Manual visual inspection |
Inspection of scintillation indices and other observables (Jiao, Hall, & Morton, 2017a; Linty, Farasin, Favenza, & Dovis, 2019; Xu, Morton, Akos & Walter, 2015). | Gives the ultimate accuracy and reliability and perfectly targets the scope of the analysis; allows an easy cross analysis of historical data and of any external aiding. | Performance depends on skills and experience of the operator; subject to human errors; requires costly and time consuming human effort; not automatic. | ✗ |
Threshold based |
Scintillation indices threshold trigger (Dubey, Wahi, & Gwal, 2006; Linty, Dovis, & Alfonsi 2018; Linty et al. 2019; Taylor, Morton, Jiao, Triplett, & Pelgrum, 2012). | Very simple implementation; low computational burden. | Low detection accuracy; vulnerable to false alarms due to multipath; requires ISMR or detrending and filtering algorithms. | ✓ |
Addition of elevation mask (Abadi, Saito, & Srigutomo, 2014; Favenza, Farasin, Linty, & Dovis, 2017) and C/N0 (Curran, Bavaro, Morrison, & Fortuny, 2015b). | Simple; capable of reducing the false alarms due to multipath. | Significant risk of discarding important measurements and to miss the detection of events. | ✓ |
Linear combination of scintillation indices and signal observables (Pelgrum et al., 2011; Vikram et al., 2011). | Simple; able to eliminate multipath. | Non-trivial tuning of parameters; limited scalability; non-neglibigle missed detection rate. | ✓ |
Elevation-azimuth masks (Atilaw, Cilliers, & Martinez, 2017; Spogli et al., 2014). | Very good multipath rejection capabilities. | Long and complex preparation phase; low scalability, tuning bounded to the specific location. | ✓ |
Non-indices based techniques |
Wavelet decomposition and transform based techniques (Fu, Han, Rizos, Knight, & Finn, 1999; Mushini, Jayachandran, Langley, MacDougall, & Pokhotelov, 2012; Ouassou, Kristiansen, Gjevestad, Jacobsen, & Andalsvik, 2016). | Overcome the problem of detrending using Butterworth filters; enhanced performance especially for phase scintillation; alternative scintilltion indices are derived. | Computationally expensive; complex implementation; require phase measurements. | ✗ |
Open-loop receivers (Romero et al., 2016). | Overcome the problem of detrending; does not require the tracking loops to be in lock condition. | Requires specific receiver implementation. | ✗ |
Machine learning based |
SVM, Fourier Transform of signal time series (Jiao, Hall, & Morton, 2017b; Jiao, Hall, & Morton, 2017c). | High accuracy; resembles manual annotation by design; fully automatic; relies on common GNSS observables. | Requires large set of labeled data for the training phase; requires predetermined elevation mask. | ✓ |
Decision Tree and correlator outputs (Linty et al., 2019). | Very high accuracy; resembles manual annotation by design; fully automatic; early detection of events; high-rate results. | Requires large set of labeled data for the training phase; vulnerable to the problem of over-fitting if the features are not properly chosen; requires access to correlator outputs. | ✓ |