search

usnistgov / psc-ns3

Public Safety Communication modeling tools based on ns-3
Other
69 stars 38 forks source link

Problematic SL-Tag tbSize data cast #20

Closed fisserL closed 3 years ago

fisserL commented 3 years ago

Hi,

a student of mine and me found a possibly quite severe bug during routine test evaluations of the PSC toolbox for NS3.

The evaluation in question is the provided wns3-2017-pssch scenario, which was introduced to evaluate the throughput for a simple sidelink communication for varying configurations.

One of the main analyzed parameters was the sub-band size, which was iterated from 2 to 50 RBs in steps of twos. The expected behavior would indicate an increasing throughput with larger sub-band sizes. (Plotted in Figure 5 of the complementary publication)

However, the data-rate calculations showed anomalies:
image

As one can see from the output, the data-rate follows changes in a "saw-tooth"-pattern w.r.t. to sub-band size, which is clearly a rather weird behavior and also different from the expected one.

A rather extensive search identified the tbSize as the main inconsistency, which was correctly reported on the MAC layer, but not on the PHY layer stats.

After some more digging, we found that the problem stemmed from a bad data type cast in the de-serialize function of the LTE SL Tag. More specifically - Line 104/105 of the lte-sl-tag.cc file.
image

The tbSize is here cast into a uint_8 which will wrap around at 255, explaining the fall in data-rate as well as the saw-tooth like behavior.

Changing the casts to their correct data type resulted in the following output of the wns3-2017-pssch script:
image

The new results fit the data from the publication and are also more in line with the expected behavior of the LTE Sidelink communication.

Maybe someone from the official team would be available to comment on this possibily quite severe bug in the LTE PSC Sidelink toolbox?

This bug was found in the most current version of this toolbox.

Thanks and much thanks for your work. Leonard Fisser

fjcintron commented 3 years ago

Thank you for submitting this issue. Our team will be reviewing this further and include the solution on the next release.

richardrouil commented 3 years ago

The issue has been resolved in release 4.0. Thank you for reporting the issue and providing a solution for it.