Network coding can remarkably improve the capacity of a network by combining the received packets at the relay or intermediate nodes. However, this mixing or combining nature of network coding makes the network particularly vulnerable to a packet pollution attack (Byzantine), which injects false information into the network. This false information is then combined with other packets with network coding to pollute the coded packet. In this thesis, first, we investigate the limitations of the existing schemes used for defense against packet pollution attacks, and analyze a recently proposed recycling polluted packet scheme, which is promising for defense against these attacks. This scheme recycles the polluted packet instead of discarding the entire polluted packet, which results in throu...[ Read more ]

Network coding can remarkably improve the capacity of a network by combining the received packets at the relay or intermediate nodes. However, this mixing or combining nature of network coding makes the network particularly vulnerable to a packet pollution attack (Byzantine), which injects false information into the network. This false information is then combined with other packets with network coding to pollute the coded packet. In this thesis, first, we investigate the limitations of the existing schemes used for defense against packet pollution attacks, and analyze a recently proposed recycling polluted packet scheme, which is promising for defense against these attacks. This scheme recycles the polluted packet instead of discarding the entire polluted packet, which results in throughput improvement. However, the framework of this scheme is not accurate for short block length, which results in loose bounds on the probability of detection error, decoding error and throughput on short block length. In this thesis, we apply new bounds to make it rigorous, so that it works well for finite block lengths. We present the new and tightened bounds on the probability of detection error, decoding error and throughput. The previous bounds were derived by assuming the use of random codes, which shed light on the theoretically achievable performance. To evaluate the practically achievable performance, we apply practical convolutional codes on the recycling approach and compare the derived performance bounds with those of the random codes.