| Author: | Rocamora, Josyl Mariela Balce |
| Title: | Fingerprinting-based indoor positioning using channel state information |
| Advisors: | Ho, Wang Hei Ivan (EIE) Mak, Man-wai (EIE) |
| Degree: | Ph.D. |
| Year: | 2023 |
| Subject: | Indoor positioning systems (Wireless localization) Fingerprints Wireless communication systems Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Electronic and Information Engineering |
| Pages: | xxxii, 204 pages : color illustrations |
| Language: | English |
| Abstract: | Techniques for indoor positioning systems (IPSs) can be categorized as range-based or range-free. Range-based methods rely on geometric mappings to approximate a location given the calculated distances or angles from multiple reference points. In contrast, range-free strategies utilize fingerprinting, wherein an acquired fingerprint data is compared to a pre-collected dataset to identify the best position estimate. Fingerprinting systems classify new data by finding the best match in a database of fingerprints using similarity functions, statistical techniques, and machine learning algorithms. These fingerprints are generated based on wireless measurements, such as received signal strength indicator (RSSI) or channel state information (CSI). An advantage of CSI is its intrinsic ability to represent multipath effects, for which RSSI is not able to identify. CSI is a complex vector representation of the wireless channel and has the potential to achieve centimeter-level positioning accuracy even with a single reference point only. Fingerprinting of CSI is preferred over other information such as RSSI as the former can exploit the effect of multipath propagation and is robust against non-line-of-sight (NLOS) channels. Recent IPS that utilize CSI consider ideal scenarios to achieve high-accuracy performance in fingerprint matching. In this thesis, we investigate CSI fingerprinting systems to provide a comprehensive research on how fingerprints are generated, how such systems localize unique points in space, how to mitigate the effect of wireless channel dynamics in the fingerprints, and how to apply deep learning and model adaption using vector embedding (VE) in the CSI domain. We first discuss the process of fingerprinting which involves offine and online phases. In offline phase, we perform a site survey by acquiring CSI from the target positions in the environment; these positions are unique points indoors. Data collection can be done using a pair of software defined radio (SDR) or a single-board computer (SBC) connected to a wireless local area network (WLAN). The labeled raw samples undergo preprocessing, which may include phase compensation, concatenation, vector normalization, dimension reduction, feature scaling, mean centering, data conversion and dimension reduction. The preprocessed fingerprints then comprise the fingerprint database that will be used for pattern matching during the online testing phase. In some algorithms, the training subset of the database is utilized for model creation. Seven CSI datasets are used for experiments such as multi-class centimeter-level positioning, indoor environment detection (IED), fingerprint quality detection (FQD), long-term temporal positioning, and model adaptation using VE. We then implement positioning algorithms based on time-reversal resonating strength (TRRS), support vector machines (SVM) and simple Gaussian classifier (SGC). TRRS is a method to compare two data akin to cosine similarity and does not need extensive model creation process due to its simplicity. SVM is a commonly used machine learning method for classifcation problems, particularly those that have a small sample size relative to the data dimension. According to our experiments on IPS, perfect centimeter-level localization can be achieved with high bandwidth fingerprints (e.g., 300 MHz) when the temporal difference between training and testing data is in the order of days. However, positioning accuracy can drop to 20% for low bandwidth cases when channel conditions are less than ideal (i.e., noisy channel). In the IED experiments, we trained a multivariate Gaussian model for each class and selected the model that returns the highest log-likelihood. By using a Gaussian distribution to model CSI fingerprints, which offer more abundant information regarding the channel dynamics than RSSI, we can exploit the variance inherent in the wireless channels. Our experiments demonstrate that SGC incurs minimal delay of less than 4 seconds and achieves high accuracy in binary room classification compared to other techniques. In particular, it achieves up to 50% and 150% performance improvement over TRRS and SVM, respectively. One essential component in achieving high accuracy is the collection of high-quality fingerprints. The quality of fingerprints may vary due to uncontrollable factors, namely, environment noise, interference, and hardware instability. For our FQD experiments, we have developed a logistic regression classifer to evaluate the quality of the collected CSI samples. The classifier is then used to sift out poor samples and only retain good ones as input to the positioning system. Systems based on TRRS and SVM show better performance when the training fingerprints are of good-quality. Channel selection using this quality detector can improve the performance of TRRS-based positioning by 8-73% and the performance of SVM-based positioning by 9-238%. Lastly, we explore the use of deep learning for recognizing long-term temporal CSI data, wherein the site survey was completed weeks before the online testing phase. Compared to implementations based on TRRS, SVM, and SGC, our deep neural network (DNN) model shows a performance improvement of up to 13.6% for multi-position classification. We also exploit vector embeddings, such as i-vectors and d-vectors, which are traditionally employed in speech processing. With d-vectors as the compact representation of CSI, storage and processing requirements can be reduced without affecting performance, facilitating deployments on resource-constrained devices in Internet-of-Things (IoT) networks. The setup based on d-vectors shows comparable performance with less storage and processing requirement, which is suitable for resource-constrained devices in IoT networks. By injecting i-vectors into a hidden layer, the DNN model originally for multi-position localization can be transformed to location-specifc DNNs to detect whether the device is static or has moved, resulting in a performance boost from 75.47% to 80.62%. This model adaptation requires a smaller number of recently collected fingerprints, as opposed to a full database. This VE approach can be utilized to adapt a DNN classifier with only a smaller number of recently collected fingerprints, as opposed to training a model from a full database. |
| Rights: | All rights reserved |
| Access: | open access |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/12381