# Characterize lasers for DWDM transmission

-January 01, 2001

In a dense wavelength-division multiplexing (DWDM) fiber-optic system, all the optical carriers exist within a narrow span of wavelengths centered around 1550 nm. Each of the many optical signals transmits with a spacing of as little as 0.8 nm from its closest neighbor. To keep these laser signals from interfering with each other, system designers must space them properly and must guarantee that adjacent signals do not drift into one another. Thus, designers need to ensure the laser diodes produce the right amount of a useful signal with the proper characteristics to communicate in a DWDM system.
Typically, DWDM systems use single longitudinal-mode laser (SLML) diodes, also called distributed feedback (DFB) lasers. To use these devices properly, designers must know their power output, operating wavelength, and linewidth, as well as the amount of noise on the laser signal. As a test engineer, you need to know how to obtain this information.
In most cases, you can use an optical spectrum analyzer (OSA) to characterize DFB lasers in the wavelength, or optical-frequency, domain. An OSA provides a tunable optical filter, typically a diffraction grating, that lets the instrument measure optical power at narrow optical-wavelength bands. So, an OSA can directly measure power and wavelength.
Due to physical limitations of the grating, however, an OSA provides a minimum resolution of about 0.08 nm, or 10 GHz, at a 1550 nm. To measure characteristics such as linewidth and noise for a DFB laser, you’ll need a higher resolution technique.
The power in a DFB laser's main spectral peak, often called the main mode, determines the power produced by the device. Peak amplitudes can range from 10 mW to 50 mW or more. An OSA measures optical power in units of dBm (decibels referred to 1 milliwatt) that you can convert to milliwatts using the equation:
dBmopt = 10 log ( Popt / 1 mW)
In an ideal laser, the main spectral peak contains all the power produced by the device, but in reality, the laser signal contains side peaks, also called side modes, that contain some power. A DFB laser’s side-mode suppression ratio (SMSR) describes the amplitude difference between the main mode and the largest side mode in decibels. A typical value is greater than 30 dB, indicating most of the power exists in the main mode. High-performance DFB lasers generate more power in the main mode, and thus they have higher SMSR values.
 Figure 1 The output of an optical spectrum analyzer shows the main peak and side-mode peaks in the light output by a DFB laser diode. The laser has a 0.58-nm peak width at –20.00 dBc. Figure 2 You can use the spectrum on a laser to measure the mode offset and the wavelength spacing (about 2.2 nm) between the two side modes adjacent to the main mode.
You’ll also need to measure the peak wavelength of a DFB laser’s main mode so you know where it transmits most of its power. Because DWDM systems pack wavelengths close together, you must know this value to within a few picometers to ensure adjacent DWDM channels don’t overlap. In addition to determining the wavelength of the main mode, you need to know its bandwidth. When using an OSA, you can set the amplitude level, relative to the peak, at which you measure the bandwidth. The text at the lower right of the instrument display shown in Figure 1 shows a bandwidth of 0.58 nm measured at –20 dBc, a typical measurement point.
Watch those side modes
A laser’s low-power spectral components also figure into DFB characterization, so you must know where they exist in the frequency domain. By using the information from the OSA, shown expanded in Figure 2 , you can determine the mode offset—the wavelength separation (typically 1 nm) between the main mode (the highest peak) and the largest side mode. And you can measure the stop band—the wavelength spacing (about 2.2 nm) between the two side modes adjacent to the main mode. Then, you can calculate the laser’s center offset, which indicates how well the main mode is centered in the stop band. The center offset equals the wavelength of the main mode minus the mean stop band wavelength. The smaller the offset, the better the laser performance. System designers need this wavelength information to establish the positions of the side modes relative to the peak output from the laser. If the side modes are present too far from the peak output, they can interfere with nearby laser signals.
The peak shown in Figure 1 appears to be about 0.2 nm wide. In fact, a laser has much a narrower peak: about 10 MHz, or 0.00008 nm for a typical 1550-nm DFB laser. The spectral peaks displayed by an OSA simply arise from the limited resolution of the instrument’s bandwidth filter. To more accurately measure laser bandwidths, often called linewidths, you can use optical-heterodyne spectrum analysis.
