A Pairwise SSD Fingerprinting Method of Smartphone Indoor Localization for Enhanced Usability

Round 1

Reviewer 1 Report

·         The work has developed PSSD to overcome the limitations of traditional RSS and SSD measurements using the smartphones. Also, it shows how the results are effected for homogeneous and heterogeneous devices. The manuscript is well written and structured. The problem statement and contribution are clearly defined. A good amount of experiment is done to establish the findings of the authors. Few comments regarding the work done:

o   Would it effect the CDF vs Error performance if the grid map intervals are i.e., 1m or 0.5m instead of 2m (mentioned in Line: 402 to 403)? In reality 2m interval could create a huge error. Is there any specific theoretical reason to choose the interval? Please clarify. You can use citations if its already used successfully.

o    Please add more explanation for the Table 2 in Line: 456 to 459. The results are convincing but, need to improve the description.

o   The authors can consider this comment as their future work if possible. Typical KNN is used in the work with Euclidean and Pearson measure. Try to used other distance based machine learning methods and check the performance. Also, I will suggest to employ Jacard and Cosine distance measures where, the metrics use scoring values to represent distances instead of square and absolute to find the changes in performance.

·         I believe novelty is in the work and it could be accepted with the minor changes.

Author Response

‍‍Dear Editors and Reviewers:

Thanks very much for your comments! Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our future research. By adding the answers/revisions to these questions to the revised version of the manuscript, we feel that the quality of the manuscript has been improved. Please find the revisions that are marked in red in the manuscript. The main corrections in the manuscript and the responds to the comments can be also found in the attached rebuttal letter. Thanks again!

Best wishes

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper proposes a space-constrained PSSD (pairwise signal strength differences) strategy to improve WiFi fingerprinting reliability, and mitigate the effect of hardware bias of different smartphone devices on positioning accuracy without requiring a calibration process. I have the following comments：

Mitigating the effect of hardware bias of different smartphone devices on positioning accuracy is not the novelty of the paper since SSD method has been addressed in the literature. Compared with the conventional SSD method, the authors use (8) instead of (6). The authors should compare the test results between (8) and (6) without PDR information to prove the effectiveness of (8).

The proposed method may lead to high computational burden. The authors should provide the computing times of RSS, SSD, and PSSD methods.

The main contribution for the paper is to use the PDR information in fingerprinting method to improve the performance. The walking speed and heading range are very important for the proposed method to select the database. How to obtain these information? Using accelerometer? If walking speed and heading range are available, simple RSS fingerprinting method + tracing method may have better performance and lower computational burden. Hybrid method with RSS and accelerometer measurements has been widely used in the literature.

It is indicated in fig.12(a) that honor 8 is a homogenous device. However, it is not supported by fig.3. It can be seen from fig.3 three handphones have the similar variation.

It is strange that the proposed method performs worse than RSS method for homogenous device. However, the proposed method have more prior information such as PDR information than RSS method.

 Signal attenuation caused by body will greatly affect the system performance. RSS measurements will different when people face or back the WIFI AP. The authors should add some analysis on this aspect.

It can be seen from table 2 that the rmse varies from 3.2 to 4.9. This is not a good positioning accuracy. With such poor performance, it is necessary to build a fingerprinting database? Could you provide the location results performed by geometric positioning method using RSS and SSD measurements?

 

Author Response

Dear Editors and Reviewers: Thanks very much for your comments! Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our future research. By adding the answers/revisions to these questions to the revised version of the manuscript, we feel that the quality of the manuscript has been improved. Please find the revisions that are marked in red in the manuscript. The main corrections in the manuscript and the responds to the comments can be also found in the attached rebuttal letter. Thanks again! Best wishes

Author Response File: Author Response.pdf

Reviewer 3 Report

The paper describes an approach for indoor localization using WiFi Received Signal Strength fingerprinting in combination with Signal Strength Difference approach and a one timestep context for the SSD. Many parts of the paper are nicely written.

Overall, many aspects covered in the paper suffer from their missing novelity. For example, step detection is state of the art. The same is true for RSS fingerprinting and WiFi SSD. What remains as new is the PSSD idea which sounds interesting.

But there are things that need clarification and improvement:

- Functions (like log in (1)) should be properly formatted as functions (non-italic, like max in (11)).

- What is the meaning of the braces in (7)?

- In (9) on the left side of the brace there's something missing?

- Where do the constants +10,-10 in (11) come from? What is their meaning?

- Generally: if a paragraph continues a sentence, it should not be indented. lines 180, 185, 207, ...


Evaluation: This is a problem...

-    It’s good to see that the approach was tested in a large building
- It's good to see the map of the building, but I would be interested in the walked paths. These paths should be added to to Fig11
- The size of the building and the training locations do not match: A builing of size 90x22m with training locations with a mean distance of 2m would require approx. 500 locations (minus unused space). Additionally, there are only 35 fingerprints in the 90m directions of the building. Something is wrong with all these numbers...
- What is the overal time required for fingerprinting (training phase)?
- The evaluation (Sec4.2) is way too superficial as it consists of only 40 lines of discussion plus one figure (Fig12, Tab2 contains basically the same information as Fig12, Tab1 shows the recognition rate of the step detection which is state of the art). Chapter 4 leaves many questions unanswered: Where do you observe large errors? What is the reason for them? How does the error vary over time during the walks? How does the walking speed influence the accuracy? Are the experiments performed in an "empty" building? What if the building is very crowded during testing and empty during training? What happens when the fingerprinting distance of 2m is increased?
-    The evaluation must be significantly improved

Author Response

Dear Editors and Reviewers: 

Thanks very much for your comments! Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our future research. By adding the answers/revisions to these questions to the revised version of the manuscript, we feel that the quality of the manuscript has been improved. Please find the revisions that are marked in red in the manuscript. The main corrections in the manuscript and the responds to the comments can be also found in the attached rebuttal letter. Thanks again!

Best wishes

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Thank you very much for answering my questions and clarifying the open points. I do not have many new comments concerning the explanations:

Point 7: If I get it correctly: The experiments do not consist of long, uninterrupted walks through all the rooms? You have many short walks, where each walk stays in its starting room?

Points 10, 11, 12: Thanks for the five pages of discussion. This was very helpful and the kind of discussion I was missing in the first version of the manuscript. However, I think it is a pity that these explanations, the figures A-C and the table are not included in the new version of the manuscript. Anything that helps me understanding the experiments will also help the normal reader. Therefore, I think that the information of these five pages of discussion can be taken over more extensively into the manuscript.

I tried to download the supplementary video, but all I get is another copy of the revised manuscript. I am not sure if you uploaded a wrong file or if there is something wrong with the mdpi system.

Author Response

Dear Reviewer,

Thanks very much for your careful review as well as the good advices on improving the manuscript. We have now made corrections according to your comments, and we hope this version can be friendlier to the normal readers as you mentioned. Thanks again!

Best wishes

Author Response File: Author Response.pdf