Which ADC Architecture Is Right for Your Application?
By Walt Kester, (firstname.lastname@example.org)
Most ADC applications today can be classified into four broad market segments: (a) data acquisition,
Figure 1. ADC architectures, applications, resolution, and sampling rates.
The classification in Figure 1 shows in a general way how these application segments and the associated typical architectures relate to ADC resolution (vertical axis) and sampling rate (horizontal axis). The dashed lines represent the approximate state of the art in mid-2005. Even though the various architectures have specifications with a good deal of overlap, the applications themselves are key to choosing the specific architecture required.
Successive-Approximation ADCs for Data Acquisition
Modern IC SAR ADCs are available in resolutions from 8 bits to 18 bits, with sampling rates up to several MHz. At this writing, state-of-the-art performance of available devices is 16 bits at 3 MSPS (AD7621) and 18 bits at 2 MSPS (AD7641). Output data is generally provided via a standard serial interface (I2C® or SPI®, for example), but some devices are available with parallel outputs (at the obvious expense of increased pin count and package size).
Figure 2. Basic successive-approximation (SAR) ADC.
The basic successive-approximation architecture is shown in Figure 2. In order to process rapidly changing signals, SAR ADCs have an input sample-and-hold (SHA) to keep the signal constant during the conversion cycle. The conversion starts with the internal D/A converter (DAC) set to midscale. The comparator determines whether the SHA output is greater or less than the DAC output, and the result (the most-significant bit (MSB) of the conversion) is stored in the successive-approximation register (SAR) as a 1 or a 0. The DAC is then set either to 1/4 scale or 3/4 scale (depending on the value of the MSB), and the comparator makes the decision for the second bit of the conversion. The result (1 or 0) is stored in the register, and the process continues until all of the bit values have been determined. At the end of the conversion process, a logic signal (EOC, DRDY, BUSY, etc.) is asserted. The acronym, SAR, which actually stands for successive-approximation registerthe logic block that controls the conversion processis universally understood as an abbreviated name for the entire architecture.
The timing diagram for a typical SAR ADC is shown in Figure 3. The functions shown are generally present in most SAR ADCs, but their exact labels can differ from device to device. Note that the data corresponding to that specific sample is available at the end of the conversion time, with no pipeline delay or latency. This makes the SAR ADC easy to use in single-shot, burst-mode, and multiplexed applications.
Figure 3. Simplified timing diagram of a SAR A/D converter.
The internal conversion process of most modern IC SAR ADCs is controlled by a high-speed clock (internal or external, depending on the ADC) that does not need to be synchronized to the CONVERT START input.
The basic algorithm used in the successive-approximation ADC conversion process can be traced back to the 1500s. It is related to the solution of a useful mathematical puzzlethe determination of an unknown weight by a minimal sequence of weighing operations (Reference 1). In this problem, as stated, the object is to determine the least number of weights which would serve to weigh an integral number of pounds from 1 lb to 40 lb using a balance scale. One solution put forth by the mathematician Tartaglia in 1556, was to use the binary series of weights 1 lb, 2 lb, 4 lb, 8 lb, 16 lb, and 32 lb (or 20, 21, 22, 23, 24, and 25). The proposed weighing algorithm is the same one that is used in modern successive-approximation ADCs. (It should be noted that this solution will actually measure unknown weights up to 63 lb (26 1) rather than 40 lb as stated in the problem).* The binary algorithm, using a balance scale, is shown in Figure 4 with an unknown weight of 45 lbs.
*Note that if ternary (base-3: 1,0,1) logic is permitted, the problem can be solved in four steps, with weights of 1, 3, 9, and 27 lbs applied on either side of the balance. Indeed, 40 lbs is then maximum with these weights.
Figure 4. Successive-approximation ADC algorithm using balance scale and binary weights.
The overall accuracy and linearity of the SAR ADC are determined primarily by the internal DACs characteristics. Early precision SAR ADCs, such as the industry-standard AD574, used DACs with laser-trimmed thin-film resistors to achieve the desired accuracy and linearity. However, the process of depositing and trimming thin-film resistors adds cost, and the thin-film resistor values may be affected after the device is subjected to the mechanical stresses of packaging.
