shown in the diagram below, which we use to calculate the radius of the plate (we show again an optional weight-saving cutout at the bottom).

The equilateral triangle that links the circles’ centers has a side length and a half-angle of 30 degrees. Thus the distance from any corner to the center of the equilateral triangle is the hypotenuse of a right triangle with angle 30 degrees and “adjacent” side equal to :

So the radius of the plate is

The inertia of the plate itself is the same as the wedge with . The inertia of the aperture hole is the same, but its distance from the pivot changes to :

The density and thickness were omitted for simplicity.

Even if the “wedge” version has its weight-saving cutout, its inertia will still be higher than that of this minimum inertia arrangement. Note, too, that the inertia in all these could be further reduced by thinning out the dotted circles—that is, they only need to be optically opaque, not structural. Also (obviously) in both cases the radius of the plate would have to be slightly larger so that the shutter opening still has some material around its entire periphery.

Neglecting friction (and the inertia of the actuating mechanism), and assuming constant torque (not likely), the acceleration of the shutter plate can be written as

with as the torque, as the angular position, and as the time. With 0 for the initial velocity, integrating twice yields

Regardless of the angle of the “wedge” plate, the angle that must be traversed for the shutter to be fully open is

and we denote it as because the time it takes to traverse this angle from 0 is the shutter lag time. The shutter will be completely closed after twice this angle is traversed, thus for that entire range the shutter is at least partially open. This is not a good measure of the exposure time because at all times except the instant that it is fully open, the aperture of the lens is partially exposed. Still, with this one-swing arrangement, anything that minimizes shutter lag will minimize exposure time, so there’s no point in trying to calculate the exposure time.

With our minimum-inertia geometry (using the numerical approximation to the inertia), the shutter lag time can be written as

This means that doubling the aperture diameter (increasing it by two stops) quadruples the lag time, quadrupling the thickness or density doubles the lag time, and quadrupling the torque cuts it in half. In reality, the plate should be thicker at the hub so it can withstand the sudden jolts of torque, and as mentioned before, the inertia could be reduced further by not making the aperture covers structural—in fact, one could think of it as a plate with three aperture holes with two adjacent ones covered with a very thin opaque material such as shim stock (one or two thousands of an inch thick). With three aperture holes, the inertia of the plate is 40% lower, and the lag time drops 23%.

To get some ideas of the forces we are talking about, lets assume that the plate is 0.5 mm thick and made of steel (8,000 kg per cubic meter). For a 300mm lens at f/11, the entrance pupil diameter is 27.3 mm, so we can use that for . The required torque to result in a lag time of 1 millisecond is 16.3 Nm (12.0 foot-pounds). If we use the inertia of our three-hole plate (ignoring the thin material to obscure two of the holes) the torque requirement drops to 9.6 Nm (7.1 foot-pounds). If we want to cut the lag time in half, we need four times as much torque. The Kodak 14″ Commercial Ektar has an Ilex #5 shutter on it, whose maximum opening is 64.1 mm. At this size, the torque required for the single-hole disk for 1 millisecond lag is 496 Nm (366 foot-pounds). A Porsche 911 Carrera puts out 370 Nm (273 foot-pounds) of torque at 4250 RPM. For this last case, at the instant the shutter is fully open, the wedge is moving at around 4200 radians per second (40,000 RPM). Note that even for a more civilized 4 millisecond lag, the disk will be spinning at 10,000 RPM when the shutter is fully open—is it even close to possible to achieve this?

More soon….

The principal advantage of the Goerz lens is that it is already f/9, and as we’ve seen in the other tests, none of the lenses so far have performed well at the larger apertures. Presumably if the lens was designed to have f/9 as the largest aperture, the design also has less compromises, since designs for larger apertures should perform at least tolerably well at the large apertures. This gives it a size and weight advantage over the two enormous enlarger lenses. The other advantage is that it is a camera lens, and thus is meant to be assembled around a shutter. Having the shutter at the aperture plane is optimal since as the shutter is opening, it is behaving like a smaller aperture—as opposed to the other extreme of a focal plane shutter which casts a shadow on the film (which is why they are curtains and not diaphragms). This may not be an issue if I get the shutter to fire quickly enough, but it is certainly “nice”. In the case of the enlarger lenses, I at first considered taking them apart and putting a shutter in there, but after seeing them, I think this is a bad idea—the Rodenstock has something over 30 blades in the diaphragm—and I know what a pain it is to put together one with just 5 blades. Thus for those lenses the shutter will have to be outside the lens somewhere. Note that by having it close to the lens it still won’t cast a sharp shadow on the film (so the effect of the images being brighter in the center from the exposure being longer there is not a huge concern) but it has to be slightly larger than it would have to be at the aperture plane. At the aperture, all rays from all object points pass through the same location, so the size can be minimized, whereas anywhere else, the shutter has to increase in size according to the “image cone” of the desired film size. Again, there is not a huge difference, because the lens to film distance is so great, but the difference becomes larger and larger as the aperture decreases in diameter (although this is counteracted by the fact that by shrinking the aperture the shutter has to be smaller anyway). If it turns out the performance of the Goerz is close to that of either of these enlarger lenses (assuming they are superior to the Kodak, of course), then that will be the obvious choice for several reasons.

