Earth Sciences / Signal Processing
/ Topic 5 of 5 · 5/5 Signal Processing Short Notes One-page CSIR NET geophysics revision: Fourier series, correlation, LTI systems, sampling, FFT, Z-transform, filters, deconvolution and fractal analysis.
9 min read Updated 16 Nov 2026
Share
Copy link
Print · PDF
Save Share this topic
csirnetnotes.com/earth-sciences/signal-processing/quick-revision
Copy link Ctrl C Pre-written message Signal Processing Short Notes — notes, formulas and practice, free. csirnetnotes.com/earth-sciences/signal-processing/quick-revision
On this page · 11 sections▾ Part C in one table Signals Fourier series and transforms Correlation LTI systems Sampling and the DFT Convolution and the Z-transform Filters Deconvolution Fractal analysis Back to the chapter Every formula and trap from Continuous and Discrete Signals, Fourier Series and Transforms , Auto- and Cross-Correlation and LTI Systems , Sampling Theorem, FFT, Z-Transform and Convolution and Filters, Deconvolution and Fractal Analysis , on one page.
Part C in one table Per the NTA information bulletin for December 2026, Earth Sciences Part C:
Given Attempt Correct Wrong Maximum 80 25 +4 −1.32 100
Only the first 25 answered questions are evaluated. Full pattern on the Earth Sciences overview .
Signals Idea Fact Energy \( E = \int\lvert x\rvert^2dt \); energy signals have zero average power Power time average of \( \lvert x\rvert^2 \); periodic signals and stationary noise \( e^{-at}u(t) \) energy \( 1/(2a) \) Sinusoid of amplitude \( A \) power \( A^2/2 \) Even/odd parts \( [x(t) \pm x(-t)]/2 \) \( \cos\Omega_0 n \) periodic only if \( \Omega_0/2\pi = m/N \) is rational; period \( N \) Discrete frequencies distinct only in \( (-\pi, \pi] \); \( \Omega = \omega\Delta t \) Sum of periodic signals period = LCM, only if the ratio of periods is rational
$$\boxed{c_n = \frac{1}{T}\int_T x(t)\,e^{-in\omega_0 t}\,dt} \qquad \boxed{P = a_0^2 + \frac{1}{2}\sum_{n\ge1}(a_n^2 + b_n^2) = \sum_n\lvert c_n\rvert^2}$$
Symmetry Series has Even cosines (and DC) Odd sines Half-wave \( x(t + T/2) = -x(t) \) odd harmonics only, no DC
Waveform Series Decay Square \( \pm A \) (odd) \( \frac{4A}{\pi}\sum_{\text{odd }n}\frac{\sin n\omega_0 t}{n} \) \( 1/n \) Sawtooth \( t \) on \( (-\pi, \pi) \) \( 2\sum\frac{(-1)^{n+1}}{n}\sin nt \) \( 1/n \) Triangle \( \lvert t\rvert \) on \( (-\pi, \pi) \) \( \frac{\pi}{2} - \frac{4}{\pi}\sum_{\text{odd}}\frac{\cos nt}{n^2} \) \( 1/n^2 \)
Dirichlet: absolutely integrable, finite extrema, finite jumps; series gives the midpoint at a jump. Gibbs overshoot: about 9% of the jump , never disappears. \( \sum_{\text{odd}}1/n^2 = \pi^2/8 \); fundamental carries about 81% of square-wave power. Pair or property Result \( \delta(t) \) 1 \( \mathrm{rect}(t/T) \) \( T\,\mathrm{sinc}(fT) \), first zeros \( \pm 1/T \) Gaussian Gaussian Comb, spacing \( \Delta t \) comb, spacing \( 1/\Delta t \) Shift by \( t_0 \) \( \times e^{-i2\pi f t_0} \): phase only Scaling \( x(at) \) \( X(f/a)/\lvert a\rvert \): short in time, wide in frequency Convolution / multiplication product / convolution Rule discrete in one domain ↔ periodic in the other
Correlation $$\boxed{R_{xy}(\tau) = \int x(t)\,y(t + \tau)\,dt = x(-\tau) * y(\tau) \;\leftrightarrow\; X^{*}(f)Y(f)}$$
