Hand Hygiene

Using the whip antenna, I did 3 power experiments, changing the power from 20 to 30, 40, and 50. We did this to see if increasing the power would clear up the data at bigger distances.

These are the averages of Power Levels 20 and 30 at the different distances:

Here are the results:

Power of 30:

From this graph, we see the data clearing up at a distance of 3.0 compared to previous experiments.

Power of 40:

Power of 50:

This graph almost looks the cleanest out of the three. On the CDF – the data for a distance of .5 is interesting- for some reason it has a strange curve.

All three of these graphs show a separation between 1 and 1.5 meters- the only question is- how high do we want the power to be in our experiments.

Information regarding base station components.

[iframe http://spreadsheets.google.com/pub?key=tHhuIrx0NBaT5O-qMxMcfug&output=html 1000 230]

Note: the base station also includes a mini-USB.

I did a quick experiment today to answer two questions I had:

1) Does the height of the anchor location make a significant difference in results? (It makes a difference- maybe not significant)

2) What does the range of RSSI received look like when the pager is stationary (so no movement of person/between locations)? (It looks much better than when there’s movement.)

I did this by setting 4 pagers out and collecting data over 8-9 minutes and the same whip antenna as in the previous experiments. Here are the results:

This graph shows that the ranges are much smaller than what we’ve been receiving, except for at position 3. In the previous experiment, the results were opposite, with position 3 having the narrowest range.

In the tall trial, I put the anchor 54″ above the ground, and the results seem to be pretty good- there is a significant separation between the near positions (1 & 2) and the far positions (3 & 4). This is good for the near/far from bed experiments we’ve been working on, if we can get this to work with movement.

The short trials also show a separation between near and far, but this separation doesn’t distinguish the difference between positions 1 and 2 as well as the tall experiment. This one does distinguish the differences between positions 3 and 4 better.

So, basically we have what we want under “perfect” conditions. Now, we need to fine tune them to work under all the conditions — so we need to add movement between locations and movement of the person wearing the pager and see if these could be what’s causing the problems.

I repeated the experiment with the whip antenna in the lab. Position 1 corresponds to “head of bed”, position 2 corresponds to “foot of bed”, position 3 corresponds to “middle of room”, and position 4 corresponds to “doorway”.

This graph is a little better than the graphs we had before. There is still a huge variance in some locations.

The CDF looks much better than the one from the hospital data- there is more of a separation between the positions. However- there is an inversion between positions 1 & 2. This could be old messages appearing when I was at position 1.

I thought these problems could be caused by the clock, so I also made plots of RSSI vs. Time to see if there was a cycle.

For the most part, there seemed to be cycling, especially near the beginning. But near the end, the data gets a little fuzzy. There are some odd sharp points though. This could mean both problems with the clock or problems with the signals.

I think we should look for more pronounced cycling & hopefully get a mote time interface to solve some of our problems. I’m not sure what we could do about the jagged points though.

After making conclusions reached in my previous post on puck batteries (see http://groklab.org/handhygiene/2010/04/20/puck-current-draw-battery-capacity/ and its comments), I’ve taken a look at the battery selection from Digikey and Mouser. Here’s what I dug up (prices are per case of fifty batteries):

• Digikey PN P340-ND – CR2450, 620mAh, 24.5x5mm, $45
• Digikey PN P121-ND – CR3032, 500mAh, 30×3.2mm, $98
• Digikey PN SY340-ND - CR2450, 610mAh, 24.5x5mm, $58

• Mouser PN 658-CR2354 – CR2354, 560mAh, 23×5.4mm, $73

I’ve done a quick search for other suppliers and none can seem to beat Digikey’s prices. But these are standard packages, so I’m sure there is someone out there somewhere dying to sell really cheap batteries.

Two problems that we may run into with these types of batteries: internal resistance and standard current ratings.  These issues are inter related. The former may shorten the life of the batteries by allowing them to dip below the mote’s threshold when it is broadcasting, probably after some of its capacity has already been used. Actual data is not spec’ed well in datasheets. The latter may imply that the battery chemically won’t be able to supply current at a fast enough rate to power the radio. Typical spec’s for standard current is around 0.2-0.5mA, while the radio seems like it draws around 20mA. It seems like a vague term, but I am guessing that this is the current draw that will provide nominal voltage and capacity. Overdrawing from the battery may cause voltage drop or decrease in life, or we simply may not be able to pull power out at that rate. Another related issue is that if the battery has a limit to the power supply, this limit might affect the radio Tx/Rx power, (thus RSSI?).

