On Active Learning and Supervised Transmission of Spectrum Sharing Based Cognitive Radios by Exploiting Hidden Primary Radio Feedback

This paper studies the wireless spectrum sharing between a pair of distributed primary radio (PR) and cognitive radio (CR) links. Assuming that the PR link adapts its transmit power and/or rate upon receiving an interference signal from the CR and such transmit adaptations are observable by the CR, this results in a new form of feedback from the PR to CR, refereed to as hidden PR feedback, whereby the CR learns the PR’s strategy for transmit adaptations without the need of a dedicated feedback channel from the PR. In this paper, we exploit the hidden PR feedback to design new learning and transmission schemes for spectrum sharing based CRs, namely active learning and supervised transmission. For active learning, the CR initiatively sends a probing signal to interfere with the PR, and from the observed PR transmit adaptations the CR estimates the channel gain from its transmitter to the PR receiver, which is essential for the CR to control its interference to the PR during the subsequent data transmission. This paper proposes a new transmission protocol for the CR to implement the active learning and the solutions to deal with various practical issues for implementation, such as time synchronization, rate estimation granularity, power measurement noise, and channel variation. Furthermore, with the acquired knowledge from active learning, the CR designs a supervised data transmission by effectively controlling the interference powers both to and from the PR, so as to achieve the optimum performance tradeoffs for the PR and CR links. Numerical results are provided to evaluate the effectiveness of the proposed schemes for CRs under different system setups.

💡 Research Summary

The paper investigates a novel spectrum‑sharing paradigm for a pair of distributed primary‑radio (PR) and cognitive‑radio (CR) links. Unlike conventional cognitive‑radio approaches that rely solely on spectrum sensing and assume a static interference model, this work exploits the fact that a PR often adapts its transmit power and/or data rate when it detects interference from a CR. Those adaptations, observable by the CR, constitute a “hidden PR feedback” channel that conveys information about the PR’s internal strategy without any dedicated signaling.

The authors first formalize the hidden‑feedback concept. The PR is modeled as possessing a predefined power‑control (and possibly rate‑control) function (f(\cdot)) that maps the instantaneous received interference‑plus‑noise level to a new transmit power. When the CR deliberately injects a probing signal, the PR reacts according to (f(\cdot)). By measuring the resulting change in the PR’s transmit power or data rate, the CR can infer the channel gain from its own transmitter to the PR receiver ((h_{c\to p})). This inference is the core of the proposed active learning phase.

The active‑learning protocol proceeds as follows: (1) the CR transmits a short, low‑power probing burst; (2) the PR, upon sensing the interference, updates its transmit power (or switches modulation/coding) according to its internal policy; (3) the CR monitors the PR’s transmission (e.g., via power‑measurement circuitry or by decoding the PR’s control channel) and records the observed adaptation; (4) using the known form of (f(\cdot)) and the PR’s SINR threshold, the CR solves a simple inverse problem to estimate (|h_{c\to p}|^{2}). The paper addresses several practical challenges: synchronization errors between probing and observation, granularity of PR’s rate adaptation (which limits the resolution of SINR inference), measurement noise on the CR’s power detector, and time‑varying channels caused by mobility. Solutions such as cross‑correlation based timing correction, Bayesian mapping of discrete rate changes to continuous SINR estimates, Kalman filtering for noisy power readings, and adaptive probing intervals shorter than the channel coherence time are presented.

Armed with an accurate estimate of the interference channel, the CR enters the supervised transmission stage. Here the CR’s objective is twofold: (i) keep the interference inflicted on the PR below a pre‑specified SINR degradation budget, and (ii) maximize its own achievable rate under that interference constraint. The authors formulate a constrained optimization problem where the CR’s transmit power (P_c) must satisfy
\