For these reasons, switched-capacitor (or charge-redistribution) DACs have become popular in newer CMOS-based SAR ADCs. The principal advantage of the switched-capacitor DAC is that the accuracy and linearity are primarily determined by high-accuracy photolithography, which establishes the capacitor plate area, hence the capacitance and the degree of matching. In addition, small capacitors can be placed in parallel with the main capacitorsto be switched in and out under control of autocalibration routinesto achieve high accuracy and linearity without the need for thin-film laser trimming. Because temperature tracking between the capacitors can be better than 1 ppm/8C, a high degree of temperature stability is achieved.
CMOS, the process of choice for modern SAR ADCs, is also the ideal process for analog switches. Therefore, input multiplexing can be added to the basic SAR ADC function relatively straightforwardly, thus allowing the integration of a complete data-acquisition system on a single chip. Additional digital functions are also easy to add to SAR-based ADCs, so features such as multiplexer sequencing, autocalibration circuitry, and more are becoming common.
Figure 5 illustrates the elements of the AD79x8 series of 1-MSPS SAR ADCs. The sequencer allows automatic conversion of the selected channels, or channels can be addressed individually if desired. Data is transferred via the serial port. SAR ADCs are popular in multichannel data-acquisition applications because they lack the pipeline delays typical in
Sigma-Delta (Σ-Δ) ADCs for Precision Industrial Measurement and Instrumentation
The basic concepts behind the Σ-Δ ADC architecture originated at Bell Labs in the 1950sin work done on experimental digital transmission systems utilizing delta modulation and differential PCM. By the end of the 1960s, the Σ-Δ architecture was well understood. However, because digital filters (then a rarity) were an integral part of the architecture, practical IC implementations did not appear until the late 1980s, when signal processing in digital CMOS became widely available. The basic concepts used in Σ-Δoversampling, noise shaping, digital filtering, and decimationare illustrated in Figure 6.
Figure 6. Noise-spectrum effects of the fundamental concepts used in Σ-Δ: oversampling, digital filtering, noise shaping, and decimation.
Figure 6A shows a noise spectrum for traditional Nyquist operation, where the ADC input signal falls between dc and fS/2, and the quantization noise is uniformly spread over the same bandwidth. In Figure 6B, the sampling frequency has been increased by a factor, K, (the oversampling ratio), but the input signal bandwidth is unchanged. The quantization noise falling outside the signal bandwidth is then removed with a digital filter. The output data rate can now be reduced (decimated) back to the original sampling rate, fS. This process of oversampling, followed by digital filtering and decimation, increases the SNR within the Nyquist bandwidth (dc to fS/2). For each doubling of K, the SNR within the dc-to-fS/2 bandwidth increases by 3 dB. Figure 6C shows the basic Σ-Δ architecture, where the traditional ADC is replaced by a Σ-Δ modulator. The effect of the modulator is to shape the quantization noise so that most of it occurs outside the bandwidth of interest, thereby greatly increasing the SNR in the dc-to-fS/2 region.
The basic first-order Σ-Δ ADC is shown in Figure 7, with the Σ-Δ modulator shown in some detail.
Figure 7. First-order sigma-delta ADC.
The heart of this basic modulator is a 1-bit ADC (comparator) and a 1-bit DAC (switch). Although there are a number of multibit Σ-Δ ADCs, those using the single-bit modulator have the obvious advantage of inherently excellent differential linearity.
The output of the modulator is a 1-bit stream of data. Because of negative feedback around the integrator, the average value of the signal at B must equal VIN. If VIN is zero (i.e., midscale), there are an equal number of 1s and 0s in the output data stream. As the input signal goes more positive, the number of 1s increases, and the number of 0s decreases. Likewise, as the input signal goes more negative, the number of 1s decreases, and the number of 0s increases. The ratio of the 1s in the output stream to the total number of samples in the same intervalthe ones densitymust therefore be proportional to the dc value of the input.