Essentially the lens decision comes down to this, because this focal length range gives a good working distance at 1:1.

As in the last test, the magnifications are approximate, based on the conjugate distances used. The target was positioned at about 45 degrees to the lens so that one image can also show the depth of field. On to the images, then….

Goerz Red-Dot Artar 305mm f/9, M=1 (working distance ~22 inches)

Images for the lens forward: f/9, f/11, f/16, f/22, f/32, f/45, f/64
Images for the lens backward: f/9, f/11, f/16, f/22, f/32, f/45, f/64

The Goerz at 1:1 has pretty impressive performance, especially considering the lens I have was made in 1955. I read that the uncoated versions were actually process lenses (designed for 1:1) but after people started using it at other magnifications, Goerz released factory-built lenses with shutters that were optimized for 1:10 (which is what I suspect what my lens is). Evidently 1:1 did not suffer much. (The source of that tid bit is here.) Apparently there are 14″ red dots out there….

But back to the images—note how the focused part of the image remains equally sharp from f/9 to f/22 (there’s slight diffraction blur at f/22), and there’s only a bit of longitudinal chromatic in the out-of-focus regions at f/9, something which disappears by f/22 (almost completely by f/16). (We must remember that the sensor of the D200 I’m using for these tests is much smaller than the intended coverage—6×6 or even 4×5; up to 4×10 or so if I try to do stereo.) In a color photograph, chromatic aberration in the defocused regions may be a problem, but in black and white, it will just add to the blur. Also remember that by the very nature of refraction’s dependence on wavelength, it is impossible to make a lens which does not have chromatic aberration in the defocused regions—different colors of light are taking different paths, but a well-corrected lens will simply have most of the colors meet in a relatively small area on the focal plane for a given magnification.

M=2 (working distance ~15 inches)

Images for the lens forward: f/9, f/11, f/16, f/22, f/32, f/45, f/64
Images for the lens backward: f/9, f/11, f/16, f/22, f/32, f/45, f/64

At this magnification, in the forward orientation, f/9 through f/22 still work well, but f/11 seems to be the best in the focal region. With the lens backward, the performance seems to improve a bit, back to the condition at M=1 where f/9 through f/16 are nearly indistinguishable, with f/22 adding a bit of diffraction blur.

Rodenstock Apo-Rodagon 300mm f/5.6, M=1 (working distance ~22 inches)

Images for the lens forward: f/5.6, f/8, f/11, f/16, f/22, f/32, f/45, f/64, f/90, f/128
Images for the lens backward: f/5.6, f/8, f/11, f/16, f/22, f/32, f/45, f/64, f/90, f/128

This Rodenstock can be considered to be a newer version of the Goerz, in that they are both apochromats, about 300mm focal length, but the Rodenstock is about 20 years newer. In the forward orientation, there is a bit of chromatic aberration in the focal region even at f/8, but it is gone by f/11, though the sharpest image is at f/16. Diffraction blur proceeds from then on. With the lens backward the performance is very similar (the focal plane evidently was not exactly in the center of the target, which makes it look much worse; it is slightly to the left of the central part).

Head-to-head against the Goerz, the Goerz wins at f/9 (versus the Rodenstock at f/8), performance is almost equal at f/11 (with the Goerz possibly having a slight edge), until at f/16, where the Rodenstock comes out ahead. Since I don’t have a good grasp of the depth of field I’ll need right now, it seems the Goerz is the winner here, on behalf that at f/11 it is not much worse than the Rodenstock at f/16 (in terms of sharpness at least) and the fact that it is smaller and is designed to have a shutter at the aperture plane.

