Skip to main content

How Windowing Affects Your Periodogram


The windowed periodogram is a widely used technique for estimating the Power Spectral Density (PSD) of a signal. It enhances the classical periodogram by mitigating spectral leakage through the application of a windowing function. This technique is essential in signal processing for accurate frequency-domain analysis.

 

Power Spectral Density (PSD)

The PSD characterizes how the power of a signal is distributed across different frequency components. For a discrete-time signal, the PSD is defined as the Fourier Transform of the signal’s autocorrelation function:

Sx(f) = FT{Rx(Ï„)}

Here, Rx(Ï„)}is the autocorrelation function.

FT : Fourier Transform

 

Classical Periodogram

The periodogram is a non-parametric PSD estimation method based on the Discrete Fourier Transform (DFT):

Px(f) = \(\frac{1}{N}\) X(f)2

Here:

  • X(f): DFT of the signal x(n)

  • N: Signal length

However, the classical periodogram suffers from spectral leakage due to abrupt truncation of the signal.

 

Windowing to Mitigate Spectral Leakage

Spectral leakage can be minimized by applying a window function to the signal before computing the DFT. The resulting PSD estimate is called the windowed periodogram:

Pw(f) = \(\frac{1}{NW}\) Xw(f)2

Here:

  • w(n): Window function

  • W: Window normalization factor

Common Window Functions

  • Rectangular Window: Equivalent to the classical periodogram.

w[n]=1, 0≤n≤N−1

w[n]=0, otherwise

Where, N is the window length

  • Hamming Window: Reduces sidelobe amplitudes, improving frequency resolution.

w[n]=0.5(1−cos(\(\frac{\ 2\pi n}{N - 1}\ \))), 0≤n≤N−1

Where, N is the window length

  • Hanning Window: Similar to Hamming but with less sidelobe attenuation.

w[n]=0.54 – 0.46cos(\(\frac{\ 2\pi n}{N - 1}\ \)), 0≤n≤N−1

Where, N is the window length

  • Blackman Window: Offers even greater sidelobe suppression but at the cost of wider main lobes.

w[n]=0.42 – 0.5(cos(\(\frac{\ 2\pi n}{N - 1}\ \)) + 0.08(cos(\(\frac{\ 4\pi n}{N - 1}\ \)), 0≤n≤N−1

Where, N is the window length

 

Implementation Steps

  1. Segment the Signal: Divide the signal into overlapping or non-overlapping segments of length N.

  2. Apply a Window Function: Multiply each segment by a window function w(n).

  3. Compute the DFT: Calculate the DFT of the windowed segments.

  4. Average the Periodograms: For overlapping segments, average the periodograms to reduce variance.

     

Properties of the Windowed Periodogram

  • Bias: Windowing introduces bias in the PSD estimate as the window modifies the signal spectrum.

  • Variance: Averaging periodograms (Welch method) reduces variance but decreases frequency resolution.

  • Trade-Off: The choice of window affects the trade-off between spectral resolution and leakage suppression.

     

    MATLAB Code

    clc;
    clear;
    close all;

    fs = 48000;
    t = 0:1/fs:0.02;
    f_ping = 12000;

    % Base sine wave
    sine_wave = sin(2*pi*f_ping*t)';

    % Apply windows
    w_rect = ones(size(sine_wave));
    w_hann = hann(length(sine_wave));
    w_hamming = hamming(length(sine_wave));
    w_blackman = blackman(length(sine_wave));

    % Windowed signals
    s_rect = sine_wave .* w_rect;
    s_hann = sine_wave .* w_hann;
    s_hamming = sine_wave .* w_hamming;
    s_blackman = sine_wave .* w_blackman;

    % FFT
    Nfft = 4096;
    f = fs*(0:Nfft/2-1)/Nfft;

    % Function to compute and normalize spectrum
    get_norm_fft = @(sig) abs(fft(sig, Nfft))/max(abs(fft(sig, Nfft)));

    S_rect = get_norm_fft(s_rect);
    S_hann = get_norm_fft(s_hann);
    S_hamming = get_norm_fft(s_hamming);
    S_blackman = get_norm_fft(s_blackman);

    % Mainlobe power (±2 bins around peak)
    mainlobe_bins = 2;

    % Function to compute power ratio
    compute_power_ratio = @(S) ...
    deal( ...
    sum(S.^2), ... % Total power
    max(1, find(S == max(S), 1)), ... % Peak bin
    @(peak_bin) sum(S(max(1,peak_bin-mainlobe_bins):min(Nfft,peak_bin+mainlobe_bins)).^2), ...
    @(total, main) 10*log10((total-main)/main) ... % dB sidelobe/mainlobe ratio
    );

    % Calculate ratios
    [total_r, peak_r, get_main_r, get_slr_r] = compute_power_ratio(S_rect);
    main_r = get_main_r(peak_r); slr_r = get_slr_r(total_r, main_r);

    [total_h, peak_h, get_main_h, get_slr_h] = compute_power_ratio(S_hann);
    main_h = get_main_h(peak_h); slr_h = get_slr_h(total_h, main_h);