The modulator also accomplishes the noise-shaping function by acting as a low-pass filter for the signal and a high-pass filter for the quantization noise. Note that the digital filter is an integral part of the Σ-Δ ADC, and it can be optimized to give excellent 50-Hz/60-Hz power-frequency rejection. However, the digital filter does introduce inherent pipeline delay, which definitely must be considered in multiplexed and servo applications. If signals are multiplexed into a Σ-Δ ADC, the digital filter must be allowed to settle to the new value before the output data is valid. Several output clock cycles are generally required for this settling. Because of the pipeline delay of the digital filter, the Σ-Δ converter cannot be operated in a single-shot or burst mode.
Although the simple first-order single-bit Σ-Δ ADC is inherently linear and monotonic because of the 1-bit ADC and 1-bit DAC, it does not provide sufficient noise shaping for high-resolution applications. Increasing the number of integrators in the modulator (similar to adding poles to a filter) provides more noise shaping at the expense of a more complex designas shown in Figure 8 for a second-order
Figure 8. Second-order Σ-Δ modulator.
A popular alternative to higher-order modulators is to use a multibit architecture, where the
While integrating architectures (dual-slope, triple-slope, etc.) are still used in applications such as digital voltmeters, the CMOS Σ-Δ ADC is the dominant converter for todays industrial measurement applications. These converters offer excellent power-line common-mode rejection and resolutions up to 24 bits as well as digital conveniences such as on-chip calibration. Many have programmable-gain amplifiers (PGAs), which allow small signals from bridge- and thermocouple transducers to be directly digitized without the need for additional external signal conditioning circuits and in-amps.
Figure 9 shows a simplified diagram of a precision load cell. This particular load cell produces 10-mV full-scale output voltage for a load of 2 kg with 5-V excitation.
The bridges common-mode output voltage is 2.5 V. The diagram shows the bridge resistance values for a 2-kg load. The output voltage for any given load is directly proportional to the excitation voltage, i.e., it is ratiometric with the supply voltage.
Figure 9. Load-cell signal-conditioning application.
A traditional approach to digitizing this low-level output would be to use an instrumentation amplifier to provide the necessary gain to drive a conventional SAR ADC of 14-bit to 18-bit resolution. Because of offset and drift considerations, an auto-zero in-amp such as the AD5555 or AD8230 is required. Appropriate filtering circuitry is needed due to the noise of the auto-zero in-amp. In addition, the output data from the SAR ADC is often averaged for further noise reduction.
An attractive alternative to the traditional in-amp/SAR ADC approach is shown in Figure 10, which uses a direct connection between the load cell and the AD7799 high resolution Σ-Δ ADC. The full-scale bridge output of 10 mV is digitized to approximately 16 noise-free bits by the ADC at a throughput rate of 4.7 Hz. (For more discussion on input-referred noise and noise-free code resolution see for Further Reading 1). Ratiometric operation eliminates the need for a precision voltage reference.
Figure 10. Load-cell signal conditioning using the AD7799 high-resolution Σ-Δ ADC.
The Σ-Δ ADC is an attractive alternative when very low-level signals must be digitized to high resolution-but the user should understand that the Σ-Δ ADC is more digitally intensive than the SAR ADC and may therefore require a somewhat longer development cycle. Evaluation boards and software can greatly assist in this process. Nevertheless, there are still many instrumentation and sensor signal-conditioning applications that can be efficiently solved with a traditional in-amp (for signal amplification and common-mode rejection) followed by a multiplexer and a SAR ADC.
Sigma-Delta ADCs for Voiceband and Audio
Digital techniques for voiceband audio began in the early days of PCM telecommunications applications in the 1940s. The early T-carrier systems used
Modern digital cellular systems utilize higher-resolution oversampled linear Σ-Δ ADCs and DACs rather than the lower-resolution companding technique. Typical SNR requirements are 60 dB to 70 dB. If companding/expanding is required for compatibility with older systems, it is done in the DSP hardware or software. Voiceband codecs (coder/decoders) having many applications other than PCM, such as speech processing, encryption, etc., are available in a variety of types.
