Relative Intensity Noise Measurement of Semiconductor Lasers

Scope

This article describes the procedure for measuring relative intensity noise (RIN) of a single-mode or near-single-mode semiconductor laser via direct detection and electronic spectrum analysis. Coverage includes the standard photodiode-plus-spectrum-analyzer configuration, shot-noise reference calibration, electronic noise floor subtraction, and the dominant measurement errors. Balanced-detector and self-homodyne RIN measurement variants used at the quantum-noise limit are outside scope.

Definition

Relative intensity noise quantifies the random fluctuations in optical power normalized to the average power. The standard definition is the one-sided power spectral density (PSD) of relative power fluctuations:

$$\text{RIN}(f) \;=\; \frac{S_{\delta P}(f)}{\langle P \rangle^2},$$

where $\langle P \rangle$ is the average optical power and $S_{\delta P}(f)$ is the PSD of the power fluctuation $\delta P(t) = P(t) - \langle P \rangle$. The units are 1/Hz, conventionally expressed in dB/Hz as $10 \log_{10}(\text{RIN})$.

The RIN spectrum of a typical semiconductor laser has three regions:

Typical RIN values for telecom semiconductor lasers:

The shot-noise limit at $\lambda = 1550$ nm and detected photocurrent $I_\text{dc}$ corresponds to a RIN floor of $\text{RIN}_\text{shot} = 2 q / I_\text{dc}$, where $q$ is the electron charge. For 1 mA detected DC photocurrent, this is $\text{RIN}_\text{shot} = 3.2 \times 10^{-16}$ Hz$^{-1} = -155$ dB/Hz.

Measurement principle

The intensity noise of the laser is detected by a photodiode and converted to a photocurrent fluctuation. The AC component of the photocurrent is separated from the DC bias by a bias-tee, amplified to overcome instrument noise, and analyzed on an electronic spectrum analyzer (ESA).

The measured noise PSD on the ESA $S_V(f)$ (in V$^2$/Hz, often displayed in dBm/Hz) contains contributions from three sources that add as variances:

$$S_V(f) \;=\; S_V^\text{laser}(f) + S_V^\text{shot}(f) + S_V^\text{elec}(f),$$

where $S_V^\text{laser}$ is the laser intensity noise of interest, $S_V^\text{shot}$ is the photodetector shot noise, and $S_V^\text{elec}$ is the combined electronic noise floor of the photodetector, amplifier, and ESA.

The shot noise floor at DC photocurrent $I_\text{dc}$ is

$$S_I^\text{shot} \;=\; 2 q I_\text{dc} \quad \text{(in A}^2\text{/Hz)}.$$

Translating to RIN:

$$\text{RIN}_\text{measured}(f) \;=\; \frac{S_I^\text{laser}(f)}{I_\text{dc}^2} \;=\; \frac{S_V^\text{measured}(f) - S_V^\text{shot}(f) - S_V^\text{elec}(f)}{G^2 R^2 I_\text{dc}^2},$$

where $R$ is the photodiode transimpedance load resistance (typically 50 Ω) and $G$ is the cumulative amplifier gain.

Correct RIN extraction requires subtracting both the electronic noise and the shot-noise floor from the measured spectrum, then normalizing by the squared DC photocurrent.

Equipment

The photodetector bandwidth must comfortably exceed the highest frequency of interest. For RIN measurements up to the relaxation oscillation peak of a typical DFB ($\sim 10$ GHz), a 20 GHz detector is appropriate. Bandwidth-limited detection produces a roll-off in the measured spectrum that is mistaken for laser RIN roll-off.

The detector responsivity at the test wavelength and the absolute optical power must be known with $<5\%$ accuracy to convert photocurrent to optical power and to verify the shot-noise reference.

Procedure

1. Configure the measurement chain

Connect the laser output to the photodetector via a calibrated optical attenuator. The photodetector AC output passes through a bias-tee (the DC arm goes to the current meter) and through the LNA to the ESA input. The photodiode bias voltage is supplied through the bias tee's DC arm.

Verify the chain by injecting a known optical power and measuring the DC photocurrent. The relation $I_\text{dc} = \mathcal{R} \cdot P$ (where $\mathcal{R}$ is the responsivity at the test wavelength) confirms detector calibration. Discrepancy indicates either responsivity drift or path loss.

