Deterministic Linear Predictors for Bit Rate Improvement

This thesis addresses the application of deterministic linear predictor s for telemetry based systems. Deterministic linear predictors are considered as a viable alternative to block transform coding in order to achieve bit rate reduction. A method of calculating the predictor coefficients of a deterministic linear predictor is presented. The system model is developed parametrically with respect to the bandwidth, polynomial order, signal class and predictor length. Two classes of signals, deterministic and random are analyzed and bounds on the performance of the deterministic linear predictor are established. A family of six pole Butterworth filters are used as presample filters in this study. The performance of the deterministic linear predictor is evaluated in terms of bit rate improvement. For the broad band data, a bit rate improvement of 4 bits has been achieved for f c=0. 20n and N=5. For input signals with varying complexity indices the bit rate improvement of 4 bits is achieved for f c ≤0.20∏ and N=6. It is shown that optimum length and order of the deterministic linear predictor exists to achieve the best bit rate reduction. Also, these relationships are shown to be subject to the bandwidth of the presample filter.