Abstract
Perfect signal recovery from discretized continuous signals has been studied under the name of the Nyquist-Shannon sampling theorem. If a signal is sufficiently sparse in a given representation domain, only a fraction of the original measurement samples can be randomly kept: this sparse vector recovery problem can be solved using Compressed Sensing (CS), which first appeared in mid-2000 and has grown in interest, reaching countless fields and creating new development paths future applications.
In this PhD research we investigate CS within defense in the GHz frequency band. This work will generalize the results of CS in terms of performance and implementation. CS is a bridge between a mathematical reconstruction problem and an applied, computationally affordable solution for numerous applications. This is interesting for ultra-wideband sensors creating large data sets. Collecting or handling large amounts of data are onerous tasks, maybe affecting the system’s performance. Three applications are studied in this PhD: non-destructive testing (NDT) for defect detection, ground penetrating radars (GPR) for landmine detection, and passive coherent location (PCL) using passive opportunistic radars for airspace surveillance.
This research has found that for NDT of certain materials using step-by-step frequency sweeps, the number of collected frequencies can be reduced down to only 20% of the original measurement volume achieving simplified and immediate reconstructions thanks to the CS algorithm known as orthogonal matching pursuit and dictionary modelling. This implies that accurate NDT measurements in the GHz band can be performed in reduced acquisition times, hence increasing the efficiency of testing or maintenance tasks.
An extension to a second (or potentially a third) measurement dimension is also possible as it is the case of GPR, whose sensor moves linearly on a platform. Additionally to reducing the frequency steps to be kept, CS can also discard complete measurements in the platform’s path and be able to reconstruct the scene satisfactorily using basis pursuit denoising. For large targets, we have found that the frequency subsampling can be reduced to only 5% of the original data or 10% of the scanning positions. For smaller targets, these figures change 15% and 20%, respectively, in combination with a modeling that reduces the effects of all intrinsic antenna-ground reverberations.
Finally, we have found that applying CS on DVB-t echoes coming back from moving targets reduces the typically huge data volumes, simplifies the reconstructed scene, enables using more channels -increasing the system’s range resolution- or possibly using multiple receivers scattered around the surveilled area for more precise target location (multi-lateration). The obtained results show that CS reduces the received data down to 0.1% for longer DVB-t sequences. Moreover, side information obtained from previous target detections can be exploited to reduce even further the amount of data needed to achieve a reconstruction.
In this PhD research we investigate CS within defense in the GHz frequency band. This work will generalize the results of CS in terms of performance and implementation. CS is a bridge between a mathematical reconstruction problem and an applied, computationally affordable solution for numerous applications. This is interesting for ultra-wideband sensors creating large data sets. Collecting or handling large amounts of data are onerous tasks, maybe affecting the system’s performance. Three applications are studied in this PhD: non-destructive testing (NDT) for defect detection, ground penetrating radars (GPR) for landmine detection, and passive coherent location (PCL) using passive opportunistic radars for airspace surveillance.
This research has found that for NDT of certain materials using step-by-step frequency sweeps, the number of collected frequencies can be reduced down to only 20% of the original measurement volume achieving simplified and immediate reconstructions thanks to the CS algorithm known as orthogonal matching pursuit and dictionary modelling. This implies that accurate NDT measurements in the GHz band can be performed in reduced acquisition times, hence increasing the efficiency of testing or maintenance tasks.
An extension to a second (or potentially a third) measurement dimension is also possible as it is the case of GPR, whose sensor moves linearly on a platform. Additionally to reducing the frequency steps to be kept, CS can also discard complete measurements in the platform’s path and be able to reconstruct the scene satisfactorily using basis pursuit denoising. For large targets, we have found that the frequency subsampling can be reduced to only 5% of the original data or 10% of the scanning positions. For smaller targets, these figures change 15% and 20%, respectively, in combination with a modeling that reduces the effects of all intrinsic antenna-ground reverberations.
Finally, we have found that applying CS on DVB-t echoes coming back from moving targets reduces the typically huge data volumes, simplifies the reconstructed scene, enables using more channels -increasing the system’s range resolution- or possibly using multiple receivers scattered around the surveilled area for more precise target location (multi-lateration). The obtained results show that CS reduces the received data down to 0.1% for longer DVB-t sequences. Moreover, side information obtained from previous target detections can be exploited to reduce even further the amount of data needed to achieve a reconstruction.
Original language | English |
---|---|
Qualification | Doctor of Engineering Sciences |
Awarding Institution |
|
Supervisors/Advisors |
|
Award date | 13 Nov 2018 |
Place of Publication | Brussels |
Publication status | Published - 2018 |