    [total_ham, peak_ham, get_main_ham, get_slr_ham] = compute_power_ratio(S_hamming);
    main_ham = get_main_ham(peak_ham); slr_ham = get_slr_ham(total_ham, main_ham);

    [total_b, peak_b, get_main_b, get_slr_b] = compute_power_ratio(S_blackman);
    main_b = get_main_b(peak_b); slr_b = get_slr_b(total_b, main_b);

    % Display Results
    fprintf('Window | Mainlobe Power | Sidelobe Power | Sidelobe/Main (dB)\n');
    fprintf('------------|----------------|----------------|--------------------\n');
    fprintf('Rectangular | %14.4f | %14.4f | %18.2f\n', main_r, total_r - main_r, slr_r);
    fprintf('Hann | %14.4f | %14.4f | %18.2f\n', main_h, total_h - main_h, slr_h);
    fprintf('Hamming | %14.4f | %14.4f | %18.2f\n', main_ham, total_ham - main_ham, slr_ham);
    fprintf('Blackman | %14.4f | %14.4f | %18.2f\n', main_b, total_b - main_b, slr_b);

    % Plot
    figure;
    plot(f, 20*log10(S_rect(1:Nfft/2)), 'k'); hold on;
    plot(f, 20*log10(S_hann(1:Nfft/2)), 'r');
    plot(f, 20*log10(S_hamming(1:Nfft/2)), 'g');
    plot(f, 20*log10(S_blackman(1:Nfft/2)), 'b');
    legend('Rectangular','Hann','Hamming','Blackman');
    xlim([f_ping-3000 f_ping+3000]); ylim([-100 5]);
    xlabel('Frequency (Hz)'); ylabel('Magnitude (dB)');
    title('Windowing Effects on Spectrum');
    grid on;

    Output 

    Window      | Mainlobe Power | Sidelobe Power | Sidelobe/Main (dB)
    ------------|----------------|----------------|--------------------
    Rectangular |         3.5771 |         4.9562 |               1.42
    Hann        |         4.3630 |         8.4370 |               2.86
    Hamming     |         4.2367 |         7.3928 |               2.42
    Blackman    |         4.4940 |        10.2410 |               3.58

     

     








Applications

  • Signal Processing: Analyzing frequency content of time-varying signals.

  • Communications: Evaluating spectrum occupancy in wireless systems.

  • Bioinformatics: Investigating periodicities in biological signals (e.g., EEG, ECG).

  • Seismology: Characterizing seismic wave frequencies.

     

    Further Reading

    1. Periodogram in MATLAB

People are good at skipping over material they already know!

View Related Topics to







Contact Us

Name

Email *

Message *

Popular Posts

BER vs SNR for M-ary QAM, M-ary PSK, QPSK, BPSK, ...(MATLAB Code + Simulator)

📘 Overview of BER and SNR 🧮 Online Simulator for BER calculation 🧮 MATLAB Code for BER calculation 📚 Further Reading 📂 View Other Topics on M-ary QAM, M-ary PSK, QPSK ... 🧮 Online Simulator for Constellation Diagram of m-ary QAM 🧮 Online Simulator for Constellation Diagram of m-ary PSK 🧮 MATLAB Code for BER calculation of ASK, FSK, and PSK 🧮 MATLAB Code for BER calculation of Alamouti Scheme 🧮 Different approaches to calculate BER vs SNR What is Bit Error Rate (BER)? The abbreviation BER stands for Bit Error Rate, which indicates how many corrupted bits are received compared to the total number of bits sent. BER = (number of bits received in error) / (total number of transmitted bits) What is Signal-to-Noise Ratio (SNR)? SNR is the ratio of signal power to noise powe...
Read more

Comparing Baseband and Passband Implementations of m-ary QAM

  Let's assume your original digital message bitstream is: 0, 0, 1, 0, 0, 0, 1, 0, 1, 1 In 4-QAM, we group them into pairs: (00), (10), (00), (10), (11). Your baseband symbols are: Symbol 1 (Bits 00): -1.00 - j1.00 Symbol 2 (Bits 10): 1.00 - j1.00 Symbol 3 (Bits 00): -1.00 - j1.00 Symbol 4 (Bits 10): 1.00 - j1.00 Symbol 5 (Bits 11): 1.00 + j1.00   To transmit these symbols over a wireless medium, we modulate this baseband signal onto a high-frequency carrier (e.g., 50 Hz). This process creates the passband signal , where the information is stored in the phase and amplitude of the sine wave. Fig 1: 4-QAM Baseband I and Q Components Fig 2: 4-QAM Passband Modulated Signal   In this example, the symbol rate is 5 symbols per second. Detailed Explanation 4-QAM Constellation Mapping In standard 4-QAM mapping, bits are converted to complex points on a grid: Bits...
Read more

