Introduction: Addressing Input Erasures in Channels with Memory
In communication theory, "erasure channels" typically model packet loss or deep fading. While most research focuses on output erasures—where signal loss occurs after the channel—this paper investigates input erasures. In this scenario, the signal is "erased" before entering a linear time-invariant (LTI) system with memory.
This distinction is vital for modeling real-world cellular phenomena like ergodic shadowing (physical obstacles blocking all nearby base stations) and dynamic user activity (bursty transmissions). Calculating the capacity of such channels often requires complex random matrix theory; the authors propose a simpler, analytical approach for two-tap channels.
Methodology and Theoretical Framework
The study focuses on a Two-Tap FIR (Finite Impulse Response) filter, where the received signal depends on the current and one previous input.
Equivalence and Matrix Analysis
The authors proved that for independent and identically distributed (i.i.d.) Gaussian input symbols, the input-erasure and output-erasure models yield identical achievable rates. By representing the channel as a tri-diagonal matrix, they observed that an erasure effectively "splits" the transmission into independent sub-blocks. This property allowed them to use recursive matrix properties to derive a rate expression in the form of an infinite sum, which is computationally more efficient than traditional integral equations.
Validation via Simulation
To validate the theoretical derivations, the researchers performed Monte-Carlo simulations (50 trials with block sizes of 200 symbols). The results demonstrated a perfect match between the analytical "Expression" and simulated data.
Key Insights and Results
- The "Splitting" Effect: For a two-tap filter, a single erasure resets the channel's memory, allowing global problems to be decomposed into manageable sub-problems.
- Filter Sensitivity: The "worst-case" filter occurs when the two taps have equal amplitude ($|h_0| = |h_1|$), maximizing interference.
- Multicell Processing (MCP) vs. Single-Cell Processing (SCP): Applying these results to a cellular uplink model revealed that MCP is significantly more robust. While SCP performance drops sharply with increased erasures, MCP maintains nearly constant throughput per active user by turning inter-cell interference into a diversity benefit.
- High-SNR Performance: In high-SNR regimes, the multiplexing gain is $(1 - q)$, where $q$ is the erasure probability.
Limitations and Future Directions
The current analytical method relies on the tri-diagonal structure of two-tap filters, making extensions to channels with longer memory (3+ taps) challenging. Additionally, the equivalence between input and output erasures only holds for i.i.d. inputs. Future work should address 2D hexagonal grids and scenarios with independent shadowing across different base stations.
Summary updated