M=2 (working distance ~16 inches)

Images for the lens forward: f/5.6, f/8, f/11, f/16, f/22, f/32, f/45, f/64, f/90, f/128
Images for the lens backward: f/5.6, f/8, f/11, f/16, f/22, f/32, f/45, f/64, f/90, f/128

At this magnification, with the lens forward, there is chromatic aberration near the focal plane even at f/11, but at f/16 and f/22 the image is very nice (with very slight blur at f/22). Perhaps what is surprising is that with the lens backward, the performance is slightly worse. I can’t help but believe it may have to do with my setup, since I’ve had problems before with the camera not being mounted rigidly enough (since it is held by the lens flange on a bellows extension).

But again, the main thing here is, can it beat the Goerz? At f/9, the Goerz has an advantage in the chromatic aberration versus the Rodenstock’s f/8. At f/11, the Goerz seems to have a slight edge in both sharpness and chromatic aberration in the defocused region, and at f/16, the Rodenstock finally seems to overtake its older adversary.

Considering that the Rodenstock performed worse in the backward orientation, while the Goerz performed better, the real comparison should be between the Rodenstock forward and the Goerz backward. As expected, the Goerz is much better at f/9 than the Rodenstock at f/8; Slightly better at f/11, and finally giving way (very slightly) at f/16. Could the Goerz be it?

Schneider Componon-S 360mm f/6.8, M=1 (working distance ~25 inches)

Images for the lens forward: f/6.8, f/8, f/11, f/16, f/22, f/32, f/45
Images for the lens backward: f/6.8, f/8, f/11, f/16, f/22, f/32, f/45

Now for the Componon. Because of its longer focal length, it has a slight working distance advantage over both the Goerz and Rodenstock, which also means my setup gets a little less rigid, and I think in this case there is a bit of motion blur in some of the images.

At this magnification, it seems to work slightly better in the forward orientation. Even at full open, it has impressive sharpness in the focal region (although the exposure is not great in that image). At f/8, the sharpness seems to be comparable to the Goerz’s f/11. At f/16, the two are almost equal, and the Rodenstock wins again.

M=2 (working distance ~18 inches)

Images for the lens forward: f/6.8, f/8, f/11, f/16, f/22, f/32, f/45
Images for the lens backward: f/6.8, f/8, f/11, f/16, f/22, f/32, f/45
(I made a mistake with the files, so there is no image for f/8.)

The Componon seems to perform better backwards here. Unfortunately, those images seem to have a lot of motion blur, and apparently I mixed something up so that f/8 is missing. The critical image at f/16, where the Rodenstock edges out the Goerz, is too blurry to make any sort of judgment. Perhaps it will be worth to repeat the test, if f/16 becomes necessary to get depth of field (remember we still have barely mentioned the needs for a super-fast flash, so the aperture needs to be as big as we can tolerate it).

Conclusion

I think it’s time to do some exposure and depth of field measurements, along with making the choice between 1:1 or 2:1 (though according to these tests it does not dictate the lens choice—which is a good thing: perhaps a convertible camera is possible). If the depth of field is enough at f/11, then it looks as though the Goerz is the way to go. Before diffraction blur starts to affect the image we only have the choice of f/16, and if we have to go there, then evidently the Rodenstock is the winner (though it may be worthwhile to re-test the Schneider).

1. The image quality must be absolutely as good as possible.
2. The working distance must be as long as tolerable.
3. The magnification must be 1 or greater.

The Kodak 14″ Commercial Ektar clearly outranks them all as far as item 2 is concerned. To get 1:1 magnification, the working distance is approximately twice the focal length (this is true for all “simple” lenses, that is, those that are not designed to have a camera-specific back-focus distance), or 28 inches (711 mm), which is confirmed by the previous round of lens tests. This means that the distance between the lens and film is also 711 mm, which would make for a terribly unwieldy camera unless fold mirrors were used. The image quality requirement just kills this lens; there are simply too many chromatic problems at all apertures. The only way to get rid of these problems is to filter the color of light allowed through the lens to a relatively narrow bandwidth, and I’m not ready to do that considering that flash illumination with extremely short pulses will be challenging enough as it is.

Items 2 and 3 together essentially eliminate the 105mm Nikon macro lens as an option. At the higher magnifications the working distance is too short—no sense in making a portable system if the insects will all be scared of it. The image quality is better than for the Kodak, but it is not stunning.

The Hasselblad 120mm f/4 macro lens is a good option; although the working distances are on the short side, the image quality is excellent.