Comparing Baseband and Passband Implementations of ASK, FSK, and PSK

📘 Overview 🧮 Baseband and Passband Implementations of ASK, FSK, and PSK 🧮 Difference betwen baseband and passband 📚 Further Reading 📂 Other Topics on Baseband and Passband ... 🧮 Baseband modulation techniques 🧮 Passband modulation techniques   Baseband modulation techniques are methods used to encode information signals onto a baseband signal (a signal with frequencies close to zero). Passband techniques shift these signals to higher carrier frequencies for transmission. Here are the common implementations: Amplitude Shift Keying (ASK) [↗] : In ASK, the amplitude of the signal is varied to represent different symbols. Binary ASK (BASK) is a common implementation where two different amplitudes represent binary values (0 and 1). ASK is simple but susceptible to noise. ASK Baseband (Digital Bits) ASK Passband (Modulated Carrier)     Fig 1:  ASK Passband Modulation (...
Read more

Amplitude, Frequency, and Phase Modulation Techniques (AM, FM, and PM)

📘 Overview 🧮 Amplitude Modulation (AM) 🧮 Online Amplitude Modulation Simulator 🧮 MATLAB Code for AM 🧮 Q & A and Summary 📚 Further Reading Amplitude Modulation (AM): The carrier signal's amplitude varies linearly with the amplitude of the message signal. An AM wave may thus be described, in the most general form, as a function of time as follows: When performing amplitude modulation (AM) with a carrier frequency of 100 Hz and a message frequency of 10 Hz, the resulting peak frequencies are as follows: 90 Hz (100 - 10 Hz), 100 Hz, and 110 Hz (100 + 10 Hz). Figure: Frequency Spectrums of AM Signal (Lower Sideband, Carrier, and Upper Sideband) A low-frequency message signal is modulated with a high-frequency carrier wave using a local oscillator to make communication possible. DSB, SSB, and VSB are common amplitude modulation techniques. We find a lot of bandwidth loss in DSB. The bandwidth of S...
Read more

Shannon Limit Explained: Negative SNR, Eb/No and Channel Capacity

Understanding Negative SNR and the Shannon Limit Understanding Negative SNR and the Shannon Limit An explanation of Signal-to-Noise Ratio (SNR), its behavior in decibels, and how Shannon's theorem defines the ultimate communication limit. Signal-to-Noise Ratio in Shannon’s Equation In Shannon's equation, the Signal-to-Noise Ratio (SNR) is defined as the signal power divided by the noise power: SNR = S / N Since both signal power and noise power are physical quantities, neither can be negative. Therefore, the SNR itself is always a positive number. However, engineers often express SNR in decibels: SNR(dB) When SNR = 1, the logarithmic value becomes: SNR(dB) = 0 When the noise power exceeds the signal power (SNR < 1), the decibel representation becomes negative. Behavior of Shannon's Capacity Equation Shannon’s channel capacity formula is: C = B log₂(1 + SNR) For SNR = 0: log₂(1 + SNR) = 0 When SNR becomes smaller (in...
Read more

Analog vs Digital Modulation Techniques | Advantages of Digital ...

Modulation Techniques Analog vs Digital Modulation In our previous discussion, we explored the necessity of modulation. In this article, we focus on the fundamental differences between analog and digital modulation. The primary distinction is that digital modulation uses a discrete digital signal to modify the carrier, whereas analog modulation uses a continuous analog signal. Advantages of Digital Modulation over Analog Modulation Bandwidth Efficiency: Digital techniques (like QAM) can transmit more data within a limited frequency range. Noise Resistance: Digital signals have superior resistance to noise because they can be perfectly regenerated. Multiplexing: It is much easier to multiplex various data types (audio, video, text) into a single digital stream. Higher SNR: Better noise immunity leads to a higher Signal-to-Noise Ratio (SNR). Increased Throughput: Modern digital techniques provide significantly higher data ...
Read more

ASK, FSK, and PSK (with MATLAB + Online Simulator)

📘 Overview 📘 Amplitude Shift Keying (ASK) 📘 Frequency Shift Keying (FSK) 📘 Phase Shift Keying (PSK) 📘 Which of the modulation techniques—ASK, FSK, or PSK—can achieve higher bit rates? 🧮 MATLAB Codes 📘 Simulator for binary ASK, FSK, and PSK Modulation 📚 Further Reading ASK or OFF ON Keying ASK is a simple (less complex) Digital Modulation Scheme where we vary the modulation signal's amplitude or voltage by the message signal's amplitude or voltage. We select two levels (two different voltage levels) for transmitting modulated message signals. For example, "+5 Volt" (upper level) and "0 Volt" (lower level). To transmit binary bit "1", the transmitter sends "+5 Volts", and for bit "0", it sends no power. The receiver uses filters to detect whether a binary "1" or "0" was transmitted. ...
Read more

Calculation of SNR from FFT bins in MATLAB

📘 Overview 🧮 MATLAB Code for Estimation of SNR from FFT bins 🧮 MATLAB Code for SNR from PSD using Kaiser Window 📚 Further Reading Here, you can find the SNR of a received signal from periodogram / FFT bins using the Kaiser operator. The beta (β) parameter characterizes the Kaiser window, which controls the trade-off between the main lobe width and the side lobe level. Steps Set up the sampling rate and time vector Compute the FFT and periodogram Calculate the frequency resolution and signal power Exclude the signal power from noise calculation Compute the noise power and SNR MATLAB Code for Estimation of SNR from FFT bins 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 ...
Read more