Accurate channel estimation (CE) techniques are highly desired to achieve the spatial diversity that is supported by multiple relays. This paper investigates a disintegrated CE technique that is unavoidably required for the space-time coding in the cooperative relay network and that can effectively work in a scalable relay network with a fixed protocol. The filter-and-forward (FF) relaying strategy is studied herein with a superimposed training method to estimate channels on different hops. To overcome disadvantage of the interference suppression method in a conventional FF relay network, a generalized filtering technique is proposed to coordinate multi-relay transmissions for reducing inter-relay interference (IRI). The generalized filtering matrix can effectively multiplex the training sequences from different relays to the destination via frequency-division multiplexing (FDM), time-division multiplexing (TDM) and code-division multiplexing (CDM) methods; unlike a conventional beamforming method being seriously constrained on the size of a relay network in which only 2 relays in dual hops are present. The improvements attained using the proposed technique are confirmed comprehensive computer simulations.