How to Achieve Ultrahigh Accuracy when Designing a 21-Bit Precision Voltage Source

Abstract

This article presents a circuit for achieving an ultrahigh accuracy voltage source. The circuit is obtained by combining two 20-bit DACs in parallel to obtain a 21-bit DAC with ±1 LSB accuracy or 0.5 ppm. The overall solution also requires precise operational amplifiers and voltage references that match the performance of the DAC. This article shows a complete problem-solving path when selecting components to achieve this accuracy. Thermal and electromagnetic interference are also considered since they can cause accuracy problems when dealing with a 21-bit DAC.

Introduction

The state of the art in digital-to-analog converters (DACs) available today has a 20-bit resolution as its maximum. This is a limiting factor for some applications requiring a higher level of accuracy such as medical imaging or mass spectrometry. This limitation can be overcome by combining high performance components in a controlled environment. In other words, the accuracy of the overall solution will be determined by the signal chain of all the components in the circuit and their placement.

The electronic circuit shown in this article shows how combining high performance components can preserve the individual accuracy toward achieving a high accuracy voltage source. This circuit uses the AD5791 together with the LTZ1000 and the AD8675/AD8676 to achieve a 1 LSB INL at 21 bits. Applications that can benefit from this level of accuracy include medical equipment capable of producing clear images of small anatomical structures. Also, this circuit can produce more accurate test and measurement equipment that can lead toward more accurate manufacturing products. The range of applications where this ultrahigh voltage source can be applied is extensive, and it can serve as a stepping stone toward developing new products that are currently limited by the accuracy of existing semiconductor products.

Applications requiring 1 LSB at 21 bits, or 0.5 ppm INL, accuracy include:

• Scientific, medical, and aerospace instrumentation
• Medical imaging systems
• Laser beam positioners
• Vibration systems


• Test and measurement
• Automatic test equipment (ATE)
• Mass spectrometry
• Source measure units (SMUs)
• Data acquisition/analyzers


• Industrial automation
• Semiconductor manufacturing
• Process automation
• Power supply control
• Advanced robotics

In test and measurement systems, the 0.5 ppm resolution and accuracy of the AD5791 improve overall equipment accuracy and granularity, enabling finer control and excitation of external sources and nano-actuators. In industrial automation, the 0.5 ppm resolution and accuracy provide the precision that is required to move, alter, or position an actuator on the nanoscale.

The AD5791 is a single 20-bit, bipolar output, unbuffered voltage output DAC. It achieves a relative accuracy specification (INL) of ±1 LSB, and it guarantees a monotonic operation with a ±1 LSB differential nonlinearity (DNL). Other important parameters include a temperature drift of 0.05 ppm/°C, peak-to-peak noise of 0.1 ppm, and long-term stability of less than 1 ppm. The internal architecture of this IC is an R-2R digital-to-analog converter made using thin film resistor matching techniques. It operates from a bipolar supply of up to 33 V, and it can be driven by a positive reference in the range of +5 V to VDD –2.5 V and a negative reference in the range of VSS 2.5 V to 0 V. It uses a 3-wire serial interface that operates at clock rates up to 35 MHz and that is compatible with standard serial peripheral interface (SPI), QSPI^™, MICROWIRE^™, and DSP interface standards.

The LTZ1000 is an ultrastable temperature-controllable reference. It provides a 7.2 V output with only 1.2 μV p-p of noise, long-term stability of 2 μV/√kHr, and temperature drifts of 0.05 ppm/°C. The part contains a buried Zener reference, a heater resistor for temperature stabilization, and a temperature sensing transistor. External components are used to set operating currents and temperature to stabilize the reference, providing maximum flexibility and ensuring the best long-term stability and noise performance. This voltage reference, with temperature stabilization, is almost insensitive to external temperature variations.

For the operational amplifiers, a low offset, low noise, and low drift operational amplifier was needed. The AD8675/AD8676 operational amplifiers were chosen for their precision rail-to-rail capabilities, featuring an ultralow offset of 12 μV, a drift of 0.6 μV/°C, a voltage noise of 2.8 nV/√Hz at 1 kHz, and input bias currents of 2 nA over the full operating temperature range.

The principle of operation for achieving a 21-bit DAC from 20-bit DACs is based on a resistor divider. The output impedance of the AD5791 is 3.4 kΩ. When connecting two outputs of two of these ICs together, the equivalent circuit becomes a resistor divider. When the code difference between the two DACs is one LSB, the output voltage of the DAC resistor divider will be half of that voltage difference, equivalent to half an LSB. In other words, this configuration allows for obtaining an equivalent 21-bit DAC by connecting the outputs of two 20-bit DACs in parallel. The interconnection diagram is shown in Figure 1. Voltage references V_REP and V_REFN are set to +10 V and –10 V respectively, then the output voltage range at V_OUT can be programmed to any voltage within that voltage range.

