Process engineers' enormous success has a way of undermining our thinking about systems architecture, often before we notice it. For instance, shrinking die size may have undermined the conventional way of looking at embedded systems with heavy electromechanical content. Convention says that in such a system-a robot or vehicle is a good example-you try to keep the control loops as physically close to the sensors and actuators as possible, in the name of signal integrity.
This way of thinking was greatly aided by the advent of inexpensive microcontrollers with decent computing power, adequate A/D conversion and decent PCM outputs. You could afford to confine the control loops for a mechanical arm, say, within the arm by putting microcontrollers on the location with the pressure and attitude sensors and the motors. The only signals that had to be exposed to the hazards of long-distance cabling were serial command strings, power and ground.
In advanced processes, however, all but the most burly 32-bit MCUs are likely to be pad-limited, encouraging designers to buy quite a bit more computing power and memory than would be necessary for a local control loop. Then all that power sitting in one place encourages thoughts about multiplexing of the control loops by multitasking the one big MCU that has essentially the same cost as one little MCU.
Unfortunately, that would mean running hopelessly vulnerable analog signals from their sources or destinations all the way to the central, shared controller. That practice promises to be a genuine nightmare for debugging and in-field reliability.
The ideal solution would be sensors and actuators that already had data converters and bus interfaces built into them. And for at least large-volume, deep-pocket designs, that is feasible, using advanced techniques for mounting die-size packages or dice on miniboards, on flexible cable material or even on the insides of sensor and motor enclosures.
As processes keep shrinking, though, the question may once again come full circle. If the smallest feasible chip-scale package today can hold a small A/D converter and a standard bus interface, tomorrow it will accommodate the addition of a substantial microcontroller. At that point we will see systems architecture swinging toward local control loops, executed by MCUs that are literally embedded in sensors and motor cases, synchronized via a standard serial bus to a central command processor. And then it will be the turn of the computer scientists to figure out how to develop control strategies for that type of distributed topology.
RSS
RSS| 
Digital| 
Mobile