Deepti and I ran some quick tests: turns out that a CR2430 (3V, 0.2mA standard current) will run the mote, and it will allow the mote to broadcast (at least with the RadioExp program). We used the same mote, alternately with two AAAs, and with one CR2430 and compared RSSI at set distances. The first test was performed with the mote horizontal, and after rough (eyeball) analysis of RSSI data, the AAA gave RSSI of around -8, while the CR240 gave RSSI of around -20. The second test was performed with the mote vertical, facing the receivers. This test gave similar (around -9) RSSI values for both batteries. Hmm….  Our test setup was on-the-fly and not very thorough, so I am guessing that the discrepancy is due to poor setup. I will plan on repeating the test in the next couple of days with a better experiment.

I did two experiments in the hospital to test the distinction between near and far from the bed in the hospital. For near the bed, I stood in two positions – near the head of the bed and near the foot of the bed. For far from, I stood in the middle of the room and just in the doorway. One experiment used 2 F antennas (near the head and near the foot) and the other one used a whip antenna.

Here are the results:

From this graph, we see that there is almost no distinction between the different locations using 2 F antennas.

From this graph, we see that near the head of the bed gives higher results than the other positions, but the distinction is still not that great.

From this graph, we see that there is almost no distinction between the different positions. But, we do see that position 4 (in the doorway) lags behind the rest of the positions, which is good.

Looking at this data, I think we’ll have to redo the experiments after we find out what could have caused this. The experiments we did in the lab showed much better distinction between locations.

An important influence on puck design (especially considering desire for low profile) is the type of battery we use. Currently we have been using Duracell Alkaline AA with a capacity of 1500mAh (the iPod mini batteries have 850mAh capacity). We expect the new puck to draw power at a faster rate than the old puck due to the force sensitive resistor interface. It will be important to know the average current draw rate of the new puck so that we can weigh that against battery time capacity and weight/size. A very useful metric would be battery energy density in terms of volume (eg, mAh/cm^3). I don’t think that this property is already tabulated somewhere out there, but if we know what kind of time capacity we would like our batteries to have (and the rate at which the batteries are drained), we can start narrowing down choices and get a general idea of what size options we will have.

I have done some quick calculations to get us in the ballpark of how fast we will be using juice in the new puck:

The mote user guide lists nominal and maximum current draws for a number of states — the three that of are interest to me are “MCU on, Tx/Rx,” “MCU idle, radio off,” and “MCU standby.” This information is available in the user guide, but I have made the following assumptions:

• The mote will draw 8.5μA (or 100μA, depending on what “sleep” means…)
• The mote will draw 21mA when it broadcasts
• The mote will be awake for 500ms when it broadcasts, 100 times a day
• The FSR conglomeration will idly draw an average of 1.5μA

I say that the mote has a duty of 500ms*100/(24*60*60 sec/day)=0.0006, so the average draw of the mote is Duty*(Tx/Rx current)+(1-Duty)*(Sleep current)=21μA. The time capacity (how long it takes for the batteries to run out) is given as Battery Capacity (mAh)/Current Draw (mA). With an iPod mini battery, this comes out to 850mAh/(21+1.5μA)=about 4 years. Maybe sleep is actually “MCU idle, radio off” for a draw off 54.5μA nominal (rounded out to 100μA just in case), in which case the time capacity is about a year. So there’s something not quite right here. Likely as the battery nears the end of its capacity, it’s voltage will drop, rendering it unusable. Maybe this method of calculation is too rough. But I’m still at least an order of magnitude off.

Anyway, as we prototype the FSR interface circuit, it may a good idea to get an empirical measurement of the rate at which the whole thing will draw current.

When we did the sensitivity/specificity experiment again, we also used the MoteMate iPhone app and the motes that wrote to flash. We used channel 14 for these motes, as opposed to channel 25 for the other motes.

Anchor 46 was in the room and Anchor 42 was at the nurse’s station. The flash data from Anchor 42 was not available for analysis.

The data in this plot shows that this experiment did not go as planned. At the Nurse’s Station – Anchor 42 should have a Higher RSSI than Anchor 46. Anchors 46 and 42 were about equal for the Nurse’s Station, and this should not be the case. The same happened in Room A. I’m not sure what happened here. The CDF’s also show this problem.

I did the experiment on the F-antenna and on the Whip Antenna again.

Here are the results:

The data for the F antenna didn’t show the separation that we were looking for. However there was a nice separation between .5 and 1 m, which we could consider in the future.

This time, the whip antenna didn’t show inversion between 5. and 1 m. The separation isn’t the best to make a single cut off value, but it’s still pretty good.