The optical-heterodyne spectrum analysis technique involves mixing an optical signal from a reference laser with an unknown optical signal, thus downconverting the optical signal into the microwave frequency spectrum. Applying the mixed optical signal to an opto-electric converter, typically a high-speed photodiode, lets you measure the mixed signal using an electronic microwave spectrum analyzer. The narrow filters in the spectrum analyzer can resolve optical frequencies down to the reference laser’s linewidth, typically 100 kHz.
In effect, multiplying, or mixing, two functions in the time domain equals cross-correlation (or convolutions, if the mathematical functions are even) of the signals in the frequency domain. Thus, the photodiode’s current represents the cross correlation of the signals from the two lasers in the frequency domain. When the linewidth of the DFB laser under test is small and compared to the linewidth of the external cavity tunable laser source, the cross correlation of laser signals produces a Lorentzian peak with a linewidth equal to the sum of the two laser linewidths. In this case, you can subtract the linewidth of the tunable laser source from the measured linewidth for a more accurate measurement.
 Figure 3 The heterodyne trace of a 1550-nm DFB laser mixed with a reference laser signal peaks at about 8.3 GHz. The spectrum analyzer measured a peak width of 150 MHz at –20 dB. Figure 4 The superimposed traces for a 1550-nm DFB laser operated at temperatures about 0.5°C apart shows the effect of temperature on wavelength.
You could measure the linewidth of the DFB laser at the –3-dB level on the heterodyne signal, but noise would interfere with an accurate measurement. To avoid this problem, measure the width of the Lorentzian peak at –20 dB below the peak. At this point, the linewidth is 10 times the actual linewidth. If measured at –40 dB, it’s 100 times the linewidth. The microwave spectrum analyzer screen in Figure 3 shows the heterodyne trace of a 1550-nm DFB laser and its peak at about 8.3 GHz. The analyzer shows a peak width of 150 MHz at –20 dB, for a laser linewidth of 150 MHz/10, or 15.0 MHz. System designers aim for narrow spectral output—typically about 3 MHz—from DFB lasers because a narrow output concentrates power in the information-carrying main mode.
You also can use the heterodyne technique to measure closely spaced laser frequencies used in DWDM systems. Figure 4 shows two superimposed traces for the same 1550-nm DFB laser operated at temperatures about 0.5°C apart. The separation between the signal peaks is 5.4 GHz, or 0.043 nm. From the two measurements, you can calculate a wavelength-tuning coefficient of 0.086 nm/°C for the laser diode. By properly measuring thermal characteristics of DFB lasers and then controlling their temperature, you can accurately space laser frequencies in a DWDM communication system.
Slight changes in laser intensity, or power, appear as noise in an optical signal and can reduce signal-to-noise ratios (SNR) and increase bit error rates. The ratio of noise to average power, essentially a dynamic-range value, yields a relative-intensity noise (RIN) value for a laser. If the noise level increases or if the average power decreases, the RIN value declines. The total system RIN is important for designers who must determine a system’s “error budget,” a measure of how much noise a system can tolerate.
You can think of RIN as a type of inverse carrier-to-noise ratio measurement. RIN represents the ratio of the mean-square optical-intensity noise to the square of the average optical power, expressed in units of dB/Hz:
RIN = & DP2> / P2 (Eq. 1)
where:
&&/font> DP2> = the mean-square optical-intensity fluctuation (in a 1-Hz bandwidth) at a specified frequency
P = the average optical power
The ratio of the squared optical powers equals the ratio of electrical power produced at an opto-electric converter. Thus, you can express RIN in terms of detected electrical powers:
RINSystem = Nelec / PAVG(elec) (Eq. 2)
where:
Nelec = the power spectral density of the photocurrent at a specific laser frequency
P AVG(elec) = the average power of the photo-current (For clarity, I’ll drop the elec subscript and present terms in electrical units, unless noted otherwise.)
