## Project Details

### Description

A first aim focuses on Signal Detection:

Signal detection mainly tries to partition the measured spectrum in frequency intervals where signal (meaningful information) is found and frequency intervals where noise (meaningless information) only has been measured. A standard technique (referred to as the energy detection method) to solve this problem is measuring the entire frequency band of interest and apply a statistical test to every frequency bin to assess the presence of signal. Although this technique works, it is considered to be slow and not very accurate as the same test is re-used for every frequency bin.

This project applies discriminant analysis for signal detection in such a way that the classical assumptions may be relaxed to meet practical measurement conditions. Indeed, we strive to modify the technique such that the advantages are: (a) no user interaction required, (b) no prior knowledge (fully blind) needed, (c) noise power is allowed to be unknown and (d) real-time analysis is possible.

A second aim deals with characterization of the probability density function of disturbing noise in measured signals:

To characterize the probability distribution of measured signals, the methodology used in practice is nearly restricted to the use of the histogram, which has a long and successful past. It is well known that it completely describes the empirical distribution of the measured signal. However alternatives to the histogram although less popular exist, still they did not found a breakthrough to the engineer's toolbox.

This project will specifically deal with such practical robustness issues in a formal way why these alternatives are not frequently used. Hence, a theoretical framework must be developed in which the finite-sample properties of these alternative techniques can be investigated. Upon exploring the practical properties, the existing techniques may be optimized to better serve the measurement community. Hence, this project's objective is to reintroduce these alternatives of the histogram to the measurement engineer's toolbox.

Signal detection mainly tries to partition the measured spectrum in frequency intervals where signal (meaningful information) is found and frequency intervals where noise (meaningless information) only has been measured. A standard technique (referred to as the energy detection method) to solve this problem is measuring the entire frequency band of interest and apply a statistical test to every frequency bin to assess the presence of signal. Although this technique works, it is considered to be slow and not very accurate as the same test is re-used for every frequency bin.

This project applies discriminant analysis for signal detection in such a way that the classical assumptions may be relaxed to meet practical measurement conditions. Indeed, we strive to modify the technique such that the advantages are: (a) no user interaction required, (b) no prior knowledge (fully blind) needed, (c) noise power is allowed to be unknown and (d) real-time analysis is possible.

A second aim deals with characterization of the probability density function of disturbing noise in measured signals:

To characterize the probability distribution of measured signals, the methodology used in practice is nearly restricted to the use of the histogram, which has a long and successful past. It is well known that it completely describes the empirical distribution of the measured signal. However alternatives to the histogram although less popular exist, still they did not found a breakthrough to the engineer's toolbox.

This project will specifically deal with such practical robustness issues in a formal way why these alternatives are not frequently used. Hence, a theoretical framework must be developed in which the finite-sample properties of these alternative techniques can be investigated. Upon exploring the practical properties, the existing techniques may be optimized to better serve the measurement community. Hence, this project's objective is to reintroduce these alternatives of the histogram to the measurement engineer's toolbox.

Acronym | GIFT137 |
---|---|

Status | Finished |

Effective start/end date | 1/09/12 → 1/09/19 |

### Keywords

- Automatic Measurement Systems
- Nonlinear Modelling
- Microwaves
- Parameter Estimation
- Instrumentation
- System Identification
- Electrical Measurements
- Telecommunications
- Nonlinear Measurements
- Electromagnetism