A novel robust feature of modulation classification for reconfigurable software radio

The key features for modulation classification must be selected. In this paper, we proposed the novel key feature, occupied bandwidth, which has robust properties of sensitive with modulation types and insensitive with SNR variation. In this paper, the numerical simulations of the five types of modulation classifiers are performed, and the prototype modulation classifier implemented as part of reconfigurable software radio evaluated using field trials. The numerical simulations for classifying both analog and digital modulation types using the same seven features including the proposed a novel feature are performed. Existing technology is able to classify reliably (accuracy >= 90%) only at SNR above 10 dB when a large range of modulation types including both digital and analog is being considered. The simulation result indicated that all five classifiers showed performance enhancement over the 95% of correct classification at the SNR of 10 dB. Especially, it was shown that minimum distance classifier and support vector classifier can achieve the probability of correct classification of 96.2% and 98.3% at the SNR of 5 dB, respectively. These good results for signals with low SNR (below 10 dB) came from the large effects of robust properties of a novel feature. The prototype of modulation classifier was implemented with one of decision tree classifiers, because of the requirement of fast processing in reconfigurable software radio. With the result of field trials, we confirmed that the performance in the prototype of modulation classifier was agreed with the numerical simulation result of decision tree classifier