A look inside the Rokbox 5 LED wash fixture

  As part of helping the Somerville Armory with some repairs and improvements to their lighting rig, I went shopping for simple LED fixtures.  My primary criterion was the best "lumens per dollar" ratio, for whatever was available with the right beam width and power appropriate for their space and stage size.  The existing rig in the performance hall consisted of a couple dozen wheezy old aluminum par-cans driven by an elderly CD80 dimmer pack, and a mysterious rats-nest of wiring connecting it all.  Two years before I had used it for the fashion show related to Somerville Open Studios, and had done a bunch of neatening-up and relabeling in the process of doping out what they had and what I could do with it.  After a year hiatus the show was slated to return, and in its early planning stages I decided to do things a little differently.  I knew where I'd *wanted* to put lights, even though what I wound up with back then was sort of a compromise with borrowed gear.  This time I figured a permanent installation could serve several needs at once:
  • Upgrade the hall's rig with nice colorful stuff
  • Put lights where I wanted them for the stage/runway
  • Serve as a nice donation to the Armory collective
and generally make a bunch of people happy, including myself, when it came time to run this thing.

Amazon seemed overrun with cheap DJ-grade gear that I knew wouldn't last or throw nearly enough light.  I cruised some of the more professional-market oriented places, comparing photometrics and prices, and LONG story short(ish), a family of PAR-style wash units from Blizzard Lighting came out on top.  In particular, the Rokbox 5 RGBAW seemed the most suited for the need -- 26-ish degree beam, more rated raw output power than most of the others, and half the price of anything with zoom capability [which I didn't need].  And *out of stock* at several places, as they've evidently been quite a popular unit.  When I called Full Compass to ask about their stock, not only did they still have a small handful left, they offered an even deeper discount price for the quantity I needed.  Awesome, *done*.  And the shipment arrived astoundingly fast.


[Images are linked to larger copies.]
Blizzard Rokbox 5 RGBAW This is what the unit looks like; it is indeed a box.  It's the bigger brother of the "Hotbox" mini-fixtures that I'd seen and even used elsewhere.  The dual yoke fits onto solid bosses on the sides with on-axis tightening knobs rather than separate tilt-locks, but a full set of reasonable friction washers is provided.  3-pin DMX, PowerCon in and out, and a fairly simple menu interface with various standalone color fades and flashes built in.  And very punchy output, drawing about 160W at full white. 

Rear panel removed The obvious rear panel has already been removed; this is what we see under it.  Fairly nice construction, with internal structures mounted with standoffs and screws instead of cheap plastic tabs.  Wiring is neatly laid out, and all the connectors are "gooped" into place for stability but not in a way that would really impede removal if needed.

Other panels removed The other panels come off easily enough.  They're all held on by small machine screws which are nicely countersunk, and more notably all the *same size*.  This makes keeping track of them easier than on many other electronic items.  The entire outer box is *steel*, not aluminum or plastic -- the unit feels quite solidly built overall, weighing close to 10 pounds, and shows some nice precision in assembly fit and finish that we normally don't expect from offshore-manufactured units.

The control and I/O panel A glimpse into the LED array
Part of the front plate stackup is a big heatsink that the LED modules are thermally coupled to.  A glimpse in through one of the corners shows a hint of how the modules are mounted and the output optics.  There are three tightly-connected layers here: the heatsink, a sub-plate that holds the output "lenslets" in place over the emitters, and the outer front faceplate sort of binding it all together.

Initially I wasn't going to bother with further disassembly in this area, thinking that there might be some careful, sealed factory alignment that I wouldn't want to disturb.


I/O panel wiring Two more screws out allows the I/O and display panel to be flipped up to view its wiring.  The A/C and DMX are simple feed-thrus as we'd expect.  The power supply appears to be a generic OEM unit that the rear half of the case might have been designed around, and it's evidently single output which leaves all the subsequent regulation down to 5 and 12 volts or whatever to live on the control boards.  The fans take in air here, and push it down through the whole heatsink fin array to an exit slot at the bottom of the unit.  They are *not* variable speed; they just run all the time so the Rokbox is a tad noisy in operation.

Here we can also see a potentially hazardous quirk these units have, pointed to by the red arrow.  This is a fourth connection to a 3-pin XLR jack.  Huh??

