Green Optical/Wireless In-Building Networks: the Physics and Principles of Design Leonid Kazovsky, Apurva S. Gowda, Ahmad R. Dhaini, Hejie Yang* and Solomon T. Abraha* Photonics and Networking Research Laboratory, Stanford University, Stanford, CA, USA Tel: +1 (650) 71-1060, Fax: +1 (650) 73-8473, e-mail: l.kazovsky@stanford.edu * Corning Incorporated, Corning, NY, USA ABSTRACT Energy efficiency is rapidly becoming an important requirement for modern networks. In this paper, we discuss the physical limits to the power consumption of wireless and optical transmission and the reason behind the existence of an optimum cell size in optical/wireless in-building networks. We show that the optimum cell size is smaller for larger bandwidths. We then discuss principles that govern the design of in-building Radioover-Fiber (RoF) distribution networks. We use theoretical models to analyze the impact of key design factors on the energy consumption of point-to-point RoF links and how their adverse effects can be mitigated. Finally, we compare the power consumption of several key in-building optical/wireless architectures based on RoF technologies, and demonstrate that centralized architectures based on RoF links can be substantially more energy efficient than baseband-over-fiber (BoF) architectures in bandwidth-limited scenarios when designed properly. Our findings also show that RoF-based architectures are energy efficient for cell sizes less than 10m. Keywords: Energy-Efficiency, Radio-over-Fiber, In-Building Networks. 1. INTRODUCTION The need to reduce the carbon footprint of telecommunication networks is widely recognized [1,]. Recent studies reveal that the power consumption of data centers and that of in-building networks (residential and office) are comparable due to the large number of buildings and the spreading Internet literacy of their users [1]. According to [3], by year 017, 55% of all IP traffic will originate from wireless and mobile devices. It has been forecasted that by year 017, two-thirds of all IP traffic will be consumer traffic from households, universities and Internet cafes. This means that the main source of extensive wireless Internet traffic will be from in-building IT networks. Thus, it is important to explore ways for reducing the energy consumption of office and residential building networks. Due to the large number of in-building networks, even a small decrease in the energy consumption in each of the individual networks should lead to a significant improvement in the overall energy consumption. Wireless communications enables mobile access to the Internet; however, it has limited bandwidth and suffers from inherent energy loss due to its broadcast nature rendering it fundamentally energy inefficient compared to directed transmission links such as optical links. In contrast, optical links offer very high-speed access to the Internet with low energy loss, but cannot offer user mobility. Consequently, optical/wireless convergence is seen as an ideal candidate for increasing the network bandwidth, decreasing its energy consumption and offering enduser mobility. Radio-over-Fiber (RoF) is a promising technology for the implementation of seamless and energyefficient optical/wireless networks []. Through this paper, we try to illustrate the physical limits of both optical and wireless transmission links and their impact on the power consumption of the network. We discuss the physics behind the existence of an optimum cell size for an optical/wireless network. We then discuss point-to-point (pp) links of Analog RoF (ARoF) and Digitized RoF (DRoF) technologies, and determine how certain key parameters affect their energy consumption using theoretical models. We also suggest certain design principles to reduce the energy consumption of these pp links. Finally, we compare the power consumption of a Baseband-over-Fiber (BoF), ARoF and DRoF based optical/wireless network and find their optimum cell sizes. The remainder of the paper is organized as follows. Section discusses the physical limits to the power consumption of a wireless and optical link and the existence of the optimum cell size. In Section 3, we discuss the effect of certain key link design parameters on the energy efficiency, as well as the design conditions for energy-efficient pp links and present the optimum cell size for several different in-building optical/wireless networks. Finally, in Section 4, we present the conclusions.. THE PHYSICS In this section the fundamental limits of the transmit energy for optical and wireless links is discussed. We start with the minimum required transmit power to receive a bit of information over the optical link. We then compare it with the minimum required transmit power to receive a bit of information over the wireless link. We build on this knowledge to demonstrate the existence of the optimum cell size.
