A hardware implementation of an autoregressive algorithm
Michael R. Smith, Theodorus J. Smit, Stacy William Nichols, S.T. Nichols, Hakan Orbay, K Campbell · Measurement Science and Technology · 1990
With short length data records, high-resolution spectral estimates can be obtained by constructing a model of the signal and using the model coefficients to calculate the spectrum. The bottlenecks associated with implementing modelling algorithms in hardware were identified using a test bed based on a general-purpose microprocessor. A comparison is made of the architecture required and speed obtained when the Burg autoregressive modelling algorithm is implemented on high-speed multiprocessor systems. The systems compared were the microprogrammable AMD29500 byte slice digital signal processor chips and the NEC mu PD77230 single-chip digital signal processor. Applications in the area of magnetic resonance imaging are discussed. Prototype systems would allow a sixteenth-order autoregressive model to be performed on a 256*128 medical image in 1 to 3 seconds.