A rule of thumb for some data compression techniques is that you can never add information back to a file. That is to say that the information lost due to lossy compression cannot be retrieved, but then, that is just a rule and some rules are meant to be broken. Over the past ten years researchers have been developing compressed sensing technology which can get more data from a signal than it actually contains, and now researchers at MIT have found a way to potentially make it commercially viable.
Compressed sensing works by actually adding information to a signal, which distorts it, but the distortions can be analyzed to uncover more information than the sensor could originally pick up. In 2006 Rice University researchers applied this by using randomly tilted micromirrors to shine light onto a single-pixel sensor. By processing the signal recorded by the one pixel though, the researchers were able to accurately reproduce an image with thousands of pixels. As impressive as that already is, the main benefit to compressed sensing is that it may allow a sensor to use considerably less power.
What MIT has done is developed a new framework for evaluating compressed-sensing methods and how they will operate on real-world hardware. Compressed sensing theory often considers the circuitry involved as being ideal, which is not true to reality. This framework should allow theorists to develop better theories and allow those who will build the necessary circuits to consider their performance before starting the expensive fabrication process.