SECTION I
INTRODUCTION & MOTIVATIONS
The ever increasing demand for high speed wireless communications mandates innovative solutions to be developed to efficiently utilize the limited spectral resources available. Conflicting quality of service (QoS) requirements by the end users makes service provisioning a challenging task, fitting examples of which are high data throughput, ubiquitous connectivity and seamless mobility. The case of systems operating over the license-free bands is even more challenging, due to presence of interference from other systems operating in the same band.
Wireless local area networks (WLANs) are by far the most widely deployed communication system operating over license-free bands. Furthermore, it is well known that as the number of WLAN stations (STA) associated with a given access point (AP) increases, so does the probability of packet collision. The resulting performance degradation significantly reduces the perceived QoS of the end users in this case. Similarly, operation of several APs in geographic proximity of each other will adversely affect operation of all those APs. The goal of this article is to introduce a novel WLAN architecture and coordination mechanism, which we name as Cognitive WLAN over Fibers (CWLANoF). Not only can this architecture extend the coverage area of a WLAN network, but also considerably improve the management capabilities and performance of such a network.
Our approach is to transform a conventional distributed IEEE 802.11 WLAN Extended Service Set (ESS) into a centralized architecture with cognitive radio (CR) capabilities. To this end multiple WLAN APs, each forming its own Basic Service Set (BSS), are replaced by passive remote antenna units (RAUs) which are connected to a centralized processing unit using broadband radio over fiber (RoF) technology [1]. The system architecture, shown in Fig. 1, will be discussed in detail. This centralized infra-structure-based architecture provides a desirable framework to facilitate CR operation and enable implementation of coordinated radio resource management (RRM).
Several studies in the literature have demonstrated the feasibility of carrying a wide-band signal through RoF systems [2], and more specifically, multiple WLAN channels over low-cost multimode optical fibers. Furthermore, the WLAN medium access control (MAC) is shown not to be affected by RoF transmissions [3]. Therefore, developing a WLAN over fiber solution, capable of operating over multiple channels is practically feasible. However, most of these studies are focused on the design of optical-electronic converter components. Furthermore, only simulcasting mode was examined in experiments on RoF with multiple WLAN channels.
An interesting capability of the proposed CWLANoF is its support for multiuser MAC. A number of studies in the literature that investigated multipacket transmission or reception for WLANs [4] are of particular interest to our framework. However, no study so far has proposed mechanisms that can provide both multipacket transmission and reception for WLANs.
While multiple RAUs can cooperate for macro diversity gains, each RAU can further utilize multiple antennas for micro diversity gains without increasing the number of fibers, if WDM is used to conned each RAU to the CogAP.
Furthermore, integration of CR capabilities in WLANs, e.g., through modification of MAC design, has recently been investigated [5], [6]. The aim of such studies is to devise methods in order to independently improve the performance of coexisting WLAN APs. However, to the best of our knowledge, no previous work has considered the CR capability for multiple co-operating receivers with distributed antennas in WLAN over fiber systems.
The salient features of CWLANoF, making it uniquely different from existing CR solutions, as well as conventional WLAN systems, can be summarized as follows.
- A set of distributed RAUs provide temporal, spatial and spectral knowledge of the radio scene for the centralized processing unit in a CWLANoF. This radio frequency (RF) knowledge in turn facilitates reliable and efficient utilization of the available industrial, scientific and medical (ISM) spectrum in real-time.
- CR capabilities, such as wideband spectrum sensing, can be incorporated in a CWLANoF based on existing WLAN standards without requiring any software or hardware upgrade at the WLAN STAs.
- An extensive array of cooperative transmission and reception techniques, including distributed multi-input-multi-output (MIMO), beamforming and transmitter/receiver diversity, can be supported in the CWLANoF framework, which are less available in conventional WLAN systems.
SECTION II
SYSTEM ARCHITECTURE
In a CWLANoF, the APs are replaced by RAUs that are connected to the centralized controller, named cognitive access point (CogAP) using a passive optical network (PON), as shown in Fig. 1. For low cost implementation, this PON employs one pair of optical fibers for duplex transmissions between an RAU and the CogAP. In the simplest case, the RAU employs a different transmit and receive antennas connected via power and low noise amplifiers to the respective fibers. If multiple transmit/receive antennas at an RAU or multiple RAUs are fed via a shared pair of optical fibers, wavelength division multiplexing (WDM) techniques can further be exploited.
