20-Bit, Linear, Low Noise, Precision, Bipolar ±10V DC Voltage Source (CN0191)

The circuit shown in Figure 1 provides a 20-bit programmable voltage with an output range of −10 V to +10 V while achieving ±1 LSB integral nonlinearity, ±1 LSB differential nonlinearity, and low noise.

The digital inputs to the circuit are serial and compatible with standard SPI, QSPI™, MICROWIRE®, and DSP interface standards. For high precision applications, the circuit can provide high accuracy and low noise performance by combining precision components such as the AD5791, AD8675, and AD8676.

The reference buffer is critical to the design because the input impedance of the DAC reference input is highly code dependent, which will cause linearity errors if the DAC reference source is not adequately buffered. The AD8676 has a high open-loop gain of 120 dB and has been verified and tested to meet the requirements of this circuit application for settling time, offset voltage, and low impedance drive capability. The AD5791 is characterized and factory calibrated to use the dual-channel AD8676 op amp to buffer its reference input, further enhancing the reliability of the companion device.

This combination of devices offers industry-leading 20-bit resolution, ±1 LSB integral nonlinearity (INL) and ±1 LSB differential nonlinearity (DNL), guaranteed monotonicity, low power consumption, small PCB size, and high cost-performance.

The digital-to-analog converter (DAC) shown in Figure 1 is the AD5791, a 20-bit, high-voltage converter with an SPI interface that provides ±1 LSB INL, ±1 LSB DNL, ​​and 7.5 nV/√Hz noise spectral density. In addition, the AD5791 features very low temperature drift (0.05 ppm/°C). The precision architecture used by the AD5791 requires forced sensing to buffer its reference voltage input to ensure that the specified linearity is achieved. The amplifiers (B1 and B2) selected to buffer the reference inputs should have low noise, low temperature drift, and low input bias current. The AD8676 amplifier is recommended for this function, which is an ultra-precision, 36 V, 2.8 nV/√Hz dual op amp with a low offset drift of 0.6 μV/°C and 2 nA input bias current. In addition, the AD5791 is characterized and factory calibrated to use this dual op amp to buffer its voltage reference input, further enhancing the reliability of the companion device.

Figure 1 shows the AD5791 configured with independent positive and negative reference voltages so that the output voltage range is from the negative reference voltage to the positive reference voltage, in this case −10 V to +10 V. The output buffer is the AD8675, which is a single-channel version of the AD8676 with low noise and low drift. The +5 V reference voltage is amplified to +10 V and −10 V using the AD8676 amplifier (A1 and A2). R2, R3, R4, and R5 in these amplifier circuits are precision metal film resistors with a tolerance of 0.01% and a temperature coefficient of 0.6 ppm/°C. To achieve the best performance over the entire temperature range, a resistor network such as the Vishay 300144 or VSR144 series can be used. The resistor values ​​are selected to be low (1 kΩ and 2 kΩ) to keep the system noise low. R1 and C1 form a low-pass filter with a cutoff frequency of approximately 10 Hz. This filter is used to attenuate the reference source noise.

Linearity Measurements
The following data further demonstrates the precision performance of the circuit shown in Figure 1. Figures 2 and 3 show the integral nonlinearity and differential nonlinearity as a function of the DAC code. It is clear from the figures that both characteristics are well within the specifications of ±1 LSB and ±1 LSB, respectively.

The total unadjusted error of this circuit is a combination of the various dc errors, namely, INL error, zero-scale error, and full-scale error. Figure 4 shows a plot of the total unadjusted error versus DAC code. The maximum error occurs at DAC codes 0 (zero-scale error) and 1,048,575 (full-scale error). This is as expected and is caused by the mismatch of resistor pairs R2 and R3, R4 and R5, and the offset errors of amplifiers A1, A2, B1, and B2 (see Figure 1).

In this example, the resistor pair is specified to have a maximum mismatch of 0.02% (typical values ​​are much lower). The amplifier offset error is 75 μV (maximum) or 0.000375% of the full-scale range, which is negligible relative to the error due to the resistor mismatch. Therefore, the expected maximum full-scale and zero-scale errors are approximately 0.02% or 210 LSBs. Figure 4 shows a measured full-scale error of 1 LSB and a measured zero-scale error of 4 LSB or 0.0003% of the full-scale range, indicating that all components perform significantly better than their specified maximum tolerances.

Figure 2. Integral nonlinearity vs. DAC code.

Figure 3. Differential nonlinearity vs. DAC code.

Figure 4. Total unadjusted error vs. DAC code

Noise Measurement
To achieve high accuracy, the peak-to-peak noise at the circuit output must be maintained below 1 LSB, which is 19.07 μV for 20-bit resolution and a 20 V peak-to-peak voltage range. Figure 5 shows the peak-to-peak noise measured over a 0.1 Hz to 10 Hz bandwidth for 10 seconds. The peak-to-peak values ​​for the three conditions are 1.48 μV (midscale output), 4.66 μV (full-scale output), and 5.45 μV (zero-scale output). The noise is lowest at midscale output, where the noise comes only from the DAC core. When the midscale code is selected, the DAC attenuates the noise contribution of each reference path.

Figure 5. Voltage noise (0.1 Hz to 10 Hz bandwidth)

However, in real applications, there will not be a high-pass cutoff frequency at 0.1 Hz to attenuate 1/f noise, but it will include frequencies down to DC in its passband; therefore, the measured peak-to-peak noise is more realistic, as shown in Figure 6. In this example, the noise at the output of the circuit was measured over 100 seconds, and the measurement fully covers frequencies down to 0.01 Hz. The upper cutoff frequency is approximately 14 Hz and is limited by the measurement setup. For the three conditions shown in Figure 6, the corresponding peak-to-peak values ​​are 4.07 μV (midscale output), 11.85 μV (full-scale output), and 15.37 μV (zero-scale output). The worst-case peak-to-peak value (15.37 μV) is roughly equivalent to 0.8 LSB.

Figure 6. Voltage noise measured over 100 seconds.

As the measurement time gets longer, lower frequencies are included and the peak-to-peak value gets larger. At lower frequencies, temperature drift and thermocouple effects become sources of error. These effects can be minimized by selecting components with lower thermal coefficients, such as the AD5791, AD8675, and AD8676, and by careful consideration of circuit construction; see the linked documents in the “Learn More” section.

The AD5791 supports a variety of output ranges, from 0 V to +5 V, up to ±10 V, and any value in between. The configuration shown in Figure 1 can be used to generate symmetrical or asymmetrical ranges as required. The reference voltage is applied to VREFP and VREFN, respectively, and the output buffer should be configured for unity gain as described in the AD5791 data sheet, setting the RBUF bit of the AD5791 internal control register to logic 1.

The AD5791 also offers a gain-of-2 mode of operation that produces a symmetrical bipolar output range from a positive reference voltage, as described in the AD5791 data sheet, eliminating the need to generate a negative reference voltage. However, this mode introduces larger full-scale and zero-scale errors. This mode is selected by setting the RBUF bit in the AD5791 internal control register to Logic 0.

The circuit shown in Figure 1 was constructed on a modified AD5791 evaluation board. Details on the AD5791 evaluation board and test methods are provided in the evaluation board user guide UG-185.