Calculation of SNR from FFT bins in MATLAB

📘Overview 💻FFT Bin Method 💻Kaiser Window 📚Further Reading

SNR Estimation Overview

In digital signal processing, estimating the Signal-to-Noise Ratio (SNR) accurately is crucial. Below, we demonstrate how to calculate SNR from periodogram and FFT bins using the Kaiser Window. The beta (β) parameter is the key—it allows you to control the trade-off between main-lobe width and side-lobe levels for precise spectral analysis.

1 Define Sampling rate and Time vector

2 Compute FFT and Periodogram PSD

3 Identify Signal Bin and Frequency resolution

4 Segment Signal Power from Noise floor

5 Logarithmic calculation of SNR in dB

Method 1: Estimation from FFT Bins

This approach uses a Hamming window to estimate SNR directly from the spectral bins.

MATLAB Source Code

clc; clear; close all;
% Parameters
fs = 8000; f_tone = 1000; N = 8192; 
t = (0:N-1)/fs;

% Generate signal + noise
signal = sin(2*pi*f_tone*t);
SNR_true_dB = 20; 
signal_power = mean(signal.^2);
noise_power = signal_power / (10^(SNR_true_dB/10));
noisy_signal = signal + sqrt(noise_power) * randn(1, N);

% Apply window
w = hamming(N)';
windowed_signal = noisy_signal .* w;
U = sum(w.^2)/N; 

% FFT and PSD
X = fft(windowed_signal);
f = (0:N-1)*fs/N;
Pxx = abs(X).^2 / (fs * N * U); 

% Find signal bin
[~, signal_bin] = min(abs(f - f_tone));
signal_bins = signal_bin-1 : signal_bin+1;
signal_power_est = sum(Pxx(signal_bins));

% Noise estimation
noise_bins = setdiff(1:N/2, signal_bins); 
noise_power_est = sum(Pxx(noise_bins));

% Estimate SNR
SNR_est_dB = 10 * log10(signal_power_est / noise_power_est);
fprintf('Estimated SNR from FFT: %.2f dB\n', SNR_est_dB);

Method 2: Using Kaiser Window

Optimized spectral estimation using the Kaiser window (Beta=38) for better side-lobe suppression.

MATLAB Source Code

clc; clear; close all;
fs = 32000;
t = 0:1/fs:1-1/fs;
x = sin(2*pi*3000*t) + 0.05*randn(size(t)); % Example Signal

N = length(x);
w = kaiser(N, 38);
[Pxx, F] = periodogram(x, w, N, fs);

% SNR Estimation
freq_resolution = abs(F(2)-F(1));
[~, target_idx] = min(abs(F - 3000)); % Find 3kHz index

% Signal Power (using bins around peak)
sig_idx = target_idx-2 : target_idx+2;
Sig_power_val = sum(Pxx(sig_idx)) * freq_resolution;
Sig_power_dB = 10*log10(Sig_power_val);

% Noise Power (excluding signal)
Noise_Pxx = Pxx;
Noise_Pxx(sig_idx) = 0; 
N_avg = sum(Noise_Pxx) / (length(Pxx) - length(sig_idx));
N_power_dB = 10*log10(N_avg * fs/2);

SNR = Sig_power_dB - N_power_dB;
fprintf('SNR = %.4f dB\n', SNR);

Power Spectral Density Calculation Using FFT

Power Spectral Density (PSD) is a fundamental metric used to characterize how the power of a signal is distributed across the frequency spectrum. While a standard Fourier Transform provides the amplitude and phase of frequency components, the PSD focuses on the intensity of power, making it an essential tool for analyzing noise, stochastic processes, and signal-to-noise ratios in communication systems.

Mathematical Logic

PSD = |FFT|² / (f_s · N)

The calculation follows a strict sequence:

Compute the FFT of the time-domain signal.
Take the Absolute Magnitude.
Square the magnitude to find raw power.
Normalize by sampling rate and sample count.

Frequency Resolution (Δf)

Δf = f_s / N

f_s (Sampling Frequency): Higher rates increase the total range but require more samples to maintain resolution.

N (Sample Count): Increasing N provides finer frequency "bins," allowing for higher precision in identifying signal peaks.

Core Principles of Spectral Estimation

Understanding PSD requires balancing the trade-offs between frequency range and clarity:

1

Amplitude vs. Power: Fourier Magnitude represents the "strength" of a frequency, whereas PSD represents its energy density. Squaring the magnitude is what shifts the analysis from voltage/amplitude levels to power levels.
2

Nyquist Limits: The sampling frequency (f_s) dictates the maximum detectable frequency. To avoid aliasing, the PSD can only accurately describe components up to f_s / 2.
3

Statistical Reliability: In real-world applications, raw PSD calculations (periodograms) can be "noisy." Techniques like Welch's Method or Bartlett's Method improve accuracy by averaging multiple segments of the signal to smooth out random fluctuations.

Why PSD is Crucial

PSD is the primary tool for identifying hidden periodicities in noisy data. By observing the "floor" of a PSD plot, engineers can determine the Noise Power, while the "peaks" reveal the presence of dominant signals, enabling the calculation of the Signal-to-Noise Ratio (SNR).

Search This Blog