Authors : Abul Hasnat, Dibyendu Barman and Satyendra Nath Mandal
Abstract: In symmetric key encryption, no matter how strong the encryption technique is there is always a great concern about the trust worthiness of the ‘key-exchange’ process and the information may be compromised once the key is exposed to intruder. This study proposes a neuro-symmetric block cipher where the key is not shared but a “weight vector” and key input file are exchanged. “Weight vector” is the updated weights of a trained artificial neural network using key input file as input and key target file as output. Turn and mix functions are used to generate 12 different keys at twelve different rounds using the key target file. During encryption, turn, shift and mix functions and XOR operations are applied between the key of first round with plain text. This process is repeated eleven rounds more on intermediate cipher text and key of the respective round to produce the cipher text. An intruder cannot decrypt the cipher text having only key-input file or the weight matrix or both until and unless he knows all of the key-input file, weight vector and structure of ANN (it is not exchanged over network) at the same time. Experimental result shows that the proposed approach is more secure in terms of cryptanalysis. For performance analysis, all statistical tests suggested by NIST and FIPS PUB-140-1 test battery have been applied on the cipher text produced by proposed algorithm and the algorithm shows reasonable good response. Finally, a comparison study is given between the popular symmetric key algorithms and the proposed algorithm. Experimental results shows that the proposed approach is more robust compared to the conventional symmetric key encryption algorithms.
Abul Hasnat, Dibyendu Barman and Satyendra Nath Mandal, 2017. A Novel Neuro-Symmetric Block Cipher. Journal of Engineering and Applied Sciences, 12: 8309-8316.