.1 Baseband-over-Fiber (BoF) Consider on-off keying transmission over an optical link with a p-i-n photodiode receiver. Assuming bits 0 and 1 are equi-probable, the minimum optical transmit power to achieve a bit error rate of P b is for an ideal quantumlimited receiver and is given by, ln(pb ) hcb P =, (1) Tx, optical λ where h is the Planck s constant, c is the speed of light, and λ is the wavelength of light. It can be shown that the required value of transmit power to maintain a required bit error rate (BER) is given by [4] αl Q 4N0B PTx, optical = e Fn, () R RF where α is the loss per unit transmission distance, L is the transmission distance, Q is the required quality factor to maintain a certain BER, R is the reponsivity of the photodiode, N 0 is the thermal noise per Hz are room temperature, B is the bandwidth, F n is noise factor and R F is the load resistance in a high-impedance receiver and feedback resistance in the transimpedance receiver. Since the quantum limit for both Amplitude Shift Keying(ASK) and Phase Shift Keying (PSK) in optical transmission is equal, we consider only ASK. 1/ Figure 1 Required Optical Transmission Power versus Figure Required Wireless Transmission Power versus Fiber Length Transmission distance Figure 1 shows the required optical transmission power to maintain a BER of 1e-3 at the receiver. As per intuition, for a higher bandwidth and longer distances, the required optical transmit power is higher.. Radio Frequency Transmission Consider a radio link that employs an isotropic antennas at the transmitter and antenna. In the absence of fading, the required wireless transmit power for a certain BER is given by 4 ) 0 F πl P Tx, wireless = ( Q BN n, for Amplitude Shift Keying (ASK) (3) λ Q BN F 0 n 4πL P Tx, wireless =, for Phase Shift Keying (PSK) (4) λ Figure shows the required wireless transmit power to maintain a BER of 1e-3. Similar to optical transmission, the required wireless transmit power increases with transmission distance and bandwidth. Also PSK shows a 3dB reduction in required wireless transmit power. Now, if we compare Fig. with Fig. 3, the required transmit power for wireless transmission is higher and increases at a much higher rate than for optical transmission. The reason is the fundamental difference in the transmission of energy from the transmitter to the receiver. The optical fiber is a waveguide and the energy is directed to the receiver. However, in wireless transmission with isotropic antennas the energy is broadcasted in all directions. Thus the loss is higher in wireless transmission. On the other hand, the broadcast nature allows the receiver to be mobile which is not possible in the case of optical transmission. Thus the idea of fiber-wireless inbuilding networks, which runs optical links to the remote units and utilizes wireless links from the remote unit to the end user, is appealing.
.3 Fiber-Wireless In-Building Networks: Transmission Power Figure 3 Wireless coverage of 100mx100m floor (r=10m) Figure 4 Total Network Transmission Power versus Number of RAUs As an example, consider a 100m x 100m floor of a building that requires complete wireless coverage as shown in Fig. 6. We assume a remote antenna unit (RAU) with its own analog and digital circuitry is connected to the head end unit (HEU) via an optical link. In this analysis, for simplicity we assume that the every cell requires a 00m optical fiber link. Since the optical transmission power remains constant, the transmission power is determined by power consumption of the power amplifier. The power consumption of the power amplifier is estimated using the drain efficiency η, and the back-off, ρ using the equation, ρ = P, (5) η P DC, Tran Tx, wireless Figure 4 plots the total transmission power required to provide complete coverage of the floor for different number of RAUs. From Fig. 4, we see that the overall transmission power increases as the number of RAUs decreases (i.e. the cell size increases). This would suggest that the most energy efficient solution is to have smallest cells possible..4 Fiber-Wireless In-Building Networks: Processing Power The foregoing discussion is incomplete since the effect processing power consumption due to the optical link transmission and the circuitry of the RAUs has not been addressed. Using a power meter, the power consumption of a Cisco x Small enterprise unit in the 80.11g transmit+receive mode was found to be ~600mW [5]. To estimate the power consumption of the optical link we assume a VCSEL [6] and a photodiode of responsivity, 1 W/A and V DD =.8V, and estimate their power consumption using the following equations, ( PTx, optical + 0.18) P DC, Laser = VDD *, (5) 0.3 PDC, Photodiode = VDD PRx, optical R. (6) (a) Quantum-limited receiver (b) p-i-n receiver Figure 5 Total Network Power (Transmission +Processing) versus Number of RAUs Figure 5a and 5b shows the total network power consumption to provide complete coverage of the floor for different number of RAUs using a Quantum-limited receiver and a p-i-n receiver respectfully. The difference