Sigma-delta ADCs and DACs also dominate the more demanding audio markets, including, for example, FM stereo, computer audio, stereo compact disc (CD), digital audio tape (DAT), and DVD audio. Total harmonic distortion plus noise (THD + N) requirements range from 60 dB to greater than 100 dB, and sampling rates range from 48 kSPS to 192 kSPS. Modern CMOS Σ-Δ ADCs and DACs can meet these requirements and also provide the additional digital functions usually associated with such applications.
Pipelined ADCs for High-Speed Applications
Today, markets that require high speed ADCs include many types of instrumentation applications (digital oscilloscopes, spectrum analyzers, and medical imaging). Also requiring high-speed converters are video, radar, communications (IF sampling, software radio, base stations, set-top boxes, etc.), and consumer electronics (digital cameras, display electronics, DVD, enhanced-definition TV, and high-definition TV).
The pipelined ADC has its origins in the subranging architecture, first used in the 1950s. A block diagram of a simple 6-bit, two-stage subranging ADC is shown in Figure 11.
Figure 11. 6-bit, two-stage subranging ADC.
The output of the SHA is digitized by the first-stage 3-bit sub-ADC (SADC)usually a flash converter. The coarse 3-bit MSB conversion is converted back to an analog signal using a
Figure 12. Residue waveform at input of second-stage SADC.
This subranging ADC can best be evaluated by examining the residue waveform at the input to the second-stage ADC, as shown in Figure 12. This waveform is typical for a low-frequency ramp signal applied to the analog input of the ADC. In order for there to be no missing codes, the residue waveform must not exceed the input range of the second-stage ADC, as shown in the ideal case of Figure 12A. This implies that both the N1-bit SADC and the
Figure 13. Missing codes due to MSB ADC nonlinearity or interstage misalignment.
This architecture, as shown, is useful for resolutions up to about 8 bits (N1 = N2 = 4); however maintaining better than 8-bit alignment between the two stages (over temperature variations, in particular) can be difficult. At this point it is worth noting that there is no particular requirementother than certain design issues beyond the scope of this discussionfor an equal number of bits per stage in the subranging architecture. In addition, there can be more than two stages. Nevertheless, the architecture as shown in Figure 11 is limited to approximately
The error-corrected subranging ADC architecture appeared in the mid-1960s as an efficient means to achieve higher resolutions, while still utilizing the basic subranging architecture. In the two-stage 6-bit subranging ADC, for example, an extra bit is added to the second-stage ADC which allows the digitization of the regions shown as X and Y in Figure 12. The extra range in the second-stage ADC allows the residue waveform to deviate from its ideal value-provided it does not exceed the range of the second-stage ADC. However, the internal SDAC must still be accurate to more than the overall resolution, N1 + N2.
A basic 6-bit subranging ADC with error correction is shown in Figure 14, with the second-stage resolution increased to 4 bits, rather than the original 3 bits. Additional logic, required to modify the results of the N1 SADC when the residue waveform falls in the X or Y overrange regions, is implemented with a simple adder in conjunction with a dc offset voltage added to the residue waveform. In this arrangement, the MSB of the second-stage SADC controls whether the MSBs are incremented by 001 or passed through unmodified.
Its worth noting that more than one correction bit can be used in the second-stage ADC, a trade-offpart of the converter design processbeyond the scope of this discussion.
The error-corrected subranging ADC shown in Figure 14 does not have a pipeline delay. The input SHA remains in the hold mode during the time required for the following events to occur: the first-stage SADC makes its decision, its output is reconstructed by the first-stage SDAC, the SDAC output is subtracted from the SHA output, amplified, and digitized by the second-stage SADC. After the digital data passes through the error correction logic and output registers, it is ready for use; and the converter is ready for another sampling-clock input.
Figure 14. 6-bit subranging error-corrected ADC, N1 = 3, N2 = 4.
