A blind classification processor for cognitive radios is realized in 40nm CMOS, featuring three-step parameter estimation for a 59× energy saving compared to an exhaustive method, and multi-algorithm feature extraction to distinguish five modulation classes: multicarrier, single-carrier PSK/QAM/MSK, and spread-spectrum signals. The chip consumes 17µJ within 2ms sensing time per classification, achieving ≥95% detection probability (PD) and ≤0.5% false-alarm rate (PFA) at 10dB SNR in a 500MHz channel.