Multi-carrier and Cognitive Networks
Contents |
Overview
Participants
Faouzi Bader, Bertrand DeVillers, Lorenza Giupponi, Christian Ibars, David Pubill, Marco Miozzo, Ana Maria Galindo, Ana Moragrega, Musbah Shaat
Activities
Multi-carrier Systems
Multicarrier transmission techniques based on FBMC were developed in the seventies to perform the conversion between PCM (Pulse Code Modulation) and FDM (Frequency Division Multiplexing) systems. In the nineties, OFDM was preferred as multicarrier scheme because it was considered simpler in concept, less complex and it had minimum latency. OFDM and its variant the OFDMA scheme are the basic communication scheme for the nowadays standards (WLAN, WiMAX, LTE, etc.).
Multicarrier communication systems have been also suggested as a candidate for cognitive radio (CR) systems due to their flexibility to allocate resources between the different SUs. As the SU and PU bands may exist side by side and their access technologies may be different, the mutual interference between the two systems is considered as a limiting factor affects the performance of both networks. OFDM based CR system suffers from high interference to the PUs due to large sidelobes of its filter frequency response. The insertion of the cyclic prefix (CP) in each OFDM symbol decreases the system capacity. The leakage among the frequency subbands has a serious impact on the performance of FFT-based spectrum sensing, and in order to combat the leakage problem of OFDM, a very tight and hard synchronizationOFDM scheme is not considered as the most appropriate physical layer for systems that must support dynamic spectrum access and also formainly in cognitive radio based networks.
Recently, it has been pointed out that by keeping the cyclic extension of each block and by moving the IFFT operation of the OFDM modulation from the transmitter to the receiver, we end up with a so-called cyclic prefixed single-carrier system with frequency domain equalization (SC-FDE), whose multiple access version is known as SC-FDMA. Solving the peak to average power ratio (PAPR) problem of OFDM, this alternative has received growing interest and been introduced in the uplink of the 3GPP LTE.
The filter bank multicarrier system (FBMC) does not require any CP extension and can overcome the spectral leakage problem by minimizing the sidelobes of each subcarrier and therefore lead to high efficiency (in terms of spectrum and interference).Moreover, efficient use of filter banks for spectrum sensing when compared with the FFT-based preiodogram and the Thomson’s multitaper (MT) spectrum sensing methods have been recently discussed in scientific literature.
So far, some attempts have been also made to introduce the filter bank multicarrier scheme (FBMC) in the radio communications arena, through proprietary schemes, in particular the IOTA technique (see: TIA-902.BBAB, “Wideband air interface Isotropic Orthogonal Transform Algorithm (IOTA) physical layer specification”, document of the Telecommunications Industry Association, 2003, and, 3GPP TSG-RAN WG1 TR25.892 (Feasibility Study of OFDM for UTRAN Enhancement. V1.1.0, June 2004). However, the full exploitation of FBMC and the optimization in the context of radio evolution, like dynamic access, has not been considered so far, as well as their combination with multiple antennas techniques.
Solving the PAPR, the Out-of-band emission problems of OFDM, SC-FDE/SC-FDMA is of special interest when channel state information is not available at the transmitter side, as it naturally exploits the multipath diversity under some conditions. In particular, with linear FDE, this characteristic has been shown to be strongly related to the receiving architecture (ZF or MMSE). This dependence has to be analyzed, mainly in a multiuser/multiantenna scenario, with special focus on decision feedback equalizer (DFE) wich has been shown to reduce the achievable rate loss with respect to OFDM.
Therefore, to reduce the gap exploitation of FBMC, and the SC-FDE/SC-FDMA and their optimization in the context of radio evolution, the main research in multicarrier of interest include, but are not limited to:
- Analysis the appropiate chunk/slot/bin shapes (time-frequency (and space)) sizes on standards based on multicarrier schemes, to increase the throughput and spectral efficiency in both cognitive and no congnitive contexts.
