A novel approach for single microphone active noise cancellation

Passive methods are costly and ineffective in noise reduction at low frequencies. Active noise control has been suggested because of these problems. Active noise control (ANC) is based on the destructive interference between the noise source waves and a controlled secondary source. In this paper, various training algorithms are compared in active cancellation of modeled sound noise using MLP neural network. Colored noise signals are used as a model of sound noise instead of noise signals from databases. An MLP neural network with different architectures is used in simulation procedure. The effect of number of neurons on the convergence speed of various training algorithms is investigated in this paper. Levenberg-Marquardt (LM), scaled conjugate gradient (SCG), BFGS quasi-Newton (BFG), resilient back-propagation (RP) and variable learning rate back-propagation (GDX) are used for training the network. Simulation results show that Levenberg-Marquardt (LM) and scaled conjugate gradient (SCG) are the fastest training algorithms.

show abstract

“…1000 samples can't help us in predicting the 1001st sample. In [11], sound noise has been modeled as a colored noise using the following method:…”

Section: Modeled Sound Noise and Neural Network Architecturementioning

confidence: 99%

Evaluating the Performance of Training Algorithms in Active Noise Control Using MLP Neural Network

Salmasi

Mahdavi-Nasab

2012

AMR

View full text Add to dashboard Cite

show abstract

“…A notch filter with a Proportional Derivative (PD) regulator has been programmed because some other studies 5–7 show that this kind of filter provides the best efficiency for the compensation for vibrations. The software used for the real‐time calculation of the coefficients of the notch filter leads to an adaptative system of compensation for vibrations.…”

Section: Active Vibration Controlmentioning

confidence: 99%