Limits...
Sparse Recovery Optimization in Wireless Sensor Networks with a Sub-Nyquist Sampling Rate.

Brunelli D, Caione C - Sensors (Basel) (2015)

Bottom Line: Compressive sensing (CS) is a new technology in digital signal processing capable of high-resolution capture of physical signals from few measurements, which promises impressive improvements in the field of wireless sensor networks (WSNs).In this work, we extensively investigate the effectiveness of compressive sensing (CS) when real COTSresource-constrained sensor nodes are used for compression, evaluating how the different parameters can affect the energy consumption and the lifetime of the device.The results are verified against a set of different kinds of sensors on several nodes used for environmental monitoring.

View Article: PubMed Central - PubMed

Affiliation: University of Trento, Via Sommarive 9, Trento I-38122, Italy. davide.brunelli@unitn.it.

ABSTRACT
Compressive sensing (CS) is a new technology in digital signal processing capable of high-resolution capture of physical signals from few measurements, which promises impressive improvements in the field of wireless sensor networks (WSNs). In this work, we extensively investigate the effectiveness of compressive sensing (CS) when real COTSresource-constrained sensor nodes are used for compression, evaluating how the different parameters can affect the energy consumption and the lifetime of the device. Using data from a real dataset, we compare an implementation of CS using dense encoding matrices, where samples are gathered at a Nyquist rate, with the reconstruction of signals sampled at a sub-Nyquist rate. The quality of recovery is addressed, and several algorithms are used for reconstruction exploiting the intra- and inter-signal correlation structures. We finally define an optimal under-sampling ratio and reconstruction algorithm capable of achieving the best reconstruction at the minimum energy spent for the compression. The results are verified against a set of different kinds of sensors on several nodes used for environmental monitoring.

No MeSH data available.


Related in: MedlinePlus

Number of CPU cycles required to compress a sample using different random Φ matrices varying the compression factor. (T1) Matrix with random 16-bit fixed-point values. (T2) Gaussian matrix generated using a Box-Muller transformation with mean zero and variance 1/M. (T3) Matrix with random floating point values. (T4) Same as T2, but the matrix is generated with the Ziggurat method. (T5) Entries of the matrix are generated from the symmetric Bernoulli distribution with . (T6) Same as T5 with P(Φjk = ±1) = 1/2.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4541899&req=5

f1-sensors-15-16654: Number of CPU cycles required to compress a sample using different random Φ matrices varying the compression factor. (T1) Matrix with random 16-bit fixed-point values. (T2) Gaussian matrix generated using a Box-Muller transformation with mean zero and variance 1/M. (T3) Matrix with random floating point values. (T4) Same as T2, but the matrix is generated with the Ziggurat method. (T5) Entries of the matrix are generated from the symmetric Bernoulli distribution with . (T6) Same as T5 with P(Φjk = ±1) = 1/2.

Mentions: In Figure 1, the number of cycles required by a microcontroller to generate the compression matrix and to perform the compression of a single sample for different kinds of measurement matrices is shown. The differences are mainly due to: (1) the computational workload required for generating the random vectors for the compression, since in some cases, the generation implies the use of complex and computationally-intensive functions, such as sqrt or log; and (2) the time spent in multiplication of the vector against the sample that, especially in the case of floating point numbers, is not negligible.


Sparse Recovery Optimization in Wireless Sensor Networks with a Sub-Nyquist Sampling Rate.

Brunelli D, Caione C - Sensors (Basel) (2015)

Number of CPU cycles required to compress a sample using different random Φ matrices varying the compression factor. (T1) Matrix with random 16-bit fixed-point values. (T2) Gaussian matrix generated using a Box-Muller transformation with mean zero and variance 1/M. (T3) Matrix with random floating point values. (T4) Same as T2, but the matrix is generated with the Ziggurat method. (T5) Entries of the matrix are generated from the symmetric Bernoulli distribution with . (T6) Same as T5 with P(Φjk = ±1) = 1/2.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4541899&req=5

f1-sensors-15-16654: Number of CPU cycles required to compress a sample using different random Φ matrices varying the compression factor. (T1) Matrix with random 16-bit fixed-point values. (T2) Gaussian matrix generated using a Box-Muller transformation with mean zero and variance 1/M. (T3) Matrix with random floating point values. (T4) Same as T2, but the matrix is generated with the Ziggurat method. (T5) Entries of the matrix are generated from the symmetric Bernoulli distribution with . (T6) Same as T5 with P(Φjk = ±1) = 1/2.
Mentions: In Figure 1, the number of cycles required by a microcontroller to generate the compression matrix and to perform the compression of a single sample for different kinds of measurement matrices is shown. The differences are mainly due to: (1) the computational workload required for generating the random vectors for the compression, since in some cases, the generation implies the use of complex and computationally-intensive functions, such as sqrt or log; and (2) the time spent in multiplication of the vector against the sample that, especially in the case of floating point numbers, is not negligible.

Bottom Line: Compressive sensing (CS) is a new technology in digital signal processing capable of high-resolution capture of physical signals from few measurements, which promises impressive improvements in the field of wireless sensor networks (WSNs).In this work, we extensively investigate the effectiveness of compressive sensing (CS) when real COTSresource-constrained sensor nodes are used for compression, evaluating how the different parameters can affect the energy consumption and the lifetime of the device.The results are verified against a set of different kinds of sensors on several nodes used for environmental monitoring.

View Article: PubMed Central - PubMed

Affiliation: University of Trento, Via Sommarive 9, Trento I-38122, Italy. davide.brunelli@unitn.it.

ABSTRACT
Compressive sensing (CS) is a new technology in digital signal processing capable of high-resolution capture of physical signals from few measurements, which promises impressive improvements in the field of wireless sensor networks (WSNs). In this work, we extensively investigate the effectiveness of compressive sensing (CS) when real COTSresource-constrained sensor nodes are used for compression, evaluating how the different parameters can affect the energy consumption and the lifetime of the device. Using data from a real dataset, we compare an implementation of CS using dense encoding matrices, where samples are gathered at a Nyquist rate, with the reconstruction of signals sampled at a sub-Nyquist rate. The quality of recovery is addressed, and several algorithms are used for reconstruction exploiting the intra- and inter-signal correlation structures. We finally define an optimal under-sampling ratio and reconstruction algorithm capable of achieving the best reconstruction at the minimum energy spent for the compression. The results are verified against a set of different kinds of sensors on several nodes used for environmental monitoring.

No MeSH data available.


Related in: MedlinePlus