Really! Just busy with some real-life stuff, namely wedding related and home renovations. I haven’t forgotten about this pick & place stuff! Lately I’ve been spending most of my project time on getting Mosquino toward an official 1.0 release. The rev2 boards just came in, so once all the parts are in I should have one ready to test soon. Here they are!

As usual, click for fullsize. Clockwise from the bottom-left are a bistable display shield, microSD shield, low power boost board (as low as 0.6V to 3.3V), Peltier shield (thermal to electricity, ~20mV to 4.1V), vibration energy harvesting shield, a stackable LiPol / thinfilm battery power shield, and of course the Mosquino mainboard itself. You may have seen early versions of some of these on the Mosquino page already, but these implement bug fixes and the latest/finalized Mosquino pinout. Can’t wait to get playing with these!

(And no, purple is not the official / final color – The PCBs were made via Laen’s sweet batch PCB service; he likes to experiment with the colors from time to time. It’s not a ripoff of Lilypad…although the Peltier board can potentially harvest from typical bodyheat gradients (>=2degC), which is an interesting development for wearable computing projects to say the least!)

cexxy version

Typical frog in a hot pot scenario; when I joined Comcast the modem lease was like $1.50 a month, and I didn’t even think about it. As of recently it’s now crept up to $7.00 a month, which kind of made me sit up in shock. How much do those things actually cost, anyway?

Answer: $16 on eBay!

Ditching the leased Comcast cable modem in favor of your own is a surprisingly simple process. In my experience, Comcast won’t even try to (intentionally) stonewall your request or tell you it can’t be done in order to keep the revenue. Unfortunately, their techs are not exactly the brightest lights in the harbor, so you might have to train them a little on how to do it. Here’s how…

Step 1: Buy the modem
Go to your favorite new or used equipment source and buy the modem. Make sure your purchase includes any necessary power cord (wall wart); if not, buy that too. Theoretically, any DOCSIS 2 or later modem will work with most cable Internet packages, but to be sure, check this list for modems tested and approved by Comcast for compatibility. Extremely fast or fancy internet packages might have special requirements. Personally, I just searched eBay for the exact model # of my existing leased modem, and bought that one. My total cost was about $21 for the modem and a 12V wall adapter.

Step 2: Plug in the modem
Before you go connecting anything, turn the new modem over and copy down the “HFC MAC” number printed on the bottom to someplace more convenient. Note, there may be several different numbers printed on the modem; the “HFC MAC” is what you want. Technically the “number” is in hexadecimal, so it can also include the letters A-F. Double- and triple-check that you copied it correctly!

Disconnect your leased modem and plug the new one in its place. Verify that the lights come on and blink just as with your old one. (It will still ‘see’ a modem signal when connected, even if it’s not activated yet.) Once it’s lighting and blinking, power-cycle your wireless router (or whatever attaches to the modem Ethernet cable) to make sure it picks up a fresh IP address from the new modem. Just to be safe, reboot your computer(s) after this to make sure the newly rebooted router gives them a fresh address too. Now your modem, wireless router and computer will be “connected” to one another as far as your home network is concerned, although they won’t be able to reach the Internet through the new modem yet.

Step 3: Activate the modem
The one and only piece of information you (and Comcast) will need for this is the “HFC MAC” number you wrote down earlier. Call the Customer Service # on your Comcast bill, and say to them:
“Hi, I’d like to use my own modem and return my leased modem.” When I did this, the main customer service gave me a separate phone # dedicated to handling this request. Call that # and repeat the request.

The Comcast person on the phone will ask for the number from your modem. Not all of them are smart or well-trained, so they may not know which number, nor tell you the correct number to provide. Whatever they say (or don’t), give the “HFC MAC” you copied earlier. Now, this is important! Have the Comcast person input the number and then recite the number back to you, to make SURE they input it correctly. This is important!

Before you hang up, start accessing Web sites and see if they start working. If the Comcast person input the # correctly, your Internet should start working again almost immediately. If not, login to your wireless router’s status page (consult its manual for how to do this) and make sure it obtained an IP address, gateway IP and DNS servers from the modem. This information may be listed under a section titled “DHCP” (a protocol for devices to request and assign IP addresses.) Try powercycling the router again while the modem is activated to make sure it gets an address.

Step 4: Return the old modem
Hopefully, everything is working now! The last thing to do is to pack up the old modem and its wall plug in a box and return it to Comcast. If it came in an official Comcast box (e.g. “self-install kit”) and you still have that box, use that box – but if not, my experience is they aren’t that picky (I used a shoe box). There is a brick-and-mortar Comcast service/payment center by my house, so I just returned it in person. If this is not an option, ask the Comcast person how to return it by mail. My experience at the Comcast payment center was very positive – just handed the modem over, they scanned a barcode on the bottom and it was automatically credited to my account. They handed back a receipt with my name/account # and the modem details on it and I was on my way. My next bill had a partial refund for the part of the month I was no longer leasing the modem. Done and done!

If all does not go well…

If your new modem isn’t delivering the Internet goods after activation, the Comcast person (billing department) will transfer you to a separate department (tech support), who have the power to ‘ping’ your modem and make sure it is visible on their end. ‘Your’ modem in this case is defined as the modem matching the HFC MAC # linked to your account, which is why it’s very important that the billing person has input the correct #, and input it correctly, BEFORE this point. Otherwise they can ‘ping’ all day and not get any result because their system is looking for the wrong damn modem! The Tech Support person has the power to ping but NOT the power to add or correct MAC #s on your account, so this sucks. Likewise, the billing department has the power to enter MAC #s, but NOT to ping the modem! If this magic number entry gets cocked up somehow, it will take a 3-way conference call between you, tech support and billing to sort it out, and not all Comcastic techs know how to pull this off with all those complicated phone buttons. I spent two hours bouncing between departments because the barely-English-speaking billing person miskeyed the # the first time.

For the insanely bored or curious…

“MAC” number stands for Medium Access Control number, which is a globally unique number (burned into the device by the manufacturer) that identifies YOUR device among the millions of others out there just like it. The “medium” referred to is the physical cable. Since your block’s local cable segment is a shared resource, this number is necessary to identify you as a paying customer and route the right bits to and from YOUR specific modem. The difference between the separate “HFC” and “CPE” MAC #s is that the HFC number (I’m told this stands for “Hybrid Fiber-Coax”, i.e. residential cable networks) is the one that’s visible on the coax (cable) side of the modem that your ISP sees, and the other (“Customer-Provided Equipment”) is the Ethernet-side number that’s visible to your equipment (e.g. wireless router). Don’t tell Comcast that number by mistake; they can’t see it on the cable end.

cexxy version

Despite the impact of work, wedding planning and Super Metroid fan-hacks (not necessarily in that order ;-) on my freetime, my scheme to design a DIY-able open pick & place system is starting to come along. So far, there is a proper vacuum placement head, a rough idea of what the software architecture might look like, and this. For those who saw the last post, you probably guessed what it was leading up to.

This is a simple proof-of-concept of a SMT tape-and-reel part feeder design. The main parts are a stepper motor and feed sprocket to advance the tape, two walls with guide slots, and a simple slider mechanism to allow the feeder to accept tapes of varying width. Please note that it does not (yet) include any mechanism for peeling and disposing of the tape covering. Suggestions and innovations in this area (as well as all others) are welcome!

All the parts were cut on a CNC mill – design files and G-code are available here. The .cb files included can be opened with the free version of CamBam. More photos / video and design details are available below.

