Both the Advanced Mobile Phone System (AMPS) and the narrowband version (NAMPS) use frequency modulation (FM) in a frequency division duplex mode. Because such systems typically use the same frequency bands in nearby cells, co-channel interference can be a significant problem. A typical digital implementation uses a quadrature demodulation algorithm to estimate the instantaneous frequencies of FM signals and recover the original modulating signal. The use of parametric spectral estimation techniques can often provide higher resolution frequency estimates for closely spaced signals, such as co-channel FM signals. A parametric method based on forward-backward linear prediction is implemented to create a model-based demodulator for AMPS and NAMPS cellular signals. This algorithm demodulates the signal by estimating the instantaneous frequency of the principle component present. In addition, improved resistance to noise and co-channel interference is gained by performing a singular value decomposition on the signal data and then estimating the parameters of the model using only the desired components of the decomposition. Examples are provided showing the performance of the demodulator in various signal environments. Experimental results show 5 to 22 dB improvement in demodulator output mean-square-error relative to a quadrature demodulator for various channel conditions including Rayleigh fading. The ability of a demodulator to reject severe co-channel interference and channel noise could allow significant increases in cellular system capacity and performance. / Master of Science