Digital Audio Noise Filtering Project

A complete, reproducible workflow for detecting and filtering noise from a voice audio signal. Uses MATLAB for signal analysis, frequency characterization, filter design, and audio/binary conversion. Implements and simulates the digital filter in Verilog with a full hardware datapath and FSM controller. The result is a practical system that processes and reconstructs real audio through a true hardware-accurate pipeline.

📦 Source code and images available at:
GitHub Repository

---

Table of Contents

• Overview
• Features
• Prerequisites
• Block Diagram
• State Machine Diagram
• How It Works
• Project Structure
• Project Structure (Without Controller)
• Example Workflow
• Usage Instructions
• Sample Result
• Credits

---

Overview

This project demonstrates a practical signal-processing pipeline:

• FFT-based noise detection and visualization
• Chebyshev Type II IIR filter design
• Fixed-point coefficient conversion (Q6.57 format: 1 sign bit + 6 integer bits + 57 fractional bits)
• Audio format conversion (Q0.15 format: 1 sign bit + 15 fractional bits) for hardware simulation
• Hardware-accurate Verilog datapath and FSM controller
• End-to-end simulation with testbench and audio reconstruction

Features

• MATLAB scripts for full signal inspection, filter synthesis, and file I/O
• Highly modular Verilog design: includes datapath, controller, register, ROM, and clipper modules
• Automated fixed-point scaling for both coefficient and audio data
• File-based Verilog testbench; no manual stimulus required
• Includes block diagrams and FSM charts for pipeline and control logic visualization

---

Prerequisites

• MATLAB R2020b or newer
• Verilog simulation tool (e.g., ModelSim or Vivado Simulator)
• Audio file in mono 16-bit PCM format (input.wav)

---

Block Diagram

System datapath architecture (Verilog implementation with controller):

---

State Machine Diagram

Controller finite-state machine (FSM) for filter sequencing:

---

How It Works

1. Noise & Signal Spectrum Analysis

• Use noise_detection.m to load your audio file and plot the FFT of both the "noise-only" segment and the full audio.
• Identify the main noise frequency and compare it against the total signal.

*FFT of noise only (first 0.5s): dominant noise ~1500Hz

*FFT of full signal: speech energy mostly at lower frequencies; noise spike still visible

2. Design & Export Filter

• Determine target filter characteristics based on plots—e.g., set a low-pass cutoff just below the noise frequency.
• Run Coeffcient_Extraction_And_Test.m:
  • The filter is synthesized using a Chebyshev Type II low-pass design, tailored to suppress the detected noise frequency (~1500 Hz).
  • Coefficients are scaled and formatted (Q6.57) for Verilog ROM.

3. Audio Format Conversion

• audio_To_Q0.15_Binary.m: converts WAV samples to 16-bit Q0.15 binary (for testbench input).

4. Verilog Digital Filter Implementation

• Datapath: Implements Direct Form I/II MAC structure, using pipelined shift registers, accumulator, and ROM coefficient lookup.
• Controller: FSM (see diagram above) synchronizes sampling, loading, and accumulation cycles.
• ROM: Stores the fixed-point coefficients used at each filter tap.
• Testbench (Digital_Filter_TB.v): Reads binary samples, feeds them to the filter, writes output to output_samples.txt.

5. Audio Output Reconstruction

• q0.15_Binary_To_Audio.m: converts the binary stream of filtered samples back to a floating-point WAV (output.wav) for audible quality verification.

---

Project Structure

File/Folder	Description
noise_detection.m	FFT analysis and noise visualization in MATLAB
Coeffcient_Extraction_And_Test.m	Filter coefficient extraction + export (Q6.57) + test that your filter works
audio_To_Q0.15_Binary.m	Convert WAV audio to Q0.15 binary samples for simulation
q0.15_Binary_To_Audio.m	Reconverts filtered binary output into WAV
Digital_Filter_TopModule.v	Top-level Verilog digital filter (controller + datapath)
Controller.v	FSM for datapath sequencing
Data_Path.v	Core multiply-accumulate datapath module
Coeffs_ROM.v	Fixed-point coefficient ROM for the digital filter
Register.v	Parameterized synchronous/asynchronous register module
Counter.v	Cycle counter for filter taps
Clipper.v	Q7.64 saturating to Q0.63 output
Digital_Filter_TB.v	Testbench for batch input and output

---

Project Structure (Without Controller)

This version implements the filter using a synchronous datapath without an explicit FSM controller, relying on always blocks and generate statements for multiplication and summation.

System Datapath Architecture (Without Controller)

File/Folder	Description
noise_detection.m	FFT analysis and noise visualization in MATLAB
Coeffcient_Extraction_And_Test.m	Filter coefficient extraction, export (Q6.57), and filter testing
audio_To_Q0.15_Binary.m	Convert WAV audio to Q0.15 binary samples for simulation
q0.15_Binary_To_Audio.m	Reconverts filtered binary output into WAV
Data_Path.v	Core multiply-accumulate datapath module (without controller)
Clipper.v	Q7.64 saturating to Q0.63 output
Tb.v	Testbench for batch input and output

---

Example Workflow

1. Analyze spectra:
   noise_detection.m → Plots noise & total signal FFT to guide filter design.

2. Design filter:
   Coeffcient_Extraction_And_Test.m → Designs & exports Q6.57 coefficients.

3. Convert audio for hardware:
   audio_To_Q0.15_Binary.m → Input WAV to binary samples.

4. Simulate in Verilog:
   Load all .v files + ROM in simulator.
   Run Digital_Filter_TB.v (or Tb.v for without controller) → Binary output produced.

5. Convert back to audio:
   q0.15_Binary_To_Audio.m → Final output WAV for quality check.

---

Usage Instructions

MATLAB

1. Place your input WAV as input.wav.
2. Run noise_detection.m to inspect noise and speech spectra.
3. Run Coeffcient_Extraction_And_Test.m for filter design/export.
4. Use audio_To_Q0.15_Binary.m to generate input_samples.txt.

Verilog

1. Add all source modules and Digital_Filter_TB.v (or Tb.v for without controller) to your simulator.
2. Ensure input_samples.txt is present in the simulation working directory.
3. Run simulation: output is saved as output_samples.txt.

Output Reconstruction

1. Use q0.15_Binary_To_Audio.m to get your final output.wav.
2. Listen and compare to verify noise attenuation and speech preservation.

---

🎧 Sample Result

After filtering, the background noise (~1500Hz tone) was successfully attenuated while preserving speech clarity.
Listen to the filtered result: output.wav

---

Credits

Developed by [Nima].
MATLAB and Verilog sources provided for educational and research purposes.

Questions or suggestions? Please open an issue or contribute!

---

Enjoy a clean signal from algorithm to hardware!