For this round of lens tests, I added a diffuser to the resolution target, so I wouldn’t get any weird diffraction effects and so that I could get much more even illumination. The diffuser is simply a small piece of white plastic that is smooth on one side and rough on the other (it is meant to diffuse fluorescent bulbs in light fixtures). I put the smooth side to the target, and, to my surprise, I could see the roughness of the backside in the images where the depth of field was long enough. I really should stop being lazy and simply make a photographic contact print of the target and mount it to something rigid.

This time I also wanted to get an idea for the depth of field at different apertures, so I imaged the target tilted at about 45 degrees relative to the focal plane. The results for the Kodak lens I won’t even show—they are very disappointing. I tested also the Hasselblad macro, and then I figured I may as well test my 135mm enlarging lens—a Schneider Componon-S 135mm f/5.6. Only after the test did I realize what spectacular lenses these are—and although the newer ones are not dirt cheap, they are not terribly expensive in the used market, especially with so many photo labs going out of business or going all digital. The results with this lens are absolutely stunning. Interestingly, it seems to work fantastically at a 1:1 ratio, even though they are claimed to be made for ratios ranging from 2:1 to 20:1.

Two notes before I discuss the results:

First, if you’re wondering why the numbers are always backwards on the target, it’s because the chrome layer from which the actual target bars are cut I always place facing the lens, so that the glass substrate does not interfere. The target is printed such that it should be read from the glass side, so it appears backwards when viewed from the chrome side. I believe this is done so that contact prints of the target (which should be done with the chrome layer directly on the emulsion) will appear in the correct handedness.

Secondly, for this round of tests I used Nikon’s Camera Control Pro software. At first I was confused about it coming with a special cable, since it only seemed to be sold in boxed editions. Then I realized there is a free trial, and indeed, the only requirement is the USB cable that comes with the camera and to set the camera’s USB interface to “PTP” (point-to-point, I suspect) mode rather than mass-storage device mode. The software is fantastic for this kind of thing, especially in changing camera settings (like non-CPU lens data) since it’s all done from the computer rather than navigating the maze of camera menus. With the ease of use I set the lens data and then used the camera in aperture-priority exposure, which worked well except for the fact that it underexposed in the smaller apertures. I adjusted the images so that they look similar, though towards the end of the test I used the exposure compensation to account for that. There are two downsides to the software: one is that it’s really easy to change the settings—which means one should remember to change everything back before disconnecting or the camera (in my case) will be left in a horrific state for a quick snapshot. Secondly, when the option to automatically display the last image is selected, the software remembers the zoom level but does not remember the relative image position, so if one is zoomed in close to examine the center of the image, the next image loaded resets the window to the top left. This is contrary to one of the most surprising feature of the D200 camera—zoom level and position in play mode is maintained as one browses through the images in the CF card. Still, the software is fantastic; the free trial is here and it also contains the link to buy the registration key directly from Nikon. Note that the trial is version 1.0, but the current version (for Windows anyway) is 1.3, and according to Nikon you have to have 1.0 installed to install 1.3 (so basically the download size is doubled). When I installed 1.0 it immediately asked if I wanted to update to 1.3.

So on to the results!

The Schneider lens was tested at nominal magnifications of 1, 2, and 3, and images were taken at each standard f-stop from f/5.6 to f/45. Nominal means I didn’t measure it, but I tried to put the lens at the right distance from the camera to obtain those magnifications. (In the M=3 case, there is a big difference in magnification between the two lens orientations.) For each setup, the lens was mounted “forwards” (meaning the side that normally faces the photo paper is facing the object) and backwards. The images are linked in a way that each setup will open a new window. So if you click on all apertures for a particular case, then you can use your browser’s forward and back buttons to navigate through the sequence.

Schneider 135mm f/5.6, M=1

Images for the lens forward: f/5.6, f/8, f/11, f/16, f/22, f/32, f/45
Images for the lens backward: f/5.6, f/8, f/11, f/16, f/22, f/32, f/45

The first thing to notice is that in all cases, sharpness at f/32 and f/45 suffers considerably due to diffraction blur. Moreover, these are typically the worst images because the longer exposures make them more prone to light leaks in my bellows-less setup.

The next thing to notice is the axial chromatic aberration at the larger apertures. The f/5.6 image is of course the one that makes this most obvious. You should notice that the blurred target portions on the left of the image (closer to the lens) have a red glow, whereas the right side has a blueish or greenish glow. At f/16, it is still there, but minutely so. At f/22 it is indiscernible, at least in the presence of JPEG compression artifacts. The magnitude of this aberration seems to be very close in both orientations of the lens, suggesting that the design is nearly symmetrical across the aperture plane. A quick look at the specifications confirms this. (Note that in the extensive Schneider specifications, the magnification is referred to as the Greek character beta.) Don’t be alarmed by the absurd list price; the lenses are much cheaper on the used market (~$150).