Unlike conventional WLANs, where each AP is equipped with its own radio modem, in a CWLAN of all radio modems and bridges are shifted to the CogAP and the RAUs are merely a distributed set of antennas with no processing capability. Further, RF signaling between RAUs and CogAP is in analog format. By centrally processing broadband RF signals received from the RAUs over the fibers, the CogAP acquires a complete picture of the radio spectrum usage in its coverage area, i.e., the radio scene. Channelization of the spectrum and the application of distributed coordinated function (DCF) of 802.11 MAC employing carrier-sense multiple access with collision avoidance (CSMA/CA) over each channel is carried out at the CogAP instead of at individual RAUs.
While multiple RAUs can cooperate for macro diversity gains, each RAU can further utilize multiple antennas for micro diversity gains without increasing the number of fibers, if WDM is used to connect each RAU to the CogAP. The aforementioned system architecture benefits from both a centralized processing unit and a distributed set of antennas. This unique system design motivates a rich set of research topics, which are elaborated upon in the following section.
SECTION III
RESEARCH ISSUES AND POTENTIAL SOLUTIONS
The availability of a comprehensive knowledge of the radio scene at CogAP, acquired by the RAUs, facilitates several innovative solutions in order to improve the capacity of CWLANoF and maintain a high level of QoS in the network.
Cognitive Radio Capabilities
The broadband RAUs enable the CogAP to acquire comprehensive radio scene knowledge in time, frequency and space domains. More specifically, the ISM band is shared by many different systems, which can potentially cause severe interference to a legacy WLAN system.
Spectrum sensing: CRs rely on spectrum sensing to identify occupied channels and thus mitigate mutual interference to coexisting systems. CR capability in a CWLANoF enhances the system performance by facilitating equal sharing in the license-free bands. By combining sensing data from dispersed RAUs at the CogAP, a cooperative spectrum sensing strategy can be implemented to cover the entire ISM hand over a relativelv large locale.
CR resource allocation: Designing practical scheduling mechanisms that can achieve efficiency and fairness in shared spectrum scenarios is of utmost importance in future communications systems. In license-free bands the challenge is to counteract the greedy spectrum access of coexisting systems as they all have an equal right in transmitting over these spectral ranges [7]. Development of a CR-enabled MAC for implementation in CogAP will alleviate the need of altering existing IEEE 802.llx MAC protocols, while sustaining a higher system capacity due to more intelligent interference management.
Spatial interference mitigation/avoidance: Most existing CR radio resource allocation schemes rely on orthogonal transmissions in the frequency domain over the shared frequency band in order to avoid interference between systems. However, the direction of arrival of interfering signals to a CWLANoF can be identified at the CogAP, enabling estimation of the spatial distribution of interference over the coverage area. The comprehensive radio scene knowledge at CogAP acquired by the distributed set of RAUs thus facilitates identification of the locations of other interfering devices (including coexisting WLAN systems or even microwave ovens). This spatial interference awareness, in turn, can increase the capacity of a CWLANoF, e.g., through usage of beamforming techniques.
SECTION IV
LOAD BALANCING
A well-known characteristic of WLANs is the formation of hotspots, whereby the non-uniform spatial distribution of STAs will overload specific APs. To alleviate this issue, load balancing will reduce the packet collision probability and enhances the network capacity. Note that although in a CWLANoF context various STAs will be attached to different RAUs, possibly even on different radio channels, it is understood that the signal processing activity of all RAUs will still be performed by the CogAP. Some interesting load balancing solutions which are facilitated by CWLANoF architecture are as follows:
Frequency planning for RAUs: The CogAP is capable of operating different RAUs independently. To reduce the interfering effect of adjacent RAUs on each other, resulting in a lower collision probability especially when dealing with hotspots, different channels are allocated to each RAU by the CogAP. Further, if a given RAU is facing a higher traffic demand, multiple channels can be allocated to meet the demand [8].
Given the centralized decision making at the CogAP, allocation of additional channels to RAUs can be dynamic and event-driven. For instance it is well-known that the packet collision probability increases with the traffic load. In that respect, collision probability at each RAU can be exploited as a triggering mechanism such that one more channel will be allocated to a specific RAU when its collision probability exceeds a certain threshold.
