TL;DR
This paper provides a rigorous mathematical framework for calculating the Extended Delivery Time (EDT)—the total time required to transmit a packet in a cognitive radio network where the primary user frequently interrupts. By deriving exact probability distributions for both continuous and periodic spectrum sensing, the authors enable a much more accurate queuing analysis for secondary users than previous models that relied only on average values.
What Problem Does This Paper Address?
In Cognitive Radio (CR), "Secondary Users" (SUs) are allowed to borrow spectrum from "Primary Users" (PUs) when the channel is idle. However, in an interweave implementation, the SU must immediately vacate the channel the moment the PU returns. This leads to a fragmented transmission process: an SU might start sending a packet, get interrupted, wait for the PU to finish, and then resume.
The total time from the start of the first transmission attempt to the successful completion of the packet is the Extended Delivery Time (EDT). While previous research often focused on the average EDT, this is insufficient for designing systems with strict latency requirements (like industrial sensors or real-time monitoring). Furthermore, standard queuing models (like M/G/1) often fail here because the "service time" for a packet depends on whether the queue was empty or full when the packet arrived—a nuance often overlooked in simpler models.
Methods
The authors model the PU's activity as a continuous-time Markov chain with exponentially distributed busy and idle periods. They analyze two sensing strategies:
- Continuous Sensing: The SU monitors the channel constantly and resumes transmission the instant the PU leaves.
- Periodic Sensing: The SU checks the channel at fixed intervals ($T_s$), which is more hardware-efficient but introduces additional "blind" waiting time.
To find the exact distribution of the EDT, the authors use Erlang distributions to model the sum of multiple random waiting periods. They derive closed-form expressions for the Probability Density Function (PDF) using modified Bessel functions for the continuous case and generalized Hyper-geometric functions for the periodic case.
Finally, they apply these distributions to an M/G/1 queuing model using a "mean value technique." Crucially, they distinguish between Type 1 packets (which arrive at a non-empty queue) and Type 2 packets (which arrive at an empty queue), as these experience different initial PU states and thus different delay characteristics.
Datasets
This is a theoretical and analytical paper. No external real-world datasets were used. Instead, the authors validated their mathematical derivations using Monte-Carlo simulations performed in Matlab. These simulations involved generating random PU "on/off" sequences and measuring the resulting delivery times for thousands of secondary packets to ensure they matched the predicted analytical curves.
Key Insights and Ideas
- The "Work-Preserving" Advantage: The study assumes a work-preserving strategy (e.g., using rateless codes), where the SU doesn't lose progress when interrupted. This allows the EDT to be modeled as a simple sum of the fixed transmission time and a random number of waiting slots.
- Sensing Interval Impact: The paper quantifies exactly how periodic sensing degrades performance. As the sensing interval $T_s$ approaches zero, the periodic sensing model mathematically converges to the continuous sensing model, providing a unified framework.
- Buffer Status Matters: One of the most important "aha" moments is the demonstration that a packet's delay depends on the buffer state upon arrival. A packet arriving to an empty queue might find the PU "on" or "off," whereas a packet waiting in line (Type 1) will always start its "service" the moment the previous packet finishes—at which point the PU is guaranteed to be "off" (in the continuous sensing case). Ignoring this distinction leads to significant errors in total delay estimation.
Key Results
- Analytical Accuracy: The derived closed-form expressions for the PDF of EDT (Equations 14 and 24 in the paper) show a "perfect match" with simulation results across various PU traffic intensities.
- Queuing Refinement: The proposed queuing analysis significantly outperforms the "conventional" M/G/1 approach. As shown in the paper's figures, the conventional model tends to underestimate delay, especially as the secondary arrival rate increases.
- Oscillation Phenomena: In the "one-shot" transmission case (very short packets), the periodic sensing PDF exhibits distinct oscillations. Each peak corresponds to a specific number of sensing intervals the SU had to wait, providing a visual representation of the discrete nature of periodic sensing.
Limitations and Future Work
- Perfect Sensing Assumption: The model assumes the SU never makes a mistake in sensing (no false alarms or missed detections). In real-world environments, sensing errors would add another layer of stochastic delay.
- Simplified Traffic: The PU traffic is modeled as a simple Poisson process. The authors note that more complex traffic models (like those with heavy tails) would be a valuable next step.
- Single Channel: The analysis is limited to a single-channel scenario. Future work could extend this to multi-channel cognitive networks where an SU might "handoff" to a different frequency entirely when interrupted.
Summary updated