The Tape Standard
The geometry of the standardized tape that holds SMT parts is documented in EIA-481-2-A, which until very recently was only available at a price too high for mortals (or from your favorite ‘alternative’ source, wink nudge). It appears that EIA disbanded at the end of 2010 and the documents are now public. Regardless, here are the parts of interest to us building a DIY tape-and-reel feeder:

Pitch (distance between sprocket holes): 4mm
Sprocket hole diameter: 1.5mm
Center of sprocket hole to edge of tape: 1.75mm
Center of sprocket hole to edge of component wells: min. 0.75mm??? (seemingly not specified; varies between manufacturers)
On opposite side of tape – ending edge of component wells to edge of tape: 0.6mm min.
Tape thickness: 0.2 ~ 0.4mm, not including covering.

The standard tape widths are 8/12/16/24/32/44/56 mm. For tapes 32mm and wider, a row of slightly elongated (same pitch) sprocket holes is added to the other side. According to various sources, the standard also says the pitch between part wells should be a multiple of 4mm and “Pin 1″ (if any) should generally be on the sprocket side and facing forward. A longer explanation of the part orientation rules is that the part should be a) widthwise (its longer dimension, if any, across the tape); b) Pin 1 toward the round sprocket holes (unless this conflicts with the first rule); c) Pin 1 facing in the direction of travel (unless this conflicts with the first 2 rules). How religiously any given vendor adheres to these rules is anybody’s guess. There is also no rule saying the ‘well’ or pocket holding each part has to remotely match the size of the part (except to prevent it being able to flip over or rotate 90 degrees during shipping), so visual positioning correction is occasionally needed for parts from particularly lazy vendors.

Feeder Design Points
With these things in mind, the shown test-design uses the following dimensions, which seem to work well with varying size tape samples I fed through it:

Sprocket thickness of 0.7874mm (.031″) – this is a ‘standard’ stock thickness in the US, so it was an easy first test. I sunk the sprocket so that it lies flush with the inside wall. With these values, the depth of the tape track is cut to 2.14376mm (0.0844 inches) to put the edge of the tape flush with the track when on the sprocket, forcing it to stay straight despite all the feed force being delivered on one side. For the ‘outside’ (non-driven side) wall, the track is cut to a depth of 1mm (0.039 inches). The width of the slot is 0.79375mm (0.03125″), as this is the closest (commonly available in the US) milling bit width that will accommodate both plastic and the slightly thicker paper tapes. To assist loading tape, a wider ‘mouth’ is cut on the input side of both sides’ slot, tapering to the slot width to guide the tape in. In reality, the mouth you see in the photos is not really large enough to be useful (it looked a lot bigger on my monitor!…). Also, the mouth on the non-drive side in the photos is on the wrong side since I completely forgot to account for this half being flipped over :p These are corrected in the downloadable files.

With the sprocket as designed, the tape track radius is 13.6mm (0.536″), and the track makes a 180 degree pass around the sprocket to eject the spent tape at the bottom of the same end it comes in on. In hand-testing some tape around the sprocket, I found it does not always sit flush against the sprocket throughout the entire 180 degrees, so the tape track is widened a bit around the sprocket to accommodate. The ~31mm sprocket diameter / number of teeth (22) was chosen purely for convenience on my part, as this is the tallest that would comfortably fit on the homebrew CNC mill I’m using for testing. In practice, a larger diameter is advisable so as not to limit the size of parts (deepness or length of the ‘wells’ vs. sprocket diameter and bend radius) that can be fed. Also, the reels themselves will be much taller than this anyway. Just keep in mind that as sprocket diameter goes up, so will the required motor torque to advance the sprocket one step, and the linear distance per step. At some point some form of gear reduction will be necessary vs. simply tacking the sprocket directly to the motor shaft.

To handle a wide (pun intended) variety of tape widths, a trio of smooth nylon PCB standoffs (0.250″ dia) are sunk and bolted into the drive side to act as sliding rails for the halves to be pulled apart to the desired width. A rubber band around both halves keeps them tensioned and in contact with the tape edges when loaded. This could definitely stand improvement, but it works for now.

Despite being a quick ‘n dirty test piece, the action of this feeder (by hand) is surprisingly smooth. The nylon spacers are pretty slippery, and the Delrin also provides a smooth, low-friction surface when machined. With the motor is a different story; these tin-can motors are 7.5deg/step and the driver board I’m using right now doesn’t really support microstepping low-current motors.

cexxy version

This is a free (open source) Python script for creating feeder sprockets for e.g. perforated tape or film advance. I wrote it for myself to generate SMD tape-and-reel feed sprockets, but it might also be useful for making replacement sprockets for 8/16/35mm film, microfilm and paper-tape systems whose original reader hardware no longer exists or is difficult to find replacement parts for. The output is a .DXF template suitable for laser cutting, 3D printing or CNC machining. “Documentation” below, but it should be pretty self-explanatory. It should work with any modern version of Python (tested on 2.6).

Download

Sprocket design goals / differences from other sprocket types

The drive sprocket’s dimensions are specified mainly by the number of teeth, width (or diameter) of the sprocket holes, and the pitch (distance between sprocket hole centers). The tape is usually advanced either tangentally to the sprocket, or partially wrapped around the sprocket. Thus the distance between the outside edges of any two teeth at any point, either tangent to the sprocket or along the circumference of the sprocket, should never exceed the distance between the outer edges of any two sprocket holes (the taper of the teeth is computed to counteract the radial splay of the teeth). Additionally, a landing area (flank) is cut at the base of the teeth matching the thickness of the tape, giving it a place to ‘catch’ when pressed against the sprocket’s inner diameter. Unlike e.g. roller chain sprockets or spur gears, no undercut (cuts below the inner diameter) is provided for rollers or a mating gear’s teeth, and no special geometry is needed along the sides of the teeth.

Some Terminology:

Pitch: The center-to-center distance between sprocket holes, and thus the sprocket teeth.

Tooth Face: The tapered portion of the tooth. In this application, the tooth taper is calculated to smoothly slide into the sprocket holes as the sprocket rotates.

Tooth Flank: The ‘upright’ (or slightly concave) base of the tooth. In this application, it should be the same height as (or slightly taller than) the tape thickness so that the tape sprockets rest fully within the flank.

Tooth Land: This is the surface left if the tip of the tooth has been blunted or “cut off”. This might be done to fit the sprocket into a particular diameter. I’m an EE; I don’t know what other dis/advantages pointy vs. blunted teeth would have in this application.

Basic usage:

Fill in all the values called for in ‘Basic Parameters’. Aside from angles, which are in degrees, use any unit of measurement you prefer (inch/mm/etc.), as long as it is consistent; output will be in the same units. If you desire a specific tooth taper angle, enter it, otherwise just press “Compute / auto angle” to suggest an angle and generate the sprocket.

Mostly, the pitch and sprocket hole width are dictated by the tape to be fed, and also drive the important diameters. You can get closer to a desired sprocket diameter by adjusting the number of teeth. The important diameters are:

Inner diameter: This is the diameter at the base of the teeth, where the bottom of the tape rests.

“Design diameter”: This is the most important diameter as far as the program is concerned, and is fully dictated by the pitch and number of teeth. The design diameter is the diameter at the top of the tooth flanks, which is the top of the tape. You could also think of this as the outside diameter of the tape if wrapped around the sprocket.

Outer diameter: This is the diameter at the tips of the teeth. By playing with the tooth angle and cutting off the tips (tooth length %), there is some leeway to constrain the outer diameter to fit the available space.

