1. Introduction

As mobile systems evolve to their fifth-generation (5G) [1, 2], the quality of experience (QoE) — the ability to interconnect and manage a plurality of mobile devices/connections per person as a device-centric “Internet of Things”— is emerging as the vital performance metric [1]. To optimize QoE, 5G networking will place increased emphasis on improved spectral efficiency and coverage via massive multiple input multiple output (MIMO) systems with hundreds of spatially-distributed antenna elements, as well as on decentralized device-to-device (D2D) connectivity that achieves ultra low latency short-range transmission via network bypass [1]. To support 4G mobile, a novel optical access segment termed mobile fronthaul (MFH) composed of fiber links carrying digitized radio traffic between centralized baseband units (BBUs) and remote radio heads (RRH) has already emerged as the leading high-speed, low-latency connectivity solution [3–6]. Since current optical MFH architectures connect each BBU with an isolated set of RRHs through a fixed optical topology, BBU pooling and virtualization have been proposed to overcome topology limitations through a cloud-based approach in which resources are dynamically moved across a single virtual BBU platform that governs many spatially-distributed RRHs [6]. This capability, along with efficient usage of BBU resources, will be crucial for bidirectional coordinated multipoint (CoMP) in 5G massive MIMO systems [1–6]. However, to virtualize many physically distinct BBUs into a single platform, ultra low latency connectivity between many BBUs will be needed [6], which may be more readily feasible with centralized rather than distributed control. Moreover, BBU pool inefficiencies can arise for CoMP with 5G massive MIMO if multiple BBUs must send/receive the same message to/from many RRHs, rather than delegating this to one BBU. Finally, BBU pooling in fixed topologies does not currently address decentralized 5G D2D, which under its current formulation cannot interact with the network infrastructure such that is restricted to cases where the communicating parties are in very near physical proximity. Longer-range (e.g. inter-cell D2D) would thus require inter-BBU 5G traffic hand-offs that would ideally be handled more locally, or entail multi-hop hand-offs deeper in the mobile backhaul network, increasing latency. To overcome these limitations that can strongly affect QoE, centralized software-defined networking (SDN)-controlled optical and electrical switching can be introduced into the optical MFH network, evolving it into a novel topology-reconfigurable architecture that is back-compatible with legacy fiber yet can enable high-speed and low-latency inter-BBU connectivity, efficient bidirectional CoMP, and inter-cell D2D with a high degree of network bypass. We propose both electrical and optical switching to enhance architecture flexibility (e.g. routing flows along different switching paths based on their granularity and/or QoE requirements), as well as redundancy (e.g. enabling multiple potential paths in case of failure). Moreover, with a SDN-based centralized control interface, such as OpenFlow [7], advanced features can be dynamically instantiated as match/action combinations in control plane flow tables, with transparency across protocols, vendors, and switch types (optical vs. electrical). In this paper, we propose and experimentally demonstrate the first SDN-controlled topology-reconfigurable optical MFH architecture for 10Gb/s peak per-cell bidirectional CoMP and inter-cell D2D connectivity in the 5G mobile era. Back-to-back latency < 7μs and a 29.6dB aggregate power budget are experimentally verified. By enabling advanced networking features, the new approach is attractive for future optical MFH.

