In applications to motion control, it is not exactly information we are encoding, but a method of controlling power without (significant) loss. A simple, linear amplifier is forced to dissipate power equal to the difference between the supply voltage and the output voltage, times the current. At several amps, and with typical supply voltages between 65 and 120 Volts, this can quickly multiply out to hundreds of watts! The worst case, in fact, is at motor stall (holding position against a load), where motor voltage = zero, so it is the entire supply voltage times the motor current dissipated in the servo amp.
The PWM scheme releases servo amps from this great disadvantage, at the cost of slightly greater complexity. Since all motors have some inductance, voltage can be applied for a short interval, and then the motor's windings can be shunted with diodes that allow current to continue to circulate. The only losses (other than the motor's own resistance) are at the moment the transistors switch, and that dropped by the diodes. A new loss can be introduced, which is eddy current heating of the motor's windings and laminations. A good output filter should be provided to reduce current fluctuations through the PWM cycle, and also to reduce electromagnetic interference.
There are several schemes to accomplish this technique. One is to switch voltage on and off, and let the current recirculate through diodes when the transistors have switched off. Another technique is to switch voltage polarity back and forth with a full-bridge switch arrangement, with 4 transistors. This technique may have better linearity, since it can go right down to an effective 0% duty cycle by having the positive and negative voltage periods precisely equal. On/Off techniques may have trouble going down extremely close to 0% duty cycles, and may jitter between minimum duty cycles of positive and negative polarity.