Some wireless networks (sensor networks, ad hoc networks) are best modeled as random geometric graphs, with a Poisson point process as the underlying node distribution and edges depending on the distances between nodes, deterministically or probabilistically. In most cases, the node degrees will be Poisson, but they won't be independent due to the geometric nature of the graph.

In the Erdös-Renyi network model, every possible edge has the same probability of existing. One can be sceptical that this complete randomness describes any real world phenomenon. But to take a closer look, one should look at the relevant properties of the ER model and compare it to those of real communication networks. E.g. we know that the degree distribution is binomial, the clustering coefficient is equal to the edge probability and the average path length - i guess this is relevant for communication networks - is in O(log n). What about the communication networks you study?

Erdös-Renyi network model is a random model that follows certain properties as Christian L. Staudt already pointed out. There are others network random models based on the degree distribution (or, more precisely, degree sequence) as Molloy-Reed model (you can find it in any web search page easily). The problem is, are you searching for a model to describe some network that you already observed, or you would like to build a network using this model? In the first case is better to study its main properties (e.g., degree distribution, clustering and average neighbours degree, k-cores, etc.). In the second case, wireless networks in the field (i.e., free of the obstructions) can be posible Erdös-Renyi graph when all nodes are within the coverage range (which is not so much expected).

As the earlier responders said - it depends (first) on the degree distribution of the system you are attempting to model. If it's a (shifted) power-law, then scale-free (but maybe not necessarily preferential attachment) is the way to go. If it is binomial, then sure, why not a Erdös-Renyi network model. For communication networks it will also depend (second) on the fact that most such real communication networks exist in $\R^3$ (or even $\R^2$) and therefore there is additional structure (e.g. clustering) - technological considerations also enforce certain other constraints (e.g. disassortativity). For communication systems, don't underestimate the fact that you are dealing with a physical system of attenuated communication channels. Then, (third) if you're thinking about *mobile* telephony...

Some wireless networks (sensor networks, ad hoc networks) are best modeled as random geometric graphs, with a Poisson point process as the underlying node distribution and edges depending on the distances between nodes, deterministically or probabilistically. In most cases, the node degrees will be Poisson, but they won't be independent due to the geometric nature of the graph.

As others have said, Erdos-Renyi random graphs are unlikely to occur in communication networks because they assume independence of the edge existence probabilities. I can't think of a situation / protocol which would achieve that. You might find this paper interesting - it shows that a network where each pair of nodes is connected with probability p (e.g. they share a key with probability p) is more highly connected than the ER random graph G(n,p).

https://www.researchgate.net/publication/1924327_Connectivity_of_the_Uniform_Random_Intersection_Graph?ev=prf_pub

Article Connectivity of the Uniform Random Intersection Graph

I think this might shed some useful information on whether networks can be modeled on the Erdos-Renyi model and can be scale free or not. It is basically a review paper, but has some interesting points that can be made use of.

http://www.sigcomm.org/sites/default/files/ccr/papers/2010/October/1880153-1880155.pdf