 
  Blizzard supports an add-on device for several of their products, named "wiCICLE", intended to carry wireless DMX between units.  The transceivers come in the form of small XLR-barrel shaped dongles, that plug into the input DMX port and have an antenna on the other end.  They are powered by the device they connect to, by means of bringing the XLR *shell* connection to +5 volts and drawing from there to ground on pin 1.  What this means is that if a cable that [erroneously, but very possible] connects its pin 1 signal ground to the metal connector shell, it shorts the Rokbox's 5V supply to ground and *smokes it*.

There's at least one forum thread taking Blizzard seriously to task for this bad engineering decision, which they already clearly acknowledge as a problem, because every unit comes with a bright red label saying   "STOP!  Check your cables"   pasted *over* the DMX input jack which you have to remove.  What they didn't do was make that power *optional* with a recessed slide switch or the like.  At worst they could have used 5-pin connectors with the extra two pins providing the power instead, which would frankly give a better chance of avoiding cable-related mishaps.


Disabling wiCICLE nonsense Here's the simplest and least destructive way to disable wiCICLE power: find the data connector on the main board and pull it.  For the yellow wire that has the obvious "5V" label in the silkscreen, use a small tool to push the barb and back that one pin out of the connector shell.  *Note 1:   That wire may be the red one instead; at least one fixture has been observed with the red and yellow wires swapped.  *Note 2:   Some units may have TWO such plugs on the main board, one routing to each of the DMX input and output jacks instead of the loop-thru running directly between the jacks.  Pull whichever one is at that end of the connector(s).  Tape the lug back along the cable so it's covered, and reconnect the rest.  No more 5V going where it shouldn't.

Blizzard is somewhat defensive about this method of providing power, claiming that they've had "very few problems" especially after putting the warning labels over the jacks.  Flawed logic: sure, I might have pulled those labels off units I own and heeded the warning, but what if they're going into *rental stock* or out on gigs with less-trained helpers?  There are plenty of situations where owners aren't going to have control over what cables get plugged into these things.  While I haven't bothered with this on the Armory units as they're going in the air as a longterm installation with known-good cables, I would do that fix for any unit headed for general use on the road.

Nuvoton control processor DMX input bias network near 75176
The control processor is a little OEM daughterboard from Nuvoton, sporting one of their ARM Cortex M0 series microcontrollers.  This appears to have a dead-simple interface -- power, a reset, and six GPIO leads.  The LED driver board is at the lower left where the inductors with blue bobbins are, and what's not shown in that picture is an unpopulated area labeled for driving the UV emitters used in the six-in-one Rokbox unit.

Near the connector where DMX data arrives we find a bog-standard 75176B transceiver chip, and we can easily dope out the surrounding bias network for the RS485 line coming in.  Yes, these units each have one -- 10K to +5 and ground.  And 100 ohm series resistors, which we've already determined aren't really necessary for "matching" line impedance.  For more on bias networks and their rationale and why connecting too many receivers that provide bias might lead to some transmission problems, see this DMX discussion.

As is typical with these units, any self-run setting is likely to begin *sending* data on the DMX bus with the expectation that slave-mode units are listening, so in a typical installation with a controller it is essential to leave all units in "Addr" mode with the correct addresses set.  This is true via wiCICLE or hardwire, of course.  Slave mode appears to go into 10-channel mode on address 001 so a master unit can control all attributes with relatively short packets.


    The emitters

Face retaining panel off I've finally braved opening up the front panel, finding that there's nothing special about it -- the two plates simply match holes and retain the lenslet flanges in fairly precise placement over the LED emitters.  The lenslet shape is somewhat parabolic on the rear, probably using total internal reflection to direct the emitter's light out the front and through a token diffusion surface. 


What's under the lenslets At DMX value 001, the emitters pop on with a gentle glow that's clearly visible but not too bright to photograph.
 
Close-up of LED module LED dies
Going very dark on shooting allows getting the detail on the emitter chips themselves.  The emitter dome is about a quarter-inch across.  As usual, it's kind of amazing how such tiny chips of stuff can belt out as much light as they do when socked with meaningful current.

