Hand Hygiene

We repeated the experiment with the spider antenna at power level 7 on channel 26.  This time we tested the vertical and horizontal positions.  The boxplot of the horizontal position is not bad, but uninspiring:

In the vertical position, though, it rocks!

The data on the left side is the cleanest that we’ve seen, showing little angular sensitivity: nice nearly horizontal lines for each distance and height out to a distance of .75m (by the way, in the lab we realized that a practical division between .5 and .75 m is probably more useful than .75m versus 1.0m.  Although it’s possible to reach at 1m, it’s not very comfortable).

Let’s break it down a bit more.  Ignoring the difference between height and angle, the data relative to depth looks like this:

There’s still some overlap, so how do the actual distributions overlap?  Here’s a set of fits, assuming the data at each distance are normally distributed:

In the end, it’s going to come down to choosing a threshold.  For that the empirical cumulative density function  at each distance may help.

This graph shows that if we set a threshold of -10 on the RSSI, we’d count about 90% of the .25 m readings, 70% of the .50m readings, and about 5% of the .75 and 1.0m values.  Not quite the 95% accuracy that we want, but we’re certainly moving in the right direction!

Here’s the results of the small 1/4-wave antenna we ordered from digikey (ANT-2.4-CW-RH-SMA-ND).  We went with a power level 7 on channel 26.  The antenna is about 1″ long (probably the dipole is coiled inside that area).  For this test, it was pointed straight downward.  The antenna looks like this:

The boxplot suggests that this won’t work for us. 

The results look pretty good at the top height (.1m down from the transmitter), but are all over the place for the lower heights except for the last one.  This pattern is consistent with a broadcast pattern that follows a narrow vertical angle of perhaps 30 degrees perpendicular to the main axis of the antenna.

That’s a shame, because it would have made a great solution for us, if it had worked.

Here’s results of a test of the on-board antenna with transmit power of 3 and channel 26.  The test is with the transmitter straight up.  The boxplot below shows angles, heights and distance. 

It looks like there is some separation between the .25m distance and the others.  Again, the power level seems low.  A boxplot of the LQI’s for this data.

This graph provides the justification for the LQI for clean data at 100.  The clean data makes a better picture than the original, but not good enough, I think.

Maybe more power would help.  This graph gets pretty unfocused on the right side.

Here’s the first test from our homemade spider antenna, which is supposed to be relatively angle independent.  It looks like this (scale in cm):

The signal is on the central dipole.  The four other blue protuberances are grounded. 

We placed the motes at the two different vertical heights (.1m and .35m) at 4 distance from the vertical axis: .25, .50, .75 and 1.0.  We used a power level of 3 on channel 26.  A boxplot of results is here:

Ideally, the data would show 4 steps, one for each distance.  In particular, we’d like to see a clear difference between .75 and 1.0m.  That would mean that the angle and height of the mote was not very important compared to the distance.  Unfortunately, this was not the case.

At the .25m distance, the extreme angles (near 0 and 180), the RSSI was very low, lower than some responses at the 1.0m distance, not a good sign.

Part of the problem may be that we used a power level of 3.  At 1 m, the data seems much more sporadic (particularly before cleaning out the RSSI values collected with poor LQI values).  However, there is little evidence that this configuration is worth pursuing.

For completion, I’ll through in the complete ANOVA table.  Not surprisingly, with 38,000 data points, everything is significant. 

General Linear Model: CleanRSSI versus Distance, Height, Angle

S = 1.12628   R-Sq = 96.72%   R-Sq(adj) = 96.71%

The new antenna apparatus is up and working.  Here’s how it looks:

The 8 motes can rotate about their long axis and about their USB.  All 8 are connected through USB to the computer.  Scripts and programs on the computer allow software to be installed and messages (including the RSSI and LQI) to be read from each of the motes simultaneously.

Data results coming soon.

The most recent design revision in the world of pager/badge circuit enclosures is a wearable name badge. It was designed to be similar in size to common pocket-clipped badges (3.25×1.5×0.6″); hold all components (ipod mini battery, charging board, mote); allow access to power switch, LED’s, USB; clip easily to a shirt pocket, belt, or pants pocket; and have a removable ‘lid’, allowing easy access to internals.