Note that the angle auto-suggest feature is currently broken (will return incorrect results). It will (usually) calculate an angle that will allow the tape to *wrap around* the sprocket at any radius from the base of the teeth, but what you really want is the tape to fit at an arbitrary angle across the teeth (specifically, the outer edges of whatever teeth it intersects while tangent to the sprocket should not exceed the outsides of the sprocket holes). For now you might have to cut a few gears and experiment, or just set the angle arbitrarily high.

Extra Options:
If you will be cutting out the sprocket on a CNC mill, outside pocketing will leave some material at the base of each tooth flank due to the diameter of the round cutter. Enabling ‘Remove cutter leftovers’ and entering the cutter diameter will add DXF points (drill hits) near the tooth edges to remove this material. Users of other fabrication methods can probably ignore this option.

If designing a sprocket in one measurement system for use in another, you can optionally select a unit conversion to be applied when writing out the DXF file. E.g. if your tape is specced in mm but your CAD/CAM software expects inches, select ‘mm to inches’ before saving the DXF.

CAVEATS:
I wrote this to solve a very specific need for one of my own projects; so very little time and debugging went into it. There is no idiot-checking. Expect errors or bizarre output if you leave necessary fields blank, mix & match units (inch/mm) arbitrarily, enter a negative number of teeth or any other physically impossible geometry. Even if you do everything correctly, there is no guarantee the output will be correct or meet your needs. Please check the results very carefully before you lay out any $$$ to have anything professionally made by a fabrication service!

Right now the arc between teeth is output as a straight line, not an arc or series of tiny lines approximating one. This should not be a huge problem for a reasonable number of teeth, but something to be aware of.

Other notes:

“Auto angle” calculation is only done if the angle field is blank: if there is a number there (including a previous auto-calculation), it will be left alone. If you have changed any parameters and want to redo “auto angle”, please delete the contents of this box.

The sprocket image shown in the program window is not to scale – it is automatically scaled to fit inside the window. It is not unusual for the sprocket to appear to change size dramatically when parameters are modified.

Square Pegs and Round Holes:
Unless you have some fancy software sweeping the sprocket teeth into 3D, you are probably making a flat gear out of flat stock, and it will have flat edges. If the sprocket holes are round, the tooth edges will contact somewhere earlier than the outside diameter of the hole, and so may need to be tweaked – especially if the material is thick relative to the holes. (See the diagram below for an exaggerated example.) Use this formula to calculate the effective tooth width that will exactly fit the hole:

w = sqrt(d^2 – t^2)

where d is the sprocket hole diameter and t is the stock thickness.

cexxy version

My page that tells you your IP address is up and running again, after a PHP configuration change by my web host knocked it out. Anyway, enjoy the glory of finding out your external IP address without getting socked by porn popups!

cexxy version

An aquarium air pump can be used as an inexpensive source of low vacuum with a small amount of tweaking. Supplies needed are:

The air pump
Screwdriver (usually) to open the air pump
Hose barb (your favorite size) for vacuum inlet
Drill
Glue (e.g. RTV/caulk, epoxy, etc.)

Of course, you could convert one by sealing up the whole thing in a big Tupperware container and punching a port through, but this method is more robust and compact.

Have a gander at the pictures below. The internals shown are pretty typical, and diabolically simple: AC wall power flows through a U-shaped electromagnet, which wiggles a small permanent magnet between the poles rapidly back and forth to pump a rubber bellows. The bellows draws air directly from the inside of the case and forces it through the output port, drawing new air into the case through some small holes or dust filter on the case somewhere. Thus, “reversing” the pump requires simply drilling your own hose barb into the case and sealing up the original vent (plus any other air leakage paths). The converted pump can be used as a vacuum pump by plugging the new port into your vacuum-needing device and letting the original port vent to atmosphere, and can still function normally as a positive-pressure pump as needed.

Common air leakage paths are around the AC cord entry, around the output port and where the screws / rubber feet go (the screws may be hiding under the rubber feet anyway). Probably the easiest thing to do is just run a nice fat bead of RTV around the entire seam between the halves of the case before putting it back together.

The pump shown pulls about 5 inches mercury (~127mmHg); most are probably in that ballpark. If your needs fall somewhere above this but well below a “real” vacuum pump (or even a disembodied fridge compressor), it might be possible to beef up the vacuum or flowrate a bit by putting 2 in series or parallel.

Some “standards”:
In the US at least (don’t know about elsewhere), the common size pumps (for 10 ~ other double digits gallon fish tanks) generally take 3/16″ flex tubing, and unmarked tubing in the fish supplies section of your local pet store is probably this size. Larger pumps with e.g. 1/4″ ports are available for large tanks, but if it doesn’t say what size tubing to use, you can probably assume 3/16. This refers to the tubing inner diameter (or the pump port’s outer diameter); the tubing outer diameter can vary significantly and is often not specced. Since you will be adding your own port, you can really make it any size you want, but sticking with 3/16 means you will have plentiful local sources of matching hose barbs, tee fittings and other parts at most any pet store.

cexxy version

This is a quick follow up to the pick & place head article, in which I actually build the darn thing :-) As in, not just fit-test the parts together and take a picture, but actually pick and place some stuff with it. I’ve been busy/lazy, so not too much to show in terms of software yet (the video you see below is running a ‘dumb’ g-code placement script). The parts list has been updated will be updated in the next couple days, along with new CamBam/DXF files.

Live test of the head as currently designed. This is bolted to my ghetto homebrew CNC mill, which is kinda slow. On a more built-for-purpose (or less ghetto) machine, performance should be much better.

The webcam’s weird eyeball-shaped case has been removed, revealing a rectangular board with sanely-spaced mounting holes in its corners. Also, a proper solenoid valve has been bolted to the head and hooked up to a reversed aquarium pump for suction. The rubber nozzles are from a sacrificial el-cheapo SMD vacuum pen ($3USD ,eBay), which comes with three sizes that fit on a standard 16-gauge needle.

Otherwise, it’s pretty much as described in the previous post.

A little bit of guts. This board holds a transistor to drive the 12VDC solenoid valve from a parallel port pin, Pololu microstepping motor driver (Allegro A4983) and 5V regulator. At the top you can make out a simple pressure sensor conditioning circuit that isn’t hooked up yet.

cexxy version

For my pick and place project, I picked up a pair of too-good-to-be-true webcams: the Cubeternet no-name UVC webcam. For this project, there is a lot to like: 2MP resolution (claimed, at least), built-in LED ring, cross-platform UVC interface, hand-adjustable focus and a legitimate glass (no polycarbonate) lens…for $16! Alas, my review of this cam is currently mixed, since one of the cameras failed after being plugged in for more than a few minutes. This particular camera – the first I tested – became warm to the touch soon after plugging in; I assumed this was normal operation and that the cam’s solid metal “eyeball” enclosure was the heatsink for a voltage regulator screwed into it. Turns out this is not normal at all; the 2nd camera does not get even slightly warm after running overnight. Now, what to do with a broken webcam? Take it apart!

Teardown photos: In here

Opening the case reveals solid components, but an unfortunately typical “Chinese toy” construction with hand-bent and soldered leads everywhere, a couple stray solder balls and liberal application of hot glue (yes, really) to hold everything in place. If you’ve ever taken apart a cheap electronic toy for soundbending, you probably know what I’m talking about. Of the identifiable ICs, there are:

Controller:
Vimicro
0342PL
TS0340B
(Integrated USB2.0 UVC camera controller in 44-pin TQFP; its manufacturer denotes it as VC0342. This is driven by a 12MHz crystal oscillator.)

