Sean McVeigh - Small Pony Pictures - December 2008
This page loosely details my first attempt at building a 400W HMI fixture using off-the-shelf parts. Once again, I have taken photos after the fact, so naturally, I don't have any in-the-making images for you, and in fact if I did, they would be quite boring anyways. Please accept this overly verbose document instead!
Why build your own HMI light? Well, it probably makes sense to start by identifying what exactly an HMI light is.
An HMI is:
First off, HMI is simply the brand name for a line of lamps produced by Osram. It stands for Hydrargium Medium arc-length Iodide. The common name for such lamps is a Metal Halide (MH) High-Intensity Discharge (HID) lamp. Many companies manufacture lines of MH HID bulbs, but for simplicity's sake, I will refer to the entire family of such bulbs as HMIs.
Incandescent tungsten lamps have a colour temperature of around 3200 Kelvin. This is a fairly warm light source, and is quite close to that of a household 100 watt bulb (actually, these are slightly more orange at around 2900K). For indoor filming, this is typically sufficient. Tungsten lamps are inexpensive, flexible, units under 2000 watts in power are plug-and-play, and they are quite easy to come by for the budget-sensitive film maker. If, however, you have exterior shots to light, or have exterior windows in your interior shot, tungsten lamps are not appropriate unless light-robbing blue (CTB) filters are placed in front of the fixtures. A 2000W tungsten lamp with a CTB gel installed only produces around as much light as a 500W unfiltered bulb -- 2 stops of light are lost, and you've already hit the current limit for a common household electrical circuit. The solution is to move to a light source that is the same colour temperature as that of daylight -- between about 5000 and 7500 kelvin, depending on the time of day. HMI lights are typically advertised as running 5500K in temperature, although if you read the technical literature, in reality the bulbs are usually rated around 6000K. Even bulbs upwards of 6500 and 7000K may be useable in exterior shots, but may require slight orange (CTO) filtration to bring them in line. At least CTO filters eat up far less light than their CTB brethren.
Another alternative for a daylight-balanced lighting instrument is the fluorescent bank. Flourescent lights have grown in popularity recently as special phosphors have been developed which nicely approximate the daylight spectrum, flicker-free electronic ballasts have alleviated shutter-speed restrictions and weight issues. However, fluorescent fixtures are more suited to soft, "wrap-around" light, as their physical size is quite large -- a 4-foot long high-output fluorescent tube is only 55 watts -- a 1000 watt fluorescent head is around 20 square feet in size. Conversely, a 1000 watt HMI globe is about 8 inches long. HMI lamps can be found in sizes up to 18,000 watts!
Wikipedia defines an incandescent light bulb as follows:
"The incandescent light bulb, incandescent lamp or incandescent light globe is a source of electric light that works by incandescence, (a general term for heat-driven light emissions which includes the simple case of black body radiation). An electric current passes through a thin filament, heating it until it produces light."
Tungsten lights give off most of their energy in the form of infrared radiation, which is why they are often termed "hot lights". Only a fraction of their rated power is released as visible light. For this reason, tungsten lights are only about a fifth to a quarter as efficient as HMI lamps. HMI lights are arc lamps. They do not generate their light by heating up a filament -- they produce light by igniting an electric plasma arc between two electrodes. A 400W HMI arc lamp produces roughly the same amount of light as a 2000W tungsten bulb. I am not entirely sure how much heat is produced by a 400W HMI bulb, but it certainly can't be more than 400 watts of heat, whereas a 2000W tungsten bulb definitely puts out far more than that (probably the equivalent of running a 1500W heater or a hair dryer). Anyone that has ever stood in front of a large "hot light" can attest to this fact.
HMI lamps do have a downside in the radiation department however, and that is that electric arcs generate a high amount of ultraviolet (UV) light. This can pose a serious health hazard, as UV radiation is the reason we don't stand out in the sun without sunscreen for any length of time. I was working in front of one of my 400W prototypes for around 15 minutes (in 5-minute spurts) and I suffered a fairly painful sunburn down one side of my face, and a case of Photokeratitis, or what welders refer to as "Arc Eye". (and I can unfortunately attest to the fact that it does in fact feel like "sand being poured into the eye"). Needless to say, any HMI fixture must incorporate a UV filter of some sort, and in fact, some larger bulbs meant for use in industrial settings and open-faced fixtures incorporate their own UV filters.
