This paper presents comparison of real-time fault signature analysis algorithms for embedded systems used in permanent magnet (PM) motor drives. The proposed algorithms are used to identify key features of every single signature in motor variables such as amplitude and phase. Three methods including Adaptive-Linear-Neuron (ADALINE), Goertzel and fault (reference) frame transformation are discussed and implementation details, CPU utilization, implementation time etc are summarized. It is shown that the fault frame transformation has the best performance and lowest CPU utilization. ADALINE also has good accuracy and calculates all signatures' amplitude and phase simultaneously whereas other methods calculate every single component one by one. The Goertzel algorithm needs higher number of samples to provide acceptable accuracy. All of these methods are developed in AC drive controller and tested on a PM motor with magnet defect fault. Comparative simulations and experimental results justify the theoretical analysis and provide comparative evaluation of the proposed algorithms.