|
|
Designing Sensor Signal Conditioning with
A new breed of programmable analog ICs is bringing greater design flexibility to sensor signal conditioning applications, allowing manufacturers to differentiate products while slashing parts inventory by using a single chip and board layout for multiple applications.
Steve Harrold, Anadigm Inc.
Until now, the main drawback of programmable SoC components was that they did not serve the analog portions of the systems in which they worked. Particularly in systems involving sensors, analog circuitry was more or less a permanent fact of life because most sensors delivered analog signals and because A/D converters (ADCs) required a conditioned input signal if good performance was to be obtained. Signal conditioning took two forms. The first limited bandwidth of input signals to less than half the ADC sampling frequency to block noise and aliasing of unwanted high-frequency signals. The second form amplified and level-shifted the signal so that it was compatible with the full-scale input range of the ADC to obtain maximum dynamic range and minimum noise in the digitized signal. For example, suppose you supply signal conditioning circuitry to two different customers. If customer A is using a sensor from company X and customer B insists on using a sensor from company Y, the signal characteristics will probably be different, and the analog circuitry will require modifications. Once, this meant that two entirely different boards were required to implement signal conditioning. But with the advent of programmable analog ICs, the same device can be used to implement both.
Programmable Parameters You can use various methods to program the reference current. It can be as simple as using an external resistor, or you can achieve more precise control by defining the current with a DAC. You can download the inputs to the DAC from a PROM at powerup, or a program running on a microprocessor can control the inputs. To obtain nonvolatility and programmability, you have to use floating-gate transistors to define the reference current. These devices have a threshold voltage that can be electrically programmed to a value that will be stable for years or until reprogrammed.
Programmable Configuration You can’t achieve this level of control simply by adjusting a bias current. Instead, you have to provide a means of altering the signal flow through the IC. One way of doing this is to build circuits from a collection of building blocks, defining the rouÆing between the blocks using programmable switches (or antifuses). The building blocks are themselves built up from a collection of components (e.g., op amps, resistors, capacitors, and switches). Using this approach, you can define a gain stage (see Figure 1) using an op amp, with one resistor connected between the signal source and the op amp inverting input (the input network) and another resistor connected between the op amp output and inverting input (the feedback network).
Connecting extra parallel resistors to the feedback network will reduce the gain of the stage. Alternatively, connecting a capacitor in place of the feedback resistor will create a low-pass filter. Configuration data that are generally downloaded from a PROM define the OPEN/CLOSED states of the switches between and inside the blocks (and thus the IC functionality and the circuit parameters). IC suppliers have devised various ways of creating configurable programmable analog ICs. The simplest uses traditional analog components in the input and feedback networks (resistors and capacitors—inductors are too large for integration except for micrüwave frequencies). This option creates a continuous-time circuit, so-called because input signals are continuously used and output signals are continuously valid. Continuous-time circuits are well suited to prefiltering to prevent aliasing when an ADC samples the signal. The downside is that the parameters of filters implemented in this way usually depend on RC time constants, and resistors and capacitors implemented in ICs have notoriously large tolerances, making it difficult to achieve consistent filter performance. One way to alleviate the problem is to modify the circuit so that time constants depend on an amplifier’s transconductance rather than a resistance. Lattice Semiconductor combines this technique with capacitor trimming to achieve filters with less than ±5% variation on their ispPAC chips. By adding diodes to the components in the input and feedback networks, you can create building blocks whose output depends on the logarithm or antilog of the input signal. Zetex has exploited this to create TRAC, a configurable programmable analog IC aimed at signal processing applications. Complex tasks (e.g., multiplication and division) are reduced to simple addition and subtraction in the log domain. Differentiation and integration functions can also be implemented, although these require external components to define the time constants. You can significantly improve the accuracy and consistency of time constants in a circuit by using switched-capacitor (SC) techniques (see Figure 2).
The switches are toggled according to the state of a clock signal (fc), and in doing so, the capacitor is alternately charged to the input voltage and then discharged. The average current is equal to I = fcCV, which is the same as would flow in a resistor of value R = 1/(fcC). In other words, the SC circuit can emulate a resistor, provided the clock frequency is higher than the signal frequency.
In fact, you can easily create SC building blocks that implement a wide range of functions, all with characteristics that depend on capacitor ratios, and thus all immune to variations in supply voltages and temperature or to aging effects. Examples include summing gain stages, oscillators, filters, rectifiers, integrators, and differentiators. You can make changes to the capacitor ratio by connecting more capacitors in parallel in the input or feedback networks. There is a downside, though: the input signal is sampled by a SC, so it’s prone to the same aliasing effects as an ADC. A continuous-time prefilter is required to ensure there are no high-frequency components above fc/2 in the signal being passed to the SC. The good news, however, is that the requirements on the filter aren’t usually too severe because the clock frequency must be well above the signal frequency for the SC-resistor equivalence to be valid, and a simple RC filter is often sufficient.
Dynamic Reconfigurability In addition (or alternatively), if the reconfiguration is done quickly enough, one piece of programmable hardware can replace multiple blocks of fixed-function hardware. This hardware multiplexing might, for example, be used in a telephone dual-tone multifrequency detector or a multisensor data logging interface. In the first case (see Figure 4), you can sequentially reprogram one filter with many different configurations, rather than using independent filters to detect the frequency of a tone. In the second example (see Figure 5), you can connect different sensors to different input pins and then sequentially reprogram the hardware to route and process individual signals according to the requirements of each sensor. Generally speaking, this capability lets you cut costs significantly.
The AN220E04 FPAA from Anadigm is a good example of an IC offering this level of programmability. Reconfiguration can be under the control of internal resources only or controlled by signals received from an external microprocessor or microcontroller. An onchip ADC
In the latter case, the microcontroller can download a succession of configuration data as required by the system algorithm. The download typically takes 50 µs, depending on the number of changes required. The new configuration is then activated by an EXECUTE command, which requires just one clock cycle. The micro can calculate new configuration data on the fly, or (for micros with limited speed and memory) it can simply download precalculated data. Software tools are available to assist in the generation of the data streams.
Conclusion
Sensor signals are often digitized as early as possible in the signal path to take advantage of the flexibility of programmable digital ICs. Now this flexibility can be obtained in the analog world without recourse to expensive ADCs and DSP chips. Using just one board that can be programmed for a multitude of applications and customers, flexible and accurate performance can be obtained in the analog world as well as the digital. Steve Harrold is Analog IC Engineering Manager, Anadigm Inc., Crewe, Cheshire, U.K.; 44 1270 531 997, steve.harrold@anadigm.com.
|
|
|
|
|
There are comments on this
