Fourier Transforms, Lock-in Detection, Box-car Integration and Modulation
CSIR NET Physical Sciences: sampling and FFT, aliasing, lock-in amplifier, phase-sensitive detection, box-car integrator, AM, FM and pulse modulation.
- 19 min
On this page · 13 sections
Prerequisites
- Fourier Series, Fourier and Laplace Transforms: definitions, standard pairs and the convolution theorem. This page applies them to instruments.
- Noise, Filtering, Shielding and Grounding: white and 1/f noise, equivalent noise bandwidth and the \( \sqrt N \) rule.
- Linear ODEs and Special Functions: Bessel functions appear in the FM spectrum.
The Hook: A Signal a Thousand Times Below the Noise
Fourier Transforms in the Laboratory
With \( F(f) = \int f(t)e^{-i2\pi ft}\,dt \), the pairs worth carrying into experiments:
| Time domain | Frequency domain | Experimental meaning |
|---|---|---|
| Rectangular pulse of width \( T \) | \( T\,\text{sinc}(\pi fT) \), first zeros at \( \pm1/T \) | a short gate or pulse has a broad spectrum |
| Gaussian, width \( \sigma_t \) | Gaussian, width \( 1/(2\pi\sigma_t) \) | minimum time–bandwidth product |
| Exponential decay \( e^{-t/\tau} \) (\( t > 0 \)) | power spectrum Lorentzian, FWHM \( \Delta f = 1/(\pi\tau) \) | lifetime broadening: a population lifetime \( \tau \) (field decaying as \( e^{-t/2\tau} \)) gives natural linewidth \( 1/(2\pi\tau) \) |
| \( \cos 2\pi f_0t \) | delta functions at \( \pm f_0 \) | a pure tone is a line |
| Dirac comb, spacing \( T_s \) | Dirac comb, spacing \( 1/T_s \) | sampling replicates the spectrum |
| Product \( f(t)g(t) \) | convolution \( F * G \) | mixing shifts spectra (modulation, lock-in) |
| Convolution \( f * g \) | product \( FG \) | a linear instrument multiplies the spectrum by its transfer function |
Two theorems used constantly: Parseval (energy is the same in both domains) and Wiener–Khinchin (the power spectral density is the Fourier transform of the autocorrelation function). Together they justify computing noise from spectra.
Time–bandwidth: \( \Delta t\,\Delta f \gtrsim 1 \) (order of magnitude; the exact constant depends on how widths are defined). A measurement lasting \( T \) cannot resolve frequencies closer than about \( 1/T \).
Sampling and the Nyquist criterion
Sampling at rate \( f_s \) replicates the spectrum at multiples of \( f_s \). A band-limited signal with highest frequency \( f_{\max} \) can be reconstructed exactly if
$$\boxed{f_s > 2f_{\max}}$$(\( f_s/2 \) is the Nyquist frequency). Components above \( f_s/2 \) alias: they appear at
$$f_{\text{alias}} = \lvert f - kf_s\rvert \quad (k \text{ the nearest integer to } f/f_s)$$At \( f_s = 1 \) kHz, inputs at 300, 700, 1300 and 1700 Hz all appear at 300 Hz, and 900 Hz appears at 100 Hz. Aliasing cannot be undone after sampling, so an analogue anti-aliasing filter goes before the ADC. The stroboscope, and wagon wheels turning backwards on film, are aliasing.
The discrete Fourier transform and FFT
\( N \) samples at interval \( \Delta t \) (record length \( T = N\Delta t \)) give \( N \) frequency points:
$$\boxed{\Delta f = \frac{1}{N\Delta t} = \frac{1}{T}, \qquad f_{\max} = \frac{f_s}{2}}$$Resolution depends only on the record length, not on the sampling rate. 1024 samples at 10 kHz give bins of 9.77 Hz over 0–5 kHz. The FFT computes the DFT in \( O(N\log N) \) operations instead of \( O(N^2) \); the radix-2 version wants \( N \) a power of 2 (zero-padding interpolates the spectrum but does not improve true resolution).
Spectral leakage: a finite record is the signal times a rectangular window, so each line is convolved with a sinc and leaks into neighbouring bins unless an integer number of periods fits the record. Windowing (Hann, Hamming, Blackman, flat-top) tapers the record ends and suppresses leakage at the cost of a wider main lobe.
