1: \begin{abstract}

2: The cloud radio access network (C-RAN) provides high spectral and

3: energy efficiency performances, low expenditures and intelligent

4: centralized system structures to operators, which has attracted

5: intense interests in both academia and industry. In this paper, a

6: hybrid coordinated multi-point transmission (H-CoMP) scheme is

7: designed for the downlink transmission in C-RANs, which fulfills the

8: flexible tradeoff between cooperation gain and fronthaul

9: consumption. The queue-aware power and rate allocation with

10: constraints of average fronthaul consumption for the delay-sensitive

11: traffic are formulated as an infinite horizon constrained partially

12: observed Markov decision process (POMDP), which takes both the

13: urgent queue state information (QSI) and the imperfect channel state

14: information at transmitters (CSIT) into account. To deal with the

15: \emph{curse of dimensionality} involved with the equivalent Bellman

16: equation, the linear approximation of post-decision value functions

17: is utilized. A stochastic gradient algorithm is presented to

18: allocate the queue-aware power and transmission rate with H-CoMP,

19: which is robust against unpredicted traffic arrivals and

20: uncertainties caused by the imperfect CSIT. Furthermore, to

21: substantially reduce the computing complexity, an online learning

22: algorithm is proposed to estimate the per-queue post-decision value

23: functions and update the Lagrange multipliers. The simulation

24: results demonstrate performance gains of the proposed stochastic

25: gradient algorithms, and confirm the asymptotical convergence of the

26: proposed online learning algorithm.

27: \end{abstract}