2. Topology re-configurability for optical MFH in the 5G mobile era

Figure 1 illustrates (a) downlink (DL) CoMP; (b) uplink (UL) CoMP; and (c) inter-cell D2D under conventional fixed topology and proposed SDN-controlled topology-reconfigurable approaches. As shown in Fig. 1(a), for a fixed topology, to accomplish DL CoMP from multiple 5G cell sites to a single mobile terminal (MT), all BBUs in charge of the respective 5G sites communicate over the legacy X2 interface (a logical point-to-point IP-based link [5]) and then transmit the same user data over parallel optical MFH links. The data is then delivered to the MT through the wireless last hop. However, with topology re-configurability, a new path can be created from BBU A to the second 5G cell site through SDN-controlled optical/electrical switching, such that DL CoMP is performed using just the computational resources of BBU A. Moreover, BBU A and BBU B can be interconnected directly through the SDN switch, with inter-BBU connectivity managed centrally by the SDN controller (not shown in Fig. 1(a) due to space constraints), reducing latency compared to distributed control. Similarly, for uplink (UL) CoMP, as per Fig. 1(b), a second uplink path to BBU A is created by the SDN-controlled switch, releasing computational resources of BBU B and off-loading the inter-BBU interface. Finally, as shown in Fig. 1(c), with a fixed topology, D2D connectivity between two MTs in different 5G cell sites cannot occur over the optical MFH segment, and may not be possible between BBU C and BBU D directly since the X2 interface would need to handle multi-Gb/s traffic as well as ultra-low latency control messaging. Instead, the inter-cell data transmission may need to exploit the mobile backhaul network, increasing transmission latency. With the topology-reconfigurable approach, however, a high-speed bypass path can be created over the optical MFH segment, enabling high-speed, low latency inter-cell D2D without burdening the backhaul, BBUs, or the inter-BBU interface.

 

Fig. 1 Optical MFH scenarios with and without topology reconfigurability: (a, left) DL CoMP; (b, center) UL CoMP; (c,right) Inter-cell D2D; MT = mobile terminal.

Download Full Size | PDF

To realize 5G optical MFH scenarios of Fig. 1, optical and electrical switching elements with an OpenFlow interface can be introduced at the central office, and configured by a logically-centralized SDN controller. To address point-to-multipoint network scenarios, the architecture should support high power budgets (as confirmed by experiment) as well as optical-layer re-configurability/tuning, which for the proposed architecture can be fully supported via pass-band tuning of optical switching elements. The proposed approach is thus compatible with both software-defined λ−tunable and λ−fixed optical MFH Tx/Rx. Finally, for the DL and UL CoMP scenarios of Figs. 1(a) and 1(b), any differential delays introduced by the SDN-controlled switching elements to the parallel optical MFH transmission paths should not exceed the latency budget determined by the cyclic prefix length of the underlying OFDM wireless signal. For LTE and LTE-Advanced systems, cyclic prefix lengths of 4.7μs and 16.6μs are allocated for regular and extended prefix cases respectively [8]. Since the scaling to 10Gb/s peak rates in 5G will be achieved via massive MIMO rather than OFDM format changes, the 4G differential delays are expected to hold true for 5G systems as well.

3. Proposed SDN-Controlled Topology-Reconfigurable Optical MFH Architecture

