Managing Single or Multi-Users Channel Allocation for the Priority Cognitive Access

Minimizing File Transfer Time in Opportunistic Spectrum Access Model

“…In the OSA literature, throughput is one of the most commonly used performance metrics. Recent studies [20]- [23], by utilizing the multi-armed-bandit (MAB) techniques, have focused on finding max-throughput channel while the SU needs to learn the unknown channel parameters on the fly. In order to transmit a file as quickly as possible, common folklore might assume that the max-throughput policy would also suggest the minimum expected file transfer time.…”

“…In addition, the file size can vary depending on the application [29]. Same channel data rate across all channels is another implicit assumption in those delay-related works [24]- [28], 1 but it doesn't reflect the realistic heterogeneous channel environment assumed in the throughput-oriented studies [20]- [23]. Clearly, allowing the SU to switch over such channels during instances of PU's interruption can further reduce the file transfer time, but to the best of our knowledge, this issue has not been fully explored.…”

“…On the other hand, MAB techniques have been extensively studied for heterogeneous channels, each with i.i.d Bernoulli distribution, in order to find the channel with maximum throughput. The widely used MAB techniques include the Bayesian approach [37], upper confidence bounds [23], thompson sampling [20] and its improvement from efficient sampling [21], and coordination approach among multiple SUs [5], [6]. In both two channel models studied in the throughput-oriented works, i.e., Bernoulli channels and Markovian channels, we will later show that "the max-throughput channel does not always minimize the file transfer time".…”

“…The cycle repeats itself until the SU transmits the entire file size F , then it immediately exits the channel in use. We omit the channel switching delay in our OSA model and the duration ∆ seconds are fully used for file transmission, which is typically assumed in order to simplify the mathematical model and design a throughputoptimal policy in the OSA literature [8], [20]- [23], [39]. Note that the duration ∆ seconds is not a randomly chosen number.…”

“…Remark 2.1. The duration ∆ does not include the sensing time for the fair comparison between our policies in Section 3 to 5 and the max-throughput policy in the same OSA model [20]- [23]. In addition, as simulated in [40], if the time duration is set to 100 ms (which is the duration of our ∆) and the target probability of accurate sensing is around 90%, the sensing time for a cognitive radio network is typically chosen to be 6 ms.…”