As far as I know, entropy of a quantized source H(X) can be viewed as the minimum number of bits needed to encode the information. My question is: Is entropy an appropriate metric of real-valued data distortion in a wireless sensor network (WSN)?
For example, there are N deployed temperature sensors in WSN. Due to spatial correlation, the measurements of near sensors have high correlation. Now we assumed the collection of data gathered by all machines follows a multi-variate gaussian distribution. Since gaussian source is continuous, we also assume that each sensor quantizes the measured temperature data with a quantization level \delta. Now the entropy of source can be calculated instead of infinity. Supposed that the resource is limited, we can only choose a subset sensors S to transmit their data, where |S| = M and M
I think this issue has been studied intensively in digitization and encoding of speech signal, audio signal and video signals. There is the waveform encoding where one encodes the signal to restore their waveform. So, they are not only decoded for preserving the entropy but also the redundancy in them.
When compression is under taken then there are lossy compression and nonlossy compression. In lossy compression some data will be discarded.
In nonlossy compression there will be no loss of information.
Compression is achieved by removing the redundancy and preserving the information content. Redundancy is removed by prediction and differential pulse code modulation. implementing such processes can lead to some slight loss in the information content such that the signal quality will suffer and gets worse,.
During transmission the compressed information can be partially lost in form of bit error rate. Here every bit in error will affect the signal quality. So, one has to preserve the information carrying bits from being lost. This is accomplished by channel coding where one adds redundancy again to the information.
I think the most suitable and robust encoding is that based on the waveform encoding. In this case the mean square error can be used to judge the distortion in the signal.
In case of nonwaveform encoding or entropy encoding one has to use the deviation in the information entropy as a measure of the quality of the signal.
Such type of encoding is used in the linear predictive coders for the speech.
Best wishes
I think this issue has been studied intensively in digitization and encoding of speech signal, audio signal and video signals. There is the waveform encoding where one encodes the signal to restore their waveform. So, they are not only decoded for preserving the entropy but also the redundancy in them.
When compression is under taken then there are lossy compression and nonlossy compression. In lossy compression some data will be discarded.
In nonlossy compression there will be no loss of information.
Compression is achieved by removing the redundancy and preserving the information content. Redundancy is removed by prediction and differential pulse code modulation. implementing such processes can lead to some slight loss in the information content such that the signal quality will suffer and gets worse,.
During transmission the compressed information can be partially lost in form of bit error rate. Here every bit in error will affect the signal quality. So, one has to preserve the information carrying bits from being lost. This is accomplished by channel coding where one adds redundancy again to the information.
I think the most suitable and robust encoding is that based on the waveform encoding. In this case the mean square error can be used to judge the distortion in the signal.
In case of nonwaveform encoding or entropy encoding one has to use the deviation in the information entropy as a measure of the quality of the signal.
Such type of encoding is used in the linear predictive coders for the speech.
Best wishes
- Badges
- Science topic