Motion sensors realized as microelectromechanicalsystem (MEMS) integrated circuits (ICs) are playing a major role in the future of cellular handsets. Not only do these MEMS accelerometers make mobile phones more usable, they can increase a handset’s reliability. Due to their increased sensitivity, reduced power consumption, reduced package size, MEMS accelerometers are being adopted more rapidly in new phone designs, supported by new functions implemented in software.

Gesture recognition is a term that describes how to add commands to a handset using movement. Examples include picking up a ringing phone and clearing an erroneously typed key. Gesture recognition can simplify the phone/human interface, but it usually helps to limit the number of stored gestures to less than five or six (the range of typical memory). The best gesture recognition systems use natural motions that require little learning or memory, such as “picking up” a ringing telephone (rather than pushing the “send” button). Using an accelerometer, one can sense that after ringing has started the handset is being moved. The movement signature of picking up the phone to the ear is measured by the accelerometer and interpreted by a microcontroller (Fig. 1).

While movements of this type ( acceleration followed by deceleration to zero shortly after, and whose positional change, as determined by double integration, is between 15 and 100 cm) are quite common, this motion occurring while the phone is ringing almost certainly signifies picking up the phone. This concept is one of the keys of reliable gesture recognition: Use context to understand the significance of the motion. The sensor requirements for this type of application require a three-axis accelerometer with a measurement range of ±2 g. The accelerometer output will be highpass filtered (probably in software) to reject acceleration due to inclination, so 0 g accuracy or stability is unimportant. The bandwidth of interest is between 1 and 50 Hz. Low noise (less than 350 µg/(Hz)^0.5] is desirable to minimize integration error.

Striking incorrect keys is common, given the small size and density of modern mobile-phone keypads. A simple gesture such as shaking a handset for one-half second can be used to clear the last entered key stroke. Longer shaking can be used to clear the last complete string typed. Again, this motion is a rather natural response and to make the algorithm more robust, designers can use the indication of keyboard use to look for the "clear" gesture. A single-or dual-axis accelerometer should be adequate for this application since human shaking will produce acceleration in every axis unless one is extremely careful to shake only in one axis. Accelerometer performance requirements are modest because high-pass filtering is used. While the actual acceleration may be up to ±10 g while shaking, a ±2 g range is sufficient since clipping should not adversely affect the shake detection algorithm.

Knowing the surroundings of a mobile telephone can help program functions that increase its usability, such as for ring control. For example, when a handset is on a table or desk where vibrate mode is not desirable, only the ringer would operate. In a meeting or setting in which a user wishes not to be disturbed, the handset could be placed face down to select silent mode so that neither the ringer or vibrate mode are on. Each one of these modes could be entered manually (using keyboard control), but it would more convenient-to have the phone do this for you automatically. An accelerometer can be used to determine the phone's orientation, and whether or not the handset is on a table or desk. So the phone can determine the desired ring mode automatically.

Handset orientation can be measured with a three-axis accelerometer. The top of a desk or table is stable and usually parallel with the Earth's surface. A handset placed on a desk will have its X, Y, or Z axes measuring approximately −1 g with the other two axes close to 0 g. (Robust algorithms will allow for some tolerance around the −1 g or 0 g measurements so that slight deviation from level, or minor 0 g bias drift due to temperature, does not perturb the algorithm.) When on a solid surface, the accelerometer will measure very little vibration. The face down position is determined by the lack of vibration and having the appropriate axis measure 1 g while the others are close to 0 g. Figure 2 shows typical waveforms from an ADXL330 MEMS accelerometer-equipped handset on a desk and in an operator's pocket.

For this application, a three-axis accelerometer with a measurement range of greater than ±1.2 g is required (±2 g is typically used). Good 0 g performance is needed, particularly 0 g stability over temperature (1 mg/°C should be sufficient) as absolute inclination of the handset is being measured. Low noise [< 350 µg/(Hz)^0.5] is needed to easily differentiate between being in a pocket or on a table.

For phone pickup sensing or clearing keypad errors with shaking, the accelerometer may be powered only when specific events occur (phone ringing or keyboard input). Low power consumption is desirable, but not necessary. For ring-style setting, however, the accelerometer will be powered most of the time, so low power operation is crucial. Three-axis accelerometers such as the model ADXL330 from Analog Devices consume as little as 200 µA (with a supply voltage of 2 V), so as to not degrade battery life excessively.

An accelerometer can be used as an input to control the system cursor or as a game input. Tilting the phone left/right or tipping it forward/backward moves the cursor left/right or up/down. This feature has been incorporated into several stand-alone games ( Nintendo's Tilt and Tumble Kirby) and game controllers (Nintendo's Wii Remote). A third (Z) axis can also be incorporated for jumping actions. Unlike the eight-position control standard on most handsets, an accelerometer can enable variable (analog) control, with cursor speed increasing with tilt inclination. Since the handset's initial position could be in just about any orientation (the user may be lying down, for example) the game is typically started with a key stroke that sets the cursor's neutral position. Since the initial position is reset every time the game or cursor control is started, accurate 0 g performance is unnecessary.

Key specifications for this application are a measurement range of at least ±1.2 g. Low noise [< 500 µg/(Hz)^0.5] is needed to prevent the appearance of the cursor trembling when the phone is on a stable surface. A bandwidth of 0 to 50 Hz is preferable (lower bandwidth will make the game appear sluggish). As games are not played continuously, very low power is advantageous, but not absolutely necessary.