Data converters

Data converters (a circuit that changes analog signals to digital representations or viceversa) perform calculations in the digital or discrete time domain more sophisticated data converters must translate the digital data to and from our inherently analog world. Analog Versus Discrete Time Signals Analog-to-digital converters, also known as A/Ds or ADCs, Convert analog signals to discrete time or digital signals. Digital-to-analog converters (D/As or DACs) perform the reverse operation.

The original analog signal (a) is filtered by an anti-aliasing filter to remove any high-frequency components that may cause an effect known as aliasing . The signal is sampled and held and then converted into a digital signal (b). Next the DAC converts the digital signal back into an analog signal (c). Note that the output of the DAC is not as "smooth" as the original signal. A low-pass filter returns the analog signal back to its original form (plus phase shift introduced from the conversions) after eliminating the higher order signal components caused by the conversion The analog signal (a) is continuous and infinite valued, continuous-time signal : a signal whose response with respect to time is uninterrupted. The signal has a continuous value for the entire segment of time for which the signal exists. By referring to the analog signal as infinite valued, we mean that the signal can possess any value between the parameters of the system. For example, in Fig., if the peak amplitude of the sine wave was +1V, then the analog signal can be any value between -1 and 1 V (such as 0.4758393848 V). Of course, measuring all the values between -1 and 1 V would require a piece of laboratory equipment with infinite precision. The digital signal in (b) is discrete with respect to time and quantized The digital signal, on the other hand, is discrete with respect to time. This means that the signal is defined for only certain or discrete periods of time. A signal that is quantized can only have certain values (as opposed to an infinitely valued analog signal) for each discrete period. Where you live the temperature in the winter stays between 0 F and 50 F Suppose you had a thermometer with only two readings, hot and cold, to record the weather patterns and plot the results. The two quantization levels can be correlated with the actual temperature as follows: If 0 F < T < 25 F Temperature is recorded as cold

If 25 F < T < 50 F Temperature is recorded as hot You take a measurement every day at noon and plot the results after one week. From Fig., it is apparent that your discretized version of the weather is not an accurate representation of the actual weather.

Now suppose that you find another thermometer with four possible temperatures (hot, warm, cool, and cold) you increase the number of readings to two per day. The quantization levels represent four equal bands of temperature If 0 F < T < 12.5 F Temperature is recorded as cold If 12.5F < T < 25 F Temperature is recorded as cool If 25 F < T < 37.5 F Temperature is recorded as warm If 37.5 F <T < 50 F Temperature is recorded as hot Here, the digital version of the weather still looks nothing like the actual weather pattern The actual weather pattern is the analog signal. It is continuous with respect to time, its value can be between 0 F and 50 F (even 33.9638483920398439 F!). Digitized signal is dependent on two things: the number of samples taken and The resolution, or number of quantization levels, of the converter. In our example, we need to increase both the number of samples and the resolution of thermometer. Suppose that finally we obtain a thermometer with 25 temperature readings and that we take a reading eight times per day. Each of the 25 quantization levels now represents a 2 F band of temperature. Observe that the digital version of the weather is approaching that of the actual analog signal. If we kept increasing both sampling time and resolution, the difference between the analog and the digital signals would become negligible This brings up another critical issue: how many samples should one take in order to accurately represent the analog signal? Suppose a sudden rainstorm swept through your town and caused a sharp decrease in temperature before returning to normal. If that storm had occurred between our sampling times, our experiment would not have shown the effects of the storm. Our sampling time was too slow to catch the change in the weather. If we had increased the number of samples, we would have recognized that something happened which caused the temperature to drop dramatically during that period. The Nyquist Criterion defines how fast the sampling rate needs to be to represent an analog signal accurately. This criterion requires that the sampling rate is at least two times the highest frequency contained in the analog signal. In our example, we need to know how quickly the weather can change and then take samples twice as fast as that value. The Nyquist Criterion can be described as

where fsampling is the sampling frequency required to accurately represent the analog signal and f MAX is the highest frequency of the sampled signal. How much resolution should we use to represent the analog signal accurately? There is no absolute criterion for this specification. Each application will have its own requirements. In our weather example, if we were only interested in following general trends, then the 25 quantization levels would more than suffice. However, if we were interested in keeping an accurate record of the temperature to within 0.5 F, we would need to double the resolution to 50 quantization levels so that each quantization level would correspond to each degree +0.5 F

Sample-and-hold (S/H) circuits are critical in converting analog signals to digital signals. The behavior of the S/H is analogous to that of a camera. Its main function is to "take a picture" of the analog signal and hold its value until the ADC can process the information. It is important to characterize the S/H circuit when performing data conversion. Ignoring this component can result in serious error, for both speed and accuracy can be limited by the S/H. Ideally, the S/H circuit should have an output shown in Fig.. Here, the analog signal is instantly captured and held until the next sampling period. However, a finite period of time is required for the sampling to occur. During the sampling period, the analog signal may continue to vary; The analog signal is "tracked during the time required to sample the signal by a circuit is called a track-and-hold, or T/H. Fig. 28.5b. It can be seen that S/H circuits operate in both static (hold mode) and dynamic (sample mode) circumstances.

Once the sampling command has been issued, the time required for the S/H to track the analog signal to within a specified tolerance is known as the acquisition time. In the worst-case scenario, the analog signal would vary from zero volts to its maximum value, Vw(max)- And the worst-case acquisition time would correspond to the time required for the output to transition from zero to Vm(max).

Since most S/H circuits use amplifiers as buffers it should be obvious that the acquisition will be a function of the amplifier's own specifications. For example, notice that if the input changes very quickly, then the output of the T/H could be limited by the amplifier's slew rate. The amplifier's stability is also extremely critical. If the amplifier is not compensated correctly, and the phase margin is too small, then a large overshoot will occur. A large overshoot requires a longer settling time for the S/H to settle within the specified tolerance. The error tolerance at the output of the S/H also depends on the amplifier's offset, gain error (ideally, the S/H should have a gain of 1) and linearity (the gain of the S/H should not vary over the input voltage range). Once the hold command is issued, the S/H faces other errors. Pedestal error occurs as a result of charge injection and clock feedthrough. Part of the charge built up in the channel of the switch is distributed onto the capacitor, thus slightly changing its voltage. Also, the clock couples onto the capacitor via overlap capacitance between the gate and the source or drain. Another error that occurs during the hold mode is called droop. This error is related to the leakage of current from the capacitor due to parasitic impedances and to the leakage through the reverse-biased diode formed by the drain of the switch. This diode leakage can be minimized by making the drain area as small as can be tolerated. Although the input impedance of the buffer amplifier is very large, the switch has a finite OFF impedance through which leakage can occur. Current can also leak through the substrate. The key to minimizing droop is increasing the value of the sampling capacitor. The trade-off, however, is increased time that's required to charge the capacitor to the value of the input signal.

A transient effect that introduces error occurs between the sample and the hold modes. A finite amount of time, referred to as aperture time, is required to disconnect the capacitor from the analog input source. The aperture time actually varies slightly as a result of noise on the hold-control signal and the value of the input signal, since the switch will not turn off until the gate voltage becomes less than the value of the input voltage less one threshold voltage drop. This effect is called aperture uncertainty or aperture jitter. As a result, if a periodic signal were being sampled repeatedly at the same points, slight variations in the hold value would result, thus creating sampling error. The amount of aperture error is directly related to the frequency of the signal and

The worst-case aperture error occurs at the zero crossing, where dV/dt is the greatest. This assumes that the S/H circuit is capable of sampling both positive and negative voltages (bipolar). The amount of error that can be tolerated is directly related to the resolution of the conversion.