Here’s a visualization of the test with .2 and .5 m with the mote worn in front. The circles are in the position relative to the Purell, which was mounted on the left edge, center of the graph The radius of the circle is proportional to signal strength. Width of the edge represents 2 standard deviations. Data from the Purell experiment below.
We expected to have the inner arc be large circles, indicating strong signal strength close to the dispenser and the outer arc be small circles representing smaller signal strength farther from the dispenser. That’s not clearly the case, although there is a trend. Note however, that not all experiments have turned out this way. Some experiments have shown differences that are more clear.
There was some concern initially that the variation of the signals would be the problem. That does not seem to be the case as the measured standard deviations are rather small compared to the size of the circles. However, our experimentation in the lab suggests that when the motes are held in a fixed position the signal strength is constant, but moving them around over an angle of about +- 30 degrees or putting objects between them easily changes the RSSI by about +-10 points.
Perhaps we should be thinking of a single sampling position as a biased estimator and think about moving the motes around a bit at each experimental position in order to get an unbiased estimator with a larger standard deviation.
I met with Ted and Phil for several hours yesterday to organize and focus this project a bit more, given that the financial resources have settled down a bit. The primary conclusion was that we need to shine a laser-like focus on the acquisition of a clear signal in the hand sanitizer application. We’re also going to get a copy of the software that they are using in the hospital to use in the lab, so we can validate that our mockup tests are consistent with the output of the software. A major difference between our tests and the software is that the software only takes one measurement every 7 seconds, whereas most of our test software takes a measurement 4 times a seconds. However, the hospital software has many critical features that we aren’t using for the hardware development, like synchronizing clocks for periodic messaging and power-saving capabilities.
We decided that the target for the hardware should be the absolute determination of who activated the hand sanitizer. The shoulder-grip length of a 95% male is 715 mm. That’s the distance from their shoulder to the middle of their hand with the arm extended directly. The 95% male forward grip reach is 835 mm. That’s the distance from the back of the shoulder to the grip point with the arm fully extended. Since the back of the shoulder is approximately equal to the position of the back of the hip, that’s a reasonable maximum distance for using the hand sanitizer, although it is not completely precise, because someone could conceivably stoop forward and extend his or her hand out to reach the sanitizer, but that would be a very awkward posture.
The minimum distance would be something similar to the forearm to fingertip length. For the 5% female, this distance is 400 mm. Subtracting distance for the front of the hip and the length of the hand might make this about 360 mm.
Thus the hand hygiene sanitizer must be able to determine whether a person is between 360mm and 835 mm of the unit when it is activated. Since it is unlike someone will be closer to the unit than the person using it, we could reasonably say that it must provide a detectable signal to a distance of 835 mm from the unit. After further discussion, we realized that this is an unrealistic constraint, since it doesn’t really specify two conditions that the mote must distinguish between. Since the target is 835 mm, let us say that we’ll try to distinguish between .75 and 1.0m.
Assume that the device is mounted on a wall and the person will be within 90 degrees directly in front of the device.
Of course the device may not be 100% accurate. For the sake of argument, let’s assume that it is accurate 95% of the time.
The device must maintain this reliability whether or not the person is wearing the pager on the front of their belt or the back.
There is probably some minimum length of time that we must assume that the person stands in front of the device to wash their hands, but I don’t know this.
There is a vertical offset between the pager mote and the mote in the dispenser. For the hand pump, the recommended height of the mount is 1.1 m above the floor. The hip height for a 5% woman is 740 mm and for a 95% male it is 1000mm. Therefore we’ll assume that the pager motes should be 0.1 to 0.35 m below the bottom of the hand pump dispenser.
We should add constraints as we figure them out.
I’ve just been doing some preliminary testing with having one receiver and multiple transmitters and noticed that the standard deviation of RSSI values gets quite high with this setup. When the number of transmitters is reduced to just one the standard deviation is reduced by about a factor of 5. This is probably because of the multiple transmitters are interfering with each other and causing some sort of problem. This is alright for now though because the real plan was to have multiple receivers and a single transmitter.
Picking up where Gray left off, we are trying to find an effective, easy way to fit all of the electronics (mote and battery charging circuit) into the supplied UIHC pagers. Gray’s design connected the boards with a Hirose 8-pin IDC connector. The connector is mounted to the battery charging board, and the leads of the ribbon wire are directly soldered to the mote at the other end.
Both boards in the pager enclosure
You can see the connector in this picture, and where the battery rests below it. In this design, there is almost zero clearance between the inside of the pager and the battery, and zero clearance between the inside of the pager and the ribbon cable (in fact, the crimp terminal of the IDC connector was partially disassembled and replaced with superglue in order to fit the cable in the enclosure).