Output from single module The emission from any given die is close to hemispherical, which gets narrowed considerably by the lenslet but the group projects a clearly positional pattern based on where each chip in the emitter sits.  This is the output from *one* lensed element, with the others masked off.

Module orientations But that's why the modules are installed in six different orientations, to mix that directionality up and overlay the outputs in a blended pattern.  The result is quite flat; there's a hint of patchiness in the amber output but overall the field is very well mixed with minimal color fringing in shadows.

    Output measurement

  One of the difficulties in the shopping process was determining net output of any given candidate fixture.  Most manufacturers of LED gear give their photometrics in terms of lux at various distances away in meters, without really taking practical beam angle into account.  This leads to a bit of math needed [read: guesswork, because these are soft-edge wash units!] to determine raw ballpark lumens, but that coupled with a notion of full-up input wattage yields an approximate basis for comparison.  I wound up calling a couple of source companies to strongly suggest that they also give their output ratings in lumens, similar to makers of things like video projectors, and let designers worry about beam geometry as a secondary concern.  In the LED market, buyers also look at lumens per watt efficacy as a metric of how well the emitters are doing and thus how hot they are likely to run, which has implications about their long-term useful life.

In pursuing that, I pointed out two fairly easy ways they could measure lumens in the lab.  A circular flat beam 13-1/2 inches in diameter is one square foot, and measuring footcandles on that plane directly translates to lumens.  Alternatively, such as in a stage setting, a circle on the deck 11 and a half feet across encloses 100 square feet, thus footcandles measured in the middle times 100 gives rough lumens.  Either method is fairly easy.


Attempt to contain the beam To set up for my own measurement, a token effort is made to contain the beam a little and make it harder-edged and less "washy", with a ring of aluminum tape run around just outside the emitters to hopefully reflect the outermost beam fringes back in toward the middle.

Lumen measuring setup Program setting 31 appears to be everything full-up, so no control console is needed for this.  I've fired it toward the house, placing the half-brighness "edge" of beam dropoff at points measured 11.5 feet apart, and the footcandle reading is taken in the middle of the beam.  Right around fifty, implying at least 5000 lumens which is about what I'd figured from the specs.  The unit is about 22 feet from the wall at this point, so the multiplying factor for these is about 0.5 and the usable field is actually a little larger than 26 degrees.  It's not a thousand-watt par64, but it's certainly holding its own here.

The output is again quite even, with perhaps just a hint of "pinking" toward the outer edge but nothing that would really stand out in a theatre setting with multiple units.  At a distance, there's no discernible color fringing at all.  Any remaining visible patchiness is completely eliminated with a cut of R114 frost taped over the face of the unit.


    Spectral analysis

  While adding amber and white to an LED unit nominally allows creation of those delicate theatrical pastel colors that should look decent on flesh, it's still not perfect color rendering.  Most LEDs are inherently peaky sources, which is one reason typical RGB output delivers such punchy color purity to our perception.  To analyze what I'll get from the Rokbox RGBAW, I've set it up inside one of the shipping boxes with the flaps forming a narrow slit across the output.  A piece of high-dispersion-angle holographic diffraction grating held over the camera lens shows a nice first-order spectral spread of what's emerging from the slit, enough for a pretty solid qualitative view of what we're getting.
 
Spectrum: red, blue, amber This shows just the red, amber, and blue components, which are quite narrow.  Especially the amber -- by itself, it's almost monochromatic, in a similar way to high-pressure sodium streetlights.  Hmm, great for realism in lighting those nighttime parking-lot scenes??
 
Spectrum: adding green Now we add the green channel, which is a little wider but rather separated from the others.  We're still not going to obtain a true yellow out of this, but can probably fake up something that looks close enough.
 
Spectrum: white LED alone This is the white emitter by itself, a phosphor-based broadband fluorescent type.  It covers red/yellow/green reasonably well but leaves a pretty serious gap in the cyan-into-blue region, so it's far from the "perfect white".
 
Full spectrum Here's everything on, where we still see that drop around the cyan area.  So while we're probably into the ninetieth percentile of color rendering, this will never achieve true daylight white.  It still looks reasonably "white" on stage with all five channels at 100%, which is quite useful because it means "full on" doesn't need any channels pulled down to look quite reasonable as a general blast of light.