Modulation
Modulation impresses a message \( m(t) \) on a carrier. In the laboratory it moves a signal away from DC and 1/f noise; in communications it shares the spectrum.
Amplitude modulation (AM)
$$s(t) = A_c\left[1 + \mu\cos\omega_mt\right]\cos\omega_ct = A_c\cos\omega_ct + \frac{\mu A_c}{2}\cos(\omega_c + \omega_m)t + \frac{\mu A_c}{2}\cos(\omega_c - \omega_m)t$$- Modulation index \( \mu = \dfrac{V_{\max} - V_{\min}}{V_{\max} + V_{\min}} \) from the envelope; \( \mu > 1 \) is over-modulation (envelope distortion).
- Bandwidth \( 2f_m \).
- Power: \( P_T = P_c\left(1 + \dfrac{\mu^2}{2}\right) \); fraction in the sidebands (which carry the information) \( \dfrac{\mu^2}{2 + \mu^2} \), at most \( 1/3 \) at \( \mu = 1 \).
Variants: DSB-SC (carrier suppressed: all power in sidebands, needs coherent detection), SSB (one sideband: half the bandwidth), VSB (vestigial, used where low frequencies matter).
Demodulation: an envelope detector (diode, then RC) needs \( 1/f_c \ll RC \ll 1/f_m \); to follow the falling envelope without “diagonal clipping” the stricter condition is \( RC \leq \sqrt{1 - \mu^2}/(\mu\omega_m) \). Synchronous (coherent) detection multiplies by a local copy of the carrier and low-pass filters, exactly what a lock-in amplifier does.
Angle modulation: FM and PM
FM: instantaneous frequency \( f_c + k_fm(t) \). For a tone, with peak deviation \( \Delta f \),
$$s(t) = A_c\cos\left(\omega_ct + \beta\sin\omega_mt\right), \qquad \beta = \frac{\Delta f}{f_m}$$The spectrum is a set of lines at \( f_c \pm nf_m \) with amplitudes \( A_cJ_n(\beta) \) (Bessel functions). The total power is constant, \( A_c^2/2 \), whatever \( \beta \). The carrier line vanishes where \( J_0(\beta) = 0 \) (first at \( \beta \approx 2.405 \)), a classic way to calibrate deviation. Carson’s rule for the bandwidth containing about 98% of the power:
$$\boxed{B \approx 2(\Delta f + f_m) = 2f_m(\beta + 1)}$$\( \beta \ll 1 \) is narrowband FM (bandwidth about \( 2f_m \), like AM); large \( \beta \) is wideband FM, which trades bandwidth for noise immunity. PM differs from FM only in that the phase, not the frequency, is proportional to \( m(t) \).
Pulse and digital modulation
| Scheme | Message carried by | Note |
|---|---|---|
| PAM | pulse amplitude | the sampled signal itself |
| PWM (PDM) | pulse width | motor drives, class-D amplifiers |
| PPM | pulse position | |
| PCM | binary code of quantised samples | \( n \) bits at \( f_s \) give \( nf_s \) bit/s; 8 bits at 8 kHz = 64 kbit/s |
| ASK, FSK, PSK (BPSK, QPSK), QAM | amplitude, frequency, phase of a carrier | digital versions of AM, FM, PM |
Phase-Sensitive Detection: The Lock-in Amplifier
How it works
graph LR
A[Experiment excited at f_ref, e.g. chopped light] --> B[Detector and preamplifier]
B --> C[Band-pass or AC coupling]
C --> D[Multiplier: phase-sensitive detector]
R[Reference oscillator at f_ref] --> P[Phase shifter]
P --> D
D --> E[Low-pass filter, time constant tau]
E --> F[DC output proportional to signal amplitude times cos theta]Let the signal be \( V_s\sin(\omega t + \theta) \) and the reference \( \sin\omega t \). Their product is
$$V_s\sin(\omega t + \theta)\sin\omega t = \frac{V_s}{2}\left[\cos\theta - \cos(2\omega t + \theta)\right]$$The low-pass filter removes the \( 2\omega \) term, leaving the DC output
$$\boxed{X = \frac{V_s}{2}\cos\theta}$$(Commercial instruments scale so that \( X = V_{\text{rms}}\cos\theta \).) Noise at a frequency \( \omega' \neq \omega \) produces a beat at \( \lvert\omega' - \omega\rvert \), which the low-pass filter removes unless \( \lvert\omega' - \omega\rvert \) lies within its bandwidth. So the lock-in is a band-pass filter centred on \( f_{\text{ref}} \) whose width is set by the output time constant, and whose centre automatically tracks the reference. No passive filter can be that narrow and that well-centred.