between the curves in Fig. 5a and 5b is negligible implying that using a quantum-limited receiver has no significant benefits. We see that there is a clear optimum cell size that has the minimum total network power consumption. The processing power consumption increases linearly with the number of RAUs which can be observed in Fig. 5a and 5b on the right of the optimum point where the increase in the power consumption is linear (processing power dominated region). On the other hand, as seen in Sub-Section.3, the transmission power consumption decreases as the number of RAUs increases. At the optimum cell size both these values are at their minimum. Since the transmission energy for the different bandwidths are different the corresponding optimum cell sizes differ (~7m for 100Mbps, ~18m for 1Gbps, ~8m for 10Gbps). Since the transmission energy increases with bandwidth, we see that the optimum cell size is smaller for larger bandwidths. Assuming frequency reuse, as we reduce the cell size, the total network bandwidth increases at the same rate as the number of cells required to cover the floor. Thus, the energy per transmitted bit will remain constant in the static energy dominated region (r<5m). By reducing the cell size, we can increase the bandwidth without an adverse effect on energy consumption. However, the capital expenditure (CAPEX) of the network will be higher. By operating at the optimum cell size, we can have the lowest possible energy per bit with a reasonable trade-off between capital expenditure and bandwidth. 3. DESIGN PRINCIPLES: RADIO-OVER-FIBER IN-BUILDING NETWORKS Figure 6 Radio-over-Fiber In-Building Network Figure 6 shows a typical in-building RoF network. The RF transmitter (TRx) and the head end unit (HEU) are located at a central location. The RF TRx modulates/demodulates the RF signal and performs the necessary signal conditioning. The HEU contains the electrical-optical-electrical (EOE) conversion circuitries required to transmit/receive the RF signal on the optical link, and the remote antenna unit (RAU) contains the EOE circuitries and the analog signal conditioning circuitries required to broadcast the signal. The figure also shows typical Analog RoF (ARoF) and Digitized RoF (DRoF) links and some of the important parameters that affect the energy consumption of these links. The radio signal generated at the HEU is transmitted over the optical link to the RAU to be broadcast on the downlink and vice versa on the uplink. ARoF and DRoF links differ in how the RF signal is transmitted over the optical link. In ARoF, the RF signal is modulated onto the optical carrier. In DRoF, the RF signal is digitized and transmitted as bits over the optical link. The loss incurred due to the optical transmission determines the required gain of the power amplifier (PA) or the maximum wireless transmission distance to maintain a required performance. In a link employing direct modulation and direct detection, the slope efficiency of the laser, s l, the fiber loss per km, α, and the responsivity of the photodetector, R, determine the electrical to optical to electrical conversion loss [7]. In the case of DRoF, the sampling frequency, f s, and the peak-to-peak voltage of the digital-to-analog converter (DAC), V pp, determine the loss due to digitization. Centralized architectures have simpler RAUs with lower circuit energy consumption (i.e., the penalty of using smaller cells and increasing the number of RAUs will be lower [8]). However, RoF links are energy-inefficient compared to BoF links, which limits the benefits of RoF-based centralized architectures. In this work, we show that RoF-based networks can be more energy efficient if designed properly. In the following two sections we discuss the major factors that affect the energy consumption of point-to-point links and extrapolate the model used to estimate the optimum cell size for a centralized architecture. 3.1 Radio-over-Fiber links In this section we discuss the effect of a few important parameters on energy consumption of the RoF point-topoint link. The parameters considered are a) Fiber length b) Slope efficiency of the laser (ARoF only) c) Nyquist Zone (DRoF only). These parameters were chosen based on their relevance in the optical link design and their