2. Measure the electronic noise floor

With the optical input fully attenuated or the laser turned off (DC photocurrent verified as zero), measure the spectrum on the ESA across the full frequency range of interest. This is $S_V^\text{elec}(f)$ — the combined detector dark, amplifier, and ESA noise floor.

Record this spectrum for later subtraction. The electronic noise floor is fixed for a given detector + amplifier + ESA configuration; small drift between calibration and measurement is acceptable.

3. Verify shot-noise-limited operation

The signal arm of any RIN measurement must be shot-noise-limited at the operating photocurrent — that is, $S_V^\text{shot} \gg S_V^\text{elec}$. To verify, illuminate the detector with a low-noise reference source (an LED or amplified spontaneous emission source) at the same DC photocurrent as the laser measurement. The measured spectrum should be approximately flat versus frequency and should match the calculated shot-noise level $2 q I_\text{dc}$.

If the measured spectrum is dominated by the electronic noise floor (within $\sim 3$ dB), the shot noise is not adequately above the electronic floor. Reduce the LNA noise figure, increase optical power (and therefore $I_\text{dc}$), or improve detector responsivity to recover headroom.

Alternative: calibrate the absolute gain $G$ of the chain by measuring the shot noise at multiple DC currents (using a variable attenuator) and fitting $S_V^\text{shot}(I_\text{dc}) - S_V^\text{elec} = 2 q I_\text{dc} G^2 R^2$. This is the shot-noise calibration sweep and is the standard absolute calibration method for RIN measurements.

4. Acquire the laser spectrum

Turn on the laser at the operating point of interest (typically biased $\geq 2 I_\text{th}$). Measure $S_V^\text{measured}(f)$ on the ESA across the full frequency range, with the same resolution bandwidth (RBW) and video bandwidth as the noise floor measurement. Record the DC photocurrent $I_\text{dc}$ simultaneously.

Optical power into the detector should be set so that $I_\text{dc}$ is in the linear range of the detector — typically 0.1–10 mA — and so that shot noise comfortably exceeds the electronic floor.

5. Compute RIN

For each frequency point:

$$\text{RIN}(f) \;=\; \frac{S_V^\text{measured}(f) - S_V^\text{elec}(f) - S_V^\text{shot}(I_\text{dc})}{G^2 R^2 I_\text{dc}^2}.$$

Equivalently, in dB/Hz with all noise quantities in dBm/Hz at 50 Ω load:

$$\text{RIN}(f) \text{ [dB/Hz]} \;=\; S_V^\text{measured} - 10 \log_{10}(I_\text{dc}^2 \cdot 50 \cdot G^2 \cdot 10^{-3}) - \text{shot-noise correction}.$$

The factor $10^{-3}$ converts mW (dBm) to W (dBW). The shot-noise correction reduces the apparent RIN by the shot-noise contribution.

For high-RIN regions (e.g., the relaxation oscillation peak), the shot-noise correction is negligible. For low-RIN regions (the floor between $1/f$ corner and the relaxation peak), the shot-noise correction is essential.

6. Report

Report:

• RIN as a function of frequency, in dB/Hz
• Operating point: bias current, optical power, temperature
• Frequency range and resolution bandwidth
• Shot-noise calibration method (sweep or reference source) and verified shot-noise floor
• Electronic noise floor in dBm/Hz across the frequency range
• DC photocurrent $I_\text{dc}$ during the measurement

Worked example

A 1550 nm DFB laser is biased at 50 mA ($I_\text{th} = 8$ mA) and coupled to an InGaAs PIN photodiode with responsivity $\mathcal{R} = 0.9$ A/W via an optical attenuator. The optical power at the detector is set to give $I_\text{dc} = 1$ mA.

Detector + LNA + ESA gain calibration via shot-noise sweep: $G = 23$ dB (factor $\approx 14.1$), with 50 Ω load. Shot-noise level at $I_\text{dc} = 1$ mA:

$$S_V^\text{shot} \;=\; 2 q I_\text{dc} \cdot R \cdot G^2 \;=\; 2 \cdot 1.6 \times 10^{-19} \cdot 10^{-3} \cdot 50 \cdot 200 \;=\; 3.2 \times 10^{-18} \text{ V}^2/\text{Hz},$$

corresponding to $-155$ dBm/Hz on the ESA.

