Analog to Digital Conversion Techniques

Page 1 of 2

by Steve Hendrix

At the heart of most portable instruments lies an analog-to-digital converter (ADC) which converts some real-world quantity, often voltage, into a digital form, which can be read by the processor and manipulated as needed to support the application. This conversion seems straightforward enough, but there are a wide variety of techniques in use to perform such a conversion, and each has its own strengths and weaknesses. As a designer, you need to be aware of the various techniques and how they can impact the rest of the design.

Flash Converters

Perhaps the most intuitive of all the ADC techniques is flash conversion. This technique uses brute force to get the job done quickly, yet it still represents the fastest conversion technique around. A string of equal-valued resistors is used to create a chain of voltages equally spaced between the limits of the ADC conversion. Each tap of this chain feeds a comparator, which produces a logic output indicating whether the input voltage is higher or lower than the selected tap. The outputs of all the comparators are then fed into a fairly large array of digital logic which converts the combined states of all the comparators into a digital output word.

The advantages of this technique focus mostly around speed. Conversion rates well up into multiple hundreds of MHz are readily available today. In addition, the inherent linearity and monotonicity are definite benefits. It is theoretically possible to build the resistor chain with unequal resistor values and thus build a custom nonlinear transfer function, but I do not know of any commercially available converters that use such a technique.

The drawbacks of flash conversion center around cost per bit of resolution. The number of analog comparators required doubles with each added bit, becoming large after 6 bits and impractically so beyond about 8 bits. Thus flash converters rarely if ever exceed 8 bits of resolution. In addition to the comparators, the number of gates required in the digital logic increases with the number of comparators on the order of N log (N). Consequently, flash converters are generally used only where flat out speed is the overriding consideration. Sampling occurs by freezing the comparator outputs and allowing the digital logic to settle, and then latching the result into an output register. Since such a converter has a very sharply defined sampling time, it is subject to all the sampling and aliasing phenomena associated with the Nyquist sampling theorem.

Tracking Converters

Where an analog signal changes relatively slowly, a tracking converter might be the answer. Such a converter dispenses with the chain of comparators and instead relies on a single comparator, while adjusting the comparison voltage. The comparison voltage is generated by some form of digital-to-analog converter, usually based on a simple R-2R resistor ladder which requires two resistors per bit of resolution. At each sample time, the input voltage is compared to the DAC output, and the DAC output is adjusted one count up or down as indicated by the comparison, so that the DAC output tracks the input voltage. The value of the DAC at the sample time is then used as the output of the ADC, since that value is made to track the input voltage.

A tracking ADC is of course much simpler than a flash converter, and grows only linearly with the number of bits, so is useful to much higher resolutions. Very fast sampling rates are theoretically possible with a tracking converter, but the allowable input signal slew rate is severely limited to no more than one count per sample time. For slowly changing signals this speed limit may be acceptable; however, other techniques have essentially relegated the tracking converter to the trash bin of history, except in some extremely specialized applications.

Integrating Converters

Commonly used for very high resolution, an integrating converter converts voltage to time and measures that time with a digital counter. A capacitor is first fully discharged and the time reset to zero. The capacitor is then charged at a constant rate while a counter increments at a constant rate. A comparator is used to detect the time at which the capacitor voltage passes the input voltage, and the counter is frozen. The value of the counter then provides a digital representation of the input voltage.

An integrating converter can provide extremely high resolution and very good linearity, but does so at the expense of speed. By simply adding more bits to the counter and slowing the charge rate of the capacitor, resolution can be extended essentially indefinitely, though practical considerations such as noise in the input signal place an upper bound on the useful resolution. Practical converters in this class range from 16 bits to 26 bits, with conversion rates measured in Hz or even fractional Hz. Despite the fact that an integrating converter samples the input signal continuously, such converters are still subject to aliasing, and indeed tend to capture the most negative peak value of the input signal plus noise, and therefore require a well-filtered input signal.

A dual-slope converter represents a further refinement of the integrating converter. A dual slope converter improves on the speed by using a fairly fast charging rate and large step sizes to achieve a first approximation of the input voltage, and then switching to a much slower discharge rate and finer step sizes to hone in on the exact value of the input voltage. Such a converter achieves higher sampling rates at the expense of a slight increase in hardware complexity, but is probably the most commonly available type of integrating converter available today.

Successive Approximation Converters

The successive approximation converter has been the mainstay of integrated converters for the past decade or two, and remains the converter of choice for applications with moderate speed and resolution requirements. A successive approximation converter proceeds with what would appear to a software engineer as a binary search. Set a reference signal at the midpoint of the input voltage range, and compare that reference voltage against the input voltage. If the input voltage is higher, the first bit of the result is a one; if lower, the first bit is a zero. Having found the first bit, move either the high or the low range limit, as indicated by the conversion result, to the previous midpoint, and repeat the search for the correct voltage within the new range using a similar mechanism. The size of the search range is cut in half at each iteration, and we acquire one more bit of the final result. Such a method is especially well suited to converters which provide a serial output, because the bits are available one at a time anyway, from most significant to least significant.

Successive approximation converters represent a comprise, middle-of-the-road solution. Typical resolutions range from 10 to 16 bits, with conversion speeds ranging from a few KHz to the MHz range. Such converters require a good quality sample-and-hold mechanism to hold their input voltage stable throughout the multitude of comparisons required to achieve a consistent result, because a change in the input voltage after the first few bits have been determined will totally befuddle the remaining bits.

Next >>

		Click here to get your listing up.