DUE to the rapid growth of the Internet, and end user's insatiable demand for broadband services, access network has emerged as a bottleneck for providing broadband services. To address different challenges in the access segment, such as capacity, cost efficiency, and support of mobile users, various optical and wireless access solutions, such as the IEEE 802.16 (WiMAX) [2] and Fiber to the Home (FTTH) [3]
[4], have been developed. Since wireless and optical access technologies are developed for different purposes, it is difficult for any single technology to address all the challenges of the access network. For example, although FTTH provides a high bandwidth, deploying fiber infrastructure all the way to the user leads to a significant cost, and the network availability is confined within the residential or business unit. Similarly, despite its ubiquitous and flexible connectivity, the limited bandwidth of wireless access network prevents simultaneous access from many users. Recently, the infrastructure wireless mesh networking [5] (or simply WMN in the rest of the paper) has emerged as a promising access solution to address cost, deployment, and mobility issues in a highly populated urban area. As illustrated in Fig. 1, the multi-hop characteristics of WMN enable a backhaul connection (connected to a gateway router) to be shared by multiple wireless mesh routers (mesh router 1, 2, and 3). Different wireless technologies such as IEEE 802.11 and 802.16 can be used for the links among the mesh routers, and the links between users and their nearby mesh routers [5]
[6]
[7]
[8]. Decreasing the number of wireless hops in such WMN by deploying more gateway routers is necessary to achieve a good performance. Because of its high capacity, low loss, and high reliability, fiber is a promising solution to provide backhaul connections to these gateway routers. Combining the complementary characteristics of optical and wireless access technologies, a hybrid optical-wireless access network can provide broadband and ubiquitous service coverage for both fixed and mobile users in a cost-effective way [1]
[9]
[10].
In this paper, we will focus on the implementation of a reconfigurable optical backhaul, and an integrated routing algorithm of the hybrid optical wireless access architecture. The rest of the paper is organized as follows. In Section II we discuss the performance improvement of WMN. In Section III we present the hybrid optical-wireless architecture. In Section IV the testbed and experimental results of the reconfigurable backhaul are demonstrated. Section V describes the integrated routing algorithm and simulation results. Section VI concludes this paper.
SECTION II.
Performance Improvement of Wireless Mesh Network
To clarify the importance of backhaul network for WMN, let's first investigate the throughput per router and overall network capacity of WMN. Consider a one-dimensional WMN as in Fig. 2, where we assume that the distance between two adjacent routers is a constant
{\bf A}
, and each router has the same transmission power and receiver sensitivity. Due to scarce frequency resource, RF channels must be reused, and transmission of one mesh router will introduce co-channel interference to others. The interference in WMN will in turn determine the throughput and capacity. For example, by applying the range model described in [11]
[12], and assuming that the transmission range (TR) is
{\rm A}< {\rm TR} <2{\rm A}
while the interference range (IR) is
2{\rm A}<{\rm IR} <3{\rm A}
as in Fig. 2, only two routers that are at least 4 hops away can transmit simultaneously, such as routers 3 and 7. Assuming that fairness is enforced by a proper algorithm, each router's transmission rate is
1/4
of the link rate. If each router has the same loading, then the throughput per router is:
{Th roughpu} t={{C\times LR}\over {K\times\sum\limits_{1}^{N}n}} \eqno{\hbox{(1)}}
View Source
{Th roughpu} t={{C\times LR}\over {K\times\sum\limits_{1}^{N}n}} \eqno{\hbox{(1)}}
And the network capacity is:
Capacity={Throughput}\times N ={{C\times LR}\over{K\times\sum\limits_{1}^{N}n}}\times N\eqno{\hbox{(2)}}
View Source
Capacity={Throughput}\times N ={{C\times LR}\over{K\times\sum\limits_{1}^{N}n}}\times N\eqno{\hbox{(2)}}
Where C is the number of channels and radio interfaces; K is the number of hops between two that can transmit simultaneously; N is the number of mesh routers served by one gateway router; and LR is the link rate.
According to equations (1) and (2), it is apparent that using more RF channels and radio interfaces per router (a higher C), more advanced coding and modulation schemes (a higher LR), and directional antenna (a lower K) will improve the throughput and capacity of WMN. These approaches, however, will eventually exhaust the wireless bandwidth or require retrofit of the deployed routers, which leads to a significant cost. Instead of resorting to wireless technologies, therefore, an effective way to improve the performance is to deploy more gateway routers in WMN. Since every gateway router has a backhaul connection, a scalable and cost-effective optical backhaul network capable of aggregating backhaul traffic is a key to addressing this requirement.