Electronic noise floor (laser off): $-162$ dBm/Hz across the band — comfortably below shot noise.

Measured spectrum with laser on at $I_\text{dc} = 1$ mA:

Subtracting shot noise and electronic noise (all in linear units), then normalizing by $I_\text{dc}^2 \cdot R \cdot G^2$:

The relaxation oscillation peak appears near 8 GHz, with a peak RIN of $-140$ dB/Hz. The off-peak RIN of $\sim -150$ dB/Hz is typical of a well-designed telecom DFB laser. The shot-noise correction is significant near the off-peak floor (where the laser RIN is only $\sim 5$ dB above shot noise) and negligible at the relaxation peak.

Sources of measurement error

Electronic noise floor not subtracted. The dominant error in low-RIN measurements. Without subtraction, the apparent floor RIN is artificially elevated and frequency-dependent following the electronic noise profile.

Shot noise not subtracted. For laser RIN within 10 dB of shot noise, omitting the shot-noise subtraction produces a systematic overestimate of laser RIN. The error becomes severe as the laser RIN approaches the shot-noise limit.

Detector bandwidth roll-off mistaken for laser roll-off. The product of detector responsivity and load impedance has a frequency response that rolls off at high frequencies. Without correction by the calibrated chain gain $G(f)$, the measured RIN above the detector 3-dB bandwidth is systematically low. Standard mitigation: measure $G(f)$ via the shot-noise sweep at each frequency.

Photodetector nonlinearity at high optical power. PIN photodiodes saturate above a power-dependent current limit, producing apparent excess noise that scales with optical power but is not laser intrinsic. Verify operation in the linear regime by checking that the shot-noise floor scales as $I_\text{dc}$, not slower.

RF pickup and 50/60 Hz line noise. External RF interference at telecom or wireless bands (cellular, Wi-Fi, instrument cooling fans) appears as discrete spurs in the RIN spectrum. Shielding the photodetector and using a Faraday cage if needed eliminates most pickup. Spurs that persist should be excluded from reported RIN.

Polarization-induced fluctuation in the optical path. If the photodetector responsivity is polarization-dependent, polarization drift in the fiber path produces fluctuations interpreted as RIN. Verify polarization stability over the measurement duration or use a polarization-insensitive detector.

Insufficient averaging at low frequencies. Low-frequency RIN ($<1$ kHz) requires long acquisition times to resolve. Standard frequency-domain measurements with $\sim$Hz-level RBW and adequate averaging are required. For $1/f$-region characterization, time-domain measurement and FFT analysis is often more efficient than direct RF spectrum measurement.

Ground loops and microphonic noise. Mechanical vibration coupling into the photodetector or amplifier produces low-frequency excess noise. Mitigation: isolated power supply, vibration isolation table, short cables.

ESA resolution bandwidth not noted. Reported RIN values must be normalized by 1 Hz (the conventional unit). ESA spectrum readings in dBm at finite RBW must be corrected by subtracting $10 \log_{10}(\text{RBW}/1~\text{Hz})$.

Validation

The measured shot-noise level should scale linearly with $I_\text{dc}$ across at least one decade. Deviation from linear scaling indicates either detector saturation, electronic-noise-dominated operation, or calibration error.

For below-threshold operation ($I < I_\text{th}$, with the laser in spontaneous emission mode), the measured RIN floor should equal $-10 \log_{10}(2 q / I_\text{dc})$, the shot-noise limit. This is an independent check on the absolute calibration of the chain.

The off-peak RIN level for a known device should agree with published values for that device type within $\pm 3$ dB. Disagreement by $>10$ dB indicates calibration error rather than novel device behavior.

References

For the foundational treatment of laser intensity noise including the shot-noise floor and the quantum origin of RIN, see Coldren, Corzine, and Mašanović (2012), chapter 4. For the relaxation oscillation contribution to RIN in semiconductor lasers, see Petermann (1991), Laser Diode Modulation and Noise. For shot-noise calibration procedures and modern RIN measurement automation, see Yariv and Yeh (2007), Photonics: Optical Electronics in Modern Communications.