Channel state information

Content sourced from Wikipedia, licensed under CC BY-SA 3.0.

Channel state information (CSI) is our knowledge about how a wireless link behaves. It describes how a signal travels from the transmitter to the receiver and includes effects like reflection, fading, and power loss with distance. To use CSI, the receiver estimates it through channel estimation and often feeds it back to the transmitter. In time-division duplex (TDD) systems, the transmitter can also estimate the channel from the reverse link.

There are two main levels of CSI: instantaneous (short-term) CSI, which tells us the current channel conditions, and statistical (long-term) CSI, which describes the overall behavior of the channel, such as typical gains and fading patterns. Instantaneous CSI enables real-time adaptation to improve reliability or data rates. Statistical CSI helps when the channel changes too quickly to track precisely.

CSI at the transmitter and at the receiver can be different. CSI at the receiver is CSIR, and CSI at the transmitter is CSIT.

In MIMO systems with multiple antennas, the channel is represented by a matrix. The received signal equals the channel matrix times the transmitted signal plus noise. If the channel were known perfectly, performance would be ideal, but in reality we have an estimate with some error.

To estimate the channel, a known training or pilot sequence is sent. The receiver uses these known signals and the received responses to infer the channel. If the channel and noise statistics are unknown, simple methods like least squares can be used. The estimation error decreases when the training signals are well designed, often requiring at least as many training symbols as there are transmit antennas.

If more information about the channel’s behavior is available, Bayesian or MMSE estimators can reduce the error further. These methods may require knowledge of how signals at different antennas and times are correlated (the channel and noise correlation).

In fast-changing channels, only statistical CSI may be reliable. In slower channels, instantaneous CSI can be estimated and used for some time before it becomes outdated.

Recently, deep learning is used to estimate CSI. Neural networks can learn to interpolate CSI across time and frequency with fewer pilot signals. CSI estimation can be data-aided (using known pilots) or blind (using only received data). Data-aided methods tend to be more accurate but require more bandwidth, while blind methods have less overhead but may be less accurate.