Abstract
A wideband fading channel is considered with causal channel state information (CSI) at the transmitter and no receiver CSI. A simple orthogonal code with energy detection rule at the receiver is shown to achieve the capacity of this channel in the limit of large bandwidth. This code transmits energy only when the channel gain is large enough. In this limit, this capacity without any receiver CSI is the same as the capacity with full receiver CSI—a phenomenon also true for dirty paper coding. For Rayleigh fading, this capacity (per unit time) is proportional to the logarithm of the bandwidth. Our coding scheme is motivated from Gel'fand-Pinsker coding and dirty paper coding. Nonetheless, for our case, only causal CSI is required at the transmitter in contrast with dirty-paper coding and Gel'fand-Pinsker coding, where non-causal CSI is required.
Introduction
In wireless communications, Channel State Information (CSI)—knowing how the environment affects the signal—is a critical resource for optimizing transmission. Conventionally, it is assumed that the receiver requires CSI to decode signals effectively, especially in wideband scenarios where bandwidth $W$ is large. In such cases, the optimal "water-filling" strategy dictates that transmission should only occur when channel gains are high. However, this typically necessitates that the transmitter also possesses channel knowledge.
This research explores a specific theoretical gap: the performance of a system where only the transmitter possesses CSI, and only in a causal manner (knowledge of the present, but not the future). While "Dirty Paper Coding" (DPC) allows a transmitter to pre-compensate for interference, it requires non-causal CSI. This paper investigates whether similar optimality can be achieved in fading channels—"writing on fading paper"—using only causal information, even when the receiver remains entirely ignorant of the channel state.
The Orthogonal Coding Strategy
The authors propose a two-part approach to bridge this gap. First, they introduce a novel orthogonal coding scheme for wideband fading channels. The total bandwidth is divided into many small sub-bands. The transmitter monitors the channel gain $|h|^2$ and only transmits energy on a specific frequency band if the gain exceeds a high threshold $\Phi$.
Second, the receiver employs a simple energy detection rule. Because the transmitter only sends pulses when the channel conditions are exceptionally favorable, the receiver does not need to know the exact channel gain or phase. Instead, it calculates the average received energy across all bands and selects the one with the maximum energy. This "wait-and-burst" strategy effectively treats the channel's randomness as a resource to be exploited rather than an obstacle to be overcome.
Key Insights
- Irrelevance of Receiver CSI: In the limit of infinite bandwidth, the lack of receiver CSI does not reduce capacity. This is significant because obtaining CSI at the receiver often incurs high energy costs for training symbols.
- Causal vs. Non-Causal Equivalence: The equivalence between causal and non-causal CSI holds when the optimal strategy relies on "rare" channel states. By waiting for these high-quality moments, the causal constraint becomes non-binding.
- Minimal Information Requirements: The transmitter does not require the exact value of the fading gain; a single bit indicating whether the gain is above the threshold $\Phi$ is sufficient for optimal performance.
Results and Implications
For Rayleigh fading, the capacity is shown to be $C \approx P \ln W$ nats per unit time, matching the capacity achieved when both the transmitter and receiver have full CSI. The authors identify the optimal transmission threshold as $\Phi = \ln w - \ln(2 \ln w)$, where $w$ is the sub-bandwidth.
Furthermore, the paper generalizes these findings to capacity per unit cost for any discrete channel with causal transmitter CSI (Theorem 3): $$\sup_u \frac{D(P_{Y|U=u} || P_{Y|U=0})}{E[b(X)|U=u]}$$ This formula relates the "distinguishability" of the output (measured by relative entropy) to the average cost of the input.
Limitations and Future Work
While these results are theoretically robust, they are derived for the asymptotic limit of large bandwidth. In practical, finite-bandwidth systems, the gap between having and not having receiver CSI may remain significant. Additionally, the model assumes independent and identically distributed (i.i.d.) fading; real-world channels often exhibit temporal or spectral correlations. Finally, while the receiver does not need CSI for decoding, the transmitter's need for CSI implies a feedback overhead that must be accounted for in practical implementations.
Summary updated