SECTION III.
Hybrid Optical-wireless Architecture
The proposed hybrid optical-wireless access network architecture is illustrated in Fig. 3. The optical backhaul consists of a WDM ring and multiple tree networks that are rooted at the ring network. The ends of the tree networks are connected to the gateway routers of WMN. Under this hybrid architecture, the upstream packets from users are aggregated at a nearby mesh router and then relayed to one of the gateway routers over WMN. Once the upstream packets reach the gateway router, they are transparently forwarded toward the central hub over the optical backhaul. In the same manner the downstream packets are transferred in the reverse direction.
The key components of the reconfigurable optical backhaul are shown in Fig. 4, where TDM-PON technologies are employed. In the central hub, multiple OLT cards are managed by the system bandwidth management module [1]. A reconfigurable control interface (RCI) is exploited to facilitate reconfiguration. Along the WDM ring, wavelength add/drop devices are used for dynamic bandwidth provisioning. At each gateway router, an optical network unit (ONU) manages the ingress and egress WMN traffic. A pair of DWDM channels is used for downstream and upstream traffic respectively.
A. Reconfigurability of the Optical Backhaul to Realize Load Balancing
Given the TDM-PON technology, network reconfigurability requires tunable transceivers to be used at ONUs as shown in Fig. 4. Tunable transceivers have been proposed for different purposes in optical access networks [13]
[14]
[15]. Due to the rapid advance of optical technology, various tunable laser technologies have been developed [15]. To optimize the cost issue, one of the promising solutions is tunable long wavelength vertically-cavity surface-emitting lasers (VCSELs) [16]. Recently an integrated fast wavelength selective photo-detection was developed [17]. Its tuning time is in nanosecond range, and its monolithic design facilitates cost reduction.
Behind the ONU, an RCI interacts with the RCI in the central hub to manage network reconfiguration. If the bandwidth management module detects unbalanced loading among OLTs [1], it will instruct the RCI to deregister some ONUs from a heavily loaded PON and register them to a lightly loaded PON through the following steps:
-
The RCI at the central hub sends the wavelength information (e.g.
\lambda 2_{{\rm d},{\rm u}}
) of a lightly loaded OLT to the RCI at the to-be-reconfigured ONU.
-
The RCI at central hub instructs the heavily loaded OLT (e.g. OLT1) to deregister the to-be-reconfigured ONU.
-
After deregistration, the RCI at the ONU tunes its tunable transceiver to the new wavelength of the designated OLT (e.g. from
\lambda 1_{{\rm u},{\rm d}}
to
\lambda 2_{{\rm u},{\rm d}}
).
-
After reconfiguration is finished, the RCI at central hub instructs the designated OLT (e.g. OLT2) to discover and register the ONU.
In the next section we will describe the testbed that implements the reconfiguration process.
B. Upgrade of Hierarchical Wireless Access Networks
The reconfigurable optical backhaul can be employed to upgrade hierarchical wireless access networks, such as the one proposed by Google and Earthlink for San Francisco city [18]. In [18], the hierarchical wireless access network consists of 3 wireless layers: wireless mesh layer, capacity injection layer, and backhaul layer, as illustrated in Fig. 5. Since the capacity injection and backhaul layers aggregate the traffic of WMN throughout the city, the capacity of these two layers that use proprietary wireless technologies will soon be exhausted as the number of users increase. By replacing the wireless links with fiber links from the backhaul and gradually to capacity injection layer and installing ONU at the aggregation tower and then gradually the access towers, the capacity can be upgraded with the transparent optical network. In this way the proposed optical backhaul can incrementally address the capacity bottlenecks of the two layers [1].
SECTION IV.