A well-known characteristic of WLANs is the formation of hotspots, whereby the non-uniform spatial distribution of ST As will overload specific APs. To alleviate this issue, load balancing will reduce the packet collision probability and enhances the network cepecity.
Example: Consider the case of two RAUs covering a given area, operating over channels with Irequencies
$f_{1}$ an d
$f_{2}$. Further, suppose RAUl detects a rate of collisions that is higher than a threshold in channel
$f_{1}$. Then the CogAP can use the “disassociation” process to force some of the STAs attached to RAUl on
$f_{1}$ to be dissociated from this channel, while simultaneously allocating a new channel,
$f_{3}$, to this RAU. Subsequently, RAUl starts transmitting beacons over channel
$f_{3}$. There exist two possibilities for any STA dissociated over
$f_{1}$. It may be able to receive beacons over channel
$f_{2}$ from RAU2 leading it to associate with RAU2 over this channel. In this manner, a portion of the traffic load of RAUl will be transferred to RAU2, creating a distributed load balancing solution. Alternately, the STA may receive beacons over channel
$f_{3}$ from RAUl, and request to associate with RAUl over this new channel. The shifting of STAs tO
$f_{3}$ at RAUl amounts to a frequency domain load balancing. This latter case is made possible particularly by the broadband RoF connections between RAUs and the CogAP. In contrast, conventional WLAN APs are generally not equipped for multi-channel operations.
Macro-diversity mechanisms: An alternative load balancing solution is to operate the dispersed set of RAUs as cooperative access nodes of a unified WLAN system. In this manner macro-diversity gains through transmitter or receiver diversity schemes can be achieved. If each RAU is equipped with multiple antennas, micro-diversity gains can further be achieved.
Let us first investigate receiver diversity techniques facilitated by the CWLANoF architecture. Packets transmitted from a given STA may be received by several RAUs. Through exploitation of maximum-ratio combining (MRC) at the CogAP, given that channel state information (CSI) for each RAU-STA link is known, the system achieves an array gain proportional to the number of RAUs due to the increased receiving antenna gain. Further, an additional diversity gain is achieved if the RAU-STA links experience independent fading. Both gains will help to reduce packet collisions, resulting in increased system throughput and reduced packet error rate (PER). Other combining techniques, such as equal gain combining (EGC), can also be utilized at the CogAP. Extensive simulation results supporting our argument can be found at [8].
Packets transmitted by the CogAP to a specific STA can also be sent via multiple RAUs. Similar to the case of receiver diversity, this approach enables various transmitter diversity techniques to be exploited. Transmitter diversity helps to even the spatial distribution of signal-to-noise ratios available at the STAs’ receivers. By reciprocity of the channel, transmitter diversity at the CogAP has the same effect as receiver diversity. However, to ensure signals from different RAUs can be added coherently at the receiving STA, the CSI information at the CogAP should be up-to-date, such that signal phases can be properly shifted for different RAUs and transmission powers can be optimally allocated across RAUs.
SECTION V
MULTIUSER MAC MECHANISM
A novel mechanism to more efficiently utilize resources of WLAN systems is multiuser MAC, whereby multiple users are served in each transmission to/from the AP. This multiuser MAC can be implemented in an array-antenna architecture, such as IEEE 802.lln standard.
Exploitation of multi-packet transmission (MPT) in WLAN MU-MAC has been proposed in [9] as shown in Fig. 2a for the case of an AP with
$K=2$ antennas. In this approach, the AP sends a multi-user request-to-send (MU-RTS) frame to
$K_{1}$ STAs. This MU-RTS contains an ordered list of STAs, determining the priority of STAs in the sequential multiuser clear-to-send (MU-CTS) stage. Each STA in the MU-RTS list feeds back their measured CSI through their own MU-CTS frames and send the frames back to the AP in the order specified in the MU-RTS. The provisioned CTS order, besides alleviating collisions, facilitates disjoining of STAs to the MPT process. To this end, each STA is assumed to be able to overhear the previous CSI reports from other ST As and can make a decision based on these reported CSI as well as its own CSI. If the MU-CTS of a given STA is not reported back to the AP, the corresponding STA will not be included in the MPT.