Dual-phase operation
A second detector with the reference shifted by 90° gives \( Y = \frac{V_s}{2}\sin\theta \). Then
$$R = \sqrt{X^2 + Y^2}, \qquad \theta = \tan^{-1}(Y/X)$$\( R \) is independent of the phase setting. Measuring \( \theta \) gives the delay of the response, so X and Y also separate in-phase (e.g. absorptive, resistive) from quadrature (dispersive, reactive) components.
Output time constant and noise bandwidth
A single RC stage (6 dB/octave) of time constant \( \tau \) has equivalent noise bandwidth
$$\text{ENBW} = \frac{1}{4\tau}$$and the output takes several \( \tau \) to settle (about \( 5\tau \) to within 1%). Steeper roll-offs (12, 18, 24 dB/octave) reduce the ENBW for the same \( \tau \) but settle more slowly. Output noise \( = e_n\sqrt{\text{ENBW}} \).
Practical points Part C uses
- Square-wave references: many lock-ins multiply by a \( \pm1 \) square wave, whose Fourier series contains \( \sin\omega t + \frac{1}{3}\sin3\omega t + \frac{1}{5}\sin5\omega t + \cdots \) (times \( 4/\pi \)). The lock-in then also responds to signal and noise at odd harmonics with weights \( 1/3, 1/5, \ldots \); a pre-filter suppresses them. With a \( \pm1 \) square reference, a sine signal gives output \( (2/\pi)V_s\cos\theta \).
- Harmonic detection: setting the reference to \( 2f \) measures the second harmonic, which for a small sinusoidal modulation of a parameter is proportional to the second derivative of the response (derivative spectroscopy, modulation spectroscopy).
- Dynamic reserve: the ratio of the largest tolerable noise to full-scale signal (60 dB means noise 1000 times full scale) before the input overloads.
- Choice of \( f_{\text{ref}} \): above the 1/f corner, away from mains frequency and its harmonics.
- Phase setting: adjust for maximum X (or use R) to get the full signal; a wrong phase reduces X by \( \cos\theta \).
The Box-car Integrator (Gated Integrator)
Some signals are short pulses at a low repetition rate: a laser-excited fluorescence decay, a time-of-flight peak, a pulsed NMR echo. Their duty cycle may be \( 10^{-5} \) (10 ns pulses at 1 kHz). A lock-in would spend almost all its time integrating noise between pulses. The box-car instead:
- Opens a gate of width \( t_g \) at a fixed delay after each trigger.
- Integrates the input only while the gate is open.
- Averages over many pulses (linear summation or exponential averaging).
Noise outside the gate is rejected completely, and averaging \( N \) pulses improves SNR by \( \sqrt N \): \( 10^4 \) pulses give a factor of 100. Two modes:
- Static (fixed delay): tracks the signal at one point of the waveform over time (e.g. while a wavelength is scanned).
- Scanning (waveform recovery): the delay is swept slowly, so the averaged output reproduces the whole waveform, with time resolution set by the gate width.
Gate width trade-off: a narrower gate gives better time resolution but collects less signal per pulse, so more pulses are needed for the same SNR. The modern equivalent is a fast digitiser with digital averaging, which records all delays at once.
Memory Tricks
Don’t Confuse
| This | vs | That |
|---|---|---|
| Sampling rate sets the maximum frequency \( f_s/2 \) | Record length sets the resolution \( 1/T \) | |
| Lock-in: frequency-domain recovery, needs a modulated signal | Box-car: time-domain recovery, needs a trigger, suits low duty cycle | |
| Lock-in X depends on the phase setting | R does not | |
| AM total power depends on \( \mu \) | FM total power is constant | |
| Envelope detection (AM, incoherent) | Synchronous detection (DSB-SC, lock-in, coherent) | |
| Zero-padding: smoother-looking spectrum | Longer record: genuinely finer resolution |
Common Mistakes
Practice
All questions below are original practice questions written for this page. They are not from any CSIR NET paper.
Level 1: Foundation
- A signal is sampled at 8 kHz. What is the highest frequency that can be represented without aliasing?
- An FFT uses 2048 samples taken at 20 kHz. What is the bin width?