Autocorrelation property Statement Symmetry even Maximum at zero lag; \( R(0) \) = energy or power Phase lost (transform is \( \lvert X\rvert^2 \)) Sinusoid \( \frac{A^2}{2}\cos 2\pi f_0\tau \) White noise \( \sigma^2\delta(\tau) \)
Correlation coefficient \( \rho = R_{xy}/\sqrt{R_{xx}(0)R_{yy}(0)} \in [-1, 1] \); \( \rho = 1 \) for a scaled, shifted copy. Delayed copy in \( y \): peak at \( \tau = + \)delay (with this convention). Wiener–Khinchin: PSD = Fourier transform of autocorrelation (wide-sense stationary process).Vibroseis: correlate record with sweep → Klauder wavelet (sweep autocorrelation, zero-phase) at each reflection.\( N \)-sample autocorrelation has \( 2N - 1 \) lags. LTI systems Property Test Linear superposition; zero in → zero out Time-invariant shifted input → equally shifted output Causal \( h[n] = 0 \) for \( n < 0 \) BIBO stable \( \sum\lvert h[n]\rvert < \infty \)
$$\boxed{y[n] = \sum_k x[k]\,h[n - k]} \qquad Y(f) = X(f)H(f)$$
Cascade \( h_1 * h_2 \) (order irrelevant); parallel \( h_1 + h_2 \); step response = running sum of \( h \). Sinusoid in → same frequency out, scaled by \( \lvert H\rvert \), shifted by \( \arg H \). Trace model \( x = w * r + n \). Random input: \( \mu_y = \mu_x H(0) \); \( S_{yy} = \lvert H\rvert^2 S_{xx} \); \( R_{xy} = h * R_{xx} \). White input: \( R_{xy}[k] = \sigma^2 h[k] \); \( \sigma_y^2 = \sigma^2\sum h^2 \). Sampling and the DFT $$\boxed{f_N = \frac{1}{2\Delta t}} \qquad \boxed{f_a = \lvert f - kf_s\rvert \in [0, f_N]} \qquad \boxed{\Delta f = \frac{1}{N\Delta t}}$$
\( \Delta t \) 1 ms 2 ms 4 ms \( f_N \) 500 Hz 250 Hz 125 Hz
Sampling theorem: \( f_s > 2f_{\max} \) (strictly); sinc interpolation recovers the signal. Anti-alias filter must act before sampling. Spatial aliasing when \( f > v_{app}/(2\Delta x) \). About 6 dB dynamic range per bit of the digitiser. Resolution about \( 1/T \) (Rayleigh); M2 vs S2 needs about 15 days. Zero padding interpolates; it does not resolve. Leakage: non-integer cycles in the window; tapers lower side lobes, widen main lobe. Real input: \( X[N - k] = X^{*}[k] \); \( N/2 + 1 \) independent bins. FFT Direct DFT \( N^2 \) complex multiplications Radix-2 FFT \( (N/2)\log_2 N \); \( N = 2^m \) \( N = 1{,}024 \) 1,048,576 vs 5,120 (about 205 times fewer)
Convolution = polynomial multiplication; length \( N + M - 1 \). DFT product = circular convolution; pad to \( \ge N + M - 1 \) for linear. Example: \( [2, 1, -1] * [1, -1, 2] = [2, -1, 2, 3, -2] \); 3-point circular \( [5, -3, 2] \). $$\boxed{X(z) = \sum_n x[n]\,z^{-n}}$$
\( x[n] \) \( X(z) \) ROC \( \delta[n - k] \) \( z^{-k} \) all \( z \) (except 0) \( u[n] \) \( 1/(1 - z^{-1}) \) \( \lvert z\rvert > 1 \) \( a^n u[n] \) \( 1/(1 - az^{-1}) \) \( \lvert z\rvert > \lvert a\rvert \) \( -a^n u[-n-1] \) \( 1/(1 - az^{-1}) \) \( \lvert z\rvert < \lvert a\rvert \)
Delay \( \to z^{-k} \); convolution \( \to \) product; reversal \( \to X(1/z) \); unit circle \( z = e^{i\Omega} \) gives the DTFT. Causal: ROC outside outermost pole. Stable: ROC includes unit circle. Causal and stable: all poles inside. FIR: always stable. IIR: check poles. Minimum phase: zeros inside \( \lvert z\rvert = 1 \) in \( z^{-1} \); roots outside \( \lvert Z\rvert = 1 \) in the geophysical unit-delay \( Z \). Dipole \( (a, b) \): minimum phase if \( \lvert a\rvert > \lvert b\rvert \).Minimum phase ↔ fastest energy build-up ↔ causal, stable inverse. Filters dB Amplitude Power −3 0.707 0.5 −6 0.5 0.25 −20 0.1 0.01