In order to increase the speed of the basic subranging ADC, the pipelined architecture shown in Figure 15 has become very popular. This pipelined ADC has a digitally corrected subranging architecturein which each of the two stages operates on the data for one-half of the conversion cycle, and then passes its residue output to the next stage in the pipeline prior to the next phase of the sampling clock. The interstage track-and-hold (T/H) serves as an analog delay lineit is timed to enter the hold mode when the first-stage conversion is complete. This allows more settling time for the internal SADCs, SDACs, and amplifiers, and allows the pipelined converter to operate at a much higher overall sampling rate than a nonpipelined version.
Figure 15. Generalized pipeline stages in a subranging ADC with error correction.
There are many design trade-offs that can be made in the design of a pipelined ADC, such as the number of stages, the number of bits per stage, number of correction bits, and the timing. In order to ensure that the digital data from the individual stages corresponding to a particular sample arrives at the error correction logic simultaneously, the appropriate number of shift registers must be added to each of the outputs of the pipelined stages. For example, if the first stage requires seven shift-register delays, the next stage will require six, the next five, etc. This adds the digital pipeline delay to the final output data, as shown in Figure 16, the timing for a typical pipelined ADC, the AD9235.
Figure 16. Timing of a typical pipelined ADC, the 12-bit, 65-MSPS AD9235.
For the 12-bit, 65-MSPS AD9235, there are seven clock cycles of pipeline delay (sometimes referred to as latency). This latency may or may not be a problem, depending upon the application. If the ADC is within a feedback control loop, latency may be a problemin the overlap area, the successive-approximation architecture would be a better choice. Latency also makes pipelined ADCs difficult to use in multiplexed applications.
However, in the bulk of applications for which frequency response is more important than settling time, the latency issue is not a real problem.
A subtle issue relating to most CMOS pipelined ADCs is their performance at low sampling rates. Because the internal timing generally is controlled by the external sampling clock, very low sampling rates extend the hold times for the internal track-and-holds to the point where excessive droop causes conversion errors. Therefore, most pipelined ADCs have a specification for minimum as well as maximum sampling rate. Obviously, this precludes operation in single-shot or burst-mode applicationswhere the SAR ADC architecture is more appropriate.
Finally, it is important to clarify the distinction between subranging and pipelined ADCs. From the discussions above, it can be seen that, although pipelined ADCs are generally subranging (with error correction, of course), subranging ADCs are not necessarily pipelined. As a matter of fact, the pipelined subranging architecture is predominant because of the demands for high sampling rates, where internal settling time is of utmost importance.
Pipelined ADCs are available today with resolutions of up to 14 bits and sampling rates over 100 MHz. They are ideal for many applications that require not only high sampling rates but high signal-to-noise ratio (SNR) and spurious-free dynamic range (SFDR). A popular application for these converters today is in software-defined radios (SDR) that are used in modern cellular telephone base stations.
Figure 17 shows a simplified diagram of a generic software radio receiver and transmitter. An essential feature is this: rather than digitize each channel separately in the receiver, the entire bandwidth containing many channels is digitized directly by the ADC. The total bandwidth can be as high as 20 MHz, depending on the air standard. The channel-filtering, tuning, and separation are performed digitally in the receive-signal processor (RSP) by a high-performance digital signal-processor (DSP).
Digitizing the frequency band at a relatively high intermediate frequency (IF) eliminates several stages of down-conversion. This leads to a lower-cost, more flexible solution in which most of the signal processing is performed digitallyrather than in the more complex analog circuitry associated with standard analog superheterodyne radio receivers. In addition, various air standards (GSM, CDMA, EDGE, etc.) can be processed by the same hardware simply by making appropriate changes in the software. Note that the transmitter in the software radio uses a transmit signal processor (TSP) and DSP to format the individual channels for transmission via the upstream DAC.
Figure 17. Generic IF-sampling wideband software radio receiver and transmitter.
The ADC requirements for the receiver are determined by the particular air standards the receiver must process. The frequencies in the bandwidth presented to the ADC consist of the desired signals as well as large-amplitude interferers or blockers. The ADC must not generate intermodulation products due to the blockers, because these unwanted products can mask smaller desired signals. The ratio of the largest expected blocker to the smallest expected signal basically determines the required spurious-free dynamic range (SFDR). In addition to high SFDR, the ADC must have a signal-to-noise ratio (SNR) compatible with the required receiver sensitivity.
