What is the right model for wireless channel interference?

In a wireline network, nodes form links with only those nodes they are wired to, and the links do not interfere with one another. In contrast, in a wireless network, signal transmissions are intrinsically broadcast, and suffer from mutual interference. In several physical layer technologies, a wireless signal is decoded by treating the sum of all the other on-going signal transmissions as noise. Hence, from a networking standpoint, there is a need to model wireless channel interference. An accurate interference model is especially important in a multi-hop network context, since there could be several simultaneous wireless transmissions. Several works in the literature have made use of simplified interference models. Some works assume a fixed range for communication and interference, while others are based on concepts like capture threshold where the desired signal strength is compared with interference from a single node at a time, rather than cumulatively. In particular, the latter model is used in ns2 which is the most common simulation tool. Under isotropic pathloss, the capture threshold model is also equivalent to the protocol model proposed by Gupta and Kumar, which is now the subject of a lot of analytical activity notably through conflict graph based problem formulations. We investigate the accuracy and appropriateness of the capture threshold based interference model, by comparing it with one based on the SINR (signal-to-interference-and-noise ratio) with additive interference calculation. We find that both in the case of random access networks, as well as in the case of scheduled networks (where node transmissions are scheduled to be completely conflict-free), a simplified interference model such as the capture threshold model, can produce significantly different results compared to an additive interference based model. Therefore, a lot of caution should be exercised before accepting or interpreting results based on simplified interference models.