If you ever wondered why your good cellphone connection just started making that scratchy noise when you stepped couple of feet away or during your daily commute your favorite NPR station produced that awful hiss when stopped at a red light and so on.. you are witnessing a wireless phenomenon called fading. Technically, there are two types of fading: multipath and shadow. The first one is mostly the cause for drastic signal change when you move very short distances. This is caused by the radio signals occasionally canceling each other when they arrive out of phase at your antenna. The second type of fading called the shadow fading is essentially caused by physical objects blocking part of the radio signal for e.g. mountains, buildings etc. My job is to make sense of these fading noises and then figure out the location of a wireless transmitter.
Interestingly, there is an application for this research. With GPS being so ubiquitous on smart phones, it is quite natural for you to assume that by turning on your GPS you can figure out where you are located. Not so fast… this is only true if you are on an open area with a clear view of the sky. But, what if you are in a new shopping mall and has to quickly figure out a way to get the hell out of this place?… you are out of luck with GPS. That’s where the indoor localization system that I am working on is supposed to help. Just like the turn-by-turn navigation on a GPS, the indoor localization technology is supposed to guide you to the exit. Though it sounds simple, the reality is a whole different matter.
Street wisdom on making this happen is to use a famous equation called Friis transmission equation which is essentially telling you that if you move away from the transmitter the signal strength drops. Unfortunately, the equation only works reliably when you are in outer space with nothing to block or reflect the wireless signals. In an indoor environment with doors, people, walls etc. there will be reflections, diffraction, refraction of radio signals. Radio signal are flying to your receiver antenna from all directions and quite often cancelling each other. Now this is awful for localization but some smart scientists have figured out a way to utilize these multiple signals to improve your WiFi speeds using a technology called MIMO. Check out IEEE 802.11 – N WiFi modems.
Coming back to my indoor localization problem, the smart people working on this problem came up with another simple idea. How about storing the signal variations over the shopping mall on to a database and then use sophisticated pattern matching techniques to figure out the position of a transmitter when a radio signal is received. The problem with this approach is that the wireless signal strength never remains the same over time on an area. If the environment changes because there are more people moving around or someone moved their new trinket kiosk into that area or some marketing guy thought that putting out that awful holiday decorations would get more foot traffic etc… the signal power will change. During the early stages of my research, I spent countless weekends collecting radio signals around the laboratory (ERL 114) @ MST to prepare for demos trying to impress some visitors with fat checks… only to fail miserably when they were present because the signal power over the lab has changed due to the presence of more people than when I originally took them. Just in case if you are curious, the image below shows the the signal power map over the lab with red areas indicating bad reception and blue areas being relatively better reception areas.
Signal strength variation over ERL 114 @ MST
Now my contribution to this technology was to turn the problem on its head and start looking at the noise in the signal rather than the signal itself. Or as I would like to say “Information from Noise”. My hypothesis was that if wireless receivers experience similar noise then they are likely to be closer to each other. Let’s bring a day to day analogy to better explain my idea. Assume you are walking aimlessly and all of a sudden it starts to rain. You are not sure where you are so you call up your friends at their home and ask them if it is raining at their place too. If some of them reply positively then there is a very high probability that you are closer to those friends than those who responded negatively. Using this information you can build a distance map from your friends and then once you know the address of some of your friends you called, you will be able to approximately find your location.
In the wireless environment, instead of bad weather like rain, I rely on the noise caused by fading, specifically the multipath fading. My wireless device queries other wireless devices in the shopping mall and ascertain if they are also observing the same multipath fading noise. Based on their response, my device builds a relative distance map of possible locations within the mall and subsequently figure my way to the exit. There are couple of caveats to this idea. Primarily, the wireless devices that you are querying cannot be very far from you. Just as in any weather system, two regions might experience the same bad weather but they might be thousands of miles apart from each other. So for multipath fading, if the wireless devices are separated by more that twice the wavelength of the radio signal, this method will fail. Additionally, the accuracy of this method is very much dependent on how finely you can measure multipath fading statistics. Some of the commercially available wireless devices do not provide very fine resolution in measuring signal power and this leads to large uncertainty in the inferred location. For the ubiquitous WiFi radio signals the wavelength is approximately 12.5 cm this limits the applicable range of localization using my method to 25 cm (9.8 inches), which is very small for any practical purpose.
Being a delusional optimist, I started looking around for other wireless technologies that might benefit from this localization method. That is how my method ended up on Radio Frequency IDentification (RFID) tags. Specifically, RFID tags that operate at wavelength 22m or larger are ideal since my method will work for a larger radius of 44 m (144 feet). You will see these RFID tags attached to lot of commercial items just like barcodes for inventory purposes. So if you have a truck full of items tagged using 22m RFID tags, my method should be able to provide a position information of the tags from a single scan. A typical case scenario is shown in the figure below.
RFID Positioning System
You can find the details about this methodology along with the mathematical derivation of multipath fading statistics and its dependence on radial distance between wireless device and all other interesting stuff in my paper that will soon be out for print @ http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=6175900&isnumber=4358975
Basheer, M.; Jagannathan, S.; , “Localization of RFID Tags using Stochastic Tunneling,” Mobile Computing, IEEE Transactions on , vol.PP, no.99, pp.1, 0