The two main issues in this design is the minimal amount of lateral clearance between the inside of the enclosure and the battery/connector; and the connector surgery required to make the ribbon cable fit inside the enclosure (which is a direct cause of the first issue). Thus the goal is to provide more clearance where the battery sits, and the most direct way to achieve this is to replace the connector.
To find a particular connector for this application, we apply the following criteria:
The size constraints lead us to connectors of the miniature or subminiature type, and the height constraint suggests we will be using a connector with a right angle header. In order to get something into our hands, Geb ordered a few Molex connector kits. The connectors of interest in this kit are branded Microlatch and Sherlock. These connectors seemed right-on in terms of our selection criteria, but there are still other options, and they will be considered later in this post.
The main difference between the Sherlock and Microlatch connectors is that Microlatch using a latching friction fit while Sherlock uses a pretty slick lever latch. The vertical Microlatch connectors are slightly (~2mm) shorter than their Sherlock counterparts; Microlatch is vertical orientation only.
Finally, it’s determined that for the pager board, the Sherlocks are the best fit of what is available. In order to be systematic and space-effecient, the board will use (2) 2-pin connectors (for the battery and USB I/O), and (1) 3-pin connector (for the mote terminals). Using these connectors should satisfy the space requirements for the pager board and allow the battery charging board to sit in the bottom of the enclosure, eliminating fit problems that Gray had experienced.
In the next revision of the battery charging board, I have replaced the connectors and enlarged the peg holes that provide a fit into the bottom of the pager enclosure.
Other Connectors
The Sherlock connectors will work quite well for the pager as it is, but it is useful to consider other options and future options for connectors. Both the Microlatch and Sherlock connectors use a 2mm pin pitch, which (at least looking at the Molex catalog) is just a bit oddball. The I/O 10-pin pinout on the mote, for example, uses 0.1″ (2.54mm) pin pitch, which seems to be far more common. This is an important consideration. It isn’t terribly relevant now because we are only using 3 pins on the mote interface, but if we plan on using more in the future, it will be far easier and more reliable to use a mounted header. Which means a different connector system. We could find a wire-based crimp connector system that will simply crimp in to other types or pitches of connectors; however, it may be easier in the long run to standardize and use 0.1″ pitch connectors in all of our designs in the future. I will later update this post with relevant model numbers of connectors that are in this pitch and meet our size requirements for other boards. For now, we will continue to use wire-to-board connectors on the battery charging board, and directly solder the other end of the wires into the mote (using a header on the mote board is impossible due to geometry).
Other Considerations
There current design of the pager board hides all of the circuit LEDs (4 on the mote, one on the battery charging circuit). This problem may be circumvented by redirecting the light using fiber optic cable, acrylic light pipe, or simple reflectors. It may be more efficient to solve this problem in the next design revision (which will be outlined in an upcoming post), but I will first work on solutions using reflectors (light pipe and fiber optics can become quickly very expensive and complicated). I don’t know how important they are to the rest of the folks for diagnostics. Also to-do is to find a nice solution to cover what used to be the face plate of the pager. In Ted’s boards, it is covered by a rectangle of laminate cut out to accommodate the USB male mate, this seems to work nice, but there might be an easy, prettier solution.
The results of this test showed that while some pucks appeared to have better RSSI values with plastic off, just about the same showed better results with the plastic on. The results of this test are ambigous and point towards the fact that it doesn’t matter whether the plastic is there or not.
The RSSI AVG ON means the plastic was on, and the RSSI AVG OFF means the plastic was off.
This test was performed using the original Purell mote as the reciever (the mote that was stationary) and a badge mote. The Purell mote was oriented vertically how is meant to be mounted inside and the badge was attached to me at about the waistline so there was a 0.3 meter height difference between the badge and the dispenser. The tests were completed with a resolution of 30 degrees and a range of -90 to 90 degrees. This range included 90 degrees to the left of dispenser (-90) to 90 degrees to the right of dispenser (90) with 0 degrees offset right in the middle. The first test was done from a distance of 0.5 meters with the badge on my front side and the data and plots can be found here:
Data, RSSI Plot, Standard Deviation Plot
From these plots you can see that the STDEV generally is larger at a lower power setting and the RSSI appears to be distributed normally. The only low point on the RSSI with this orientation is at 90 degrees to the left of the dispenser, and even this isn’t that much of a deviation from the rest of the values.
The next test was done with the mote at a distance of 0.2 meters from the Purell dispenser with the badge again on my front side, the data and plots can be found here:
Data, RSSI Plot, Standard Deviation Plot
With this test the RSSI value again appears to drop when approaching from the left, but as in the last test not by very much. Other than that the RSSI plot is pretty even from all directions with little difference elsewhere. The standard deviation plot though shows a wild jump at power level 15 that could be from bad reads, but overall the trend is that at lower power levels there is greater variance in the RSSI values seen.
