Virtual Analog modeling aims to digitally replicate analog audio equipment, allowing practitioners and musicians to utilize their gear in the digital domain. Specifically, audio circuits, such as preamplifiers and distortion pedals, are often composed of networks with multiple transistors, which, as multi-port elements, pose significant challenges for emulation. In this thesis, we extend a method recently proposed for the emulation of Bipolar Junction Transistors (BJTs) to account for Metal-Oxide-Semiconductor Field-Effect Transistors (MOSFETs). The method leverages a vector definition of waves within the framework of Wave Digital Filters to achieve an explicit realization of a circuit containing a single multi-port nonlinear element. We evaluate the performance of various architectures, including Multi-Layer Perceptrons and ResNets, and apply the proposed methodology to model the Obsidian Overdrive pedal. In order to carry out the modeling, we divide the original schematics into three stages, each of which contains a single MOSFET, and we then use them in cascade to emulate the overall processing chain. Our results demonstrate strong performance when applying the method to circuits containing MOSFETs, paving the way for the emulation of circuits with other multi-port nonlinear elements.
La modellizzazione Virtual Analog mira a replicare digitalmente le apparecchiature audio analogiche, consentendo a tecnici e musicisti di utilizzare le loro attrezzature nel dominio digitale. In particolare, i circuiti audio, come preamplificatori e pedali di distorsione, sono spesso costituiti da reti con più transistor, che, come elementi multi-porta, presentano notevoli sfide nella loro emulazione. In questa tesi, estendiamo un metodo recentemente proposto per l'emulazione di Transistor a Giunzione Bipolare (BJT) ai Transistor a Effetto di Campo Metallo-Ossido-Semiconduttore (MOSFET). Il metodo sfrutta una definizione vettoriale delle onde nell'ambito dei Wave Digital Filters per ottenere una realizzazione esplicita di un circuito contenente un singolo elemento non lineare multi-porta. Valutiamo le prestazioni di varie architetture, tra cui Multi-Layer Perceptrons e ResNets, e applichiamo la metodologia proposta per modellare il pedale distorsore Obsidian Overdrive. Per effettuare la modellazione, gli schemi originali sono suddivisi in tre stadi, ciascuno dei quali contenente un singolo MOSFET, utilizzandoli in cascata per emulare l'intera catena di processo del segnale. I risultati dimostrano buona accuratezza nell'applicazione del metodo ai circuiti contenenti MOSFET, aprendo la strada all'emulazione di circuiti con altri elementi non lineari multi-porta.
Wave digital neural network-based models of MOSFETs for virtual analog applications
Ferrè, Marco
2023/2024
Abstract
Virtual Analog modeling aims to digitally replicate analog audio equipment, allowing practitioners and musicians to utilize their gear in the digital domain. Specifically, audio circuits, such as preamplifiers and distortion pedals, are often composed of networks with multiple transistors, which, as multi-port elements, pose significant challenges for emulation. In this thesis, we extend a method recently proposed for the emulation of Bipolar Junction Transistors (BJTs) to account for Metal-Oxide-Semiconductor Field-Effect Transistors (MOSFETs). The method leverages a vector definition of waves within the framework of Wave Digital Filters to achieve an explicit realization of a circuit containing a single multi-port nonlinear element. We evaluate the performance of various architectures, including Multi-Layer Perceptrons and ResNets, and apply the proposed methodology to model the Obsidian Overdrive pedal. In order to carry out the modeling, we divide the original schematics into three stages, each of which contains a single MOSFET, and we then use them in cascade to emulate the overall processing chain. Our results demonstrate strong performance when applying the method to circuits containing MOSFETs, paving the way for the emulation of circuits with other multi-port nonlinear elements.File | Dimensione | Formato | |
---|---|---|---|
Article_Thesis___Wave_Digital_Neural_Network_based_Models_of_MOSFETs_for_Virtual_Analog_Applications.pdf
non accessibile
Dimensione
961.3 kB
Formato
Adobe PDF
|
961.3 kB | Adobe PDF | Visualizza/Apri |
Executive_Summary___Wave_Digital_Neural_Network_based_Models_of_MOSFETs_for_Virtual_Analog_Applications.pdf
non accessibile
Dimensione
547.86 kB
Formato
Adobe PDF
|
547.86 kB | 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/10589/223020