$$\boxed{\lvert H(f)\rvert^2 = \frac{1}{1 + (f/f_c)^{2n}}}\quad\text{(Butterworth: } -3 \text{ dB at } f_c,\ 6n \text{ dB/octave)}$$
Ideal low-pass: \( h = 2f_c\,\mathrm{sinc}(2f_c t) \), non-causal and infinite; truncation causes Gibbs ripple. FIR: finite, always stable, linear phase if symmetric (delay \( (M-1)/2 \)). IIR: recursive, poles, cheaper for sharp cut-offs. Phase: zero (no shift, forward–backward), linear (constant delay), minimum (causal, phase distortion). 3-point average: \( H = (1 + 2\cos\Omega)/3 \), zero at \( f_s/3 \). First difference: \( \lvert H\rvert = 2\lvert\sin(\Omega/2)\rvert \). Wiener normal equations \( \sum_j f_j r[i - j] = g[i] \): Toeplitz, solved by Levinson recursion (about \( M^2 \) operations). Deconvolution Method Key point Frequency-domain inverse \( W^{*}/(\lvert W\rvert^2 + \varepsilon) \) to avoid dividing by small \( W \) Dipole inverse \( (1, -a) \to (1, a, a^2, \dots) \) if \( \lvert a\rvert < 1 \) Spiking (Wiener) assumes white reflectivity, minimum-phase wavelet, low noise, stationarity Prewhitening add a small percentage to \( r[0] \): stability, less noise boost Predictive, lag \( \alpha = 1 \) same as spiking Predictive, larger \( \alpha \) keeps wavelet front, removes periodic multiples Water reverberation period \( 2d/v \); Backus operator \( (1 + RZ^n)^2 \) Homomorphic log spectrum → cepstrum; separate by quefrency
2-point spiking filter for \( (1, -0.5) \): \( (0.952, 0.381) \), output \( (0.952, -0.095, -0.190) \), error energy about 0.048 (less than 0.0625 for the truncated exact inverse). 150 m water at 1,500 m/s: 0.2 s, 50 samples at 4 ms. Fractal analysis $$\boxed{D = \frac{\log N}{\log(1/r)}} \qquad \boxed{L(r) \propto r^{1 - D}} \qquad \boxed{N(r) \propto r^{-D}}$$
Object \( D \) Cantor set 0.631 Koch curve 1.262 Sierpinski triangle 1.585 Sierpinski carpet 1.893 Menger sponge 2.727
Self-affine profile (fBm) Relation Spectral slope \( S(f) \propto f^{-\beta} \), \( \beta = 2H + 1 \) Dimension \( D = 2 - H = (5 - \beta)/2 \); surface \( D = 3 - H \) Brownian \( H = 0.5 \), \( \beta = 2 \), \( D = 1.5 \) Persistent / antipersistent \( H > 0.5 \) / \( H < 0.5 \) Noise colours white \( \beta = 0 \), pink \( \beta = 1 \), Brownian \( \beta = 2 \)
Semivariogram \( \gamma(h) \propto h^{2H} \); rescaled range \( R/S \propto n^H \). Gutenberg–Richter with constant stress drop: \( N(>r) \propto r^{-2b} \), so \( D = 2b \) (about 2 for \( b \approx 1 \)). Divider worked case: 100 steps at 10 km, 238 at 5 km → \( D \approx 1.25 \). Claim a fractal only over at least one to two decades of scale. ⚠ Trap · Traps: fractals
The divider slope is \( 1 - D \), not \( D \). \( H \) measures smoothness and \( D \) roughness: higher \( H \) means lower \( D \). Use \( D = 2 - H \) for a profile and \( 3 - H \) for a surface.
Back to the chapter Return to the Signal Processing overview , or revise the seismic background in Seismology Basics and Earth’s Internal Structure .
Advertisement Advertisement