M=2

Images for the lens forward: f/5.6, f/8, f/11, f/16, f/22, f/32, f/45
Images for the lens backward: f/5.6, f/8, f/11, f/16, f/22, f/32, f/45

In this case the chromatic performance at the larger apertures is similar to the Hasselblad macro lens’s. However it seems to disappear by f/16 rather than f/22 as in the M=1 case. Interestingly, the lens seems to perform better facing the object than the other way around. This is especially true at the full-open aperture.

Note that magnification as I write about it here is optical magnification, that is, the size of the image divided by the size of the object. In the brochure for the enlarger lenses, Schneider says they are designed to be used at 2x to 20x magnification, which implies M=0.5 to M=0.05 according to my notation (replace M with for their notation). This confused me at first; in fact, I had to go back and edit the post after I realized it.

So, not to confuse anyone, but as far as the lens is concerned, when the it is facing the object the magnification is 2 whereas when it’s backwards it’s 0.5—in other words, the lens backward at an optical magnification of 2 should perform better than the lens forward. The opposite seems true at the larger apertures, and at the smaller ones they are indistinguishable.

M=3

Images for the lens forward: f/5.6, f/8, f/11, f/16, f/22, f/32, f/45
Images for the lens backward: f/5.6, f/8, f/11, f/16, f/22, f/32, f/45

This case (that of the lens backwards) has published MTF curves by Schneider. With our M=3 and the lens facing backwards we have the equivalent situation to Schneider’s (I’m not sure what the minus sign implies). I believe that in the plots, the x-axis is the relative (radial) height of the ray on the image. Note, for example, that decreasing the aperture from f/8 to f/11 does not increase the resolution at the center but it does make a difference at the edges.

In my tests there is a clear difference in the two lens orientations, with the backwards lens performing better. The higher the optical magnification with the lens facing forward, the farther we are from its design space. The behavior seems to be similar to the previous case; that is, by f/16 any chromatic problems are gone. These are the worst images by far simply because I do not have bellows long enough (and not enough cardboard boxes lying around) to block out stray light. I chose not to pursue better images because I was tired and because this particular lens still has a shorter working distance than I want.

Hasselblad Makro-Planar CF 120mm f/4, M<1

The magnification in this case is less than 1; in fact, it is less than in the previous test where the lens was tested at M~0.8. Since the target is tilted, I won’t try to guess what it is, but it can be compared directly to the images above in terms of scale.

f/4, f/5.6, f/8, f/11, f16, f/22, f/32

You should notice immediately that even at f/5.6 the chromatic aberration is much worse than in the case for the Schneider lens at M=1. This is precisely why the performance of the Schneider surprised me so much, since this Zeiss lens is no joke. The chromatic aberration is apparent all the way up to (but excluding) f/22, where diffraction blur is already affecting the image. It is impossible to compare sharpness between cases at different magnifications, but I think it’s safe to say both these lenses are very sharp.

Chromatic aberrations are obviously a problem in color images, even if the aberration is only present in the out-of-focus regions. In black-and-white pictures, all it really does is increase blur, so if it is not present at the focal plane, then there’s no reason to discard a lens completely because of it. Although I can certainly see myself doing plenty of black and white with this camera (seeing as how I can print nice prints in my garage/darkroom), I’m not going to allow a loss in performance and justify it with that. Thus so far the first choice lens has become the Schneider 135mm enlarger lens, both because of it’s superb performance and its reasonable working distance at M=1. Still, I believe the camera could benefit from a longer focal length (thus giving a longer working distance). The surprising performance of this lens has also enticed me to test an old Goerz 12-inch f/9 apochromatic “red dot” lens, since an apochromatic lens is better corrected for chromatic aberrations than simple achromats (which is everything that doesn’t say “apo” in front of it).

But this has also taken me back to eBay in search of longer focal length Schneiders and any long-focal length apochromatic enlarger lenses I can find.

As a side note, I guess the performance of an enlarger lens being so great should not have surprised me. Schneider’s philosophy, at least, is that the enlarger lens should out-perform the image-recording lens.

In summary, I will test my 12 inch Goerz next, since that will have a nice working distance of around 20 inches at M=1 (and it is a camera lens and thus meant to have a shutter mounted between the front and back cells). Based on that, I may have to spend some time on eBay before settling on a lens and start working on the other parts.