Precision temperature control with relatively high-power heaters driven directly from unregulated "120-V" ac lines is an efficient, inexpensive way to manage large thermal loads. A number of annoying complications can render this straightforward method problematic, however.

Among the worst of these troubles are the ubiquitous, random, and large variations in the ac mains. Variations in the RMS line voltage of ±5% (and worse) are ever present, which can make accurate temperature control very difficult. Every 1% twitch in RMS heater voltage will translate to a 2% lurch in square-law (V_{H}^{2}/R_{H}) heating. Temperature regulation may go entirely to pot before thermal excursions resulting from these random power changes can be caught and corrected by an unaided thermal control loop. Two possible solutions for this problem are pre-regulating the heater supply (expensive for big loads) and compensating for the variations.

This thermostat serves in a thermal control application that offers a perfect example of intolerance for temperature-variation, high-resolution, and optical resonant cavities (etalons) used in tunable-laser development *(see the figure)*. Etalons employ the principle of optical interference to precisely measure the wavelength of light in a laser beam. To do this accurately, the internal dimensions of the etalon must remain constant to within a tolerance of tens of nanometers. Despite the use of low thermal expansion materials in etalon fabrication, such extreme dimensional stability can only happen if the temperature of the etalon is rigidly controlled. For this reason, the design stability for this thermostat circuit is ±0.01°C.

The principles underlying this circuit's thermal control loop are described in *"Take Back Half: A Novel Integrating Temperature-Control Algorithm," *Electronic Design*, Dec. 4, 2000, p. 132*. Here, a Kelvin-connected platinum resistance temperature detector (PRTD), operates in conjunction with the bridge network and voltage reference VR1. Together they produce a temperature-sensitive voltage, V_{RTD}, and a setpoint voltage, V_{P1}.

The temperature-setpoint error voltage (V_{RTD }− V_{P1}) is input to the TBH integrator A1. After this, the integrator output is scaled by the adjustable R7 × C1 time-constant, buffered by A2, and output as I1 to pulse-width-modulator A4. Therefore, if V_{RTD} < V_{P1} (i.e., temperature > setpoint), V_{C2} will ramp up. This causes the heater duty factor (H_{PWM}) to ramp down and the heater, R_{H}, to cool off. If V_{RTD} > V_{P1} (i.e., temperature < setpoint), H_{PWM} will ramp (and R_{H} will heat) up.

Meanwhile, crossed-diodes Q2/Q3 and comparator A3 track the sign of the V_{P1 }− V_{RTD} difference. A3's output goes high when V_{P1} > V_{RTD} and low when V_{P1} < V_{RTD}. Temperature-setpoint crossings will cause the S2/S3 cross-connected CMOS switches to merge the charges on capacitors C2 and C1. This allows the TBH convergence-forcing bisection (described in the TBH article mentioned above) to go into effect.

Feed-forward compensation for potentially pesky 120-V variations occurs via sampling of the heater supply voltage (V_{H}) by S1 and the R2/R4 network. Compensation is achieved by the A4 PWM oscillator. It closes a feedback loop through S1, which strives to adjust H_{PWM} to maintain the charge balance on C4. For this to happen, I1, the heater power-control signal from the A1/A2 error-integrator, must be balanced by I2, the average current sourced to C4 by S1. Because I2 = H_{PWM}(V_{H }− 75)/R2, at balance H_{PWM} = I1 × R2/(V_{H }− 75). So for any given value of I1, H_{PWM} is inversely proportional to (V_{H }− 75).

H_{PWM} then changes by −2% for each +1% deviation of V_{H} from its nominal value of 150 V. For example, let R_{H} = 100 Ω, V_{H} = 150 V, and H_{PWM} = 50%. Then P_{H} = H_{PWM }× V_{H}^{2}/R_{H} = 0.5 × 150^{2}/100 = 112.5 W. Now suppose V_{H} were to suddenly increase by 10 V (V_{H} = 160 V). Without the feed-forward compensation feature, P_{H} would jump by more than 15 W to 0.5 × 160^{2}/100 = 128 W. But instead, the feed-forward compensation kicks in to drop H_{PWM} to 44%, limiting P_{H} to 0.44 × 160^{2}/100 = 112.6 W. Consequently, the unwanted P_{H} excursion is reduced to an insignificant 0.1 W.