How to choose analog-signal-chain components

Signal conditioning can prepare a sensor’s output for digitization.

In a previous series, we looked at the analog-to-digital converter (ADCs) and sources of error that occur within the device. Of course, errors can creep in upstream of the ADC along the analog signal chain as the signal to be digitized is acquired and conditioned.

Figure 1. A measurement system with digital signal processing will often respond to a sensor input by driving an actuator.

Q: What sort of conditioning?
A: It depends on the signal. Typically, you’ll have a low-level signal from a sensor such as a thermocouple, as shown in Figure 1, that needs amplification to drive the ADC. Alternatively, if you’re monitoring voltage fluctuations on a 480-VAC three-phase line, your signal conditioning will include attenuation and isolation. In either case, you also may need filtering.

Figure 2. Using an analog filter on a noisy signal can help make full use of an ADC’s dynamic range.

Q: Wait, if we are digitizing the signal, why not do the filtering digitally, as described in a recent series on convolution?
A: We certainly could do that. As we discussed in that earlier series, a digital moving-average convolution filter is effective at removing Gaussian white noise. However, digital filtering is not always optimal. For example, consider the black trace in Figure 2. It appears to be quite noisy, with peaks extending to nearly 2 V in each direction, but we can’t quantify the noise in advance. If we want to digitize the black signal directly, we need an ADC with a ±4-V analog input range. If it turns out that the actual signal of interest is the red trace in Figure 2, then it would be much more cost-effective to use an analog filter and feed the 2-V peak-to-peak filtered red signal into the ADC so we can take full advantage of the ADC’s dynamic range and maximize our signal-to-noise ratio (SNR).

Q: What else should I know about analog vs. digital filters?
A: Compared with analog filters, digital filters generally have less ripple in the passband, more attenuation in the stopband, and sharper roll-off. In contrast, analog filters are always faster, because any digital filter will have some latency. In addition, an analog filter will have a wider dynamic range. For example, a 12-bit ADC will have a maximum count of 4,095 bits. We know from Equation 1 in our post on effective number of bits (ENOB) that the RMS quantization error is one least significant bit divided by the square root of 12, or 0.29, giving us a dynamic range of 14,120. In contrast, an operational amplifier with a maximum output of 20 V and internal noise of 2 µV has a dynamic range of 10 million.^[1]

Q: What happens after the signal is digitized?
A: It goes to a processor, which may provide additional filtering and then make decisions based on the signal level. A typical setup may have multiple sensors feeding the processor. The processor communicates system status remotely, via an Ethernet or other wired connection or, as shown in the figure, a wireless link. The processor may also act based on sensor inputs. It may, for example, reduce a motor’s speed in response to rising temperatures by sending a digital code to a digital-to-analog converter (DAC), which in turn drives a power amplifier that controls the motor.

Q: How do we build the signal-conditioning block in Figure 1?
A: The basic building blocks in the analog signal chain include the op amp, the instrumentation amplifier, the transimpedance amplifier, and the transconductance amplifier. We’ll take a closer look next time.