EEPROM:
Turbo IC, Inc.
TU24C64CSF
29830628
(64-Kbit I2C EEPROM in SOIC-8. Contains USB descriptor strings referencing “Vimicro Corp. Venus USB2.0 Camera” and “Sirius USB2.0 Camera (Audio)”. The remainder of the data most likely consists of imager-specific register initialization values. Here is a dump of the EEPROM contents in ASCII HEX format, or in raw format.)

Voltage regulator:
Kingbor 6206A
0947/33
(Ho-hum, 3.3V 3-terminal regulator.)

Imager:
QM20MMS
4TS28CPT30
Typical “big glass plate” CMOS image sensor; this is the partnumber silkscreened on the bottom of it, but the Google turns up very little information and certainly no datasheets. An user on a Chinese message board says it is a 2MP imager made by Micron.

Misc:
There are six very bright LEDs hand-soldered into the board and bent into position; an electret microphone is also glued into the case. A handful of what appear to be discrete transistors/FETs deliver power to the LEDs and may serve a purpose switching/sequencing power to the imager.

In the images of the controller side, you can see a big solder blob dangling precariously off one of the FETs onto the PCB. While it’s not clear if this one was the culprit, this blob or one of a couple similar ones are the most likely cause of failure. Despite all this, the lens assembly is all glass as claimed, and seems to be of much higher quality relative to the rest of the guts. The minimum focus distance is well below 1 inch. On another bright side (no pun intended), the LEDs are bright as hell, adjustable via an analog thumbwheel on the USB cable, and holes in the four corners of the square board can allow easy attachment to the placement head. The untimely death of one of the cameras is certainly discouraging, and given the internals can’t be cleanly written off as a fluke. Still, even assuming a 50% failure rate, doubling up on these cams is still a good bit cheaper than the nearest name-brand equivalent.

cexxy version

Quick ‘n dirty (but working!) image segmenter for randomly-strewn part identification. About 1 page worth of scripting takes an image of objects on background, determines which part is the background, determines the outside contour of each object and numbers each as a separate object. Now that it’s known where to look for one specific object, the task of identifying that object (or just matching it to another just like it) becomes a whole lot simpler. Combined with the auto-aligner, this reduces a “naive” (bruteforce cross-correlation between needle and haystack images) image matcher to only having to scan against 4 orientations (90-degree rotations) to find which has Pin 1 in the right place (and whether it’s the same part, etc.) Hopefully as I dig deeper into opencv, there is a less-naive algorithm builtin for this that does not rely on contrast/color historgrams: most electronic parts basically consist of a flat black body and shiny reflective metal leads (i.e. appearing the same color as your light source and/or the background, and/or whatever happens to be nearby at the moment). Edge-based stuff still seems like a better approach, though I would welcome being proven wrong if it means not having to write the identifier from scratch myself :-)

Steps in brief:
The first image was taken using the actual webcam that will be attached to the pick n place head, looking at a handful of representative parts on a piece of white paper. This image was dumbly processed using a Sobel edge-detector (it’s builtin to Gimp and I was feeling lazy), Gaussian blur to expand the soon-to-be-resulting mask around the part a little and close any gaps in the edge-detection result, and finally threshold the result to produce the second image. The goal in these steps is to produce a closed-form contour blob for each part that’s at least as wide as the part, while minimizing stray blobs from random noise / dirt specs / etc. (internal, fully-enclosed blobs/noise due to part features/markings is OK). Finally, opencv’s FindContours function is run (mode=CV_RETR_EXTERNAL) on the resulting image, returning a vector that contains a polygonal approximation of each external contour found. Each discrete (non-touching) contour blob is returned separately, that is, every part in the frame is now effectively tagged and numbered!

There are a couple noise points identified in the image above. Better-chosen constants for the initial image operations (threshold, blur radius, …) may help, but I’ll probably end up having it measure the area of the blobs and throw away any that’s too small to possibly contain a valid part. Switching to a more advanced edge-detector, e.g. Canny, may help too. In any case, the full image matcher should figure it out eventually :-)

Code Demo – basically ripped straight from the pyopencv examples
Segmentation example – requires Python (2.6) and opencv 2.1.0 / pyopencv.

cexxy version

This weekend I got some parts in and put together a preliminary placement head for my open-source pick ‘n place project. My requirements are that it be buildable with off-the-shelf parts (ideally same-source, to save on shipping) and no special equipment, allow +/-180 degree rotation while maintaining an undisturbed vacuum, and support interchanging of the “tools” (vacuum needles). All that’s really needed to build this are the discrete parts shown, a bit of drillable plastic (e.g. Delrin) for the base material, and a drill. A drill press would be handy (a CNC mill *really* handy, and not such an out-there thing to have considering you are probably retrofitting this onto one).

This head consists of a hollow rotary shaft with a Luer lock fitting on one end, right-angle flexible tubing barb on the other end, and a large toothbelt (notched belt, timing belt) gear in between. The shaft is held in place but allowed to rotate by a pair of bearings, and the rotation is provided by a small stepper motor at the other end of the toothbelt. The gear ratio is approximately 5.2:1, providing a rotational resolution of about 0.35 degrees/step with a common 200 step/rev stepper motor (if no microstepping is used). Finally, just to the left of it is a 1024×768 Webcam with manually adjustable focus and a ring of built-in LEDs for lighting. The webcam mounting is definitely not ideal, given the camera’s weird eyeball-like shape. Tentative plan is to lash it down with some string, align it nicely with respect to the CNC table, then backfill the opening the camera’s butt sits in with epoxy.

The hardest part was finding a combination of parts that would all fit together nicely. Currently, the fits are mostly exact to “pretty damn good”, but a bit of adhesive is needed to join them permanently.

Parts List
Unless otherwise noted, all of these parts were sourced from Small Parts Inc. in the US due to the large selection and an actually competent parametric search engine, which was a great help in finding combinations of mutually-fittable parts. Accordingly, measures are in Imperial unless noted otherwise (that’s just how they come here).

Misc. Parts
2x 2″ x 4.5″, 1/4″ thick pieces of Delrin or similar
4x 1/2″, #4-40 bolts and nuts (assuming 2 bolts for stepper motor)
4x 1/2″, #8-32 bolts for standoffs
1x Webcam, manual focus, hi-res and builtin lighting strongly recommended
1x 3mm shaft stepper motor*
1x Method of attaching to your mill – a bit of aluminum angle bracket, or a dovetail, etc., depending on your mill.
1x 3-way air solenoid valve
1x Vacuum source (see here for a cheap one)
Air tubing and appropriate couplings to your solenoid valve. The head’s air path terminates in a 3/32″ hose barb, so you’ll want a 3/32″ to (whatever) barb adapter, or a 3/32 that screws directly into your solenoid. Those using metric are on your own :-) but will probably have an easier time of it anyway.

* The timing pulleys and stepper motors only come in a handful of diameters (imperial for the pulleys and usually metric for the small motors), so a 3mm (.118″) motor onto a .125″ ID pulley was the closest I could come up with in a reasonable amount of effort. Any small, el-cheapo permanent magnet (“tin can”) stepper motor should work here, but sourcing it may be annoying. Off-the-shelf 3mm-shaft motors I found are Jameco’s ValuePro 42BY48H08, Anaheim Automation’s TSM42 series, and Portescap’s 42/44 series e.g. 42×048 and 42S100. This surplus stepper is also worth a look, but you’ll have to remove a pre-attached plastic(?) gearhead. Beware, many of these smaller motors are 7.5deg/step, so even with the gear reduction, you will probably want to look at microstepping them to ensure adequate rotational resolution. Also, most of the companies selling them have no online click-and-buy ordering; you’ll have to phone up a salesdouche at the least, hope that you are worth their time to buy One Lousy Motor, and possibly haggle (“Request a Quote”). How companies that don’t know what their product costs stay in business is beyond me, but that’s a rant for another day.

