Regardless of its many personality flaws, the ordinary bi-metallic thermocouple (TC) remains the dominant technology for industrial temperature sensing. TCs are popular because of their durability, simplicity, and low cost, not to mention their pedigree (they have been around since John Seebeck discovered them in 1822).
These sensors continue to thrive despite a multitude of irksome signal-conditioning challenges associated with their use. Some of the worst TC quirks include lowly millivolt-amplitude signal levels, cold-junction compensation (CJC) requirements, and nonlinearity that can become extreme at low temperature (which is sometimes the most problematic of all).
The signal-conditioning circuit in Figure 1 addresses TC foibles with a combination of three features, including chopper-stabilized “zero-drift” amplification (which is derived from A1, Linear Technology’s LTC1049). It also relies on ΔVBE cold-junction compensation (using Q1, any generic small-signal npn). An aggressive, digitally controlled potentiometer (the Xicor X9C103 DCP) linearization loop is employed as well. When joined with a standard type-N TC, the circuit becomes a “reasonably” accurate Kelvin-scale analog thermometer. It offers a 10-mV/K output scale factor and untrimmed ±3% full-scale error from 4K (the temperature of boiling liquid helium) to 373K (boiling H2O).
Not all of the NIST standard TC types (B, E, J, K, N, R, S, and T) will work at cryogenic temperatures. One that does, however, is the type-N TC (Ni-Cr-Si alloy versus Ni-Si-Mg). Type-N TCs are characterized by temperatures ranging from 3K to 1570K (−70°C to 1370°C). They also have a room-temperature Seebeck coefficient of approximately 26 µV/K that’s “relatively” constant (±10%, which is good for a TC) from 140K to 370K.
Over this 230° temperature span, VOUT = \\[VTC + (26 µV × TQ1)\\]/0.0026 = 0.01 V/K. Here, VTC is the TC Seebeck signal, while TQ1 indicates the absolute temperature of CJC Q1. Also, 26 µV × TQ1 represents the temperature-compensating ΔVBE signal generated by Q1, S2, S3, and the S2-modulated Q1 collector-current ratio, (R1 + R2)/R1.
Error-curve A in Figure 2 illustrates the performance of a linear TC approximation. Although acceptable for the upper portion of the temperature range, the simple first-order TC model goes to pot at colder temperatures, ending up more than 100° off at 4K! This is because the type-N Seebeck coefficient starts to nose-dive around 140K as it falls from 26 mV/K at 273K, to 20 µV/K at 140K, and less than 2 µV/K near 0K. It’s at this point that the DCP-based linearization algorithm kicks in.
The linearization applied by the circuit in Figure 1 works by subtracting a 0- to 2.8-mV progressive correction voltage (VDCP in Figure 2) from A1’s (VTC + 26 µV/KTQ1) input signal when VOUT is less than 1.4 V. According to the rules for calculating VDCP correction, if VOUT is greater than 1.4 V (i.e., T > 140K), then comparator A2 holds the DCP’s up/down count-direction line low. Consequently, the DCP wiper sticks at ground, forcing VDCP to zero.
But when VOUT is less than 1.4 V, A2 selectively enables DCP upcounting. By doing so, it closes a feedback loop that maintains VDCP = 0.002(1.4 − VOUT) and thereby establishes VOUT = (VTC + (26 µV × TQ1) − 0.0028)/0.0006. The net error of the composite response is never more than ±10° away from the correct temperature over the entire 0K to 373K temperature span (Fig. 2, curve C).
For both the DCP’s INC terminal and the S2/S3 CJC synchronous modulator/demodulator, the 208-Hz clock signal is supplied by the multivibrator implemented by A3/S1. The 208-Hz signal on S3 enhances the circuit’s immunity to 60-Hz pickup, which is further improved by the grounded-TC input configuration used.
VDCP ripple, caused by dither from the DCP wiper, is filtered by the linearization-loop capacitors, C1 and C2. Such filtering provides beneficial interpolation of the 100 discrete DCP steps. It also enables an effective continuum of VDCP correction levels. Resistors used in the circuit should be 1% or better and have low temperature coefficients to preserve the thermometer’s no-trim error performance. The total thermometer power consumption is typically 10 mW.