Spatial Filtering in Sensor Networks with Computation Codes

A data collection problem in sensor networks is formulated in which the number of channel uses per source sample is greater than one. An example of this problem is given in which the objective of the data collector is to compute a filtered and downsampled version of the sensor field. In this regime, it is shown that uncoded transmission is not appropriate and that strategies based on separating source and channel coding perform poorly. By using a novel coding strategy based on computation codes, the power-distortion tradeoff becomes more favorable than that from separation.