Physical layer models for node cooperation study small groups of nodes whose operation is analyzed and optimized in isolation from the rest of the wireless network. Current analytical tools provide limited insight into how such techniques impact large wireless networks, where interference from other nodes in the network is a significant concern. Consequently, very little is known about how these techniques impact higher layers, which, in practice, manage this interference. A recurring theme in many such techniques is signal superposition, where different transmissions on a common communication medium mutually interfere at one or more receivers. Each receiver recovers its message(s) of interest by optimally exploiting its knowledge of the codebooks of different interferers. Carefully designed signal superposition techniques are in fact optimal for certain types of one-to-many ("broadcast" ), many-to-one ("multiple-access" ) and certain cases of relay-aided communication. These techniques stand in contrast to more traditional orthogonal schemes that are designed specifically to avoid such interference. We examine signal superposition strategies in two canonical cases: broadcast and multiple-access. Our investigations involve a combination of theoretical analysis and experimental prototyping. In our theoretical study we employ tools from stochastic geometry to analyze the medium access problem in networks composed of node clusters with local broadcast or multiple-access influencing each another through interference. We show that in networks composed of many randomly-placed clusters, each with local broadcast or multiple-access, orthogonal schemes offer useful properties such as their flexibility in adapting their spatial re-use to each receiver (broadcast) or a smaller spatial contention (multiple-access). We show how a single broadcast cluster can be realized by designing the first known prototype of a superposition-coded wireless system using off-the-shelf channel codes and experimentally demonstrate the spectral efficiency gains over time division multiplexing for a fixed error rate. Furthermore, we use this prototype to show that the coding gain (rather than the spectral efficiency gain) from superposition codes significantly improves link reliability without the need to increase transmit power or bandwidth, opening up the possibility of novel medium access protocols that can leverage superposition codes.