Figure 2 illustrates the proposed topology-reconfigurable 5G optical MFH architecture. At the central office, the BBU pool features connections to the electrical SDN-controlled OpenFlow switch, 10⁺Gb/s software-defined (SD) optical Tx/Rx array, and any-to-any optical switch, all under the domain of an SDN controller and API. Through dynamic control of SDN-enabled Tx and Rx switch ports (a)-(F) and SDN-enabled optical Tx ports (1)-(9), advanced features such as CoMP, low-latency inter-cell D2D transmission and inter-BBU connectivity are enabled. As shown in Fig. 2, a FlowMod control message with a single match and two actions can be used to instantiate DL CoMP, by which the DL user data from BBU A is sent from Tx port (a) of the OpenFlow switch to dual Rx ports (i) and (k). This scenario models both DL CoMP from two transmitters to a single user at the aggregate peak rate, and DL CoMP of the same aggregate message to multiple users at sub-peak per-user rates. Two single-match, single-action FlowMod messages can be exploited to realize UL CoMP from a single user to two transmitters with comparable UL channels by merging uplink traffic from Tx ports (j) and (l) onto a common Rx port (d) for processing in BBU B. Moreover, a single match-action FlowMod linking designated BBU Tx/Rx ports can be invoked to establish low latency, high-speed (10Gb/s) inter-BBU connectivity, and also perform load balancing (BBU C→ BBU D in Fig. 2). Optical DL CoMP (Fig. 2, solid green lines) can also be enabled via SDN control of the any-to-any optical switch that differs from a conventional optical switch in its capability to support optical DL multicasting as well as loopback functionality. Figure 3 shows the detailed any-to-any optical switch structure for a 4 × 4 case of DL/UL tunable wavelength sets λ₁-λ₄ and λ₅-λ_8, respectively, with port labels in Fig. 3 matched to those of Fig. 2. To enable optical DL CoMP, port A of the 1:2 WSS in Fig. 3 is configured to output DL wavelength λ₁ in lieu of inactive DL wavelength λ₄ by using a FlowMod control message extended to support λ-based WSS pass-band tuning (Fig. 3, first 5G feature). Moreover, by de-activating DL λ₃ and re-tuning UL wavelength λ₅ to λ₃ using an extended FlowMod message, a low latency loopback path for inter-cell D2D (Fig. 3, second 5G feature) is created in the optical switch by the 1:2 colorless optical couplers (Fig. 3). Low latency, energy-efficient inter-cell D2D can thus be achieved by optical loopback through ports D and E without accessing higher layers (Fig. 2). The DL optical pre-amplifier in Fig. 3 is placed at the DL output of the 1:2 optical coupler, such that the loopback optical signal will undergo a stage of optical amplification in the inter-cell D2D configuration, resulting in equivalent effective optical power budget and BER performance as the DL transmission case. Inter-cell D2D connectivity can also be implemented by linking Tx/Rx ports (n) and (o) of the electrical SDN switch (Fig. 2), albeit at the expense of higher latency due to the multi-layer O/E/O data path. For higher port counts, optical loopback can moreover be enabled by SDN control of larger-scale optical switches, which we confirmed by experiment. We also note that although the Ethernet format was selected for this experiment due to equipment availability, i.e. compatibility with the available SDN-controlled OpenFlow switch and the available latency measurement hardware, the proposed architecture does not preclude use of alternate protocols (e.g. CPRI, IP, OTN, etc.).

