Blind modulation classification (MC) is an intermediate step between signal detection and demodulation, with application in both commercial (such as software defined radio) and military communication systems (such as electronic surveillance and electronic warfare). There are two general approaches that can be used for the solution of the MC problem, one is the decision theoretic approach and the other is the statistical pattern recognition approach. Both are under investigation in the CCSPR.

Decision Theoretic Approach
The MC problem is viewed as a multiple-hypothesis testing problem (choosing from a candidate list of modulations, based on the observed waveform), and likelihood techniques are used for its solution. A novel quasi-hybrid likelihood ratio test (qHLRT) classifier is proposed, which rely on low-complexity yet accurate parameter estimators. The qHLRT classifier is implemented with less computational burden than other optimal (ALRT) or sub-optimal (GLRT, HLRT) likelihood based approaches for MC solutions, and still achieves a reasonable performance. Antenna arrays, which exploits spatial diversity, is an effective add-on to improve the MC performance.

Statistical Pattern Recognition Approach
This approach is also considered as feature extraction and decision-making method. Eight-order cyclic cumulant (CC) -based features are investigated for classifying real- and complex-valued constellations. Features based on the n-th order CCs (n=4,6,8), robust to carrier phase and timing offset, as well as frequency offset and phase jitter, are proposed for QAM recognition. The minimum Euclidian distance is also employed for decision.