Context. Satellite communication foresees the operation at higher frequencies, thanks to higher data-rates and BW. Current systems operate mainly at Ka-band, with V and W likely to be the next step. Because of this, the necessity to retrieve correctly attenuation profiles arises, with rain assuming a relevant role. Current models are characterized either by limited validity (i.e. Drufuca) in frequencies or the need of a physical-based breakdown of components (i.e. TAFS). Aims. The main goal is to assess the capability of high-f attenuation prediction via fre- quency scaling exploiting artificial neural networks (ANNs), trying to avoid the need of more theoretical approaches. Methods. After preprocessing, multiple fully connected neural networks have been devel- oped using Matlab function fitrnet, in every frequency scaling case (distinguished by starting and target frequency). Rain rate, and meteo data have also been used. Finally, it has been developed both a single configuration and a dual one, which handles the data differently according to the SSI parameter. Results. Neural network performed a good prediction of the upper attenuation, with accu- racy increasing with the addition of meteorological data. Also, train/test error remained consistently low (i.e. good overall fit). Further improvements were obtained thanks to dual networks, which handled better the rainy conditions.
Contesto. L’incremento di data-rate e BW richiede l’uso di frequenze operative sempre più alte nel campo delle telecomunicazioni satellitari. I sistemi attuali operano in nella banda Ka, mentre si prospetta un passaggio nelle bande V e W. Per questa ragione, la necessità di prevedere correttamente i profili di attenuazione è focale, con la pioggia che ha una particolare rilevanza. I modelli correnti sono limitati o dalla validità ristretta in frequenza (ex. Drufuca) o devono analizzare ogni contributo singolarmente (ex. TAFS). Obiettivi. L’obiettivo è valutare le capacità di predizione dell’attenuazione, tramite fre- quency scaling, tramite gli ANN, cercando di eliminare approcci più teorici. Metodi. Dopo il preprocessing, network multipli sono stati sviluppati con la funzione di Matlab fitrnet, in ogni combinazione adottata (distinte dall’attenuazione di partenza e di arrivo). Inoltre, sono stati usati dati meteo e rateo di pioggia. Infine, sono stati sviluppati sia un network con una configurazione unica che uno doppio, il quale analizza i dati in modo differentemente a seconda del valore di SSI. Risultati. Il network si è dimostrato in generale valido nella predizione della attenuazione a frequenza maggiore, con l’accuratezza incrementata qualora si aggiungessero meteo e rateo di pioggia come input, con errori della macchina consistentemente bassi. Ulteriori incrementi in prestazioni sono stati ottenuti grazie all’uso del network doppio.
Tropospheric attenuation frequency scaling at EHF: a machine learning approach
Murer, Riccardo Giovanni
2023/2024
Abstract
Context. Satellite communication foresees the operation at higher frequencies, thanks to higher data-rates and BW. Current systems operate mainly at Ka-band, with V and W likely to be the next step. Because of this, the necessity to retrieve correctly attenuation profiles arises, with rain assuming a relevant role. Current models are characterized either by limited validity (i.e. Drufuca) in frequencies or the need of a physical-based breakdown of components (i.e. TAFS). Aims. The main goal is to assess the capability of high-f attenuation prediction via fre- quency scaling exploiting artificial neural networks (ANNs), trying to avoid the need of more theoretical approaches. Methods. After preprocessing, multiple fully connected neural networks have been devel- oped using Matlab function fitrnet, in every frequency scaling case (distinguished by starting and target frequency). Rain rate, and meteo data have also been used. Finally, it has been developed both a single configuration and a dual one, which handles the data differently according to the SSI parameter. Results. Neural network performed a good prediction of the upper attenuation, with accu- racy increasing with the addition of meteorological data. Also, train/test error remained consistently low (i.e. good overall fit). Further improvements were obtained thanks to dual networks, which handled better the rainy conditions.File | Dimensione | Formato | |
---|---|---|---|
Thesis_991790_Murer.pdf
accessibile in internet per tutti
Dimensione
6.85 MB
Formato
Adobe PDF
|
6.85 MB | Adobe PDF | Visualizza/Apri |
Executive_Summary_991790_Murer.pdf
accessibile in internet per tutti
Dimensione
2.45 MB
Formato
Adobe PDF
|
2.45 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/235993