Continued demands for increased bandwidth have resulted in the deployment of fiber-optic-based networks. Solid-state lasers drive these fiber-optic lines, which can provide high information density. Highly packed data schemes such as DWDM (dense-wavelength-division multiplexing) involve driving a fiber with multiple lasers to obtain large, multichannel data streams. The narrow channel spacing relies on precise control of laser wavelength to within 0.1 nm. Lasers can provide this precision, but temperature variation influences operation. The intensity of a laser peaks sharply versus wavelength, implying that control within 0.1 nm of laser wavelength is necessary to maintain performance (Figure 1a). A typical laser wave length-versus-temperature plot exhibits a 0.1-nm/°C slope, which means that, although temperature facilitates tuning laser wavelength, the temperature must not vary after the laser wavelength has stabilized (Figure 1b). Typically, temperature control of 0.1°C is necessary to maintain laser operation well within 0.1 nm.

A temperature controller for this application must meet some unusual requirements. Most notably, because of ambient-temperature variations and laser-operation uncertainties, the controller must either source or remove heat to maintain control. Peltier-based TECs (thermoelectric coolers), permit this type of control, but the controller must be truly bidirectional. The controller's heat-flow control must not have a dead zone or untoward dynamics in the 'hot-to-cold' transition region. Additionally, the temperature controller must be a precision device that can maintain control well inside 0.1°C over time and temperature variations. Laser-based-system packaging is compact, so that size with efficient operation is important to avoid excessive heat dissipation. Finally, the controller must operate from a single low-voltage source, and the controller's power delivery—presumably a switched-mode type—must not corrupt the supply with noise.

Temperature controller meets unusual demands

The TEC temperature controller in Figure 2 meets these demands. The circuit includes the input, which includes a DAC and thermistor; a controller stage; and the output stage, which includes the TEC. The LTC1658 DAC, IC₁, and the thermistor form a bridge, and IC₂ amplifies the output. The LTC1923 controller, IC₃, is a PWM that provides appropriately modulated and phased drive to the power-output stage.

The laser is an electrically delicate and expensive load. As such, the controller provides a variety of monitoring, limiting, and overload-protection capabilities: soft-start and overcurrent protection, TEC-voltage sense, current sense, and 'out-of-bounds' temperature sensing. Aberrant operation results in circuit shutdown, preventing laser-module damage. Two other features promote system-level compatibility. A PLL-based oscillator permits reliable clock synchronization of multiple controller ICs in multilaser systems.

Finally, the switched-mode power delivery to the TEC is efficient, but special design considerations are necessary to ensure that switching-related noise does not reflect back into the host's power supply. The controller includes edge slew limiting, which minimizes switching-related harmonics by slowing down the power stages' transition times. This feature greatly reduces high-frequency harmonic content, which prevents excessive switching-related noise from corrupting the power supply or the laser (Reference 1). The switched-mode-power-output stage, an H-bridge type, permits efficient bidirectional drive to the TEC, allowing either heating or cooling of the laser. The thermistor, TEC, and laser, which the manufacturer packages as one module, have tight thermal coupling.

The DAC allows you to adjust the temperature setpoint to any individual laser's optimum operating point, which manufacturers normally specify for each laser. Controller gain and bandwidth adjustments optimize the thermal-loop response for best temperature stability.

Thermal-loop considerations

The key to high-performance temperature control is matching the controller's gain bandwidth to the thermal feedback path. Theoretically, this matching is a simple matter using conventional servo-feedback techniques. Practically, the long time constants and uncertain delays inherent in thermal systems present a challenge. Both servo systems and oscillators are feedback systems, but one is supposed to oscillate, and the other is not. This unfortunate relationship is very apparent in thermal-control systems.

A model of the thermal-control loop is simply a network of resistors and capacitors (Figure 3). The resistors are equivalent to the thermal resistance, and the capacitors are equivalent to thermal capacity. In Figure 3, the TEC, TEC-sensor interface, and the sensor all have RC factors that contribute to a lumped delay in the system's ability to respond. To prevent oscillation, gain-bandwidth limiting is necessary to account for this delay. Because high gain bandwidth is desirable for good control, delays need to be as low as possible. Laser-module purveyors presumably address this issue during manufacturing.

The model also includes insulation between the controlled environment and the uncontrolled ambient. The function of insulation is to keep the loss-rate low so that the temperature-control device can keep up with the losses. For any given system, the higher the ratio between the TEC-sensor time constants and the insulation time constants, the better the performance of the control loop (see sidebar 'Practical considerations in TEC-based control loops').

Optimize the temperature-control loop

Temperature-control-loop optimization begins with thermal characterization of the laser module. As mentioned, the ratio between the TEC-sensor and insulation time constants is important. Determination of this information places realistic bounds on achievable controller gain bandwidth. When you subject a typical laser module to a 40°C step change in ambient temperature, the ambient-to-sensor lag, measured in minutes, exhibits a classic first-order response (Figure 4). The figure plots the laser module's internal temperature, monitored by its thermistor, versus time with no power to the TEC.

