Source Number Estimation in Array Signal Processing – AIC Method Comparison

Source number estimation in array signal processing represents a critical challenge within signal analysis. Widely adopted methodologies include AIC, MDL, HQ, and EDC. The AIC method employs information-theoretic principles, conducting model selection by minimizing both residual sum of squares and a penalty term. In code implementations, this typically involves computing eigenvalues from the sample covariance matrix and identifying the minimum AIC value across candidate source counts. The MDL approach follows Bayesian criteria, selecting models through minimal data encoding length calculations, where the penalty term increases logarithmically with sample size to ensure consistency. HQ method serves as an enhanced version of AIC, particularly effective for small-sample scenarios through modified penalty coefficients that reduce overestimation tendencies. EDC technique operates on minimum description length principles, determining optimal source numbers by minimizing the required bits for signal description, with implementations often featuring flexible penalty functions adaptable to various data conditions. Through comprehensive comparison of these methods' strengths and limitations, practitioners can select the most appropriate source number estimation technique based on specific application requirements, such as array configuration, signal-to-noise ratio, and computational constraints.