Dear Minh Gia Hoang,

I suggest you ti see links and attached files on topic.

https://arxiv.org/pdf/2012.01563.pdf

https://tel.archives-ouvertes.fr/tel-01499325/document

https://www.science.gov/topicpages/k/kalman+filter+algorithms

https://towardsdatascience.com/a-simple-kalman-filter-implementation-e13f75987195

https://www.semanticscholar.org/paper/Acquisition%2C-Filtering-and-Analysis-of-Positional-Kautz/d1cb91011d7205df34ba7b50019e4e0dc7b38a18

http://paduaresearch.cab.unipd.it/6731/3/matteo_danieletto_tesi.pdf

Best regards

Dear Minh Gia Hoang , If you could apply Kalman filter for tracking applications on IoT nodes as multi-nodes(multiagent) switching topology, it will be sufficient and will be a novel work. if you interested please find relevant literature in this domain. If you feel difficulty please feel free to contact me. Thanks

Hello. Your question lacks detail. If you are implementing a Kalman filter in a multi-sensor network, then it probably is not a Kalman filter, but some form of information fusion across mulitiple sensor nodes. The fact that you wish to use fixed point arithmetic suggests that you are seeking to create local state and covariance estimates at each node and share these estimates. Possibly you wish to fuse them into a global state estimate. These are all design decisions that impact overall system performance and local/global processor requirements.

If you really are doing local (node-based) state estimation, rather than just transmitting raw measurements to a central processor, make sure you use a square root filter or you will lose a lot of numerical precision. The original reference for square root filtering and smoothing is G. J. Bierman's 1977 book "Factorization methods for discrete sequentiall estimation." There are other books that deal with specific multi-sensor tracking problems. Have a look at the Artech House library catalogue.

Please recommend this answer if you find it useful.