You can characterize the TEC sensor's lumped delay by operating the laser module in Figure 2's circuit with the gain set at maximum and with no compensation capacitor. The result is large-signal oscillation due to thermal lag dominating the loop (Figure 5a). This figure presents a great deal of valuable information. When a circuit 'doesn't work' because 'it oscillates,' whether at millihertz or gigahertz, four burning questions should immediately dominate the pending investigation: What are the oscillation frequency, the amplitude, the duty cycle, and the waveshape? The solution to the problem invariably resides in the answers to these queries. Just stare thoughtfully at the waveform, and the truth will bloom.

In this case, the frequency, which TEC-sensor lag primarily determines, limits how much loop bandwidth you can achieve. The high ratio of this frequency to the laser module's thermal time constant—the lag characteristic in Figure 4—means that simple, dominant-pole loop compensation is effective. The saturation-limited waveshape suggests that excessive gain is driving the loop into full cooling and heating states. Finally, the asymmetric duty cycle reflects the TEC's differing thermal efficiency in the cooling and heating modes.

Reducing the controller's gain bandwidth from the extremes of Figure 5a produced Figure 5b's display. This waveform results from a small step change in the temperature setpoint of approximately 0.1°C. Gain bandwidth is still excessively high, producing a damped, ringing response that lasts longer than 2 minutes. The loop is just marginally stable. Figure 5c's test conditions are identical to Figure 5b's, but the gain bandwidth is significantly smaller. The response is still not optimal, but settling occurs in approximately 4.5 seconds, or approximately 25 timesfaster than the previous case. Figure 5d's response, taken at further reduced gain-bandwidth settings, is nearly critically damped and settles cleanly in about 2 seconds. A laser module optimized in this fashion easily attenuates external temperature shifts by a factor of thousands without overshoots or excessive lags.

Further, although substantial thermal differences exist between various laser modules, some generalized guidelines on gain-bandwidth values are possible (see the sidebar). A dc gain of 1000 is sufficient for this application's required temperature control, with bandwidth below 1 Hz providing adequate loop stability. Figure 2's suggested gain and bandwidth values reflect these conclusions, although stability testing is mandatory in all cases.

Verify temperature stability

After you optimize the loop, you can measure the temperature stability by monitoring the thermistor-bridge offset with a stable, calibrated differential amplifier. Note that this measurement monitors thermistor stability. The laser's temperature stability is somewhat different due to slight thermal decoupling and variations in laser power dissipation.

Figure 6 records ±1 millidegree baseline stability over 50 sec in the cooling mode. A more stringent test measures longer term stability with significant variations in ambient temperature. Figure 7a's strip-chart recording measures cooling-mode stability against an environment that steps 20°C above ambient every hour over nine hours. (Yes, such archaic devices as strip charts are still useful.) The data shows a variation of 0.008°C, indicating a thermal gain of 2500. (The term 'thermal gain' is temperature-control-aficionado jargon for the ratio of ambient-to-controlled temperature variation.) The 0.0025°C baseline tilt over the nine-hour plot length derives from varying ambient temperature. Figure 7b uses identical test conditions as Figure 7a, except that the controller operates in the heating mode. The TEC's higher heating-mode efficiency furnishes greater thermal gain, resulting in a fourfold stability improvement to about 0.002°C variation. Baseline tilt, just detectable, shows a similar fourfold improvement versus Figure 7a.

This level of performance ensures the desired stable-laser characteristics. Thermistor aging characteristics primarily determine temperature stability over years.

The switched-mode power delivery to the TEC provides efficient operation but raises concerns about noise injected back into the host system via the power supply. In particular, the switching edge's high-frequency harmonic content can corrupt the power supply, causing system-level problems. Such 'reflected' noise can be troublesome. The LTC1923 avoids these issues by controlling the slew of its switching edges, minimizing high-frequency harmonic content (Reference 1). This slowing of switching transients typically reduces efficiency by only 1 to 2%, which is a small penalty for the greatly improved noise performance. Figure 8a shows noise and ripple at the 5V supply with slew control in use. A ripple of 12 mV in amplitude is usually not a concern, as opposed to the high-frequency transition-related components, which are much lower in amplitude. Figure 8b, a time and amplitude expansion of Figure 8a's display, more clearly studies the high-frequency residue. High-frequency amplitude, measured at center screen, is about 1 mV. A good way to measure the effectiveness of slew limiting is by disabling it. High-frequency content jumps to nearly 10 mV, or almost 10 times worse performance (Figure 8c). Leave that slew limiting in there.

This level of noise reduction is suitable for most applications. Some special cases may require even lower reflected noise, and you can use a simple LC filter in these cases (Figure 9). Combined with the LTC1923's slew limiting, this filter provides vanishingly small reflected ripple and high-frequency harmonics. With this filter in place, the ripple is only approximately 1 mV, and the high-frequency content is at submillivolt levels (Figure 10a). Figure 10b expands the time scale to examine the high-frequency remnants. The amplitude is 500 μV, or approximately one-third of Figure 8b's reading. As before, you can measure the effectiveness of slew limiting by disabling it. The result is a 4.4-times increase in high-frequency content to approximately 2.2 mV (Figure 10c). So, as before, if you want to achieve the lowest reflected noise, leave that slew limiting in there.