The total system noise, NT(f), at the receiver output results from three fundamental contributions: laser-intensity noise, thermal noise, and photonic shot noise. Laser-intensity noise, NL(f), refers to the noise generated by intensity fluctuations due primarily to spontaneous light emissions that depend on the structure of the laser. The presence of external light feedback or reflections into the laser will increase this noise. Typically, the intensity noise reaches a maximum at a laser diode’s center wavelength. Photonic shot noise, Nq, comes from noise created by the random arrival of photons from a source. The photonic shot noise is independent of frequency. (Operating conditions, such as bias level and modulation frequency, also directly affect the noise level, but those considerations go beyond this discussion. [Ref. 1])
You obtain the total system noise, NT(f), expressed in units of W/Hz, by adding the individual noise contributions:
NT(f) = N L(f) + Nq + Nth(f) (Eq. 3)
When the intensity noise, NL(f), from the laser far exceeds the sum of shot and thermal-noise, system noise essentially equals the intensity noise. Consider a 1-mW laser that produces system noise of –145 dBm/Hz, thermal noise of –168 dBm/Hz, and shot noise of –169 dBm/Hz. Converting these values to linear terms and subtracting the shot and thermal noise from the total yields the laser intensity noise. For this example, that noise amounts to only 0.04 dB less than the system noise. Thus, the system noise for this laser arises almost exclusively from laser-intensity noise.
For the shot-noise value and thermal-noise value to contribute more than 1 dB to the total noise power, the laser-intensity noise would have to decrease by 15 dB, from –145 dBm/Hz to about –160 dBm/Hz. Thus, to improve system performance further, the laser manufacturer would have to decrease the laser’s intensity noise. But as intensity noise goes down, shot noise and thermal noise will predominate if they’re about 5 to 10 dBm/Hz larger than the other noise term.
You can find the relative intensity noise for the laser itself (RINLaser ) from the system’s intensity noise value (RINSystem ) by substituting the individual terms in Equation 3 for Nelec in Equation 2:
RINSystem =
RINLaser + Nq / PAVG + Nth (f)/ P AVG (Eq. 4)
And by rearranging the equation, you can determine RIN noise for the laser:
RINLaser =
RINSystem – Nq / PAVG – Nth (f) / PAVG (Eq. 5)
You can calculate thermal noise using the noise figure specified for the measurement system—the amplifier and electronics that follow the photodiode. A typical value for a measurement system’s thermal noise is –174 dBm/Hz, and using low noise amplifiers will help reduce thermal noise.
You can take shot noise and thermal noise into account using the following formula:
RINLaser =
RINSystem – Nth / (r P AVG (opt))2 RL – 2q / r PAVG (opt)
(Eq. 6)
where:
r = responsivity; r v/RL
PAVG(opt) = average optical power
RL = the load resistance of the input amplifier on the microwave spectrum analyzer
q = the electron charge, 1.60 x 10–19 coulomb
Note that voltage responsivity, rv, is commonly defined at the reference plane of the photodiode. If an amplifier is integral to the opto-electric converter, as is the case in some OSAs, the responsivity is given from the reference plane located after the amplifier. For such cases, the equation for reponsivity should include the gain, Gv, of the amplifier:
r = rv / G vRL
(A complete derivation of Equation 6)
 Figure 5 A plot of RINLaser vs. RINSystem for five detected optical power levels from 0.1 mW to 2.0 mW shows the relationship between total system noise and laser noise less the thermal and shot noise at each power level.
Figure 5 provides a plot of RINLaser vs. RINSystem for five detected optical power levels from 0.1 mW to 2.0 mW. The graph shows the relationship between total system noise and laser noise less the thermal and shot noise at each power level, assuming a fixed noise-figure value of 8 dB for thermal noise and responsivity of the photodetector (0.8 A/W or 40 V/W into 50 V). Thus, at a detected output of 0.1 mW, the laser system has a system RIN of about –127 dB, and a laser RIN of –130 dB. In this case, the laser contributes most of the noise to the signal.
Be careful when using this subtraction method to determine RINLaser . When you subtract one small number from another, small errors can have large effects. Errors in the amplitude accuracy of the photodiode’s frequency response can also cause exaggerated effects. T&MW
Reference
1. Sobol, H., “The Application of Microwave Techniques in Lightwave Systems,” IEEE Journal of Lightwave Technology, Vol. LT-5, March 1987. pp. 293–299.