If you want to skip that hassle, I’ve taken an alternative approach and simply bodged a surplus 2mm-shaft motor up to a 3mm shaft by gluing a short piece of thicker steel tubing onto the existing shaft (see partlist).

Design Files
CamBam drawings with machining operations for the top and bottom plates. The machining ops assume 1/4″ thick plastic, 1/8″ endmill for most cuts, and .166″ (#19) drill for the #8-32 bolt holes into the standoffs. It’s designed for the Cubeternet webcam (or equivalent eyeball-cam) and a stepper motor with 42mm center-to-center mounting holes. You can get the free CamBam at http://www.cambam.co.uk/.

Note, I made some small improvements to these files after the above prototype was carved, so what you see in the file will not match it exactly. In particular, the standoffs were moved to more optimal places and a feature has been added allowing the motor to be slid to remove any slack in the belt.

Assembly
The assembly should be pretty self-explanatory. One bearing on each of the Delrin plates (inset the ball bearing if you can; otherwise gluing it down should be fine). The hollow shaft goes thru the bearings, Luer adapter and hose barb go on either end (use adhesive, just don’t clog the air path with it). The motor shaft center should sit just a hair (50-100 mils?) over 1.75″ from the hollow shaft. On mine, an online calculator produced the 1.75 figure, but the belt turned out to be a bit loose once built this way. If possible, make an elongated hole for one of the motor screws so the motor position can be adjusted to tension the belt. You could probably also insert a peg somewhere in the belt path to push it inward and take up the slack. With the camera focus set such that the largest part you will ever populate can fit in-frame, find the resulting camera height (distance from part) and set the depth of the shaft so that the camera is “in focus” maybe an inch or two above the placement position (i.e. with the needle touching the board). This will allow the head to focus on parts without touching the needle down.

For my build, I used instant glue to attach the Luer adapter to the needle shaft, and hot-melt glue to tack down the large pulley and hose barb. None of these parts should be seeing significant force; if they are, you’re Doing It Wrong and the hot glue should hopefully break loose before something more important does. Using a non-permanent adhesive for the hose barb and pulley also allows this assembly to be disassembled later if needed. The shaft itself is free-floating and the large pulley rests against the ball bearing due to gravity. This will prevent damage due to crashing the needle into the table or too-tall part, but if you experience problems with the shaft riding up on its own, try adding a bit of extra weight or put a dab of adhesive where the shaft passes through the lower bearing.

Suggested “v2″ improvements
–Move air valve (if/when one is specified) onto head to minimize air volume between valve and head. Needed? (may be beneficial for reversed aquarium pumps or other weak vacuum sources)
–Bump detect: rather than firmly adhere the shaft into the bearing, allow it to float up and down, normally resting by gravity with the large pulley against the ball bearing. Place a contact switch just above the pulley: if the head/part contacts the surface with more than minimal force (enough to lift the shaft), contact switch is triggered. This could be used to halt the machine if a bump was not expected. If the switch’s trigger position is reliable enough, it could be used intentionally to automatically determine component heights.
–Probe function: There is a conductive metal path from the needlepoint all the way up to and including the bearing outer race, so it would be easy to touch a contact here and use the needle to probe for any conductive objects (e.g. find the tabletop if it is metal, or some capacitive shenanigans for PCBs/etc.). Useful?

cexxy version

So, there’s some really cool, empowering stuff going down these days with regard to manufacturing. Cartesian machines (i.e. CNC mills) are relatively simple to build from off-the-shelf parts; there are a bajillion people doing this and plenty of ready-made open-source designs available. More recently, hobbyists have gotten in on designing open-source rapid prototypers (3D printers); as a result, designs have now crossed the sub-$1000 threshold off-the-shelf, and you can even build a GPL’ed 3D printer that can almost replicate itself!

One thing that I haven’t seen cross the blood-brain barrier of proprietary commercial systems is pick-and-place machines that can assemble electronics. These things are badass; full of automated win and articulating robotic arms, but they’re also damned expensive: the crap ones start at >$10k and use literally a fishing-lure-and-weight type arrangement to peel back the tape covering tape-and-reel parts, so you have to keep resetting the weights. Those with more advanced / less manual feeders scale skyward from there. And of course, the software end of these things, especially machine vision algorithms to place parts more accurately, is some serious $ecret $auce. So… let’s change this!

Most of the “big stuff” is straightforward: The PCB layout software generates a list of coordinates for each part. A small vacuum needle mounted on a Cartesian head picks up each part from a known location, rotates it 90/180/etc. degrees as needed, and sets it down at its coordinates. It does not even need to be 100% accurate: surface tension of the solder during reflow will pull most minorly misaligned parts back into place.

The big barriers are:

1) Low cost / self-manufacturable feed mechanisms:
Electronic parts are packaged in several different ways, most commonly tape-and-reel, plastic tubes, or in trays. Each has a different, maybe cumbersome, way of knowing the location of the next part in the package and freeing the part from the package. Picking up stuff and putting it down is easy compared to dealing with the wide variety of tape and tube sizes reliably. Oh, and if your board uses 50 unique parts, you need 50 feeders. Hence the emphasis on making them cheap and mass-self-produceable, e.g. by CNC or casting or 3D printing.

and/or…

2) Machine Vision
For larger parts, once the first part is successfully picked (e.g. by human intervention), it is enough to know how many parts per inch of tape, advance the tape a known amount per pick, then just grab blindly for the part and plant it at its destination coordinates. But for smaller and finer parts, this is not accurate enough: the parts can be slightly off-center or crooked in their tape wells, and this becomes significant as the part size decreases. Professional machines use a set of cameras and image processing algorithms to recognize the part, find its dead center and correct any rotational error. In theory, a suitably good vision system would allow you to peel back the tape and just sprinkle the parts on the table, forgoing feed mechanisms entirely at the expense of some small manual labor. Actually programming this algorithm on the other hand…

Another nice thing to have would be:

3) Automatic needle swapping. Many more advanced CNC mills are able to spit out their current tool, e.g. a specific size drill bit (in a known location in a tool rack) and pick up a different tool. It would be nice for the pick and place machine to be able to change to smaller and larger needles/suction cups to handle large and small parts seamlessly. If not, placements can be sorted, e.g. smallest to largest, so that the needle only has to be manually changed a couple times.

I’ve made some very initial feasibility-study stabs at building such a machine, and begun building a bit of hardware. I created a separate page for this project with more detailed specs/documentation and progress so far:
Pick and Place Project

cexxy version

It’s like third grade all over again, except that I could theoretically have ice cream for breakfast if I wanted to.

cexxy version

UPDATE: For Veo Stingray drivers, try these first. In case they disappear… Stingray 300V (Win98/ME/2000/XP) and Stingray 323V (Win2k/XP only). The one sold by AllElectronics uses the 323V driver.

I picked up an old “Veo Stingray” camera from surplus dealer AllElectronics. These things are pretty junk by modern standards (320×240 resolution, unsightly rounded “looking through a tube” image), but it does have the variable (manual) focus I needed for an imaging project, and the price was right. As for drivers… the company that makes this thing seems to have evaporated, and the particular variant (USB product ID 808B) seems to have never been heard of by anyone, even though there are some identical cameras with slightly different internal hardware (and different Product ID) floating around. They all are (were?) manufactured by Xirlink.

This particular variant can be identifed by USB VID 0545 (Xirlink), Product ID 808B.