Following the MU-RTS/CTS sounding, the AP can use MPT to serve K2 STAs, where
$K_{2}\leq K$) simultaneously. As no multi-packet reception (MPR) capability is provisioned in [9], these
$K_{2}$ STAs will follow a sequential acknowledgment phase. We propose to extend this framework to a MU-MAC for OFDM-based systems supporting both MPT as well as MPR to simultaneously receive MU-CTS and ACK frames. This novel MU-MAC, shown in Fig. 2b, helps reduce the overhead of MU-RTS/CTS sounding and sequential ACKs by using MPR. Multi-packet reception effectively collapses multiple control frames, such as MU-CTS or ACK, into a single control frame, and thus eliminates unnecessary inter-frame spaces and backoff times. Therefore, a higher MAC efficiency can be achieved, especially for short frames as pertinent to voiceover-IP application for instance. Simultaneous reception of MU-CTS frames also keeps CSI more up-to-date than the sequential reception scheme owing to the reduced delay between the MU-RTS and data packets. Numerical results verifying the superior performance of joint MPT-MPR is elaborated in [10].
SECTION IX
ARCHITECTURE AND IMPLEMENTATION DETAILS
An overview of the SDR-based testbed platform for CWLANoF is shown in Fig. 3. The centralized multi-channel SDR platform consists of RF front-ends, a powerful processing unit based on digital signal processors (DSPs) and field-pro-grammable gate arrays (FPGAs), and a number of ADC/DACs to translate signals between the analog domain and digital domain without information loss. The STA is a portable small form factor DSP-based SDR platform with sufficient signal processing capability.
The CogAP is implemented in a chassis as illustrated in Fig. 4. Corresponding RF frontends connect to the ADC/DAC units through coaxial cables for analog signal feed and digital cables for control purposes. Our chosen SDR platform uses the Peripheral Component Interconnect (PCI) or compact PCI bus for interconnecting various units. DSP programs and FPGA designs are uploaded from a host personal computer to the processing unit via high speed Ethernet ports. The personal computer also serves as a control console of the SDR platform during the experiment set-up.
There is no off-the-shelf low cost RF - to- IF (Intermediate Frequency) module available with a 75-MHz bandwidth; therefore, two overlapping 40- MHz channels are configured to cover the 75-MHz North American ISM band spectrum. RF front-ends simultaneously translate four such 75- MHz channels into sixteen baseband signals. The ADC unit then sample and quantize these signals and feed them to the processing unit for processing in the digital domain.
Processing units of popular commercial SDR platforms are based on either general-purpose processors such as those employed in personal computers, or special-purpose processors such as FPGA, OSP, and their combination.
Cable-connected RAUs are needed for a comparative study to examine the advantage of radio over fiber systems over traditional cable-fed wireless access systems. They are also useful in experimental setups in which the RAU is not located at a great distance from the hub. We now elaborate on several important issues on the SDR platform.
Processlna Unit
—Processing units of popular commercial SDR platforms are based on either general-purpose processors such as those employed in personal computers, or special-purpose processors such as FPGA, DSP, and their combination.
Software for general-purpose processors requires the least programming specialization, provides the best code-reuse and an easy upgrade path to more capable processors via direct replacement of the processor. However, it consumes much processing resources to realize a real-time channel decoder (e.g., convolutional code, Turbo code or low-density parity check code) or accomplish MIMO signal processing using a general-purpose processor. The data link between the personal computer and ADC/DAC units can only provide 480 Mb/s, which is hardly enough for sampling a 20-MHz WLAN channel, not to mention the entire 75-MHz ISM band.
FPGA/DSP-based processing units are capable of parallel operations such as channel decoder, filters, and other advanced signal processing tasks. Therefore, in spite of widely available open sourced, general-purpose processor-based SDR platforms, we chose a FPGA/DSP-based SDR platform. The processing unit contains multiple DSPs, yielding enough processing capability for our research needs. Each DSP in the processing unit also contains one Viterbi and Turbo channel decoder to ease wireless application development. This hybrid DSP /FPGA structure enables rapid prototyping and real-time processing capability.
ADC/DAC Units
The SDR platform can be synchronized to a common clock since both RF front-ends and all units contain reference clock input and output ports,’ thus, up to 8-by-8 MIMO applications can be supported when two RF front-ends are synchronized.
—The ADC and DAC units provide a sufficient number of ADC/DACs with good resolution to handle four 75-MHz channels in parallel. There are sixteen l4-bit 105 million samples per second (MSPS) ADCs and sixteen 14-bit 480 MSPS DACs. The data-path bottleneck is the path between ADC/DACs and the processing unit, for which the star topology is preferred over bus topology. In our testbed, the data path between each ADC/DAC unit and the processing unit is an 8-Gb/s full duplex ultra-high-speed link. This allows advanced joint-pro-cessing algorithms to be implemented at the processing unit.