- A dual-phase lock-in reads \( X = 3 \) µV and \( Y = 4 \) µV. What are \( R \) and \( \theta \)?
Answers
- 4 kHz (the Nyquist frequency).
- 9.77 Hz. \( T = 2048/20\,000 = 0.1024 \) s; \( \Delta f = 1/T \).
- \( R = 5 \) µV, \( \theta = 53.1^\circ \).
Level 2: Part B Style
Practice Q1. A 900 Hz sine wave is sampled at 1 kHz without an anti-aliasing filter. The reconstructed signal appears at:
(a) 900 Hz (b) 500 Hz (c) 100 Hz (d) 1900 Hz
Answer: (c) 100 Hz
\( \lvert 900 - 1 \times 1000\rvert = 100 \) Hz. The 900 Hz tone is above the 500 Hz Nyquist frequency and folds back.
Practice Q2. A 100 W carrier is amplitude-modulated with index 0.5. The total transmitted power and the fraction in the sidebands are:
(a) 112.5 W, 11.1% (b) 125 W, 20% (c) 150 W, 33% (d) 106.25 W, 5.9%
Answer: (a) 112.5 W and 11.1%
\( P_T = 100(1 + 0.25/2) = 112.5 \) W; sideband fraction \( 12.5/112.5 = 0.111 \). Option (b) uses \( \mu^2 \) instead of \( \mu^2/2 \) (125 W, 20%); option (c) uses \( \mu \) (150 W, 33%); option (d) counts only one sideband.
Practice Q3. An FM broadcast signal has peak deviation 75 kHz and maximum modulating frequency 15 kHz. Carson’s bandwidth is:
(a) 90 kHz (b) 150 kHz (c) 180 kHz (d) 30 kHz
Answer: (c) 180 kHz
\( B = 2(75 + 15) = 180 \) kHz; \( \beta = 5 \), wideband FM.
Practice Q4. A lock-in amplifier uses a single-pole output filter with \( \tau = 1 \) s. The input white-noise density is 1 µV/\( \sqrt{\text{Hz}} \). The rms noise at the output, referred to the input, is about:
(a) 0.25 µV (b) 0.5 µV (c) 1 µV (d) 316 µV
Answer: (b) 0.5 µV
ENBW \( = 1/(4\tau) = 0.25 \) Hz; noise \( = 1 \times \sqrt{0.25} = 0.5 \) µV. Option (a) forgets the square root; option (d) is the noise in a 100 kHz bandwidth without the lock-in.
Level 3: Part C Style
Practice Q5. A system is sampled at \( f_s = 1 \) kHz. Input components at 300 Hz, 700 Hz, 1300 Hz and 2300 Hz are present. Consider:
- All four appear at 300 Hz in the sampled spectrum.
- A 450 Hz anti-aliasing low-pass filter before the ADC would leave only the genuine 300 Hz component (apart from filter leakage).
- Doubling the record length at the same \( f_s \) halves the FFT bin width.
- Doubling \( f_s \) to 2 kHz, keeping the number of samples fixed, would make the 700 Hz component appear correctly and would also halve the bin width.
Which are correct?
(a) 1, 2 and 3 only (b) 1 and 4 only (c) 2, 3 and 4 only (d) 1, 2, 3 and 4
Answer: (a) 1, 2 and 3 only
Statement 1: \( \lvert700 - 1000\rvert = 300 \), \( \lvert1300 - 1000\rvert = 300 \), \( \lvert2300 - 2000\rvert = 300 \). True.
Statement 2: the filter removes everything above 450 Hz before sampling, so nothing can fold. True.
Statement 3: \( \Delta f = 1/T \). True.
Statement 4: at 2 kHz the Nyquist frequency is 1 kHz, so 700 Hz is represented correctly; but with \( N \) fixed the record is half as long, so \( \Delta f \) doubles. False.
Practice Q6. A chopped optical signal gives a square wave of peak-to-peak amplitude \( 2V_0 \) (switching between \( +V_0 \) and \( -V_0 \) after AC coupling) at the reference frequency. A lock-in amplifier that reports the rms amplitude of the fundamental is correctly phased. Consider:
- It reads about \( 0.90V_0 \).
- It reads \( V_0/\sqrt2 \approx 0.71V_0 \), the rms of a sine of amplitude \( V_0 \).
- If the phase is then shifted by 90°, X reads zero and R is unchanged.
