There’s another dimension to the THD versus linearity problem as well. This plot represents the linearity error for every code for an 18-bit precision ADC. Using Equation 1, we can calculate that the THD we would expect from this converter is –126 dB, which is quite good!
When comparing precision industrial analog-to-digital converters (ADCs) to those designed for audio, many engineers notice that the audio parts tend to be much less expensive than the precision converters while offering superior performance in some specifications such as dynamic range. Audio converters generally use delta-sigma architectures, and some of the most precise precision converters use the same architecture. With such similarities, it’s quite tempting to use audio converters in precision applications.
If dynamic range and dynamic specifications are most important to your design, audio converters may provide the best value for you. In some industrial applications, such as vibration analysis, or seismic and geophysical analysis, dynamic performance is key. Specifications such as total harmonic distortion (THD), spurious-free dynamic range (SFDR), and signal-to-noise ratio (SNR) are the most important, while offset and gain errors are not.
The difference between audio and precision delta-sigma converters lies primarily in their performance at dc. However, audio-specific devices generally don’t perform all that well near dc, which is where most precision applications have some of their most demanding requirements. That doesn’t mean you can’t use them, but there’s a number of things to consider before you attempt this.
Know Your Parameters
Because dc accuracy isn’t that important in audio, it’s difficult to find audio converters that specify any of the important parameters related to dc accuracy, such as offset, gain error, or drift. A quick survey of some audio converters did find a few that specify a number of these, but only as typical numbers, which means they’re characterized but not tested in manufacturing. Testing these parameters is a time-consuming and thus costly process that precision data converters must go through.
Temperature stability and initial accuracy generally aren’t considerations for audio devices, but they are critical for most precision measurement applications. Another cost-saving measure that audio converters use is an internal voltage reference, but no information about its initial accuracy or drift characteristics can be found in most audio converter data sheets. Knowing this information is essential in precision applications, because the reference determines the overall accuracy of the system. Without this information, calculating the system error budget is next to impossible.
High-resolution industrial converters usually include a chopper-stabilized front end to maintain dc accuracy, while audio parts do not. Furthermore, the digital filters in many audio devices do not pass dc, but have a high pass filter down at low frequencies. This would prevent their use completely for dc measurements.
If the particular converter in question can work at dc, or the application doesn’t really require operating at dc itself, there is still the concern about calculating your error budget. Because audio converters rarely specify linearity, you are left trying to extrapolate or infer integral non-linearity (INL) from total harmonic distortion plus noise (THD+N) numbers.
THD is certainly related to linearity—a converter with good linearity usually has low THD. Quantifying the linearity from the THD, though, is difficult if not impossible. Equation 1 shows the relationship between THD and linearity:
In this equation, N is the number of codes spanned in the transfer function, EL(i) is the linearity error at each code, and ERMS is the full-scale signal. If we know the linearity error at every code along the ADC transfer function, Equation 1 should allow us to calculate THD.
If we wish to predict THD plus noise (THD+N, or also called SINAD, signal-to-noise-and-distortion), then we must also include quantization noise (EQ(i)) and random noise (EN(i)) in this equation, as shown in Equation 2:
From these equations, you can see that given a set of linearity data, you can predict THD. But given a THD number, you cannot determine the linearity errors directly, since THD is an integrated measurement of the linearity.
There’s another dimension to the THD versus linearity problem as well. The plot shown in the figure is the linearity error for every code for an 18-bit precision ADC. Using Equation 1, we can calculate that the THD we would expect from this converter is –126 dB, which is quite good!
Examining the datasheet from this device shows that the THD specification is only –116 dB. This is 10 dB different from the linearity calculation predicted. What could account for this difference?
A Closer Look
First, understand that to measure such low THD, the sine wave signal source must be exceptionally good. This requires a very high-quality signal generator, and even with high-quality generators, filtering often is done on the signal source before it is sent to the ADC to ensure that the THD of the input signal is as low as possible. This is quite challenging, and the results shown in the data sheet specification may be a limitation of the measurement system.
The second factor to consider is that linearity is really a dc specification and is usually measured using a dc servo loop or very low-frequency sine wave signals. THD, however, is a dynamic measurement, and dynamic signals place different conditions on the ADC. Distortion effects in the sample-and-hold circuitry, from the amplifier used as well as the voltage coefficient versus frequency performance of the capacitors, will degrade THD performance from that predicted from the dc linearity.
If you feel that an audio converter’s dc performance is adequate for your application, there are still a couple of other things to consider. Precision industrial converters offer a variety of digital interface options. They may have a parallel interface, or a serial interface based on a common protocol like serial peripheral interface (SPI) or I2C.
Audio converters tend to be serial interfaces only, which often are I2S type interfaces. These interfaces tend to be less flexible in their speed and interfacing options. This could create problems in the overall system implementation if using standard microcontrollers, which may have SPI or I2C peripherals, but probably not I2S.
Lastly, note that audio parts, due to their consumer nature, generally have a shorter life cycle than industrial parts, which typically offer 15- to 30-year life cycles.
Precision applications depend upon knowing the accuracy of the measurement system, and factors such as offset, gain error, and drift are critical. Data converters designed for these applications cost a bit more than their audio counterparts because of the extra circuitry used to minimize these errors and the extra testing needed to ensure meeting the device’s specifications.
If your application requirements can tolerate some uncertainty in those factors, but demand good dynamic specifications, then an audio ADC may be a good choice. As the system designer, it’s up to you to weigh the benefits of both before selecting a data converter.