RF Front-Ends
—Since in-house development of RF front-ends is costly and time-consuming, we chose off-the-shelf RF front-ends under the following considerations.
Independent RF Carrier Tuning and Transmitting/Receiving Gain Control
—Each channel in the RF front-end can be independently tuned over the 2.4 GHz ISM band with full control on receiving and transmitting gain through a standard control cable connected from the processing unit. The RF front-end also supports 5 G Hz Unlicensed National Information Infrastructure (U-NII) band operations which offer additional flexibility for future use.
In-Phase/Quadrature (I/Q) Separation
—We can separate I/Q components in the analog or digital domain, each having their own pros and cons. Analog I/Q separation requires ADCs with a lower sampling rate and therefore a less expensive implementation. However, it suffers from direct current offset, carrier leakage and I/Q imbalance problems. Digital I/Q separation can mitigate these problems by first translating a RF signal to an intermediate frequency (IF) signal, sampling the IF signal with a high sampling rate, and then digitally separating the I/Q components. Digital I/Q separation, however, requires an RF-IF module supporting 75 MHz band-width, which is costly. To avoid procuring expensive customized RF-IF modules, we choose to use analog I/Q separation initially, while specifying a standard control interface between RF front-ends and the processing unit. Another benefit of a standard control interface for RF frontends is that we are not tied with single RF front-end provider.
Synchronization
—The SDR platform can be synchronized to a common clock since both RF front-ends and all units contain reference clock input and output ports; thus, up to 8-by-8 MIMO applications can be supported when two RF front-ends are synchronized.
SDR Platform of Mobile Stations
—Ideally the STAs should follow the same design as the CogAP for the convenience of software development and future research needs. However, under a tight budget, mobile STAs are built on a single channel small form factor platform, including one stand-alone processing unit with an ADC/DAC add-on module providing two 14-bit 125 MSPS ADCs and two 14-bit 500 MSPS DACs. The data path between the ADC/DAC module and the processing unit is a 12-Gb/s full-duplex ultra-high-speed link. Both the ADC/DAC module and the processing unit provide reference clock input and output ports, enabling the platform to be easily synchronized.
Test Plan
—Following the research targets described earlier, we plan to carry out extensive experiments as will be discussed next.
Cooperative Spectrum Sensing and RRM in the Entire 75-MHz ISM Band
—At an early stage of our test plan, only uplink is considered since only the CogAP is able to handle the entire ISM band in real time. Once the non-WLAN channel usage in the ISM band is detected and identified, the frequency plan of WLAN can be changed to avoid the interference to WLAN. Another way is to mitigate interference through co-channel-interference mitigation techniques that are facilitated by the distributed antennas.
Single-User Distributed MIMO WLAN with Single 20-MHz WLAN Channel
—We shall examine the feasibility of receiver-based and transmitter-based MRC, where we need to verify that the phase jitter and other nonlinear effects of the DAS will not affect successful coherent addition at the target STA.
Multi-User Distributed MIMO WLAN with Single 20-MHz WLAN Channel
—We shall examine the feasibility of space-division multiple access (SDMA) and MIMO precoding. Both synchronous and asynchronous channel use will be considered, where asynchronous channel usage indicates that transmitting STAs have different packet sizes and therefore different channel access periods. The experimental results will be compared for both cable-fed and fiber-fed DAS to observe the effects of non-linearity of optical components. We will also examine the practicability of using SDMA and MIMO precoding in a heterogeneous 802.11 network where 802.11g and 802.11n STAs coexist.
Multiple 20-MHz WLAN Channels in the ISM Band
—In the presence of multiple WLAN channels, we expect stronger nonlinearity effects of optical components on MRC and CCI mitigation. Operations of multi-channel MAC and multiple wireless services over the WLAN will be verified in the later stage of our testbed deployment. Our final goal is to utilize the entire ISM band as a common pool of radio resources. The enabling techniques include cooperative spectrum sensing, advanced RRM techniques, OFDMA or other novel physical and MAC layer technologies.
This work was supported in part by the Canadian Natural Sciences and Engineering Research Council.