- Use in LTE/WiMAX-like Non-OFDM (NOFDM) or FBMC systems in cognitive based systems.
- Design of efficient RRM algorithms for up/downlink based cognitive networks (NOFDM, FBMC vs. OFDM(A)).
- Minimum frequency gap between adjacent bursts in the uplink using the FBMC to minimize multiuser interference and comparison with regular OFDM(A) scheme mainly in cognitive radio context.
- Optimisation of the FBMC carrier shape to deal with ICI effect problem.
Informations on recent results in the field are available in AT publications website.
Cognitive Radio Networks
Another point of interest is to study how self-organization concepts can be applied to multicarrier networks in order to pursue autonomous and self-optimized systems capable of reducing the burden over the network operator core network. Some examples of self-organized multi-carrier systems, which aims at autonomously managing the radio resources for coexistence in the spectrum band with other systems or networks, are the cognitive radio networks, based e.g. on IEEE 802.22, and the femtocell networks, based e.g., on LTE or Wimax.
The state of the art of radio resource management in multicarrier networks is quite vast, however, contributions focusing on taking advantages of machine learning schemes for designing truly cognitive approaches are still to be deeply investigated, especially in distributed network architectures, where the intelligence of the network is not centralized on nodes such as Radio Network Controllers (RNC). The taxonomy of access management of different coexisting links can generally be divided into the following approaches:
- Fixed Access: Distinct chunks of spectrum are clearly assigned to each link. Note that this does not preclude the use of opportunistic or cognitive RRM techniques within each link.
- Opportunistic Access: The less important link utilizes the bands of the more important link(s) in an opportunistic fashion which is typically triggered by the interference between coexisting links exceeding a prior set threshold.
- Cognitive Access: The threshold decision engine of opportunistic techniques is replaced here by a more sophisticated decision engine, which typically learns from past experiences and takes into account anticipated decisions.
Cognitive access is generally superior to opportunistic access, which in turn is superior to fixed access. We will therefore henceforth concentrate on cognitive principles only.
The high-level operational cycle of cognitive radios relies, in the context of RRM, on the following elements:
- Acquisition: The acquisition unit provides quintessential information of the surrounding environment, such as spectrum occupancy or interference temperature. .
- Intelligent Decision: The core of a cognitive radio is without doubt the intelligent decision engine, which typically learns from past experiences gathered from e.g. dynamics of interference or statistics of spectral occupancy. Based on some intelligent algorithms, it then draws decisions on choice of band and resource block, transmission power, etc.
- Action: With the decision taken, an important aspect of the cognitive radio is to ensure that the intelligent decisions are being carried out, which is typically handled by a suitably reconfigurable software defined radio (SDR), some policy enforcement protocols, among others.
The progress beyond the stat of the art proposed for the future years is first to evaluate the applicability of the machine learning based schemes described in the previous section to multicarrier systems and to model by means of game theoretic schemes the performances of the distributed multicarrier systems at the equilibrium point. However, many issues remain open.
The first problem related with reinforcement learning based algorithms, especially when implemented in a decentralized setting, as it is the case in a distributed femto network or a cognitive radio network is the speed and accuracy of convergence to the prior set targets. Efforts are actually undertaken focusing how to make nodes with greater experience share their expert knowledge with other nodes.
The second problem is related with the so called dynamics of learning. According to the paradigm of independent learning, each node learns its policy independently of the other nodes, but then the transition model depends on the policy of the other learning nodes. The third problem is related with the memory requirements of the proposed approach and to the discrete state space representation, which is generally assumed in literature. Since the expert knowledge of the Q-learning approach is commonly uploaded in Q-table, scalability issues may arise when the state or action spaces increases. Different techniques are here explored and investigated in the context of neuronal networks to estimate the Q-learning function. The Game theory, Markov decision process, and Fuzzy logic approaches are also valuable tools used to reduce the complexity of the representation mechanism.
Informations on recent results in the field are available in AT publications website.