Here is a solution that might work (but see update above first): This camera and several IBM PC Cameras use the same or similar chipsets (Sunplus SPCA5xx)…with a small tweak, the IBM camera driver can also be tricked into supporting the ‘808B’ Stingray (maybe others?) by adding its VID/PID to the driver .INF file. Kinda like slipping your cuckoo egg into another nest.

This file includes the IBM driver and tweaked .INF. The following devices are supported:

%XIRLINK.DeviceDesc%=Xirlink.Device,USB\VID_0545&PID_8002
%XIRLINK.DeviceDesc%=Xirlink.Device,USB\VID_0545&PID_8080
%XIRLINK.DeviceDesc%=Xirlink.Device,USB\VID_0545&PID_8009
%XIRLINK.DeviceDesc%=Xirlink.Device,USB\VID_0545&PID_808B

To install, follow the README.TXT included.

Win2K/XP Driver for Veo Stingray (808B) and IBM V5000

Linux?
It looks like several open-source driver projects may support the Veo Stingray, IBM PC Cameras and similar SPCA50xx variants (not to mention the classic Dakota Digitals). Note, if your exact ID is not listed as supported, you may be able to get it working with a tweak similar to the above. Have a look at:

http://xirlinkwebcam.sourceforge.net/
http://www.linux-usb.org/ibmcam/ (older project)
In addition, the package ‘gspca’ may work for these and many other SPCAxxx-based (and other!) webcams. See here for the gspca/spca5xx project and a list of supported cameras.

cexxy version

So there we are, minding our own business, when an angry demon springs to life in the basement. He is on a rampage, pounding against the walls with all his might, THUMP THUMP THUMP THUMP BANG BANG BANG. We race downstairs. Turns out it is only the Sears Kenmore front-loading washing machine’s internals beating themselves to death during the final spin.

After ruling out an unbalanced load (even fully empty, BOOM BOOM BOOM), I lookup the symptoms online. After a few thousand pages of angry ex-customers’ rants, without even lifting a wrench, there is no doubt what the failure is.

All available evidence confirms that this machine is DESIGNED TO FAIL EVERY FIVE YEARS*. The fuse is an innocent-looking metal bracket on the back of the spinning clothes basket known as a spider; its purpose is to support the basket on the drive shaft leading back to the motor. This spider, an uncoated pot-metal part made of a brittle Al+Fe alloy, is fully exposed to and tumbles its way through all the detergent, bleach, dirty laundry water, etc. After a few short years of this, it corrodes through and shatters during the spin cycle! Of course, you can’t buy just this part; you have to plunk down for the entire basket/etc. assembly at $200 a pop**. Guess who now has a nice spare stainless steel basket laying around to turn compost in, heh.

My brief description does not do this problem justice, but there is an excellent video from a former Kenmore owner detailing this spider issue and several other, likely intentional, design flaws.

So, you probably guessed what I did next – took it apart. The insides of the plastic wash tub are a bitch to get at (nondestructively); the whole thing has to be carefully unhooked from a pair of shocks and large springs, liberated from half a dozen hard-to-reach hoses and tubes with fiddly retaining clips, and the glued-on rubber seal cut away from the front of the machine. There are also cinderblock weights (yes, really!) on the tub which ought to be removed unless Hulk Hogan is helping you lift it out. On mine, the bearing surfaces were also so rusted, it took beating the living hell out of the shaft with a big hammer to separate the assembly from the tub.

Here is what I found inside:

One rotted, broken-ass spider. Surprised? Also note all the "used to be spider" white crud packed in beneath the bearing.

One rusted-out shaft for a rotted, broken ass-spider.

One reaction vessel washtub for a rotted, broken-ass spider. Yuck, my clothes were in there? The white crap looking like crushed concrete at the bottom is corroded bits of spider that haven’t yet ground away at the pump impeller while making their way to the city’s sewer. You can also see where the grease has begun leaking out from the cheap rubber lip seal that’s supposed to be keeping water out of the bearings.

One rusted-out hole for a rusted-out driveshaft to grind slide into. "That’s not my mating surface, baby!"

</p>

All this because Sears/Kenmore*** could have easily solved this with $1 worth of engine paint (etc.) to coat this fiddly metal bracket, but chose not to. Also, like in the video above, my outer tub has a nice deep gouge line where the screws in the wildly-flailing basket tore it up. Luckily we were home when it went, and our catlike reflexes caught it before they could carve all the way through.

*says everyone on the Internets. Don’t McLibel me.

** and if you want to save the hassle of retaining clips patwanging across your basement and the Hulk sneaking looks up your skirt, and have the new part installed by an authorized repairman… there’s a reason they have those commercials with the repairman playing Solitaire all day because no support calls come in… it really is cheaper to buy a new machine. ($75 diagnostic visit + $450-$700 repair quotes vs. $500-600 for current units.)

***technically, Sears/Kenmore just buy the appliances from other OEMs and put their names on them. This particular model (417.42142100) is made by Electrolux; others are rebadged Frigidaire or Whirlpool units. Unfortunately, all seem to share a similar design, right down to the fail timer.

cexxy version

We got a kitty last weekend. Like many new additions of the human kind, we hadn’t really planned on it*, but now we are in love with her. Her name is Spirit (she comes pre-named).

She had no trouble settling into the new house, except for spooking out at the noise from the radiators at first. She’s such a little attention whore – follows the hyoominz all around, and if we are on the couch, sits down next to someone and headbutts them ’til she gets belleh rubs.

So far she has also discovered that Christmas tree ornaments swing and make a tinkling noise when swatted at, knitting needles move and click while in use, and must therefore be attacked, and – OMG WIIMOTE CURSOR!

*Somewhere between “rescue mission” and “interstate feline trafficking” does the truth lie.

cexxy version

An update, 1/28/2011: It turns out there was a legitimate problem on the site after all! Specifically, hacked by spammers at some point and filled with invisible links & keywords. Skip down to the comments for the details, and be aware that the rant that follows is based on a fairly complete lack of information! (Said information being over a year in coming, but I have to say that hearing it from the man in charge himself pretty solidly exceeds my expectations….and makes me hang my head in shame a little for even thinking of defecting to Bing.)

So, as of sometime within the last few weeks or so, I am a convicted Google Spammer! As you may guess from this tone, I have not actually done or considered any such thing. But as tends to happen when you take all humans out of the loop and put all faith in the smartness of an infallible Algorithm, you throw your fair share of babies out with the bathwater, and today, one of those babies is me. I discovered this (not like I would have gotten an email or anything) when I happened to Google the name of my product to get to its mainpage (much faster than typing the full URL, which is longish), and it wasn’t there. With some further digging, I found my entire domain and all subdomains have been blacklisted from Google’s index.

That particular level of digging leads to a tool called Google Webmaster Tools, which, after proving myself the owner of the site, returns this:

This site may be in violation of Google’s quality guidelines. More Details
Pages on your site may not appear in Google search results pages due to violations of the Google webmaster guidelines. Please review our webmaster guidelines and modify your site so that it meets those guidelines. Once your site meets our guidelines, you can request reconsideration and we’ll evaluate your site.
Submit a reconsideration request

Full stop.

These webmaster guidelines cover “quality” (i.e. don’t spam the index)–including such helpful first-grade spam no-nos as not stuffing pages with invisible keywords, bogus META tags, providing special fake pages to known search bots, and other stuff that might have worked on Lycos in 1996–as well as various style-guide suggestions, admonishing webmasters to be on their best grammar, and even going as far as discussing size and placement of images on the page. Is a webmaster really expected to perform feng shui to stay in Google’s good graces?

