Is it possible to implement either a reliable or an unreliable (process) failure detector using an unreliable communication channel?

What will be an ideal response?

An unreliable failure detector can be constructed from an unreliable channel – all that changes from use of a reliable channel is that dropped messages may increase the number of false suspicions of process failure.

A reliable failure detector requires a synchronous system. It cannot be built on an unreliable channel since a dropped message and a failed process cannot be distinguished – unless the unreliability of the channel can be masked while providing a guaranteed upper bound on message delivery times. A channel that dropped messages with some probability but, say, guaranteed that at least one message in a hundred was not dropped could, in principle, be used to create a reliable failure detector.