Constrained Autoregressive (CAR) Model

In this paper, we propose a novel autoregressive (AR), constrained autoregressive (CAR) model for various signal modeling applications. CAR model is based on constraining one of the model parameters of the autoregressive model. This helps obtain a modified or desired AR spectrum for the signal. Constraining different AR parameters or changing the values of a particular parameter results in dissimilar AR spectrum for the signal. The value of this constrained parameter can be used for externally controlling the gain or improving the spectral resolution between two peaks in the spectrum. In this work, a0 parameter is constrained and different values are assigned for this coefficient. This changes the spectral gain. This CAR model is tested for applications such as pitch detection in a speech signal and detection of QRS complex in an electrocardiogram (ECG) signal. A higher gain CAR error filter improves the efficiency of the pitch detection algorithms or QRS complex detection.