Error propagation in protocol stacks

Error control is frequently performed at multiple levels of a protocol stack. To better understand the overall performance of an entire stack from the error control perspective, we consider an idealized model consisting of two layers. The lower layer transmits cells directly over a noisy radio channel, modeled as a Gilbert channel. These cells consist of a header (H) that is distinct from the payload (P). A cell is correctly received if and only if both H and P are correct. When cell errors occur, the lower layer attempts to recover the erroneous cells through a selective repeat ARQ protocol. The higher layer protocol entity, called a connection-fragment, is composed of L cells and is successful when all its L cells are correctly received. During a prolonged outage, a large number of consecutive cells may be in error. Perceiving a lack of activity, the higher layer entity resets the connection-fragment after some time. For example, such a run of errors may indicate a loss of sync and induce a reset. We assume that the connection is reset once H+1 header errors have occurred in a row. After some time, the connection-fragment is re-established and all cells, including those that were correctly received in the prematurely aborted previous connection-fragment attempt, are retransmitted. We study the time needed to complete a connection, including packet retransmissions at the lower layer and possible reinitialization of the connection at the higher layer.