- If a sine reference at \( 3f \) is used instead, R reads one third of the fundamental reading.
Which are correct?
(a) 1, 3 and 4 only (b) 2 and 3 only (c) 1 and 3 only (d) 2, 3 and 4 only
Answer: (a) 1, 3 and 4 only
Statement 1: the square wave’s fundamental has amplitude \( 4V_0/\pi \); its rms is \( 4V_0/(\pi\sqrt2) = 0.900V_0 \). True.
Statement 2: the square wave’s fundamental is larger than \( V_0 \), not smaller. False.
Statement 3: \( X = R\cos\theta \) goes to zero at 90°; \( R = \sqrt{X^2 + Y^2} \) does not depend on phase. True.
Statement 4: a square wave’s third harmonic has amplitude \( 4V_0/(3\pi) \), one third of the fundamental. True.
Practice Q7. Fluorescence excited by 10 ns laser pulses at 1 kHz decays with a lifetime of 50 ns. The single-shot SNR at the peak is 0.5. Consider:
- The duty cycle of the signal is of order \( 10^{-4} \), so a box-car integrator is better suited than a lock-in amplifier.
- To reach SNR 50 at a fixed delay, about \( 10^4 \) pulses must be averaged, taking about 10 s.
- A gate of 5 ns, scanned in delay from 0 to 300 ns, can recover the decay curve.
- The natural (Lorentzian) linewidth, FWHM, corresponding to the 50 ns excited-state lifetime is about 3.2 MHz.
Which are correct?
(a) 1 and 2 only (b) 1, 2 and 3 only (c) 2, 3 and 4 only (d) 1, 2, 3 and 4
Answer: (d) 1, 2, 3 and 4
Statement 1: the signal lasts a few lifetimes, roughly 100–200 ns, out of every 1 ms: duty cycle of order \( 10^{-4} \). The box-car gates out the noise in between. True.
Statement 2: SNR must rise by 100, so \( N = 100^2 = 10^4 \) pulses at 1 kHz: 10 s. True.
Statement 3: scanning mode with a gate much shorter than the 50 ns lifetime reconstructs the waveform. True.
Statement 4: the population decays as \( e^{-t/\tau} \), so the emitted field decays as \( e^{-t/2\tau} \) and its power spectrum has FWHM \( \Delta f = 1/(2\pi\tau) = 1/(2\pi \times 5 \times 10^{-8}\ \text{s}) \approx 3.2 \) MHz. True. Using \( 1/(\pi\tau) \) (6.4 MHz) treats the intensity decay as if it were the field decay, a common slip.
Quick Revision Table
| Item | Result |
|---|---|
| Nyquist | \( f_s > 2f_{\max} \); alias at \( \lvert f - kf_s\rvert \) |
| DFT resolution | \( \Delta f = 1/(N\Delta t) = 1/T \) |
| FFT cost | \( O(N\log N) \) |
| Decay ↔ Lorentzian | field \( e^{-t/\tau} \): FWHM \( 1/(\pi\tau) \); population lifetime \( \tau \): natural width \( 1/(2\pi\tau) \) |
| Rect ↔ sinc | first zero at \( 1/T \) |
| AM | sidebands \( f_c \pm f_m \), \( P_T = P_c(1 + \mu^2/2) \), sideband fraction \( \le 1/3 \) |
| Envelope detector | \( 1/f_c \ll RC \le \sqrt{1 - \mu^2}/(\mu\omega_m) \) |
| FM | \( \beta = \Delta f/f_m \), lines \( J_n(\beta) \), Carson \( 2(\Delta f + f_m) \) |
| PCM rate | bits × sampling rate |
| Lock-in | \( X = R\cos\theta \), \( Y = R\sin\theta \), \( R = \sqrt{X^2 + Y^2} \) |
| ENBW (single RC) | \( 1/(4\tau) \) |
| Square reference | also detects odd harmonics at \( 1/3, 1/5, \ldots \) |
| Box-car | gate + delay + average; SNR \( \times\sqrt N \); for low duty cycle |
Key Takeaways
What’s Next
- High-Frequency Devices: the oscillators, mixers and detectors that generate and demodulate carriers at microwave frequencies.
- Noise, Filtering, Shielding and Grounding for the noise models used here.
- Fourier Series, Fourier and Laplace Transforms for the mathematics.
- Electronics and Experimental Methods (Advanced): Quick Revision.