Publication about SPLAT's propagation models

[der-stefan]

https://www.researchgate.net/publication/261061432_Comparison_of_Longley-Rice_ITM_and_ITWOM_propagation_models_for_DTV_and_FM_broadcasting

[davidsrsb]

What was the conclusion?
My own experience with 380 MHz TETRA is that ITM gave much more accurate predictions than ITWOM for cells between 3km and 50 km radius using SRTM3 DTED

[dBitech]

SRTM3 likely does not have adequate resolution to attribute all of the differences between ITM and ITWOM, Further, without good detail on the specific obstructions, the model is likely not well seeded. (I have a couple of other comparison studies somewhere that discuss this in more detail). But from what I recall, ITOWM has Root Mean Square Error of less than 2dB over measured results, and a std dev of around 5dB, which is nearly a magnitude better overall than most other models.

[hoche]

It was John Magliacane's wish to revert to ITM for Splat 1.4.3, which is the main reason I changed it for that and why it's currently the default. This paper came up in the discussion. ITM is a lot faster and yields "pretty good" results for casual usage; is that reason enough?

My feeling is that more physical studies like this need to be done, and we need some more characterization of why one might be better than another. I went through Sid's code and Sid's whitepaper line-by-line trying to understand his changes and they all seemed pretty reasonable, but we really could use some more in-the-field evaluation.

We definitely should have some high-level overview of the models and their strengths and weaknesses though.

[hoche]

Oh yeah; Sid's whitepaper is out of date and not well organized or indexed. It really needs to be rewritten. It was one of the reasons I was trying to get in touch with him.

[dBitech]

From https://ieeexplore.ieee.org/document/8767299 we see it reported that ... "the RSSI values calculated using the RF planning tool for ITWOM are closest to the values obtained from the real-life LoRaWAN. Moreover, we also show evidence that the choice of a propagation model in an RF planning tool has to be made with care, mainly due to the terrain conditions of the area where the network and the sensors are deployed."

[dBitech]

It's also worth noting that the coverage predictions utilizing splat's ITM more closely resemble the coverage predictions shown by commercial programs such as PathLoss or TAP than those that are produced by Radio Mobile.

[dBitech]

I'm just going to drop this here in case we did not know about these reference standards of propagation models as produced by NTIA https://github.com/NTIA

[dBitech]

Once we get into adding land cover, the following method for simplification http://summit.sfu.ca/system/files/iritems1/19751/08966944.pdf may be of use.

[der-stefan]

According to J. D. Parsons' "Mobile Radio Propagation Channel" 2nd ed. there is an urban factor similar to the Okumura model which can be used in urban areas. Anyhow, we first have to implement landuse input in order to reach out for an appropriate model.

[der-stefan]

There is another interesting (quite short) Master thesis about speeding up SPLAT!:
http://www.visuallmr.com/Documentation/PathLossModels/UsageOf.Lr.Itm.Itwom.pdf

[davidsrsb]

A quick read shows that OpenMP back then gave the same boost that Splat 1.5 has. A pity that the parallel code did not get more widely known at the time.
Presumably a land cover aware model would be slower than ITM. The commercial propagation software that I have tried all have the land use as another raster file

[dBitech]

we may be able to leverage some knowledge from a talk Matt Godbolt gave at CPPcon in 2019 https://www.youtube.com/watch?v=HG6c4Kwbv4I with regards to OOP, DoD, Functional with respect to Path Tracing this may allow us to improve upon performance, it may also allow us to think about implementing some more advanced predictions which take multipath into consideration.