For the measurements shown in this article, the hardware used connects two off-the-shelf AD5791 evaluation boards (ordering information EVAL-AD5791). The boards share the same voltage reference, the LTZ1000 module, which is mounted on only one evaluation module. The connection of the reference between the two boards is established using three twisted wires. Additionally, an extra wire is used to connect the outputs of the two DACs. The performance demonstrated in this article could be further improved by mounting the two AD5791 DACs on the same board and ensuring short connections between components using optimized PCB traces.

During the collection of linearity data, the results were affected by the presence of external radiated noise at low frequencies (below 1 MHz). This noise primarily originated from the proximity of the boards used for testing to the power supply and other nearby instruments. To mitigate this noise, all the hardware was enclosed in an electromagnetic field (EMF) blocking enclosure, effectively shielding the hardware under test from external radiated noise. The enclosure used is shown in Figure 2.

• Variations in ambient temperature can also affect measurements. The stability of the voltage reference can be affected if it does not operate at a constant temperature. The LTZ1000 addresses this issue. This voltage reference incorporates an internal resistor, and, with the use of external components and a feedback loop, it regulates the temperature of the die. The internal temperature of the component remains constant, ensuring that variations in the external ambient temperature have no effect on voltage output stability.
• Active components, including power supply devices, can cause changes in the output voltages of the supply rails and affect the output voltage of the DAC. The impact of voltage supply variations on the output voltage is reflected in the DCPSRR specification of the DAC. The operational amplifiers used for the references and output buffer also exhibit temperature dependency.
• For high accuracy applications, special attention should be given to the selection of resistors. It is important to choose resistors with a low temperature coefficient, ideally around 0.01%, and, if possible, operate the system at a constant temperature to minimize variations in resistance.

Changes in the external temperature of the voltage reference IC result in proportional fluctuations in the output voltage due to the temperature coefficient drift. The impact of these fluctuations on the integral nonlinearity (INL) is illustrated in Figure 3. The INL graph was obtained using the ADR445 reference at room temperature without the use of an EMF blocking enclosure. The board used for testing contained resistors with a typical temperature coefficient of 3 ppm/K. The observed jumps in the INL are attributed to temperature changes in the room, such as variations in the number of people present and the cycling of the air conditioning system. The measurement was conducted over a duration of approximately 24 hours.

To minimize temperature variations during the tests, options are available, such as using a temperature forcing machine that can provide a highly stable temperature throughout the duration of the tests. To maintain simplicity, the same EMF blocking enclosure used to isolate the boards from external radiated electromagnetic noise was utilized to maintain a relatively stable temperature during the testing. The power dissipation of the boards was calculated to be below 0.5 W, resulting in an interior temperature range of 25°C to 30°C within the EMF blocking enclosure throughout the duration of the tests.

Having identified all the elements that can impact the output voltage of a signal chain DAC, the next step is to program the two DACs to effectively obtain a 21-bit DAC. From a digital perspective, when dealing with a given 21-bit code, the DAC code needs to be split into two halves. If the original code is even, the remainder of the division is zero. If the original 21-bit code is an odd number, the remainder of the division is one. In this case, one DAC should be programmed with the result of the division, while the other DAC should be programmed with the split code plus one. An example is shown in Table 1.

This concept can be extended further by dividing the LSB size of the AD5791. For example, for a 22-bit DAC, the output of four DACs shall be connected in parallel. From a performance perspective, the main concern is the noise, especially considering that for a 20 V voltage span, the LSB size is 4.77 μV. No measurements have been conducted at this level in this article. To evaluate this circuit, a dedicated board with four DACs mounted would be necessary.

Results

Figure 4 shows the INL performance of the two AD5791 DACs connected to achieve 21-bit INL accuracy. These results show an INL below ±1 LSB, which was the goal of this exercise. These results were obtained under controlled temperature conditions, with the entire setup enclosed in an EMF blocking enclosure.

Figure 5 displays the DNL for the 21-bit configuration, demonstrating its monotonicity. The DNL results exhibit a discrete number of valid DNL codes. This is attributed to the limitations of the DVM used, considering that the LSB size of the 21-bit configuration is only 9.53 μV.

These results were obtained using the 3458A DVM, an 8.5-digit voltmeter, and a standard laboratory power supply. The voltage measurements were taken with a time window of 20 ms, corresponding to the 50 Hz frequency of the mains in Europe, where the measurements were carried out.

Conclusion

Combining two AD5791 20-bit DACs makes it possible to achieve a 21-bit DAC performance with 1 LSB INL. However, it is crucial to pay attention to the entire signal chain to minimize accuracy errors. Additionally, external factors such as temperature and electromagnetic interference can impact the output of the system.

For further work, it is recommended to build this circuit on a PCB to enhance signal integrity and minimize external noise coupling. In addition to INL and DNL measurements, it is recommended to measure other parameters, such as noise figures to further extend the characterization of the system. To facilitate this, the use of a new dedicated PCB is recommended.

About The Authors