A Localization and Tracking System Using Single WiFi Link

Round 1

Reviewer 1 Report

The authors approach the problem of indoor localization by proposing a full system that includes the 3D-MUSIC algorithm (with improved computational efficiency). The paper does not remain only on simulation but provides experimental validation.

Overall, the proposed system seems to be working, however, due to the lack of some information as well as some parts that are completely missing, the final results obtained via experiments are questionable. (I am not saying that they are wrong, it is just that the lack of information makes the final results hang in the air.).

Editorial remarks:

1) The introduction is well-written. The state-of-the-art is nicely summarized. The shortcomings of the SotA are well-presented, and the goal of this paper is well-established.

2) I think that the CSI modeling part can be summarized. By now, almost all the discussion provided in this part is known in the literature (the fact that delay creates a phase change over subcarriers, Doppler creates a phase change over range bins and AoA creates a phase change over antennas).

3) On page 7, 5th line from the bottom, "estimations of TOA, TOF, and Doppler": TOA should be replaced by AOA.

Scientific remarks:

1) The construction of the covariance matrix is key when we are dealing with 2D-3D MUSIC. It is not a simple covariance matrix that yields U_n, which contains the eigenvectors that span the noise subspace. The covariance matrix needs to follow a certain structure. However, the authors do not explain anything about how they are constructing the covariance matrix. This needs a mathematical description.

2) After the construction of the covariance matrix, and its SVD, we end up with U = [U_s U_n], where U_s spans the signal subspace. However, the authors do not mention anything regarding how they choose the size of these subspaces, i.e., estimation of the number of signals. This is a crucial step for obvious reasons, and it should not be overlooked. 

3) The authors have done indoor experiments. It is known that indoor scenarios bring ghost targets (due to the multipath components). However, the trajectory lines seem to be very clear, i.e., there is only one trajectory which is the one following the ground truth. How did the authors remove the ghost targets? The system should have detected the ghost targets and followed them for at least a few snapshots, and then the adaptive Kalman filter can discard that trajectory if the ghost target is not following a "meaningful" trajectory. But we do not see any of these. How did the authors handle these problems?

4) Why does the TOF start from -10ns which corresponds to negative ranges? What is the reason for this?

5) The most important contribution of this work seems to be the construction of 3D-MUSIC and how efficiently it is solved by the novel method. However, that novel method is explained only by words as follows: "More specifically, the search range for AOA is from the previously estimated AOA value minus a parametric value to the previously estimated AOA value plus the same parametric value.". First, the sentence is grammatically broken. It is really not understandable. Second, this is not a convincing way to explain a contribution; it needs an algorithmic/mathematical explanation. 

6) The authors claim that the computational complexity is reduced by a significant amount: from a few hours to tens of seconds. However, computational complexity should not be described based on how long it takes to solve a problem since that metric is very much system dependent (the OS kernel, the CPU/GPU power, the implementation itself whether it uses MATLAB, C/C++, if the OS kernel is busy with other stuff or just this computation, etc.). A fair proof of an improved computational complexity should come in the form of mathematics, not words. Given the complexity of the proposed system, it can be quite challenging (or even impossible) to obtain the computational complexity in O(.) notation. However, the current explanation is far from solid.

7) The criterion provided in (11) comes out of nowhere. Why is this criterion effective? How did the authors find this criterion?

8) Providing CDF of errors is quite counter-intuitive. These types of comparisons are done often with RMSE or Probability of Detection. For instance, analyzing Fig. 11 can tell us that the system is not making any velocity errors above 3 m/s. while 70% of the time, the error is at least 1 m/s? Please consider providing more explanation for these graphs.

9)  The authors speak about the improved computation of 3D-MUSIC. However, they are not speaking about the computational complexity of building the covariance matrix and then computing its SVD. If the number of samples is quite large (which seems to be the case here), both the covariance matrix and its SVD will take a lot of "time". Even though the solution with 3D-MUSIC is improved, maybe the rest of the system is still inefficient?

 

 

 

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments to the Author

 

This paper proposes a location and tracking system using a single WiFi link based on channel state information. It is an interesting research and is an important topic. However, there are several points that needs to be addressed to improve the quality of the manuscript.

 

Suggestions to improve the quality of the paper are provided below:

 

1)     The current list of applications (highlighted in the second paragraph of Section 1) not comprehensive enough and lacks the necessary references. Some examples of the applications missed out include emergency management, smart energy management and HVAC controls and occupancy detection. Please review and reference these established works to highlight the important applications where indoor positioning leveraged.

Indoor localisation for building emergency management

Filippoupolitis, A. et al. (2016, December). Bluetooth low energy based occupancy detection for emergency management. In 2016 15th international conference on ubiquitous computing and communications and 2016 International Symposium on Cyberspace and Security (IUCC-CSS) (pp. 31-38). IEEE.

 

Indoor localisation for smart energy management

Tekler, Z.D. et al. 2022. Plug-Mate: An IoT-based occupancy-driven plug load management system in smart buildings. Building and Environment, p.109472.

 

Indoor localisation for smart HVAC controls

Balaji, B. et al., 2013, November. Sentinel: occupancy based HVAC actuation using existing WiFi infrastructure within commercial buildings. In Proceedings of the 11th ACM Conference on Embedded Networked Sensor Systems (pp. 1-14).

 

Indoor localisation for occupancy prediction

Tekler, Z.D. et al., 2022. Occupancy prediction using deep learning approaches across multiple space types: A minimum sensing strategy. Building and Environment, 226, p.109689.

 

2)     In the introduction, authors mentioned very briefly about the limitations of different indoor  positioning technologies adopted by past studies. However, the comparison between WiFi and Bluetooth Low Energy (BLE) should be further elaborated since these are very established technologies for localization and crowd monitoring. For instance, WiFi technologies tend to result in inaccurate estimation of occupancy due to the presence of multiple WiFi-enabled devices and has a high false-positive detection rate in densely populated areas. On the other hand, BLE technologies are more power-efficient than WiFi technologies. Please refer to the reference below on the disadvantages and advantages of each technology to strengthen the manscript’s talking points.

 

Tekler, Z.D., et al., 2020. A scalable Bluetooth Low Energy approach to identify occupancy patterns and profiles in office spaces. Building and Environment, 171, p.106681.

 

3)     The different steps in Algorithm 1 should be expressed using mathematical notations instead of using words.

 

4)     The authors should also consider including the line trajectory generated by Widar2.0 in Figure 14.

 

5)     It was mentioned in the last paragraph of Section 3 that the maximum error of the proposed algorithm is greater than Widar2.0 (i.e., high error variance) as the former only uses the previous estimation and current observation value to determine final position and to achieve real-time tracking. However, it is still unclear why the path matching algorithm, Hampel filtering and smoothing filtering (used for improving stability) cannot be applied to the data that has been collected so far.

 

6)     For the section on “The influence of environments on tracking accuracy”, it would be helpful to include a layout of the three test environments so that the readers get a better understanding of the rooms’ layout and the possible effects of the multipath phenomenon.

 

7)     Regarding the section on “The influence of the shapes of trajectory on tracking accuracy”, I am not sure why the overall error of the vertical line is the smallest when it seem to deviate very significantly from the ground truth as shown in Figure 14. Please elaborate further.

 

8)     In the Conclusions and Discussion section, please mention the limitations of the existing approach and elaborate on how can they be improved in the future works.

 

 

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Authors addressed my comments

Reviewer 2 Report

Thank you for addressing my comments and concerns. This manuscript is ready for publication. Great job!