Signal Reconstruction

Contents Home The Problem The ARMA Model Autocorrelation What We Did The Data Conclusions [A Word on Decoding] Acknowledgements Who we are

After the signal is sent through the channel in the manner described, there is still the task of reconstructing the recorded voice at the other end. Because of time constraints, this project centers on the encoding side alone, which is still enough to require a good understanding of all concepts and procedures involved. The decoding is very similar, however.    The process begins with the error signal that is being received, the difference between the prediction and the original signal. In some implementations, as we have pointed out, it is possible that the system is not using the error, but instead using locally generated white noise with the same variance as the error signal on the encoder side. This is done in cases where quality is not as important as quick transmission; the resultant voice is mostly mechanical, but understandable.    Regardless of the choice, the rest of the procedure is the same. The error or noise is multiplied by the Gain transmitted by the encoder. It is then added to the predicted value of the signal at that point, which has been calculated the same way as before, multiplying the prediction coefficients transmitted by the encoder by the previous values of the final output signal. As in the encoder, the prediction system assumes that the values previous to the first sample are all zero, and starts using the calculated values as they are produced.    In this way, each speech segment is reconstructed, and the signal can then be put back together, recovering the message.

Copyright (c) 2000 by the Oracle Gang.