effect on the loss. Although the results reported in this work do not present an exhaustive comparison of energy consumption between ARoF and DRoF with respect to all possible parameters, they do provide us with insights into the effect of loss on energy efficiency and several factors that can be leveraged to design energy efficient systems. Table 1. Parameters and values used in analysis Parameter Symbol Value Unit Thermal Noise at room temperature N 0-04 dbw/hz Load Resistance R l 50 Ω Receiver Noise figure F n 3.3 db Signal Bandwidth B 0 MHz Drain Efficiency η 0 % Amplifier back-off ρ 10 db Carrier Frequency f c.4 GHz Fiber loss per km α 0. db/km Photodetector Responsivity R 0.65 A/W Slope Efficiency of the Laser s l 0.3 W/A Sampling Frequency f s 300 MHz 3.1.1 Fiber Length Figure 7 Energy per bit versus fiber length PAPR f V c IF pp 10 10 sin f P s RF, DRoF = R l P e αl s RF ARoF l R, P in Figure 8 Energy per bit versus laser slope efficiency, (7) =. (8) Figure 7 shows the energy per bit of a RoF point-to-point link versus fiber length. As can be observed, unlike ARoF, energy per bit for DRoF does not change significantly as the fiber length increases. For DRoF, the RF power at the RAU is determined by f s and V pp as given by Eq. 7, where f IF is the intermediate frequency of the reconstructed radio signal and PAPR is the peak to average power ratio of the OFDM signal. This RF power is not affected by any optical link loss. However, the RF power loss in ARoF increases as fiber length increases (as demonstrated through Eq.8 [7], where L is the length of the fiber and P in is the input radio signal power). For a fiber length of km and α=0.db/km, the loss is about 14dB. This implies that for optical transmission with high loss or long-reach fiber scenarios (e.g., backhaul networks), it may be more energy efficient to use DRoF. From Figure 7, we observe that DRoF is more energy efficient for the parameter values are as shown in Table 1. However, the exact cross-over points depend on several parameters such as the noise, amplifier gain, bit resolution, etc. 3.1. Laser Slope Efficiency Equation 8 highlights the importance of the slope efficiency of the laser. In Figure 8, we plot the energy per bit of an ARoF point-to-point link versus the slope efficiency of the laser. We observe that by improving s l from 0.3 W/A to 0.6W/A, the energy per bit is reduces by ~76%. There may be an increase in power consumption of the laser to improve slope efficiency, which may reduce the energy savings. However, if the power consumption of the link is dominated by the power consumption of the amplifier, using a laser with a higher slope efficiency will still lead to significant savings.
3.1.3 Nyquist Zone Figure 9 Reconstructed Spectrum of 0MHz wireless Figure 10 Energy per bit versus Nyquist Zone for DRoF signal after sampling (f s =300MHz) In Figure 9, the spectrum of reconstructed digitized RF signal is plotted along with the frequency response of a zero-order hold circuit. There is a large loss in the RF power due to the sinc behaviour of the DAC. The amount of loss depends on the Nyquist zone in which the carrier frequency exists. Figure 10 plots the energy per bit for reconstructing the RF signal from different Nyquist zones. The Nyquist zone can be reduced by increasing the sampling frequency or via up-conversion using a local oscillator at the RAU instead of using the DAC. In this study, we consider the latter. Lower Nyquist zones have lower energy per bit due to lower loss. 3. Radio-Over-Fiber In-Building Networks (a) (b) (c) Figure 11 Green Architectures (a) ARCH1 (b) ARCH (c) ARCH3 (a) (b) Figure 1 Power consumption of the architectures versus cell size (a) s l =0.3W/A and upconversion using DAC (b) s l =0.6W/A and upconversion using a local oscillator at the RAU Figure 11 shows the five architectures that we consider and analyze. We describe them as follows: 1. ARCH1, distributed processing using BoF: This is the legacy architecture (Fig. 8a). It employs wireless access points, each with its own processing circuitry, connected to the Main Distribution Frame (MDF) via 1Gbps optical Ethernet links.. ARCH, centralized and dedicated analog processing and centralized and shared digital processing using ARoF: In this architecture, all RAUs share a DSP unit but have dedicated analog transceiver chains at the MDF (Fig. 8b). ARCH employs ARoF links between the RAU and the MDF.