The final test was done with the mote at a distance of 0.2 meters, but this time it was stapped to my backside to simulate the inteference of a person. The data and plots can be found here:
Data, RSSI Plot, Standard Deviation Plot
The plots again show a drop when approaching from the left, but this time at about a 30 degree offset. There is also a STDEV spike at this offset for all power levels except 5.
These final plots and data sheets are for the overall results of the test when at a power level of 15:
This test explored receiving and sending power variations between and within motes for one specific configuration. The purpose was to determine if, when a particular power or RSSI level was chosen in practice, that level was reliable for other motes and how much variation there might be for a given arrangement. One mote was used as a reference. Fourteen other motes, including 4 in badges, 5 in pucks, and 5 unencased, were used as test subjects.
One mote was fixed in place and each of other motes were placed, one at a time, at the same fixed distance and orientation from the first mote. Both motes were at the same height (46″ above ground) and 24″ between from each other. Both motes were placed with the USB facing upwards and the circuit boards facing each other (battery packs facing away).
The OnePowerBlinkRadio program was loaded on both motes, with the reference mote being identified as mote #1 and the other as mote #2. This program caused each mote to broadcast a signal at power level 5 on channel 11. Each mote also listens for broadcasts and when it hears a transmission, it report the number of the transmitting mote and the RSSI value of the reception. This report broadcast is picked up by the serialListen program and logged to a file. Each mote was tested against the fixed mote for one trial, with approximately 120 communications each way.
The results suggest that motes receiving identical signals report similar RSSI values (42.87, 40.06, 45, 46.48, 39.71, 39.80, 36.2, 36.47, 34.05, 41.05, 44.96, 41.61, 43.07, 40.93) with an average response of 40.9 and a standard deviation of 3.7. Within a trial, the motes were remarkably consistent with a standard deviation of just 0.64.
Reports from the broadcasts heard by the reference mote were similar. These were pooled with an average value of 39.43 and a standard deviation of 4.36.
Reports from the motes in the pucks were notably different from the others. It appears that most could receive signals normally, but many of their broadcasts were not heard by the receiver. However, one had its batteries replaced just before the experiment and its results were normal. We’re double checking those values.
[from an earlier draft] Several guesses as to why this may be is that either the batteries were low or the plastic somehow interfered. Also, not only were these mathematical values different, the number of packets received by the receiver from the hockey pucks was quite a bit lower in 3/5 cases (one was down to around 20, from the expected 120). These results point us in the direction of more testing of the hockey puck style motes and their respective sensitivity.
This test consisted of separating the motes by 5 different distances and getting the LQI and RSSI values. We did this to see if one or the other has more consistent and relative results. Also, this test was to check the sensitivity of LQI because some articles say that for the distances we are dealing with the RSSI value will provide the proper resolution/sensitivity. Because only 5 distances were tested there are only 5 data points (the averages) on each plot. This low resolution makes the plots look bad there is an interesting trend of both the RSSI and LQI dropping in the middle of the range and higher at both ends. The expected line would be increasing as the motes get closer to each other, but every orientation showed that around 1 meter both quality indicators dipped a little. Well onto the plots…
The first two plots are of the averages and standard deviations with an orientation of the two motes vertical with the usbs pointing up, battery packs facing away and a height off the ground of 40 inches.
A link to the data is included here http://spreadsheets.google.com/ccc?key=0AoolpjQMTdWpdFVoWnBQX0RJQXF6cmxseU95Z2ZMY3c&hl=en
The next two plots are the averages and standard deviations with an orientation of the two motes lying horizontally on the ground with the usbs facing each other
A link to this data is included here http://spreadsheets.google.com/ccc?key=0AoolpjQMTdWpdHU4dzRUaWpnVm9CTGh2c2hieEZMN3c&hl=en
The next two plots are with the orientation such that the motes were horizontal on the ground with the usbs facing away from each other
A link to this data is included here http://spreadsheets.google.com/ccc?key=0AoolpjQMTdWpdFNMSTlrUHlSdEo5Wkd4c0VJeXkyNFE&hl=en
This time there was one mote held vertical with the usb pointing upwards while the other was horizontal. The two motes were placed .55 meters away from each other and the horizontal mote was spun around 360 degrees with a resolution of 10 degrees. This was a preliminary test because we have not checked for any other wireless activity in the area yet so there may be some sort of
interference. Here is a plot of the rssi for several different power levels at each angle, with zero degrees when the horizontal usb is pointing towards the vertical mote: 
This second plot is of the standard deviation of the rssi at certain power levels. It is easy to see that that most of the standard deviation appears to be random, but there is a spike at zero degrees. Now that I think about it, the point marked 0/360 degrees is when the horizontal motes antenna is pointed away from the vertical, which probably explains the low rssi at these points. 
Powered by WordPress
