Dirty Paper Coding (DPC)

The core mechanism of DPC involves the transmitter applying a pre-subtraction operation to the intended signal. By using knowledge of Channel State Information (CSI), the transmitter encodes the message such that the known interference from other users' signals is effectively canceled out before transmission. This is analogous to writing on a dirty sheet of paper where the writer knows the exact location and color of the dirt stains and can adjust the ink accordingly, making the message perfectly legible to a reader who does not know where the dirt is.

DPC is the optimal transmission strategy for a Gaussian MIMO broadcast channel. It achieves the sum-rate capacity region, meaning no other coding scheme can provide a higher total data rate. This is a fundamental result in information theory, proven by Max Costa in 1983. The technique demonstrates that, counter-intuitively, interference known non-causally to the transmitter does not reduce the channel's capacity at all.

The concept originates from Costa's seminal paper, which established the mathematical equivalence between a point-to-point channel with known interference and a standard interference-free AWGN channel. Key implications:

• No power penalty: The transmit power required is the same as if the interference did not exist.
• Non-causal knowledge: The transmitter must know the interfering signal perfectly before encoding.
• Receiver blindness: The receiver does not need to know the interference structure to decode its message.

In a multi-user MIMO setting, DPC is implemented via a successive encoding process. Users are ordered, and the signal for each subsequent user is encoded by treating the signals of all previously encoded users as known, non-causally available interference. This process is the dual of Successive Interference Cancellation (SIC) at the receiver. The encoding order critically impacts the achievable rates for individual users, allowing for rate balancing.

True Costa-style DPC requires infinite-length, random codebooks, making it unrealizable in practice. The most common practical approximation is Tomlinson-Harashima Precoding (THP). THP replaces the complex binning scheme with a simple modulo arithmetic operation at the transmitter and receiver. This non-linear technique effectively constrains the transmit signal power while pre-canceling interference, offering a significant performance gain over linear precoding like Zero-Forcing.

Another class of practical DPC implementations involves vector perturbation. The transmitter adds a carefully chosen, integer-valued perturbation vector to the data symbols before linear precoding. This perturbation shapes the transmitted signal to minimize the power penalty associated with channel inversion. Finding the optimal perturbation vector is a closest lattice point search problem, solvable via sphere encoding algorithms, and approaches the theoretical DPC capacity for large constellations.