Since an incandescent tungsten filament nicely approximates a black-body radiator, it has a nice continuous, uniform spectrum. That is to say that colours are more or less equally represented in it's spectrum (balancing for the 3200K peak, of course). The problem with discharge lamps is that they are not black-body radiators, but instead rely on a carefully formulated mix of ionized gases (or phosphors in a fluorescent tube) to approximate a continuous spectrum. For a really good example of this, go and stand under a sodium vapour lamp (very narrow reddish-yellow spectrum) and try to distinguish between a green shirt and a blue shirt. The trick, then is to introduce more elements into the atmosphere inside the bulb which can fill in the gaps in the spectrum when ionized in the plasma arc. The Colour Rendering Index (CRI) of an HMI (or fluorescent) bulb quantifies how continuous its spectrum is, and in turn how well colours are reproduced when illuminated with it. Generally, cinema-grade bulbs are in the 90+ CRI range, although anything above 80 or 85 is considered good. If the CRI of a bulb is below 80, I would not recommend using it for cinema. Fluorescent lamps have traditionally been shunned due to the relatively large green spike in their spectrum (mainly due to the mercury content) and accordingly low CRI. New phosphor blends have brought high-CRI (95+) tubes into mainstream use.
One of the most readily evident downsides of running an HMI arc lamp (and any type of discharge lamp, including fluorescents) is that they are not self-ballasting like tungsten lamps are. An external ballast is required to limit and regulate the electrical current through the arc. An arc lamp ballast traditionally has consisted of a large heavy transformer, requiring an alternating (AC) current supply. In recent years, electronic ballasts, which are just switching-mode current-limiting power supplies, have displaced their bulkier magnetic counterparts as the ballasts of choice. The second electrical component required to run an arc lamp is an ignitor. The ignitor produces very high voltage pulses to ionize the atmosphere between the electrodes in the bulb and permit the low-impedance path necessary to strike the arc. Ignition voltage depends on the electrode gap -- the larger the gap, the higher the voltage required to strike the arc. Smaller HMI globes in the 150-1000 watt range typically require only a couple kilovolts to ignite. For this reason, it is important to treat the electrical components in an HMI system with respect, as they can and likely will kill you if mishandled.
WARNING: Math Content
As mentioned before, incandescent lamps operate by heating a (typically) tungsten filament to produce light. Since there is negligible cooling of the filament between peaks on the mains AC power line, the lamp continues radiating light even at the zero-crossings of the power waveform. Unless you are filming in the thousands of frames per second, this cooling (and thus flicker) is not visible. For this reason, tungsten lamps are considered continuous light sources.
Since magnetic ballasts (transformers) were the primary power supply for HMI lamps, and the most readily abundant source of AC power is the mains power grid at 60Hz (or 50Hz in Europe), the arc is not a continuous light source. The electrical current crosses the zero-mark in a 60Hz AC circuit 120 times per second, leading to an arc frequency of 120Hz. While not visible to the unaided eye, an intermittent camera movement will experience a noticeable flicker at a beat rate equal to the absolute difference between the shutter frequency and that of the arc lamp (|Fshutter-Farc|). If the arc frequency is an integer multiple of the shutter speed, then no flicker will be seen. For this reason, 24fps is considered an "HMI-Safe" shutter speed (24 x 5 = 120). This should intuitively make sense, because every time the shutter is open (in a 180-degree shutter system) 2.5 pulses of the arc occur, and every time the shutter is closed, 2.5 pulses occur. If the camera were operating at 25fps in a 60Hz system, this would not be the case, and a flicker rate of 5Hz (25x5 - 120) would be observed. You have probably experienced this if you shoot video under magnetically-ballasted fluorescent lights, since the NTSC video framerate is 29.97fps and 120 - 29.97x4 = 0.12Hz. This results in a slow (once every 8.3 seconds) pulsating brightness that I personally find irritating in most amateur home videos, but that for some reason most people never seem to notice. You will experience the same limitations using an HMI lamp with a magnetic ballast. For a list of HMI-safe shutter speeds, see these handy links from Panavision: 60Hz HMI-safe speeds and 50Hz HMI-safe speeds. The use of an electronic ballast eliminates these problems and allows you to film at any framerate, typically up in to the thousands of frames per second. Electronic ballasts are usually found in 2 varieties -- the low-frequency square wave type, and the high-frequency type.