Experimental Testbed of the Reconfigurable Optical Backhaul Network
To demonstrate the reconfiguration scheme and evaluate its performance, we have implemented an experimental testbed based on commercially available devices. The testbed is illustrated in Fig. 6, which consists of two OLTs with fixed optical transceivers and one ONU with a tunable transceiver. We assume that the PON1 and PON2 are heavily and lightly loaded, respectively, and the ONU is originally registered to PON1. The transient response of tuning wavelength from 1591nm to 1586nm is shown in Fig. 7(b), showing that it takes
33.6\mu {\rm s}
for the filter to stabilize after the control voltage is changed. Since the tuning time is much longer than that of the tunable transmitter, the tunable filter dominates the overall ONU reconfiguration period. We thus programmed a reconfiguration period of 50us at the RCI's in FPGA.
To realize the reconfiguration process for load balancing, we implemented the state diagrams for the RCI at the central hub, heavily and lightly loaded OLTs, reconfigurable ONU, and RCI at the reconfigurable ONU according to the discovery, de-registration, registration processes defined in EPON standard. The state machines are summarized in Fig. 8, which are programmed in the two FPGAs. Fig 9 demonstrates the events that took place during the experiment, which include the ONU deregistration from the OLT1 reconfiguration period, and the ONU discovery and registration performed by the OLT2. Note that in Fig. 9 the message propagation delay is insignificant compared to the reconfiguration period that is in ten's of
\mu {\rm s}
scale.
SECTION V.
Integrated Routing Algorithm for Load Balancing
In addition to the proposed hybrid optical-wireless network, we also propose an integrated routing algorithm that is adaptive to traffic demands via dynamic load balancing. The integrated routing algorithm is described as follows:
-
Wireless link state update: Each mesh router periodically probes the states of the links with its neighbors. The link state information will be broadcast and reach the gateways.
-
Local WMN route calculation. Based on the link state information, gateways calculate the optimum route for each local router using a shortest path algorithm.
-
Route cost report: Each ONU reports the routes selected by the gateways and the associated cost to the central hub.
-
Gateway selection: For each mesh router, the route assignment module at the central office selects the gateway with lowest cost.
-
Congestion monitoring: At the ONU, the flow rate of each nearby router is monitored and a capacity table is continuously updated to reflect the loading of WMN.
-
Load balancing. If congestion is detected at any ONU, the ONU will send congestion report to the central office. Route costs to the WMN router through neighboring gateways will be calculated and the route with the lowest cost will be selected.
-
Flow rerouting. Packets to the congested WMN router are rerouted through neighboring gateways.
To evaluate the network performance, numerical simulation is conducted with the NS-2 simulator. The WMN in our simulation consists of 4 gateways and 192 mesh routers as shown in Fig. 10(a), where the distance between any two adjacent nodes is 100m. We use 802.11b protocol in wireless transmission whose data rate is 11 Mbps. Poisson traffic and two-ray radio propagation is assumed throughout the simulation. We also assume constant transmission power and receiver sensitivity among all nodes, and the transmission and interference ranges are 120m and 180m respectively. Each node has only one radio interface and omni-directional antenna is used. Note that in this simulation scenario, the shortest path routing uses hop count as a cost metric and hence is just minimum hop routing. We place uniformly distributed background traffic flow to each mesh router, with the aggregated flow rate out of each gateway to 48 mesh routers being 1 % of the 802.11b's date rate. Assume there is a hot-zone composed of 15 mesh routers near one of the gateways (as shown in the shade region in Fig. 10(a)). We place additional loading in the hot-zone. The congestion threshold of the hybrid routing algorithm is set at 8% of the 802.11b's data rate (0.88 Mbps), i.e. after the aggregated flow rate out of the gateway router exceeds 0.88 Mbps, the corresponding ONU will report congestion to the central hub. The simulation results in Fig. 10(b) and (c) show that after the additional overall loading in the hot-zone exceeds 0.77Mbps (in the presence of 0.11 Mbps background traffic), flows to the boundary routers are shifted to other three gateways to balance the loading. As a result, the throughput and delay are improved for about 25%. It confirms our expectation that the integrated routing algorithm will outperform the minimum hop routing.
In this paper we have proposed a novel reconfigurable optical backhaul that aims to address the scalability issue in WMN and leads to a hybrid optical wireless access network. Both the TDM-PON and WDM ring technologies are leveraged due to their cost-effectiveness and technological maturity. Reconfigurability in the optical backhaul is useful for load balancing to improve efficiency and performance, and we have built an experimental testbed to demonstrate its feasibility. We have also proposed an integrated routing algorithm for this hybrid optical and wireless architecture, which is adaptive to traffic demands with dynamic load balancing.