Gradient-Based Real-Time Spectrum Sensing at Low SNR One of the main steps of enabling dynamic spectrum allocation for cognitive radios is spectrum sensing. Two critical problems in sensing a wide bandwidth are variations in noise power and degraded performance at a low SNR. The common FCME approach may lead to missed detection in sensing a wideband. The improved gradient-based energy detection method proposed here satisfactorily addresses the problem of choosing an appropriate threshold in such a scenario. The sensing bandwidth is divided into smaller subbands, and the gradient of the mean of subbands (GMSB) is computed and summed over past data samples in each subband. Applying a histogram-based threshold on the GMSB enables the accurate distinction of signal and noise without perfect knowledge of the noise floor. The theory of the proposed algorithm has been validated with numerical simulations over a wide range of low SNR values at different signal bandwidths. The detection performance of the approach is more effective than the FCME. The technique is tested over real-time wideband sensing in the GSM and CDMA bands.