Comment

1 February 2002

Classic control algorithm with a twist

Laser manufacturers need sensitive instruments that provide highly accurate measurements.

By Roland Lowe and John Gibbons

Fiber-optic communication systems use solid-state lasers to generate light beams that carry thousands of voice and data channels.

These lasers require tight temperature control with thermoelectric coolers (TECs), but manual tuning to arrive at optimal proportional-integral-derivative (PID) coefficients is tedious and time consuming.

One new solution uses autotuning software that quickly finds coefficients optimized for either minimum overshoot (for best overtemperature protection) or minimum settling time (for fastest time to temperature).

IF TEMPERATURE VARIES

Current technology does not offer an economical substitute for a TEC to control laser diode temperatures in fiber-optic communications systems. The laser diode's sensitivity to temperature changes demands active cooling.

If the temperature varies, the laser diode's dominant output wavelength and light intensity might change, leading to signal overlap, cross-talk problems, and inadequate carrier amplitude.

Therefore, the laser diode modules (LDMs) for these systems thoroughly test in production with the laser diode mounted on a TEC.

Before LDM production tests begin, the TEC and laser must have a stable set-point temperature.

For each temperature set point, the LDM manufacturer wants to arrive at the optimum combination of PID coefficients.

This is complicated because different LDM designs having different thermal masses typically test on the same production line, with each one requiring a different set of PID coefficients.

Fine-tuning these coefficient values manually can take hours. Given that LDMs usually test at multiple temperature set points, manufacturers frequently average PID coefficients for the various set points to save setup time. This means the coefficients are less than optimal for all test temperature settings.

ACCEPTABLY DAMP RESPONSE

The manual tuning process for the TEC temperature algorithm is difficult because a PID controller output signal (W) comes from:

where P is the proportional gain of the controller, D is the derivative gain, and I is the integral gain parameter.

As the value of P is increased, the system responds faster to changes in set point but becomes progressively underdamped and eventually unstable. However, D can adjust to achieve a critically damped response to changes in the set point. Too little damping results in overshoot and ringing, whereas too much causes an overly slow response.

The effect of the integral term is to change the TEC power until the steady-state value of the temperature error (T_S–T₀) is zero. Provided I is kept sufficiently small, parameter values can be found that give an acceptably damped response, with the temperature error tending to zero whenever the set point (T_S) changed by a step or linear time ramp.

Depending on the relative values of the PID parameters, response can optimize for minimum overshoot or minimum time to reach the set-point temperature. LDM manufacturers have different priorities under different circumstances, and they may need the ability to do either.

Depending on the control algorithm, different control and tuning techniques can work. A common design in PID controllers uses the Ziegler-Nichols (Z-N) algorithm, which can satisfy a given steady state and transient response characteristic.

This method calculates the PID feedback gains using parameters obtained from an experimental step response of the system.

It consists of increasing the proportional feedback gain, with the derivative and integral gains set to zero, until a sustained oscillation appears in the step response of the system.

The oscillation period and corresponding proportional gain serve to determine the appropriate PID gains using a lookup table.

Ziegler and Nichols described both open-loop and closed-loop tuning techniques.

In the open-loop method, the controller operates in manual mode without feedback. Then with the integral and derivative actions shut off and the controller in automatic mode, closed-loop tuning takes place.

Both methods involve experimentation.

PUSH THE SYSTEM TO TEMPERATURE

To eliminate experimentation and manual tuning, we modified the Z-N algorithm. It greatly reduces test setup time and lets the TEC reach stable set points in the shortest time possible or reach set point with minimum overshoot.

The unit accomplishes this by applying the open-loop (no feedback) Z-N tuning method, then applying the extracted system timing parameters to the modified Z-N tuning equations.

There are two sets of Z-N equations derived for two different tuning criteria: The first set is tuned to create a minimum settling time response, which pushes the system to the final temperature quickly but at the expense of some temperature overshoot.

The second tuning option approaches the set point less aggressively with minimal temperature overshoot but at the expense of a longer settling time. Both sets of tuning constants are created each time the user runs the autotune algorithm and can be used as is or read back from the unit for further tweaking if the user so desires.

Typical tuning time for the LDMs that were tested was about four minutes.

Minimum settling time and minimum overshoot modes allow this software's use with a variety of load types and active cooling or heating devices. Control outputs (up to 50 watts) maintain constant temperature, current, voltage, and sensor resistance.

For example, when controlling a large area TEC in a test fixture, the minimum overshoot mode protects the device under test from overtemperature damage. For set points that do not approach the maximum specified temperature, using the minimum settling time mode shortens the time it takes to reach a stable set point.

The differential aspect of the instrument's PID control reduces the required waiting time between measurements at various temperatures.

The temperature set-point range of –50° to +225°C covers most requirements for production testing of cooled optical components and subassemblies. Compared with TEC controllers that use less sophisticated P-I loops and hardware control mechanisms, the software-based, fully digital PID control provides greater temperature stability. IT

Behind the byline

Roland Lowe is the lead applications engineer, and John Gibbons is a senior analog design engineer at Keithley Instruments, Inc. in Cleveland. Lowe has a BSEE from Rensselaer Polytechnic Institute and an MSEE from Cleveland State University. Gibbons received his BSEE and MSEE degrees from Case Western Reserve University.

Related Links