Another requirement is that the ADC meet the SFDR and SNR specifications at the desired IF frequency. The basic concept of IF sampling is shown in Figure 18, where a 20-MHz band of signals is digitized at a rate of 60 MSPS. Note how the IF sampling process shifts the signal from the third Nyquist zone to baseband without the need for analog down-conversion. The signal bandwidth of interest is centered in the third Nyquist zone at an IF frequency of 75 MHz. The numbers chosen in this example are somewhat arbitrary, but they serve to illustrate the concept of undersampling. These applications place severe requirements on the ADC performance, especially with respect to SNR and SFDR. Modern pipelined ADCs, such as the 14-bit, 80-MSPS AD9444, can meet these demanding requirements. For instance, the AD9444 has an SFDR of 97 dBc and an SNR of 73 dB with a 70-MHz IF input. The input bandwidth of the AD9444 is 650 MHz. Other 14-bit ADCs optimized for SFDR and/or SNR are the AD9445 and AD9446.
Figure 18. Sampling a 20-MHz BW signal with an IF frequency of 75 MHz at a sampling rate of 60 MSPS.
Successive-approximation is the architecture of choice for nearly all multiplexed data acquisition systems, as well as many instrumentation applications. The SAR ADC is relatively easy to use, has no pipeline delay, and is available with resolutions to 18 bits and sampling rates up to 3 MSPS.
For a wide variety of industrial measurement applications, the sigma-delta ADC is ideal; it is available in resolutions from 12 bits to 24 bits. Sigma-delta ADCs are suitable for a wide variety of sensor-conditioning, energy-monitoring, and motor-control applications. In many cases, the high resolution and the addition of on-chip PGAs allow a direct connection between the sensor and the ADC without the need for an instrumentation amplifier or other conditioning circuitry.
The Σ-Δ ADC and DAC, easily integrated into ICs containing a high degree of digital functionality, also dominate the voiceband and audio markets. The inherent oversampling in these converters greatly relaxes the requirements on the ADC antialiasing filter and the DAC reconstruction filter.
For sampling rates greater than approximately 5 MSPS, the pipelined architecture dominates. These applications typically require resolutions up to 14 bits with high SFDR and SNR at sampling frequencies ranging from 5 MSPS to greater than 100 MSPS. This class of ADCs is used in many types of instrumentation, including digital oscilloscopes, spectrum analyzers, and medical imaging. Other applications are video, radar, and communications applicationsincluding IF sampling, software radio, base stations, and set-top boxesand consumer electronics equipment, such as digital cameras, display electronics, DVDs, enhanced definition TVs, and high-definition TVs.
The use of manufacturers selection guides and parametric search engines, coupled with a fundamental knowledge of the three basic architectures, should help the designer select the proper ADC for the application. The use of manufacturers evaluation boards makes the process much easier. The Analog Devices ADIsimADC® program allows the customer to evaluate the dynamic performance of the ADC without the need for any hardware. The required software and the ADC models (and many other analog and digital design aids) are free downloads at http://www.analog.com. This tool can be extremely valuable in the selection process.
Not to be overlooked is the proper design of the ADC input-, output-, and sampling-clock circuitry. Data sheets and application notes should be consulted regarding these important issues. Finally, and equally critical to achieving a successful mixed-signal design, are layout, grounding, and decoupling. For a detailed treatment of these and other design issues, the reader is encouraged to consult the two comprehensive texts listed in the Further Reading as well as the Analog Devices website, http://www.analog.com.
2. Walt Jung, Editor, Op Amp Applications Handbook, Published by Newnes, an imprint of Elsevier, 2005, ISBN: 0-7506-7844-5.
3. For more information on products or applications, please visit the Analog Devices, Inc., website: http://www.analog.com.
Copyright 1995- Analog Devices, Inc. All rights reserved.