indoor localization of UHF RFID tags
For many commercial applications it is of interest to identify and localize objects. The most traditional way of identifying objects is to use labels with a printed barcode and attach them to the objects of interest. However, these barcodes need direct line of sight between label and scanner, which can be a burden for some applications, for example when labeled products are packaged or the labels are integrated into objects.When a RF transceiver is attached to the object, identification can be done with the help of a wireless communication system and does not require line-of-sight. Technology that uses RF to identify objects is known as RFID and the objects are labeled with so-called RFID tags.
Equipping tags with fully functional transceivers requires a lot of energy and therefore batteries, increasing size and cost. Technological advances have made it possible to produce tags without batteries, so-called passive tags, which can harvest energy from an incoming RF signal. The RF signal from the reader can also be used to transmit information from reader to tag. To communicate back to the reader, these passive RFID tags use backscattering. The antenna of a tag reflects a certain amount of the energy it receives. By modulating this reflection, tags are able to transmit information. Many RFID systems have been developed. Long range (typically a couple of meters) and battery-less tags are best supported in the UHF band. The industry standard protocol for passive UHF RFID tags became known as EPCglobal Class 1 Generation 2. Over the years, these Generation 2 tags have become very cheap (less than 10 cents) and are currently used commercially for a wide range of applications. These applications range from identifying cars that enter a parking garage to individual item level tracking of clothes in a retail supply chain, from the manufacturer right up to the counter of a shop.
The advantage of non line-of-sight identification of tags can also be a disadvantage as tagged objects can be anywhere within the read range of the reader. For logistics companies tracking a parcel between distribution hubs, this coarse localization is not a problem. However, when tracking items that are moved from a storage room inside a shop to the shop floor, the ill-defined read range becomes problematic. The read range is difficult to model and to measure as UHF signals are susceptible to reflections, especially within an indoor environment with a lot of reflecting objects. A lot of research has been performed to determine the exact location of a tag, based on different signal characteristics like signal strength and phase. This thesis describes possible solutions to the localization problem.
By measuring the phase difference between the transmitted continuous wave and the received backscatter from the tag at different frequencies, it is possible to estimate the distance between the reader and the tag. By measuring the distance to three readers it is possible to determine the location of a tag with the help of trilateration. The location of the readers has to be known and any error in the distance measurement influences the location estimate. To overcome the influence of the environment on the distance estimate, this thesis suggests the use of reference tags and the K-Nearest Neighbors (KNN) algorithm to derive a location. A localization experiment is done and our phase-based algorithm is compared with a KNN algorithm based on received signal strength. The results in terms of average localization error are similar, about 0.4 m. Furthermore, some experiments are used to determine whether the use of phase measurements for the detection of a moving tag in a portal application is viable. The results show that with state-of-the-art readers tags moving at walking speed cannot be read fast enough to track the phase. However, with improved readers that can scan the phase at least two times fast, tags that move at walking speed can be detected.
Multiple readers and reference tags are expensive to install in a commercial environment. Another approach is to use an array of multiple antennas, a so-called phased array. When a signal is received by two different antennas there will be a time delay between the signals dependent on the DOA of the signal. If we assume that the signals under investigation are narrowband, this time difference becomes a phase difference, hence the name phased array. So, by measuring the phase differences between antennas, the DOA can be determined. In case of two antennas the phase difference translates directly into a DOA. For an array with more antennas the MUSIC or EPSRIT algorithm can be used to estimate the DOA.
The observation that a tag is within the near field of a phased array leads to the fact that there is an extra phase difference depending on the distance of the tag to the array. This distance can be estimated with the help of 2-dimensional estimation algorithms. Experiments are used to validate this approach in a real environment. An average angle error of 3 to 4 degrees and a range error in the order of 0.3 m is measured. This range error is comparable to the 0.4 m achieved by the setup described before. Differences with measurements in an anechoic room show that the performance of the system heavily depends on the environment, as the average errors decrease to 1 degree and 0.2 m, respectively. By combining DOA measurements from multiple phased arrays and the help of trilateration, a location estimate can be made. To decrease the energy consumption of a multi-array system, this thesis explores the use of heavily quantized signals instead of the high resolution signals used in the near field experiments. The DOA estimation algorithms are based on correlations between the different array channels. By using single bit quantized signals, errors are introduced in these correlations, which can be corrected by making some assumptions about the shape of the received signals.
Experiments in an anechoic room show that the suggested correction can decrease the average error of single-bit quantized signals from 4 degrees back to the unquantized average error of 1 degree. Experiments in a realistic environment show that by using single-bit quantized signals, the DOA estimation degrades from 4 to 6 degrees. If this increase in error is permissible for the application, it is possible to construct an array to estimate the DOA without the need for high resolution ADC, which saves on computational cost and power.
Overall, this thesis shows that there are many options to localize Generation 2 tags. However, due to the complex environment with severe multipath effects, localization of Generation 2 tags still remains an open problem.