3. ARCH3, centralized and dedicated analog processing and centralized and shared digital processing DRoF: As illustrated in Fig. 8c, this architecture is similar to ARCH, but it employs DRoF links between the RAU and the MDF. Figure 1a compares the power consumption of the architectures to cover a 100mx100m floor using typical design parameters (e.g., s l =0.3W/A and upconversion using DAC). Fig. 1b considers a scenario where the lasers slope efficiency is higher (s l =0.6W/A) for ARCH and upconversion is done using a local oscillator at the RAU in ARCH3. The mitigation techniques suggested result in significant energy savings, and change the optimum cell size. For ARCH3, the optimum cell size is ~10m and for ARCH it is ~5m. If the cell size is required to be around 10m to achieve the required network capacity, then, for the parameters considered, ARCH3 is the most energy efficient solution. Thus if designed properly, RoF-based architectures can be the most energy efficient, in bandwidth-limited scenarios. These results differ from the results in [] since the model we apply takes into account the impact of the DAC and different link designs. 4. CONCLUSIONS In this paper, we show that the physical limit to energy consumption for wireless communications is much higher than that of optical communications due to its broadcast nature. We demonstrate the existence of an optimum cell size due to the trade-off between transmission power and processing power. The adverse impact of loss on the energy consumption of point-to-point Radio-over-Fiber (RoF) links due to key design parameters was discussed and we show that these effects can be mitigated by simple design principles that reduce the loss due to EOE conversion. The optimum cell size for Radio-over-Fiber based architectures was evaluated to be ~10m. In bandwidth-limited scenarios and for cell sizes up to 10m, RoF-based architectures can be more energy efficient if the mitigation techniques are applied. ACKNOWLEDGEMENTS The authors would like to thank Corning Incorporated, USA for their support and guidance of this work. Ahmad R. Dhaini is partially supported by the Natural Sciences and Engineering Research Council of Canada (NSERC) and by the National Science Foundation (NSF) via Grants CNS-067085 and CNS-1111374. REFERENCES [1] L. Kazovsky, T. Ayhan, A. Gowda, K. Albeyoglu, H. Yang, and A. Ng'oma, "How to Design an Energyefficient Fiber-Wireless Network," in OFC/NFOEC 013, OSA Technical Digest (online) (Optical Society of America, 013), paper OM3D.5. [] Kazovsky, Leonid G.; Ayhan, Tolga; Gowda, Apurva S.; Dhaini, Ahmad R.; Ng'oma, Anthony; Vetter, Peter, "Green in-building networks: The future convergence of green, optical and wireless technologies," ICTON 013, vol., no., pp.1,5, 3-7 June 013. [3] 'Cisco Visual Networking Index: Forecast and Methodology, 01 017'. [Online] [4] Agrawal, Govind P., Fiber-Optic Communication Systems, John Wiley & Sons, Inc. 010. [5] Kadir M. Albeyoglu, Personal Communications [6] Carlsson, C.; Martinsson, H.; Schatz, R.; Halonen, J.; Larsson, A., "Analog modulation properties of oxide confined VCSELs at microwave frequencies," Lightwave Technology, Journal of, vol.0, no.9, pp.1740,1749, Sep 00. [7] Cox, Charles Howard, ''Analog Optical Links: Theory and Practice'', Cambridge, UK: Cambridge University Press, 004. [8] Yang, Y, Investigation on digitized RF transport over fiber, PhD thesis, Engineering, Department of Electrical and Electronic Engineering, The University of Melbourne, 011.