Hand Hygiene

This test includes two sets of data. The first set is the test performed with the antenna pointing down in the customary fashion. The second test is performed with the antenna pointing upwards. This second test was performed because, as you can most likely see from the plots, during the first test the top mote closest to the broadcaster had abnormally low RSSI. This is most likely because the mote is above the ground plane and little to no signal can reach it. The offsets for these tests were with the low motes at 0.045m below the broadcaster and the high motes at 0.240m above the broadcaster. Because some ended up below and some ended up above the broadcaster we face an issue because the 7MM antenna has a ground plane that, at closer distances, will have very little broadcasting strength above.

This plot is a boxplot of the data with the mote pointing down in the normal way:

CDF of the data with the mote pointing down in the normal way:

CDF by height of the data with the mote pointing down in the normal way:

From this last plot it appears that the 0.25 and 0.5 distances are swapped for the low and high motes. This is what makes the CDF plot of all the data so inconclusive and causes the closest high mote to appear far away. This is the main reason why we ran the second test. And the data

Boxplot of the data with the mote pointing up:

CDF of the data with the mote pointing up:

If we can somehow clean up the closest mote values the 0.5m CDF is almost a completely vertical line. This is awesome because it provides a clean cut RSSI value that says ‘this mote is x meters away’ and can help discern between far and close motes. This test again had a tanking of the closest high motes RSSI values. We are going to look into this because the ground plane should not be causing this issue. The expected results of the top row would be when like the bottom row of the tests conducted using heights for badge motes because the distances are similar.

This first boxplot is a boxplot of another test with the 7MM Antenna with the same orientation as before with the heights 0.1m and 0.35m. When I came into run this test I noticed the test setup had the distances 0.2m and 0.45m and this is what appears to be the heights the last test was run at. Now this may not have an effect on the data, but this could be what made Mike’s test look so much different. Because of this I just ran two tests, one with the heights we want (0.1 and 0.35) and one with the heights the setup was left at (0.2 and 0.45). The following plots are of this data.

Boxplot of data with heights 0.1 and 0.35:

CDF of data with heights 0.1 and 0.35:

These data and plots appear to show that the subtle height differences between the tests appear to have little effect on the results. We are still getting good separation from 0.25m from all other distances.  The following plots are of the altered height test setup

Boxplot of data with heights 0.2 and 0.45:

CDF of data with heights 0.2 and 0.45:

With these altered heights we are still getting excellent separation from 0.25m to 1m, but the differences between 0.25m and 0.5m or 0.75m have gotten smaller and there is a little more overlap.

These two tests verify the results of previous 7MM Antenna tests such as 7MM1 and 7MM3 / 7MMG2

The original test of the 7mm antenna which seemed so promising was performed at a power level of 20. Our recent attempt to verify the results was performed at a power level of 5. In order to find out where we stand with the 7mm antenna, two additional tests were performed. The first was a replication of the power level 20 test. Second, a test was performed with an auxiliary ground plane beneath the antenna to see if there is a significant effect on performance.

The results of the first test:

And the results for the same test with an auxiliary ground plane attached:

Although the ground plane data seems a little cleaner, both sets of data appear to indicate that this antenna is capable of performing at the level we require. This would suggest that going foreword the 7mm antenna is the favorite for inclusion in the next generation of design. Also, we recognize that a higher transmit power provides cleaner data.

The first test with this antenna, 7MM Test One, showed that we could maybe get some separation between distances and that the antenna was not angular dependent – except in some extreme cases. We decided to run the test again to verify these results but repeatedly came up with something different. The data from this second test show a clear dependency on angle and allow for no real separation between distances.

My first thoughts may be that the original test was just a fluke and it doesn’t represent the real data set. This has happened before, but never to this degree. Because the difference in spread is so much we are questioning whether this mote is the right choice. Here is a boxplot of the RSSI vs. Distance Height Angle that provides a summary of the data and allows you to see the angular dependency:

From this boxplot you can see that the RSSI appears to dip around 90 degrees at almost every distance and height. This is bad because if the antenna is angular dependent then if a person approaches from the wrong angle they might not get picked up, but a person passing by and not using the hand cleaner would.

The following plot is a CDF of this data:

From this plot it is appearant that there is not a good cutoff value to seperate a total distance of even 1m. That means that you cannot choose an RSSI cutoff such that you could get almost all the 0.25m values and none of the 1m values. There is just too much overlap

My second guess for why this is happening is that the broadcasting mote may have a different orientation than the first test. This is a possibility because I didn’t run the first test and couldn’t find any decent documentation on how it was oriented then. Luckily the person who ran the test is available, but we still need better documentation, or just better consistency at documentation.