 

Fig. 2 Proposed SDN-controlled topology-reconfigurable optical MFH architecture for bidirectional (DL/UL) CoMP and low latency inter-cell D2D. Tx = transmitter; Rx = receiver.

Download Full Size | PDF

 

Fig. 3 Any-to-any optical switch architecture for optical multicast and loopback functions. OC = optical coupler; WSS = wavelength selective switch.

Download Full Size | PDF

3. Experimental setup and results

Figure 4 shows the experimental setup. An extended OpenFlow 1.0 controller and API running on a PC was linked via TCP/Ethernet connections to a NEC ProgrammableFlow PF5240 [9] electrical SDN switch, 4 × 4 any-to-any optical switch (Fig. 3), and two arrays of four 10G SD Tx/Rx each. To further scale data rate, 100G-class transceivers may be substituted for the 10G SD Tx/Rx used in this experiment. To obtain accurate latency measurements, a 10Gb/s Ethernet (10GE) traffic generator and tester (TGT) was connected to 10GE ports of the PF5240 SDN switch and to optical client interfaces of the SD Tx/Rx arrays. Latency for both bidirectional CoMP and inter-cell connectivity (i.e. loopback) scenarios was assessed after calibrating the latency bias from the TGT and 10G Tx/Rx. The10G SD Tx included a tunable C-band DFB laser, 11.1Gb/s NRZ OOK encoder and FEC encoder (7% overhead, FEC limit BER = 3 × 10⁻³), while each SD 10GRx comprised a photodiode (PD) and FEC decoder used to measure BER based on 6.6 × 10¹¹ bits. For DL CoMP via the electrical SDN switch shown by Fig. 4(i), FlowMod with a single match and two actions was invoked, while DL CoMP using the optical switch, Fig. 4(ii), was enabled by shifting the pass-band of 1:2 WSS port A. It is noted that the optical DL CoMP scenario investigated here corresponds to digital multicasting over the optical MFH link rather than analog photonics-aided beam-forming for the wireless air interface, as in [5]. The UL CoMP scenario of Fig. 4(iii) was implemented with two single-match, single-action FlowMod messages that merged 2 × 5Gb/s UL flows onto a common 10GE output port. For inter-cell D2D connectivity using the electrical SDN switch, Fig. 4(iv), a single FlowMod match-action linked switch input-output ports in a loopback configuration, while an extended FlowMod exploiting the VLAN ID field for λ-tuning [7] was sufficient for optical inter-cell D2D via optical loopback as per Figs. 4(v), 4(v,a), and 4(v,b). Back-to-back transmission latency for optical inter-cell D2D was also measured using a 96 × 96 SDN-controlled optical MEMS switch, producing virtually identical results. In this experiment, dynamic switch configuration times were dominated by the optical switch (~ms level), yet with novel monolithic silicon photonics designs, <2.5μs dynamic configuration has been demonstrated for large scale optical MEMS [10]. To assess the power budget and BER performance of the proposed architecture, λ₁- λ₄ DL and λ₅- λ₈ UL 10GE channels with λ allocation and power spectrum as per Fig. 4(ii) were de-multiplexed in the optical switch, followed by single-channel DL transmission over 10km SSMF (representative reach for optical MFH [4]) and a 1:512 passive split (27dB attenuation). Due to the UL Rx-side optical pre-amp which resulted in an effective 33dB power budget, only DL transmission was power limited, such that DL BER constrained performance. Figure 5(a) plots the maximum back-to-back transmission latency (i.e. end-to-end packet delay) vs. packet size for the DL/UL CoMP of Fig. 4 with 0km SSMF, showing that the proposed DL optical switching approach enables ≤4μs latency even with maximum 10GE packet size (1518 bytes), while in the DL electrical switching case, latency rises to 6.3μs. This is attributed to electrical SDN switch buffering delay that grows with packet size and also affects DL optical CoMP latency since all DL traffic traverses the electrical SDN switch (Fig. 4). As also shown in Fig. 5(a), the UL CoMP latency rises slightly higher to 6.8μs due to time-domain traffic multiplexing in the electrical SDN switch, in which congestion control time grows with packet size. Nonetheless, the worst-case UL back-to-back latency remains <7μs, leaving an ample >100μs latency budget for the completion of remaining signal processing functions subject to the <150μs aggregate back-to-back delay targets for optical MFH systems [4]. The maximum differential latency between the dual DL CoMP paths was also measured at maximum 10GE packet size for both the optical and electrical cases and was found to be 0.02μs and 2.49μs, respectively, while for UL CoMP, the maximum differential latency was slightly higher at 3.6μs. Nonetheless, both DL and UL CoMP differential latency remained well under 4.7μs, satisfying the differential latency budget given by the regular cyclic prefix length in 4G and beyond wireless transmission systems [8]. The BER for performance-limiting DL 10GE transmission is shown in Fig. 5(b), confirming the FEC limit was met at −24.6dBm received power (in this case corresponding to 10km SSMF + 1:512 passive split.) A 29.6dB power budget was thus achieved with no penalty with respect to optical back-to-back. Figure 5(c) plots back-to-back latency for optical vs. electrical loopback (0km SSMF), with 0.03μs and 3.4μs maximum latencies observed respectively, highlighting an order-of-magnitude reduction via the novel optical approach. Moreover, all-optical switching enabled constant latency vs. packet size as opposed to electrical switching. This is quite important since larger packet sizes are typically used to achieve efficiency in high speed links. The trends of Fig. 5(a) and Fig. 5(c) show that for even larger packets (e.g. 9kB jumbo frames), the advantages of largely-optical or all-optical switching would become even more pronounced. To satisfy cost, scalability, and power budget requirements, better optical switch integration with fast dynamic configuration (~ms-ns level depending on flow dynamicity), high port counts and integrated optical gain is needed. In this regard, novel silicon photonics-based solutions may be quite attractive [10].

 

Fig. 4 Experimental setup of 5G SDN-controlled topology-reconfigurable optical MFH architecture; optical spectrum at (ii): 0.1nm resolution.

Download Full Size | PDF

 

Fig. 5 (a, left) DL/UL CoMP latency; (b, center) DL BER; (c, right) Inter-cell D2D latency.

Download Full Size | PDF

4. Conclusions

To our best knowledge, we have experimentally demonstrated the first SDN-controlled topology-reconfigurable optical MFH architecture for bidirectional CoMP and inter-cell D2D connectivity in the 5G mobile era. 10Gb/s peak per-cell rates with <7μs back-to-back transmission latency and a 29.6dB power budget have been experimentally confirmed. By enabling advanced networking features, the new approach is attractive for future optical MFH.