Low-frequency square wave ballasts operate, as the name implies, at a frequency of somewhere around 50Hz up to a couple hundred Hz. Depending on the complexity of the circuit, they may produce square waves at the mains frequency (50Hz or 60Hz). The reason these waveforms produce no visible flicker is that the edges of the waveform are square -- that is to say that they transition from their positive peak to their negative peak very rapidly (on the micro-seconds scale) ie. the time that the arc remains extinguished is minimal. Another way to think of it is as follows: If you take the absolute value of an AC sine-waveform (ie. the current flowing through the bulb), you end up with a series of humps which very noticeably increase in magnitude, and then decrease to zero before bouncing back again -- this is the source of flicker visible in magnetic ballasted discharge lamps. If you do the same to a square wave of the same frequency, you will find that you have more or less a straight line at the peak power level with extremely narrow dips to zero and back. It is because these "ticks" in the waveform are so short (micro-seconds) that the flicker is not noticeable unless you are operating at a framerate where the exposure period starts to approach the duration of these ticks -- ie. one frame out of every hundred may appear noticeaby darker than the others because it coincides with a tick in the current waveform. The downside of using square-wave ballasts is that square waves can be expressed as a harmonic series of frequencies. These harmonic frequencies caused by the hard edges of the square wave can lead to resonance inside the HMI globe itself, causing it to audibly "sing". For this reason, many square wave ballasts have a switch to select between the pure square wave mode of operation, and a mode whereby the edges of the square wave are rounded off to minimize the harmonics. Now before you shout out "but now the lamp will start to flicker at off-speeds because it's getting closer to a sine wave again", let me remind you that we only care about sound on-set when we are shooting at sync speed -- ie. 24fps in North America. Since we are shooting at sync speed, and the HMI is now running at a fixed 60Hz frequency, we will not experience any flicker.
High-frequency ballasts do not necessarily operate with a square wave output, but that is unimportant. The way flicker is eliminated in a high-freqency ballast is simply by bumping the current waveform up into the thousands of hertz range. 30KHz is not uncommon for such ballasts. Now, obviously not all framerates divide evenly into 60,000 (multiply by 2 for the frequency of peaks), so instead we ignore the beat frequency, and focus on the difference in the number of pulses seen per frame. Think of it as follows: if you are running at 29.97fps, and 29.97 divides into 60000 2002.002 times, this means that there are 1001.001 pulses of light per frame (assuming a 180-degree shutter) For a different shutter angle (A), multiply your result by A/180. So out of every 999 frames, 998 have 1001 pulses of light, and one has 1002 pulses (998x1001 + 1002 is divisible into 30000). The upper bound on the difference in brightness between 1001 and 1002 pulses per frame is therefore a factor 1002/1001 =~ 1.001. The base-2 logarithm of 1.001 is around 0.00144 or approximately 1/700th of an F-stop, which is effectively zero for our purposes. A simple algorithm for computing this magnitude accurately is beyond the scope of this document, and depends largely on the shape of the waveform, as integration is required. Suffice it to say that I believe the number of F-stops of flicker should be bounded by the following equation for all periodic waveforms:
It is a fact that not all HMI lamps can be switched off and then back on again without sufficient time for the bulbs to cool down (typically 5-10 minutes for bulbs under 1000W). The hot electrodes will not strike an arc without greatly increasing the ignition voltage (10x or more). For this reason, special hot-restrikeable (HR) bulbs and ignitors must be selected if the ability to re-power lamps is a requirement. 30,000 volts is not uncommon for a typical 575W HR bulb. Due to the very high voltages that can be involved, larger insulated bulb sockets may be required for hot-restrikeable HMI bulbs to minimize arcing between the pins.
While this may sound like a plus, it can be very frustrating for the designer on a budget to decide which particular socket style and bulb size to go with. Start by selecting a desired output power. Then determine which bulbs are available in that range. Determine which socket styles these bulbs require (many of these sockets can be hard to come by). Determine if you need hot-restrike ability, then check if the bulbs you settled on come in an HR flavour. They will likely require you to re-select your socket. And now maybe it turns out that you need to re-size your ignitor. And so on, and so forth. Running out and buying a ballast because it seemed like a good deal may leave you in a situation where only one brand of bulb is compatible with it, and only in a socket size that is available from a wholesaler in Korea. So please, make sure you've located a source for all of your components before beginning a DIY HMI project. I'll provide some links later on.