Unfortunately there’s been an onslaught of spam (50 messages in the last 48 hours) and I’ve had to change the permissions so that you need to be registered to comment.  Hopefully that will take care of it

I spoke with Shane at Howard Electronics about what we might like to set up an electronics prototyping setup.  He said that the stencil and oven approach will probably be more trouble than we want.  Instead he recommended direct soldering.

Specifically, he recommended the JBC Tools Dual Soldering Station DI3000TP.  He said the tweezer grips are excellent for placing the small resistors.  For the chips, he recommended using a drag solder technique.  The system is $995.

He also said that a video monitor will make it much easier to get reliable results.  He recommended the HEI-VM-LCD system — an 8″ display would probably be fine.  That $2400.  We’d need a stand, so perhaps the HEI-VM-BS for $2740.

The whole assembly kit would be about $3700.

I’ve been thinking about the possibility of making motes ourselves.  We have the open source circuit and board design for the telos.  We could get the boards prototyped from one of the board building houses listed here, for example, such as Precision Technologies, which is geographically close (though I’m not sure that makes much difference).  I suspect that the ultimate cost of these would be less than $5 per board.

We can get the chips from Newark or Digikey.  I did a quick check and this would require a day or two of research to get exactly the right equivalences, but I came up with a total cost of $35 for the parts, not including all the resistors and capacitors, which are typically a couple cents each.

Mike and Tim were looking into modifying the Gerber files directly, which would allow us to do relatively simple things, like changing the layout of the USB to put on a USB-mini.  To do bigger modifications, like incorporating the USB charging circuit and adding a connector for the iPod batteries, we should probably redraw the circuit in a software package like Eagle.  The freeware version of Eagle that we currently have only supports 2-layer boards and the mote is a 4-layer board.  We could go with the non-profit standard use for $125.

We’ll also print out a gerber file to make a 3-mil or 4-mil mylar stencil, which we can do either at the Engineering Prototyping shop on their laser printer, or order one for $25 from Pololu.

We can place the components on the board with tweezers under a microscope, then reflow the solder in our skillet or a $500 reflow oven.

The full simulation of the Axial Mode Helical Antenna showed some interesting behavior. Antenna was set up as though mounted directly above a dispenser.

Orientation of antenna for this test.

There were several positions where signal was blocked.

The lowest, closest position showed very poor reception. My initial reaction is that this is due to the mass of liquid mounted beneath the antenna. This is a potential problem with the helical “directional from above” concept. To correct this, the antenna would have to be mounted higher, further out from the “wall,” and angled back toward the dispenser. This mounting creates its own set of problems by introducing additional hardware, but is interesting in it’s potential for quality results.

You also see that at the top, the two furthest out motes were cut off fairly effectively. In the ECDF you can see that the response was good with the exception of the lower, nearest mote.

Exhibits fairly good separation between ranges.

Good separation at the target 0.50/0.75 range, however the performance at 0.25 negates the usability of this particular setup.

I plan to test a few additional configurations of the directional antenna to see if we can more closely replicate the proof of concept results.

We’re getting a little behind in documentation for all the experiments we’ve been running, and it is time to figure out which direction we want to pursue. 

The best results so far were obtained with the directional antenna, the 7mm antenna, then the spider antenna.  The first and 3rd were home made.  The second would be the easiest to incorporate into the current system.

Here’s a summary of the different tests we’ve conducted.

Onboard F-antenna:  This is our most studied antenna.  Here’s what we documented:

Face to Face at 0.5, 1.0 and 2m, power 10, 15 and 20.  Conclusion – Things are going to be complicated.
Rotating motes at 1.5m, power 20.  Conclusion – Orientation matters.
Pairwise compatibility (redo) (best)- facing, 2m, power 20.  Comparison of signal strength of seven different motes. Conclusion – the motes are effectively interchangeable. The best test used 14 motes, 24″ apart and measured a between standard deviation of 3.7 and within of 0.64 RSSI value.
Mote orientation – horizontal motes, 2m apart, 30-degree increments, power 10 and 20.  Conclusions – Large variation (30-60, 25-50 RSSI) in signal strength as a function of orientation.
Mote Power level – vertical orientation, 2 rotations, 2 heights, 0.5, 1.0 and 2.0m apart, power levels 1-31.  Conclusion – the power level can be tweaked to separate various regions of the RSSI, such as providing greater separation between distances or orientations.
Obstacle Interferance (inside puck) – one horizontal, one vertical, distance .55m. various obstacles:  person, wire, roomba (like a laptop).  Conclusion — these obstacles didn’t have much impact on the signal strength (< 1 or 2), except for the roomba, when put close to the mote.  In the inside versus outside puck conditions, we find that the plastic of the puck makes a small, but insignificant difference in the signal.
Second orientation test – one mote horizontal, the other vertical, .55m separation, vertical mote rotated in 30 degree increments about 360 degrees.  Conclusion – in this configuration, the broadcast pattern is fairly reliable and smooth across a variety of power levels. 
Angle of approach test – A worn mote versus the Purell.  The mote is worn in front at distances of .2 and .5 m and behind at a distance of .2.  Conclusion:  When behind at extreme angles, it was difficult to separate the signals.
Full angle-height test (repeat with vertical transmitter) – Puck and badge orientation.  The first test with full statistics at 2 heights, 4 distances and 10 angles between 0 and 180.  The results indicate that it will be very difficult, if not impossible, to determine mote distance from the puck or vertical mote with a single RSSI reading at power level 3.  Perhaps .25m at the high level can be distinguished from others, but that is it.

Spider Antenna, Horizontal
Full angle-height test – Fully crossed 2 heights, 4 distances, 10 angles.  Power level 3.  Conclusion: Some separation of .25 high straight ahead from others, but other levels indistinguishable.  Probably power level too low.

Spider Antenna, Vertical (repeat at power level 10, repeat at levels 10 and 20, multiple powers)- Fully crossed 2 heights, 4 distances, 10 angles.  Power level 7. Conclusion:  With one threshold we could get about 90% of the .25 m readings, 70% of the .50m readings, and about 5% of the .75 and 1.0m values. 

By changing the power levels, you can get separation among different distances, but these are different between the top and lower levels.

At power level 10, the cutoff between close and far is less pronounced.  At power level 20 we get wonderful separation between .25 and .5 m, probably at the 95% confidence level, but almost no separation between .5 and .75.

Spider Antenna, 45 Degrees. 
We did this test to see if we could angle the “sweet spot” of the antenna through the center of the target area.  We did 2 heights, 4 distances and 10 angles.  Conclusion:  Significant angular dependence (not surprising).  If you look only at the regions near the front of the transmitter, there is good separation between the .25, .5-.75, and 1 m ranges, but the middle two ranges were not well distinguished. 

Quarter Wave Antenna, vertical
Vertical small antenna through SMA connector.  Fully crossed 2 heights, 4 distances, 10 angles.  Power level 7.  Conclusion.  Works well at the high heights, clear distinction.  However, lower motes couldn’t get the signal.  Probably the signal pattern cuts them out.

Quarter Wave Right Angle Antenna, vertical
Right angle, larger, 1/4-wave antenna pointing downward.  Standard heights and angles, power 10.  Exhibits surprisingly high directional sensitivity in the vertical position where we expected that it would be symmetrical.

7MM Antenna
This antenna is soldered directly to the board.  Power level 20.  Test through 2 heights, 4 distances, 10 angles.  Conclusion — Amazingly smooth progression of data.  The CDFs of the results suggest that a 95% confidence interval would be hard to find, but the results are so “sensible and well behaved” that this really seems like a promising antenna to pursue.

Purell Devices – Motes in standard orientation and rotation about Ted’s and our prototype Purells.  Ted’s prototype has high signals for several orientations at the .25 level and lower values elsewhere, but there is not much consistency.  With the newer prototype with the F-antenna there is more consistency at .25 and .5m, but there are still bad orientations.

Directional Antenna
Homemade helix, about 6″ long and 3″ in diameter.  Tested with 2 heights, 4 distances and 10 angles.  Conclusion:  Crystal clear distinction among the different distances of .25, .5 and .75.

The concept of using a directional antenna to eliminate broadcast reception in selected regions is nothing new. This quick test shows that by using a directional antenna to create a cone of reception beneath a sanitizer one could create an RSSI drop off far more dramatic than the drop off due to standard wave propagation seen in an omnidirectional antenna. An additional bonus of the axial mode helical selection is that it’s circular polarization should produce very little variance with change in position relative to the broadcaster. This will need to be verified experimentally.

Note separation of 0.50 and 0.75 meters.

Again, distinct separation between ranges becomes possible with a directional antenna.