/FIRMan

Finding the optimum filter order for minimal distortion with Manhattan distances.

Primary LanguageMATLAB

FIRMan

This repository contains some experimental data I found interesting for my 44DSP Mini-Project. Submitted on February 1, 2021.

Abstract

The role of a FIR is to capture and remove noise signals. The purpose of this study is to identify the optimum FIR filter specifications which prioritises speed and the prevention of distortion. A pair of vocal signals were given, with one of the signals contaminated with a single frequency signal. Different specifications of FIR were tested for real-time capability with minimal distortion. Low-pass filter type is found to be the fastest (176 ms) and with equally minimal distortion as the band-stop filter type (183 ms). However, every filter type produced a latecy lower than 250 ms and is considered an acceptable latency for good quality communication by Krasniqi et. al.

Results

Removing the single-frequency noise was simple but while I was looking for the optimum filter order for the different filters, I observed a chaotic relationship between similitude and filter order. I calculated similitude by computing the Manhattan distance between the original and filtered signal, where a smaller Manhattan distance would imply that the filtered signal more closely resembles the original.

Potential Improvements

Looking back, instead of finding the lowest Manhattan distance, computing the mean distance may have weakened the weight of any compounded errors and represented the similitude results better. Might have been good to try cosine similarity as well.