While there are clearly many advantages to using HMI lamps for cinema lighting, the complexities involved should now make it apparent why it is that they command a higher price than a simple tungsten filament bulb that just needs to be connected with 2 wires directly to the mains AC. If you want to inexpensively shoot with daylight-balanced fixtures, I can heartily recommend assembling a fluorescent bank which takes perhaps 4 x 4-foot tubes and has an electronic ballast. Many off-the-shelf shop-lights sport these ballasts, but rest assured that if yours doesnt, you can easily change out a magnetic one for a purchased electronic one. The bonus with such a fixture is that you can simply swap tubes to change from tungsten balanced lighting to daylight balanced.
If, however, you need a hard light that is daylight balanced and are less than enthusiastic about the 2 stops of light loss incurred by filtering a tungsten unit, then read on.
My first order of business was deciding what power range I wanted. Typical small HMI units fall into the following 3 ratings: 150W, 575W, and 1200W. Actually, 150W may be more like 200W, but that is unimportant, as I decided it probably wasn't worthwhile constructing such a small unit. Don't be fooled by the wattages though, as if you recall, HMI lamps put out something like 4-5x the amount of light of a comparably rated tungsten unit, so a 150W HMI is more akin to a 650W tungsten unit. The 1200W size (5000W tungsten equivalent) seemed a bit too large to be immediately practical, and probably would command a more substantial price tag. For this reason, I settled on the mid-range power level of 575W.
It turns out that while 575W and 1200W are very common bulb sizes in the cinema world, they are almost unheard of in the commercial ballast market. After exhaustively researching affordable off-the-shelf ballast units, and finding that the cheapest were well over $1000 on the second-hand market, I decided that a re-evaluation of my plan was in order. Some simple reading up on the subject of HID ballasts (instead of HMI) turns up a wealth of inexpensive units available in the 400W, 600W, and 1000W sizes. These ballasts are mainly designed to drive high-pressure sodium (HPS) lamps, and HPS-replacement metal-halide lamps, and seem to be geared towards the aquarium and hydroponics markets. The trick then becomes finding the appropriate bulb to match up with it.
Well, after some hunting around Osram's website, I found the HQI line of metal halide HID lamps could be operated either from HQI-type ballasts, or NAV (sodium vapour) ballasts, and are intended for use as HPS-type-replacements. If you look up at the lights in a parking lot nowadays, or the lights high up in the rafters of any big-box store, I guarantee you are very likely looking at an HQI-type bulb. You are probably also looking at a bulb coupled to an inexpensive magnetic ballast and ignitor. The HQI line of bulbs can be found in Osram's Discharge Lamps Overview document. More lamps may be found by browsing the catalog at catalog.myosram.com. Other resources include the catalog at Philips Lighting and the commercial products catalog at GE Lighting. While many bulbs are available, it is sometimes hard to strike a balance between power, size, UV-filtering, CRI, and colour temperature. I settled on the HQI-TS 400W/D in an Fc2 base. It is a double-ended 5200/5600K bulb with a CRI of 93 and a life expectancy of 10,000 hours (cinema HMI bulbs are often rated for only 500-1000 hours, thought I think that is probably due to the fact that they want to try and guarantee a certain quality of light and that as they age, the spectrum starts to shift). The downside of going with this bulb is that it is fairly large physically at just over 20cm in length, it lacks UV filtration (which is common in HID lamps), and it can only be operated +/- 45 degrees from the horizontal. I purchased it from the homebrew projector suppliers diy-beamer.com for 48 Euros including the Fc2 socket.
In this photo are pictured the Osram HQI-TS 400W/D, the Osram HMI 575W/SE, and the GE CSR400SE/HR. The HQI bulb is almost twice the size of the single-ended GE CSR400, which will be going into my next unit. As it stands, the 20cm bulb plus the socket means I need to find a fixture that will accommodate 9" of bulb internally.
