TL;DR
This paper investigates the noncoherent capacity of continuous-time fading channels, a crucial metric for wireless communication. It demonstrates that for practical SNR levels, the capacity is surprisingly robust and closely resembles that of a simple AWGN channel, regardless of the fine details of the fading process, which previously caused significant sensitivity.
What Problem Does This Paper Address?
In wireless communication, the "capacity" of a channel defines the maximum rate at which information can be reliably transmitted. For fading channels, where the signal strength fluctuates, this capacity is notoriously difficult to analyze, especially when the receiver doesn't know the exact state of the channel (noncoherent setting). Previous research on discrete-time fading channels showed that the noncoherent capacity's behavior at high signal-to-noise ratios (SNR) was highly sensitive to the precise mathematical model of the fading process, specifically the "support" of its power spectral density (PSD). This sensitivity was problematic because these fine details are often unmeasurable in practice, making the theoretical results less useful for engineers. This paper aims to determine if this sensitivity persists in more realistic continuous-time fading channels and if it impacts capacity at SNR values relevant to real-world systems.
Methods
The authors tackle this challenge by developing a novel approach to analyze continuous-time channels. Standard methods involve discretizing the channel, but this often requires knowledge of the channel's random singular functions, which are unavailable in the noncoherent setting. Instead, they discretize the channel by projecting the input and output signals onto a specific set of functions called a Weyl-Heisenberg (WH) set. This method preserves the channel's stationarity and provides a mathematically tractable way to handle mutual information. A key innovation is their definition of an "underspread" channel, which is more robust and physically meaningful than previous definitions based on compact support. They then derive a lower bound on the noncoherent capacity using this discretized channel and by treating the "self-interference" (a consequence of the discretization) as additive noise. This lower bound is then compared to an upper bound derived from the capacity of a standard AWGN channel.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- Robustness at Practical SNRs: The most significant finding is that for practical SNR values, the noncoherent capacity of underspread Rayleigh-fading channels is largely insensitive to the precise shape of the channel's scattering function. This is a major departure from previous discrete-time models where such sensitivity was a significant issue.
- A New Definition of "Underspread": The paper introduces a more practical and robust definition of an "underspread" channel, based on a small fraction of the scattering function's volume being outside a specific rectangle, rather than requiring compact support. This is crucial for making theoretical results applicable to real-world scenarios.
- Weyl-Heisenberg Sets for Tractability: The use of Weyl-Heisenberg sets for discretization is a novel and effective technique. It allows for a mathematically manageable analysis of continuous-time random LTV channels, a notoriously difficult problem.
- Self-Interference as Noise: Treating the self-interference introduced by the WH discretization as additive noise is a clever simplification that allows for a concrete lower bound on capacity. This approach highlights the trade-off between signal-space dimensions and interference reduction.
- Trade-off in WH Set Design: The paper implicitly suggests a trade-off in designing the WH sets: a slight increase in the product of time and frequency resolution (TF) beyond 1 improves the signal-to-interference ratio (SIR) but can reduce the number of available signal-space dimensions, impacting capacity at very high SNRs.
Key Results
The paper demonstrates that for continuous-time Rayleigh-fading channels satisfying the WSSUS and underspread assumptions (with ΔH << 1 and ε << 1), the noncoherent capacity is very close to the capacity of an AWGN channel with the same SNR and bandwidth, for all SNR values of practical interest. This holds true regardless of whether the channel's scattering function is compactly supported or not. Specifically, the capacity is shown to grow logarithmically with SNR up to a certain point (p < 1/(ΔH + ε)), beyond which the rate of growth slows down. The analysis suggests that for practical wireless channels, which are typically highly underspread, the lack of channel knowledge at the receiver has a minimal impact on capacity.
Limitations and Future Work
The authors acknowledge that their analysis relies on treating self-interference as additive noise, which provides a lower bound. A more complete analysis would require upper-bounding the capacity with perfect channel state information at the receiver, which is challenging due to self-interference. They suggest that advanced tools from large random matrix theory might be needed for this future work. Additionally, while they provide a criterion for designing WH sets, the optimization problem to find the ideal set is analytically complex, and they resort to numerical optimization for a specific pulse shape. Future work could explore more efficient optimization techniques.
Summary updated