Anyway, I’m at quite a loss to explain why I would be banned from Google, as getting an entire high-ranking site removed from a search index seems like something that would require some pretty big-ticket shenanigans. Of course, this is The Algorithm we’re talking about; there seems to be no indication that a live human was involved in this decision*.

Of course, any of the usual SEO tricks would fit the bill. But I don’t engage in any of that (for here, the boards or the main site), and the only “optimization” I do to this blog’s traffic is to post something interesting once in a while. (Really, since everything on this server is ad-less and free anyway, the only thing More Traffic can get me is a bigger bandwidth bill.) I do know that Google will display warnings / block content if it detects a site has been compromised, but a thorough dig through the server-side files indicates this is not the case, either.

So what’s left are some straws to grasp at:

• Legitimate incoming links from adult-ish sites and adult-ish search queries for my side project
• “Bad Words” or links appearing in message board posts (forget this piddly blog, what the rest of the free world knows cexx.org for is one of the oldest continuously running spyware help forums in existence. Posts here–especially diagnostic logfiles posted by affected users– contain references to bad programs/sites, links to same, and repetitive content (e.g. Windows registry trees) that occasionally generate false positives. A human would easily see that we are helping people rather than spamming the world; The Algorithm may not do as well.)
• Occasional comment spams slipping through in UGC (“User-generated content”, the current buzzword for “stuff the site owners didn’t write themselves”, such as blog comments, message board posts, every video on Youtube, etc.) As you can see from the numbers showing at the bottom of this page, comment spam is as fundamental to the Internet as the threat of rickroll, and the filter’s doing pretty well at blocking them. If transient v1@gra comment spams were grounds for being delisted, half the internet wouldn’t show up. Then again, for those of us who have not used Lycos et al lately, how would we google-addicts know if half the internet wasn’t showing up in our searches…?)
• My content being scraped and appearing on third-party spamblogs (yes, it happens. I–of all people!– have fired off a couple DMCA takedown demands in the last couple years, but really, for splogs on splog-friendly Korean ISPs this has about as much effect as firing off complaints for every email spam your receive. These automated scrapers usually end up scraping from someone with deeper pockets and much better arm-twisting power at some point, and the problem (for lil old me) solves itself.)
• Old pages/posts, dead links, occasional bad grammer or speling mistakes?
• Maybe Google are still mad at me for exposing a huge bug in their search some years ago? (In theory, this would make it trivial for someone to determine whether they ranked higher than a competitor, or see how a specific tweak to their keywords/etc. affected their ranking. But since I’m pretty sure no humans were actually involved in this, I kinda have to rule this hypothesis out…)
• Statistically anomalous distribution in keyword content of sites that link to mine?
• Statistically anomalous distribution of topics I discuss, tag clouds, etc. (or as mentioned earlier, help forum posts)
• Someone I’ve pissed off in the past robo-submitting my URL to the automated “report a spammer” page?**
• Googlebombs or other shenanigans performed (maliciously or not) by third-party sites?
• Wild Conspiracy Theories (paid off by a malware company? Malware authors have been trying to block their victims from being able to reach help forums such as cexx.org’s for years; maybe the’ve found a way to up the ante. Or maybe Sergey bought my Trance Vibrator and didn’t like it.)

The possibility that any site could be delisted by the actions of third-party sites (e.g. competitors) is simply disturbing. As unlikely as I’d hope it to be, Google’s complete secrecy regarding its delisting criteria (even I, after proving myself the legal owner, can’t get boo about what’s going on with my own site) makes such a scenario impossible to rule out. For what it’s worth, Google does explicitly mention links to “bad neighborhoods” in this Guidelines page, and some sites by and for the SEO people (who presumably know their stuff, this being their entire business model) seem to think this does apply to incoming links as well.
That is unacceptable.

If I haven’t gotten to the bottom of it soon, my only choice might just be to block Google from the site outright (why pay for the bandwidth their crawler uses if we are being excluded from the results?) and personally wean myself from Google search, for whatever that is worth as a personal stance. Is my best alternative really “Bing, and it’s done“?? Google, you really put me between a rock and a hard place.

* nor, based on analysis of the server logs, the “Reconsideration request” you can submit via the Webmaster tools thingy. Unless there are human reviewers lurking in an underground bunker somewhere disguised as residential cable customers from Peoria, or a vast distributed network of speed-readers who are each assigned one line to read, the speed of the hit-streams identifiable as coming from Google during the time of said review easily beat my personal best, and, much like the MTV Music Awards, show no evidence of human intelligence.

** From said page: “If you believe that another site is abusing Google’s quality guidelines, please report that site at https://www.google.com/webmasters/tools/spamreport. Google prefers developing scalable and automated solutions to problems, so we attempt to minimize hand-to-hand spam fighting. The spam reports we receive are used to create scalable algorithms that recognize and block future spam attempts.” Great, The Algorithm is now in charge of deciding who the humans on the Web are.

cexxy version

I was reading the Thinkgeek catalog a while back and an interesting gadget caught my eye: “Wi-Spy”, a 2.4GHz spectrum analyzer on a USB stick. Coming from the world where “spectrum analyzer” refers to a big benchtop box with a name like Le Croy_{right, we can’t afford those} Agilent on it, my first thought was, “A good spectrum analyzer in the GHz range is like a gazillion bucks; what’s really inside this thing?”

Not much, it turns out. After seeing the guts I would not use this device for anything remotely demanding accuracy, but on the bright side, it would be pretty straightforward to make your own that works just as well! Possibly with parts that are already on your board (*cough*XBee).

According to this review and a couple other tech sites, the innards of the original are nothing more than a Cypress WirelessUSB SoC (CYWUSB6934) and USB interface. According to this updated review, the latest version switches to TI/Chipcon’s CC2500 radio ($2.15 USD in quantity) and retails for $399.

The “WirelessUSB” and CC2500 radios are inexpensive 2.4GHz radio chips used in devices such as cordless mice, game controllers and remote controls. What it has in common with many other digital radios is an RSSI (Received Signal Strength Indication) register, which can be read out to provide a measure of how much signal (i.e. interference) is present on any given channel. Wireless gadgets normally use such a mechanism to find the cleanest channel to broadcast on. Most any radio chip with an RSSI readout should work here: XBee, TI/Chipcon’s CC25xx parts, Nordic, etc. I have a big nut going right now for Nordic’s nRF24L01, but unfortunately this does not have an RSSI measurement to speak of (there is a 1-bit “collision detect” flag that warns if the RF energy on the current channel is above a set threshold).

So essentially, the firmware running on the dongle consists of:

for (channel=0; channel<255; channel++)
{
WriteReg(RF_CHANNEL, channel);
signal=ReadReg(RSSI);
send_usb(signal);
}

Hell, why didn’t I think of selling $2 of chip and 4 lines of code for $400? ;-) (I’ve even used those old Cypress radios before on a work project – Unigen’s ready-made JUNO module shown here is about 7 bucks.)

The DIY-Spy

As proof of concept (and a cool toy for anyone who has one of these lying around), I have implemented a working Wi-Fi spectrum analyzer on TI’s ez430-RF2500 development kit ($50), a 2-part USB dongle which consists essentially of a CC2500 radio strapped to an MSP430 low-power microcontroller (detachable bottom half) and a USB interface which enumerates as a virtual serial port (top half). The top half doubles as a standalone MSP430 programmer, so this kit is a great cheap way to get started playing with them.

Being the lazy efficient person that I am, I grabbed mspgcc and an existing ez430 RF demo program (JM Kikori’s RF2500 test application – raw transceiver) and just tossed the infinite RSSI loop into the middle of it.