The Mole Richardson model 412 2000W fresnel lamp has a 10" lens on the front, and at least 10" of clearance inside the body. It just so happens that the kind folks at Pyramid-Films have several available for just $150 including barndoors and scrims. Be warned though, this is a fairly large lamp. I was more accustomed to the size of 650W fresnels and 2K open-face tungsten lamps, and I expected it to be not much larger than the 2K open-face.
Well, such is the burden that I will bear for going with the longer-lived, inexpensive bulb that runs off an inexpensive electronic ballast. Which brings me to the next stage in the selection process.
There are many brands of inexpensive electronic ballasts out there: Ice Cap, Galaxy, Lumatek, etc. Unfortunately, many of the HQI ballasts you will come across are rated for 220V as many of these HQI-type bulbs are geared towards the European market. Also be wary of ignition voltage ratings. Some of these ballasts require an external ignitor to strike a bulb. After sifting through pages of online reviews and running into many dead-ends specification-wise, I decided to go with the Lumatek 400W ballast. No one could really determine if it would power the HQI globe from Osram, but since it was rated to run HPS lamps and seemed to have enough oomph in the ignition voltage department, I figured for the price, it was worth a try. Moreover, the HQI-TS spec sheet put the current requirement right in the range of this ballast. When the ballast arrived, I found an instruction sheet enclosed which indicated that when troubleshooting ignition problems, it was quite probable that your European-style bulb was not compatible with said ballast.
I mounted the Fc2 sockets onto a wooden base to test it out, since I hadn't yet received my Mole 412. Wired the ballast up to the socket using the supplied shielded cable, installed the bulb, and plugged the thing in. Well, it lit up right away, flickered a bit for 30 seconds, and danced through a variety of shades of blue and green before finally settling in and burning a very bright daylight. I used this setup outside to work on my car one night and the neighbours were sufficiently stunned. From inside the house looking out my front door, I was tricked into thinking it was still daylight outside. It also came on a shoot shortly after and worked nicely as an interior fill off a bounce board in a scene that had a window in the shot.
Some people have asked me how much heat the ballast produces, and I would describe it as cool enough to pick up and relocate, but warm enough that you probably wouldn't want to be tasked with holding it for the duration of a shoot. The hottest I have measured it is 45 degrees Celsius. Also, my heel can attest to the fact that those radiator fins are kind of sharp! I would estimate it is dissipating between 20 and 30 watts.
The Mole 412 finally arrived and I immediately went to work on removing the existing 2000W socket. Check out the size of that 2K bulb. Seems a fair bit larger than the double-ended 2K bulb I had in my Arrilite, but that is probably due to the heatsinking requirements of the single-ended base.
It was a trivial exercise to disassemble the fixture by removing 4 screws that secure the base to the can. The base houses the mechanical assembly and sled on which the socket/bulb and reflector travel back and forth to adjust the beam pattern. The base has to be twisted when removed in order for the reflector to clear the opening. It's a nice big reflector by the way -- somewhere on the order of 8" across. Reflectors in a fresnel fixture are circular, and the bulb is placed at the focal point of the sphere. In this arrangement, any light that leaves the bulb heading to the back of the lamp is reflected right back through the center of the bulb again. Since a fresnel fixture does not depend on parabolic focusing of the beam by the reflector, but instead by the positioning of the bulb with respect to the fresnel lens itself, this reflector arrangement eventually made sense to me. (you are effectively doubling the forward-facing light output of the bulb by bouncing the would-be-lost beams heading in the opposite direction).
Here you can see the yoke I fabricated to mount the sockets inside the lamp. I just bolted some scrap aluminum angles to the sockets and bent a strip of brass into a U-shape which mated up with some more aluminum scraps that were screwed to the sled. I know you're thinking that this is not a U-shaped yoke, but I assure you it was quite nicely formed and visually appealing until I tried to slide the base back onto the can. The socket and mirror would not fit through the bottom of the can so I had to raise the base of the yoke up a couple of inches to clear the narrow throat (see the reflector mount for an example of the required geometry). The high-voltage leads from the sockets were routed around behind the reflector and out the existing wiring path in the base of the fixture. There was a very heavy duty in-line switch housed on the base, which I removed since I figured it would not be suitable for the high-voltage ignition pulse produced by the ballast. Also, in the event that the switch was left open, I don't think the ballast would be happy about driving into such a high-impedance load. Instead, an in-line switch will be connected on the mains AC side of the ballast.
