In recent years, thanks to machine learning flourishing, countless studies have attempted to go beyond content-agnostic message transmission and take care of messages’ purpose and semantic meaning. Due to the promised efficiency and reliability advancements that would be obtained by switching to semantic communication, we decided to study V2V communication systems (more specifically, V2V communication for the “cooperative perception” application) to detect possible approaches for semantic empowering. Within the cooperative perception application, systems mostly have two difficulties: sharing the massive captured data and managing data transmissions. Therefore, we studied this application-based communication system from two distinct aspects: semantic-empowered physical layer and network managing. In the scale of the physical layer, our objective was to find the proper goal-oriented encoding/decoding modules that can extract and recover the semantic features. By analysing some proposed DNN learning-based pairs of encoder/decoder that address mentioned challenges, we found the great opportunity of employing these modules in the V2V systems to obtain significant compression rates with the limited accuracy declining. Furthermore, in the scale of the network managing, we proposed a novel factor graph-based planning system that predicts impending collisions by exploiting the spatial grabbed data of the traffic environment. Concerning the network condition, this system employs combinatorial optimization to pick the best massage combinations to avoid potential collisions. Although we designed a program for assessing our method, the complete assessment was not feasible due to the shortage of time. Nevertheless, this study reveals an existing gap between the conventional and goal-based criteria used in network management. Our proposed method does not just have the potential to be executed in real scenarios but also promotes the idea of making goal-oriented strategies. The new intelligent networks, with smart entities and more specific ultimate goals, can utilize the available data in the network to control it more efficiently with respect to communication goals. Indeed, our proposed system evidently identifies this issue and addresses it by presenting an innovative way to make a goal-oriented network.
Negli ultimi anni, grazie allo sviluppo del machine learning, numerosi studi hanno cercato di andare oltre una semplice trasmissione del messaggio al fine di occuparsi anche dell’obiettivo del messaggio e del suo significato semantico. Grazie ai miglioramenti in quanto a efficienza e affidabilità ottenuti da questo switching semantico, abbiamo deciso di studiare i sistemi con comunicazione V2V (in particolare, comunicazione V2V per l’applicazione di percezione cooperativa) al fine di rilevare possibili vie per migliorare la comunicazione tramite la semantica. All’interno dell’applicazione di percezione cooperativa, i sistemi hanno generalmente due difficoltà, cioè condividere la grande quantità di dati e gestire la trasmissione. Quindi, abbiamo studiato questo sistema di comunicazione basato su un’applicazione secondo due aspetti distinti: miglioramento semantico del livello fisico e gestione della rete. Dal punto di vista del livello fisico, il nostro obiettivo era quello di trovare i corretti moduli di encoding e decoding orientati all’obiettivo che possano estrarre e recuperare l’informazione semantica. Analizzando le coppie di encoder/decoder di tipo DNN basato sull’apprendimento che si occupano di questa sfida, abbiamo trovato l’ottima possibilità di utilizzare questi moduli nei sistemi V2V al fine di ottenere una compressione significativa limitando la perdita di accuratezza. Inoltre, per quanto riguarda la gestione della rete, abbiamo proposto un nuovo fattore basato sui grafi per il sistema di planning in grado di predire i prossimi conflitti sfruttando i dati catturati. Considerando le condizioni della rete, questo sistema usa l’ottimizzazione combinatoria per trovare la combinazione di messaggi migliori per evitare dei probabili conflitti. Siamo riusciti a scrivere un programma per provare il nostro metodo; tuttavia, il lavoro finale non è ancora completo per mancanza di tempo. Nonostante tutto, questo studio mostra un’evidente differenza tra il metodo convenzionale e quello basato sull’obiettivo per gestire la condivisione di messaggi in una rete. Il metodo proposto da noi non ha soltanto la possibilità di esecuzione in uno scenario reale, ma cerca anche di promuovere l’idea di considerare gli obiettivi durante la comunicazione. Le nuove reti intelligenti, con eventualmente degli obiettivi specifici, possono gestire in modo più efficiente le risorse della rete
Exploring the different aspects of V2V communication systems for specifying the semantic empowering capabilities regarding cooperative perception application
SOURCHAP, ALI
2021/2022
Abstract
In recent years, thanks to machine learning flourishing, countless studies have attempted to go beyond content-agnostic message transmission and take care of messages’ purpose and semantic meaning. Due to the promised efficiency and reliability advancements that would be obtained by switching to semantic communication, we decided to study V2V communication systems (more specifically, V2V communication for the “cooperative perception” application) to detect possible approaches for semantic empowering. Within the cooperative perception application, systems mostly have two difficulties: sharing the massive captured data and managing data transmissions. Therefore, we studied this application-based communication system from two distinct aspects: semantic-empowered physical layer and network managing. In the scale of the physical layer, our objective was to find the proper goal-oriented encoding/decoding modules that can extract and recover the semantic features. By analysing some proposed DNN learning-based pairs of encoder/decoder that address mentioned challenges, we found the great opportunity of employing these modules in the V2V systems to obtain significant compression rates with the limited accuracy declining. Furthermore, in the scale of the network managing, we proposed a novel factor graph-based planning system that predicts impending collisions by exploiting the spatial grabbed data of the traffic environment. Concerning the network condition, this system employs combinatorial optimization to pick the best massage combinations to avoid potential collisions. Although we designed a program for assessing our method, the complete assessment was not feasible due to the shortage of time. Nevertheless, this study reveals an existing gap between the conventional and goal-based criteria used in network management. Our proposed method does not just have the potential to be executed in real scenarios but also promotes the idea of making goal-oriented strategies. The new intelligent networks, with smart entities and more specific ultimate goals, can utilize the available data in the network to control it more efficiently with respect to communication goals. Indeed, our proposed system evidently identifies this issue and addresses it by presenting an innovative way to make a goal-oriented network.| File | Dimensione | Formato | |
|---|---|---|---|
|
2022_10_Sourchap.pdf
Open Access dal 20/09/2023
Dimensione
7.59 MB
Formato
Adobe PDF
|
7.59 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/10589/195696