Research Idea: Calibrating and investigating Veins' max interference distance setting
The max interference distance is an important tuning parameter for Veins' performance. It trades correctness for speed, and that is fine as long as it is done deliberately. But typically it is not given much thought, and that may be a problem. The default value of 2600m was set after the introduction of the antenna models (see commit b8637c811). It was not revised after the changes to the sensitivity and noise levels of veins 5.0 and 5.1 (see: bloessl2019case).
The value should be re-considered when changing on of the following variables (not exhaustive, there may be more):
mac1609_4.txPower(also: antenna gain)phy80211p.minPowerLevelphy80211p.noiseFloor- anything that goes into the computation and evaluation of the SINR, actually, so even the MCS
Ideally, the simulation outcome should not change compared to an infinite distance, while the simulation runs (much) faster. The simulation outcome meaning which vehicle receives which message at what time, also considering interference. Especially the pure reception of messages without present interferers should not depend on the setting at all. That is why it is called max interference distance: The rationale being that at some distance the signal becomes too weak to have influence as (one of many) interferer(s). But this is way beyond the range at which it could be detected and decoded at all.
But right now there is no real way to know which value to use and what the consequences are. This is due to the interaction with the other variables listed above. E.g., with an increase in transmit power, the same signal at the same distance will have more power and thus more interference. But by how much that changes the outcome is hard to tell.
An more understandable option would be to give a max interference power. Assuming free space path loss, this could be used to compute a distance automatically. One would still have to consider other factors and variables, but everything on the sending side would be covered by this.
A further step would be to even include the reception side by using a maximum interference ratio, as in SIR.
But this would need a reference signal to compare to (maybe at phy80211p.minPowerLevel or some other barely decodable minimum).
Alternatively, the max interference ratio could be relative to the phy80211p.noiseFloor.
Aside from these different configuration parameter formulations, a calibration mechanism could help. It could even help in developing and validating said different formulations. Such a calibration mechanism would run simulations with different max interference settings and record the simulation outcome. It would then compare the outcomes of the different runs and quantify if and how significant the outcomes differed. This could be done by comparing against a golden run with infinite max interference distance. Or by starting with a low max interference distance and increasing the value until there are no more (or little) changes. The first option is probably the safest, telling at which distance there is no difference at all, while the second may be faster. Care has to be taken though to make the simulation run long enough and with enough communication. Otherwise effects like multi-message interference could be overlooked. This should also consider the mobility and layout of the simulation scenario – an urban city with dominant building obstacle shadowing may need vastly different max interference settings than a highway.
Said calibration mechanism could be run with any simulation configuration to automatically asses the max interference setting. Authors can then decide their individual trade-off between correctness and speed. Furthermore, existing models could be checked for validity, as they may have too small max interference distances.
Maybe I’ll consider this for a student’s thesis topic or a paper of my own.
Update: Before the integration of the antenna models, there already had been a dynamic computation of the interference distance (see commit b8637c811). Now it might be possible to re-activat the old formula and autment it with the best-case gain of the used antenna models.