We present "concatenated" network coding technique that incorporates multi-level coding across a network of relays in multi-user, multi-hop noisy wireless networks. Parity bits are generated by the relays based on the noisy codewords of previous nodes and transmitted to the next nodes until the information reaches its final destination. A key feature of the proposed approach is that more powerful code can be constructed, hence more errors can be corrected, as the information travels through the network. It can provide a large diversity gain for a large set of sources (users) with limited number of relays and simple decoding complexity at the destination. The probability of decoding error and normalized throughput are analyzed, and the optimal network topology and energy allocation that maximizes the normalized throughput is investigated. Asymptotic number of relays per source node for maximizing the normalized throughput is analyzed.