    Control strategy

  The easiest way to run these units from a desk is in 5-channel mode, directly mapping RGBAW to what the board is sending.  The other modes seem to just be confusing and not really worth fighting with.  Another complaint about these units seen around the net is that their output response is somewhat nonlinear with respect to DMX value.  They do seem somewhat fast off the bottom with quite a lot of the output on by the 50% point, and not a whole lot left in the second half of control.  One could presumably compensate for this by setting a "slow bottom" dimmer profile on the relevant channel if the board supports it.  For theatrical work it is also definitely worth setting dimming "mode 1" or higher on the Rokbox itself, which applies smoothing to all brightness changes and approximates filament delay in conventional fixtures.  Setting "mode 0" will allow instantaneous on/off changes and strobing for club environments and such.
 
Studying the processor I/O port pins The question is, does the nonlinearity come from the control firmware or just the response of the LEDs themselves?  It's thus worth taking a look at how the devices are driven.  A couple of minutes with a scope shows that the ports on the processor board are used in a totally straightforward fashion: R,G,B,A,W map directly to PWM signals output on the P1,P2,P3,P4,P5 pins.  P6 appears to carry something related to the DMX data, but it's unclear how that would get multiplexed in a six-channel UV-capable unit.

Let's look at the blue signal on P3.

 
Scope trace of blue modulation The main signal is a low level on the blue, but there's a little bit of crosstalk hash from the red and green whose levels are up a little higher.  [Or maybe my ground connection is crap...]  The positive width seems to vary in precise linear accord with where the slider on the console is, going from no pulses at all at DMX value 0 to very near 100% duty at DMX 255.  Qualitative visual output doesn't quite follow this, however, and is best shown in a couple of short videos to capture the time dynamics of what's going on.
Nonlin.mov [682k, 28s] demonstrates a couple of things: the temporal smoothing of dimming mode 1, and a subjectively nonlinear brightness in the unit output dimly reflected in the surface of the scope screen.  The sound is of the "blackout" button being toggled on the board, and the non-instantaneous drift of the resulting PWM is clear to see.  Modes above 1 make that even slower.  Then, the channel slider is moved slowly up and down.  There is an additional visual nonlinearity from the camera dimming itself down as overall light level increases; pay attention to how bright the 'scope front panel looks above maybe 50%.

Startmod.mov [83k, 10s] is on a shorter time scale and shows the port output from the first very few DMX values increasing from zero and again, the reflection of how that looks from the unit.  No idea why the first pulse is shaped the way it is; perhaps something to do with how the port hardware gets re-enabled for the start of every pulse.  The MCU has plenty of time to do other things in between pulses, and may put the port back in tri-state mode when not actively outputting.  But that's getting in a little too deep here; we have the piece of info we need about the output curve, and how we might compensate for it on a desk that can do dimmer profiles.  Unfortunately, the configurable dimming modes only affect speed, and don't offer any internal curve changes on their own.  Adding that would be a relatively simple matter of firmware and associated documentation.

In practice when shot onto a stage, the nonlinearity is not a big issue.  For a baseline of 100-ish footcandles read at a particular spot, taking the channels to 50% leaves about 60 footcandles on deck.  So sure, the output curve is slightly "bowed upward" but certainly not in any unreasonable way, and anyone who's inordinately fussy about it can apply a gentle compensation profile as desired.


Plenty of clearance under yoke for connections And check this out: because of where the power and data connectors are located, there's plenty of clearance to swing the yoke *over* them, even with the long "switchcraft" style DMX heads.  Try that with a Selador / Lustr from ETC, or a lot of other fixtures with relatively compact yokes!  The natural hang of the box with the thumb-screws loose is about 45 degrees downward, too, so in a typical hung position there's relatively little likelihood of tilt drift if the screws aren't torqued in very hard.

Just don't run them sitting upright on a floor or table and firing horizontally, e.g. the position this one is in now, because the entire air exit would be blocked off.


  Overall, I would say "nice job" from Blizzard and their collaborators.  The wiCICLE thing needs some work and output linearity could be better, but on early examination their *BOX line seems to be a pretty robust platform with professional-grade build quality and attention to detail.  Now we get to see how it holds up over time in a fairly busy event space. 

_H*   170428