Here you can see the unit fired up with the lens open. I have not yet mounted a UV filter on this, but my plan is to pick up a sheet of Rosco Cinegel #3114 Tough UV Filter for under $10 at B&H. The filter will be cut to size and mounted to the rear of the fresnel element, inside the fiture. The temperature inside the fixture has not yet exceeded 100 degrees Celsius, so I am quite confident it will not melt or burst into flame.
Okay, I reused that photo at the head of the document, but I figure it belongs at the end too. I need to burn the unit in and do some flicker tests with the fixture oriented in different positions, but since the bulb is mounted horizontally, I fully expect to be able to rotate this thing completely about the yoke axis. The +/- 45 degrees-from-horizontal rating on the burn position for this bulb (P45 in Osram-speak) should not be an issue since I can not tilt the fixture from side-to-side anyways.
Well, that's it for my "large" 400W fixture. Total cost was under $500. Considering the cheapest 575W unit on the market is probably cool-lights CL-MF0575 at $1600, I think this has been worthwhile, if not somewhat bulkier. Heck, the 1000W Lumatek ballast plus a 1000W HQI-TS bulb would fit nicely inside this Mole 412 housing, providing a reasonable alternative to this budget 1200W unit. You probably don't even want to know what K5600 sells their 400W units for (over $4k), although those are PARs and not Fresnels. Still, HMIs are terribly overpriced.
As shown in an earlier photo, I am currently in posession of some 400W and 575W cine HMI globes. I tried to power the GE CSR400 using the Lumatek ballast, but it will not power up fully. It strikes, ramps for about 20 seconds, and then goes out. Looking at the specs for this lamp, it seems it requires on the order of 7A compared to just under 4A for the HQI-TS bulb. This is because it operates at a lower voltage due to the shorter electrode gap in the bulb (producing a better point source too I might add). The Lumatek ballast simply will not source that much current, so it shuts down. It is possible that the 1000W Lumatek with a current limit of 8.45A will power it, but I have a feeling it will end up driving it over-current. Also, for the price, it is worth investigating other options.
I managed to locate a ballast manufacturer in Germany that produces HMI ballasts and after much discussion, decided to order one of their 400/575 ballasts plus the appropriate ignitor which I can use to power standard 400W and 575W HMI globes. The downside of their product, I was told, was that they required cooling to operate correctly, and for this reason (fans), they had not yet been able to break into the lucrative cinema lighting market. I assured them that I was doing some R&D on just such a unit, and they were most helpful with specs, and locating critical thermal points on the board, etc. I have even run my 400W globe with only a single PC fan aimed at the unit. I have not yet tried it with no airflow, but perhaps I will give it a shot. The fans I am playing with are fairly small and very quiet. Since the ballast will be used some distance from the lamphead itself, I don't believe this fan noise will pose a problem. It is also possible that with enough heatsinking, the fan will be unnecessary. More research to follow. Unfortunately, Schiederwerk will not sell directly to the public unless you are ordering at least 10 or 20 units, so I had to order from one of their vendors, Professional Lamps Inc in New Jersey. The PVG5-57LC 400/575W ballast came to $450 and the ignitor was $99. So, already this is quite a bit more costly than the Lumatek solution, but will handily power both a 400W and a 575W lamphead.
Here is the ballast and ignitor powering the GE CSR400 in a GZZ-9.5 socket (exotic and tough to find, but K5600 will actually sell them to you). I am still waiting on a G22 socket so that I can fire up the Osram HMI 575, but I don't expect any difficulties. These globes should handily fit within a much much smaller fixture. Something on the order of a 650 or 1K fresnel. Of course, I may get creative and mount one in a PAR fixture like the Joker Bug.
As stated earlier, please exercise extreme caution when working around high-powered electronics. There are so many ways you can kill yourself with this stuff, from touching the wrong capacitor by mistake, to forgetting a ballast is plugged in and grabbing a wire, to holding an insufficiently insulated high-voltage (ignitor) lead, to improperly grounding your power supplies, to exploding HMI globes (yes, they explode violently, complete with high-temperature shards), to suffering severe sunburn (and getting cancer). If you don't know for sure what you are doing, don't do it.