Fig 1 Enclosure	Fig 2 Lid
Fig 3 Prototyped enclosure and lid	Fig 5 Component layout
Fig 4 Closeup of lip

Figures 1 and 2 illustrate the drafted design. The lid features a tension clip that clips into the slot on the left side of the enclosure in Fig 1. Notice the ‘lip’ around the inside edge of the enclosure in Fig 1. This is where the lid rests. The boards (mote and battery charging circuit) have been arranged with the battery in an ‘over/under’ fashion as illustrated in Fig 5. Note that without mounting hardware, the battery charging board and the mote would be displaced from each other by about a quarter inch. The connectors on the underside of the charging board clear the mote, connecting to the mote USB, mote I/O, and the battery. Their wires would run through the dead space between the boards and the inside surfaces of the enclosure. The boards mount with screws and grommets to the three holes visible in Fig 3. The fit would (theoretically) be tight enough to provide a friction fit for the battery.

There are a few problems with this design:

• It’s still pretty bulky in size. At more than a half an inch thick, it would be a pretty hefty nametag to wear in normal nametag fashion clipped to a shirt pocket (the rectangular size is decently in nametag range). The limiting factor in the thickness of this design are the connectors and USB receptacle being on either side of the charging board and the clearance required between the mote and charging board (to protect components and electrical traces).
• It’s pretty bulk in weight distribution. The package weighs approximately 40 grams. This may not seem significant, but when it is attached by clip on one side of your shirt pocket, the wearer will be stuck with a sagging shirt/pocket. Pretty uncomfortable.
• The prototype produced by the IATL extrusion printer does not have enough resolution to reliably produce the model’s features. While the technicians said they could go down to about 0.015″ (or about 0.02″ reliably), features such as the lid ‘lip’ on the enclosure (0.03″ faces) were not resolute at all (fuzzily indicated in Fig 4). Tolerances between the lid and the enclosure were not consistent. Also, while the lid is designed to be 0.07″ thick, it has very little resilience and stiffness and would not attach to the enclosure, even if it did fit.
  • The technician at the IATL (Daniel Langstraatd) said that they have another printer with much finer resolution that could probably produce the prototype. However, this material must be coated in a glue to seal and strength it, adding an arbitrary amount of thickness to each edge.

After speaking with Geb and Matias we have decided to forgo this design and concentrate on something that is perhaps thinner but wider and could easily fit in a shirt or side pocket (a la I Phone form factor). Top-down profile size is not as large of an issue if we can make the thing easily to slip or tuck away. Also considered is a two-part ‘hinge’ design: it basically functions as a nearly pocket-sized clip, with one side on the outside of the pocket containing the mote, and the other side hanging over into the pocket, containing the battery a/o charging board. This would be easy to attach to a pocket (although not much else) and would likely provide a comfortable weight distribution. It is unlikely that the rapid prototyping centers would be able to produce reliable, resolute parts on this scale, suggesting that bodies of this thinness would be produced from aluminum. Matias spent about five minutes trying to convince me that if we could polish it shiny enough, we could make somebody put a brick in their pocket.

For now, concentration will go to finalizing the second-gen pager board design.

Ted forwarded a nice link from Hack A Day about populating SMT PCB boards.

Charging board rect. footprint: 1.65×2.6
Daughter board: 0.57×0.8
Daughter board elevation: 0.115
USB elevation: 0.271
Ipod mini battery: 1.6×1.25
Battery elevation: 0.175
Mote: 2.6×1.27
Mote elevation: 0.36
E-shop board thickness: 0.06
Mote thickness: 0.04 (0.2 w/ components)
Battery thickness: 0.175
Sherlock 2-pin: 0.3×0.4
Sherlock 3-pin: 0.38×0.4

(All dimensions in inches. Elevation measure from top of charging board to top of other board)

I was researching the type of plastic that the “hockey pucks” use and it is acrylonitrile butadiene styrene (ABS). It has a melting temperature of 221 * F and is relatively close to acrylic plastic, which has a melting temperature of 212*F. During this process I was also researching how to best polish the plastic. One way included sanding down to 1000 grit sandpaper and then adding a wax on the surface. Another idea was flame polishing. This was first tried and burned the plastic and I plan on making another trial. Another idea was using a chemical compound from a manufacturer for ABS plastic. I will then apply each type of polishing to different parts of the “Hockey Puck” and cast a mold. to see the differences.

Although we’re still collecting data, it is beginning to look like the RSSI signal is messy for reasons that are going to be tricky to clean up. We have a number of ideas that we can pursue:

1. Track the badge-dispenser signal for longer periods and a faster rate, hope that we get enough enough experimental power to reliably distinguish the mean signal strength at .75 and 1.0 m.
2. Change the antenna design within the dispenser unit to provide a more robust signal within the range of interest.
3. Use a pointing antenna mounted above the dispenser to create a zone of detectability right around the dispenser.
4. Add another transducer, like a sonar sensor to provide alternative distance estimate.

Here are the strength and weaknesses of each approach.
(more…)