Conclusions
Conclusions |
|
ContentsHome
|
With the "aaah" signal, the amplitude of the transmitted signal was decreased by half, translating to a one bit reduction per transmitted sample. The "ooh" signal prediction performed even better, decreasing by 75% or two bits. This was the simplest possible model, there are many ways to improve upon its performance. Within the linear domain there are several other methods, including covariance, modified covariance, and the Burg method. If the mean value of the signal is nonzero, looking at the MA (moving average) part of the ARMA method as well as the auto regressive would improve the prediction. An adaptive filter with error feedback could create a better prediction by looking at previous mispredictions to correct future predictions. The signal could also be analyzed in the frequency domain rather than the time domain to reduce the computational complexity. For a more ambitious improvement, one could try implementing this compression in real time.
ApplicationsLinear signal prediction has several useful applications in many familiar technologies, including seismic prediction of earthquakes, Doppler prediction of storm movement over time, and Radar tracking devices and missile guidance systems for military use. The specific linear prediction application explored here is used in cell phones, voice over IP, and low bandwidth submarine transmission. This form of compression is used primarily in low bandwidth applications, but other compression schemes use similar methods (e.g. Real-Audio, streaming video, and control of Internet traffic) |