Here is the main addition (simple & sweet). Forgive Wordpress’ complete mangling of the formatting.


static void dump_rssi()
{
int i;
char rssi;
for(i=0; i<256; i++)
{
hal_spi_radio_writeReg( MRFI_CC2500_SPI_REG_CHANNR, i); // Channel number. Default spacing is ~200KHz/channel
hal_spi_radio_cmdStrobe( 0x34 ); // enter Rx mode (not needed except to start autocal)
rssi = (hal_spi_radio_readReg(MRFI_CC2500_SPI_REG_RSSI));
uart0_polled_putc((rssi&0xFE) | (i==0)); // Cheap speed hack: write upper 7 bits of RSSI value (throw away LSB). Use LSB to signal start of 256-channel RSSI byte list
hal_spi_radio_cmdStrobe( 0x36 ); // enter IDLE mode (not needed except for autocal)
}
}


The only real divergence of this from the pseudocode loop above is that the CC2500 requires periodic (or frequent, depending who you ask) recalibration, which can be set to be performed automatically on certain conditions (such as entering Rx mode). This setting is enabled, and the code above toggles between Rx and Idle modes between each channel change to trigger auto-calibration. To make things faster (for me as well as the gadget ;-), rather than perform any kind of real serial handshaking to keep the dongle and the graphing script in sync, I just truncate the least significant bit of the RSSI reading (1/2dBm) and use it to flag the start of the 256-channel sweep.

A more efficient approach would be to perform such calibration once per channel and cache the results (writing them back along with each channel change), which would in theory allow for much faster channel changes. But it’s really a moot point since here the 9600-baud connection to the virtual COM port (MSP430 <–> USB controller) is the limiter of maximum speed.

Most of the time in this project went into tweaking around with the CC2500’s fifty billion or so configuration register bits, not all of which are really documented. In the end I just installed their win32 configuration wizard for the undocumented voodoo (e.g. correct IF setting) and annoying math, and hand-tweaked the remaining settings. For this I set the base frequency to 2400MHz, and the channel spacing to ~ 405KHz, providing the frequency range 2400MHz to 2505MHz across the 256 channel settings. This corresponds to the 2.4GHz ISM band and a little extra. All automatic gain control features are disabled.

Testing
I’ve actually been sitting on this project now for a couple months, since I wasn’t getting the results I expected initially (sitting close to my WiFi access point) and had no good way to test it. Luckily, as it turns out I was tasked with developing some RF gadgets (nRF24L01-based) at work, which could easily be set up to output a raw carrier wave at a programmable frequency. Sure enough, testing revealed some bugs in my initial CC2500 register settings, now corrected.

The video above shows the resulting spectrum output as the nRF’s naked carrier is swept through a handful of frequencies. As it progresses, it should also become evident where my WiFi access point is broadcasting. The green lines show instantaneous signal, the red shows the peak value at each frequency, and the blue shows a running average. The display script is written in Processing.

Downloads

Fairly undocumented; use at your own risk.
diyspy.zip – Source code and .elf binary for ez430 kit (use mspgcc to build from source), .inf file for virtual COM port (win32 only), and Processing script for display.

cexxy version

Blinking RGB LED throwies adorn the “Steaming Globe” sculpture near the Kendall/MIT station just after the 4th of July fireworks show. No idea how this could have happened.

Kids had a blast pulling them off and throwing them back onto the globe (and each other!). The whole rest of the night as we walked around, we saw blinking kids.

cexxy version

Heh… I’ve just completed NIH certification to administer human test protocols. I feel like that’s got to be worth a few mad scientist points.

So, I’m working on this project at work involving research into new forms of computer interface peripherals. (No, not that one.) As part of the project, we are required to test these interface concepts with live humans, survey them and see how well it works for them. The survey is positively harmless: basically like handing someone a shiny new cellphone set to vibrate, and programmed to vibrate in a different pattern depending who’s calling. Then run each pattern and see if the user can use it to successfully identify the caller. To make things simple, we had proposed to do these tests with our own employees (everyone pretty much said, “hell yeah, I’d like to play with that” when I described the project, and it sure as hell beats pulling in random strangers from the Craigslists to nose around in our secret lab).

Oh, for the catch: it’s a DARPA-funded project, and this required usability survey stuff constitutes Human Testing. To do any Human Testing, you need to draft a Human Test Protocol detailing exactly what you hope to get out of the research, intricate details of the exact test(s) that will be performed (down to the specific questions you will ask, and in what order) and the exact hardware involved in each test (note, this hardware hasn’t yet been designed by the time these forms/tests must be finalized…also remember that the entire point of these tests is to guide the design of said hardware, which now must be set in stone long before the test). So much for iterative development, that’s for sure. Add to that lengthy Informed Consent forms, confidentiality procedures, SOPs, HIPAA forms, risk assessments relative to daily activities (like, say, using your cell phone…), Conflict of Interest forms, Anti-Coercion Forms, and the whole shebang must be submitted to, and approved by, an IRB (Institutional Review Board), then the IRB’s approval approved by the governmental Office of Something Something Something. The package we submitted went up to Appendix V. (Not V as in roman numeral 5, V as in A thru… . I kid you not.). We each also had to take this lengthy “Protecting Human Research Participants” online course. Not that I’m not completely for protecting my human research participants, but come on – we’re asking them simple gadget usability questions, not giving them experimental drugs.

We got feedback from the IRB today, our protocol is NOT yet approved. They need more forms submitted and a bunch more specific details carved in stone.

I shouldn’t be complaining. Poor Jeff, my officemate, is handling the vast majority of this process in concert with a PhD consultant who is actually used to dealing with this crap. It just burns having to scramble around slapping together super-crude prototypes to test on-the-sneak to get even some rough semblance of how to structure the tests, not to mention the hardware, to get any kind of meaningful results (screwing the schedules on my other projects), how much of the project’s fixed budget will be swallowed up by this approval process instead of, say, actual R&D, and just how pointless the tests will be when it’s all said and done. Our money shot is 3 days’ worth of usability testing at the end of the program, predetermined at the start of the program.

Oh yeah, and letting your own staff participate instead of random losers from off the street? BIG MISTAKE. Far from simplifying the random Craigslist idiot, asking strangers medical screening questions, losers breaking our equipment, NDA, and losers selling our tech secrets to the Commies cans of worms, it actually hugely complicates things. Now you need to prove to the IRB, and the governmental IRB approval approvers, that employees’ participation in the research is indeed voluntary (not, “hey, participate in my research study or ur fired”), and all the steps you have taken to ensure that all participation is voluntary, that no one has the capability to bribe or coerce anyone into participating, that supervisors/bosses cannot determine who participated, etc.

Some notes, in case I EVER have to do this again:

Do Not promise, or get involved in a project where they will make you promise, human testing (no matter how harmless it seems) for any government project.

Do Not promise human testing -of anything- on a Phase I SBIR budget.

Do not write, review, modify or proofread human test protocols after playing Portal. Now the whole thing is in GLaDOS Voice. Likewise for the Big Red (supercolliding super-) Button that stops the test. (”The effects prolonged exposure to the e-stop are not a part of this test.”) I’m incredibly tempted to slip something into the protocol promising cake after the study, but then I’d probably have to get FDA approval too :-(

HIPAA stands for Highly Increased Paperwork and Annoyance Act.

(And yes, technically I *could* construct a test sequence in such a way as to make the prototype play music. But I haven’t gotten around to it yet choose not to. Yeah, that’s it.)

cexxy version