System Level Performance of Millimeter-wave Access Link for Outdoor Coverage

Transcription

1 13 IEEE Wireless Communications and Networking Conference (WCNC): PHY System Level Performance of Millimeter-wave Access Link for Outdoor Coverage Mohamed Abouelseoud and Gregg Charlton InterDigital, King of Prussia, PA 196, USA Abstract The increased demand on data and the scarcity of available bandwidth motivate research for new technologies beyond 4G. In this paper we provide a system level study of a new cellular architecture that incorporates millimeter wave technology ( GHz) for the access link. A system level simulation is carried out for a university campus and an urban environment and the sensitivity to various design parameters has been studied. The objective of such a system is to provide up to X the wireless network capacity that is delivered today without a significant increase in energy or deployment costs. I. INTRODUCTION The exponential growth of data demand, increased use of smart-phones and tablets and the emergence of new wireless applications with rich multimedia content create a need for a new synergetic strategies capable delivering this huge demand. The use of small cells increases the spacial reuse and reduces transmit power which leads to greater throughput in the network. The main downside is the need for interference management and the cost associated with deploying many small cell base stations. Another viable solution is to explore higher frequency bands where huge amounts of spectrum are available with potentially very affordable licensing costs. The bandwidth of the GHz unlicensed spectrum alone is about 7 GHz (depending on country), which is more than all available bandwidth in the traditional cellular spectrum. The availability of this bandwidth opens the door for new applications that need higher network capacity and power savings by trading spectrum for power and using simpler waveforms. There has been a lot of work in both academia and industry investigating the use of GHz carriers, for example [1] [4] for indoor coverage and [5] for outdoor coverage. There have been other activities in the industrial standards as well, for example, IEEE 2.11ad, WirelessHD, ECMA-387, IEEE c and the Wireless Gigabit Alliance (WiGig). There are, however, many challenges associated with the GHz band. The high molecular oxygen absorption is a concern for long links. Another limiting factor is due to the dbm FCC EIRP limit. Furthermore, the GHz band is unlicensed, making it difficult to predict how well different radio access technologies (RATs) may coexist and if operators will consider the band reliable enough. Another big concern is the propagation channel itself. The millimeter wave (mmw) carriers have near optical properties with high penetration Fig. 1: The proposed system architecture losses, high reflection losses, and little diffraction, leading to line of sight (LOS) dominated coverage. It is expected that the mmw base stations (mbs) to be deployed in heavily crowded areas where the LOS between the mb and the user equipment (UE) can be easily blocked. The proposed architecture, Figure 1, includes new small mmw base stations (mbs) that are overlaid on a cellular network. The mbs are denser than the enbs and self-backhaul using a mmw MESH network to the enb (or other wired access). The phased array antennas create narrow steerable beams that provide links with low interference to both access links and backhaul transmissions. Each such node is also connected to a coverage network on a traditional carrier (e.g., a cellular network) to provide mobility management, security, and other control functions. In this paper we provide a system level study for the outdoor access link of a mmw system. We simulate the physical layer of such system and carry out sensitivity studies on multiple design parameters. Deployments in a university campus and an urban area are studied and we evaluate the effect of the directivity of the Tx and Rx antenna, the density of the mbs, modulation schemes, system threshold, inter-site distance and mb location placement. II. SIMULATOR DESIGN The simulator comprises two main modules. The first is the cell planning tool, WINPROP from AWE Communications [6], which is capable of creating outdoor scenarios in which the user is able to specify the RF properties of the various building materials and vegetation areas. Based on ray /13/$ IEEE 4146

2 No of Beams Sep. Angle ( ) No of Elements (NxN) Gain (dbi) 3dB beamwidth ( ) Crossover Point ( ) Spacing (λ) Crossover Point Gain (db) TABLE I: Base station antenna patterns specification (a) 3D view (b) mbs deployments Fig. 2: University college campus mbs deployments tracing, the tool determines the delay, signal strength, and type of interaction the path rays encounter from transmitter to receiver for each path arriving at a user-defined grid point while taking the transmit antenna pattern into account. The second module is a system simulator written in Matlab. This tool uses data from WINPROP to create channel models that are used to estimate the throughput achieved by UEs. The UEs locations are randomly drawn from the set of pre-generated grid points. Each dropped UE is randomly oriented in azimuth and elevation.the simulator models the impacts of the transmit power, inter-cell interference, receive antenna gain, modulation scheme, scheduler parameters, etc. The statistics are then collected to generate an estimate of the overall network performance. A. Channel Model The aforementioned WINPROP tool is configured to run in a deterministic fashion in which individual rays (LOS, reflected, and diffracted) are traced between each transmitter and the entire set of user-defined grid points. The signal strength, delay and angle of arrival for each ray are function of the reflection and diffraction angles, building materials, propagation model, transmit antenna pattern etc. The received power/delay combination allows for the creation of a tapped delay profile. A fading profile is applied to each of the identified rays to capture the fast fading environment. The fading profile follows a Rician distribution rather than a Rayleigh distribution with k- factor of because of the LOS characteristics of the channel. B. Tx and Rx Antenna Design and Beam Assignments To generate the mb s desired antenna patterns we use a uniform rectangular planar phase array (URA) while varying the number of elements. The UE is associated to one of the possible beams that the array could generate. Each URA is assumed to cover in azimuth. The design parameters for the tested patterns are given in Table I. The gain of the Rx antenna depends on the received ray AoA. An ideal omnidirectional antenna and a directional antenna have been tested. The omni-directional antenna has a uniform dbi gain in all directions while the directional antenna is a 4x4 URA with element spacing equal to half the wavelength. The antenna array generates 5 beams in the elevation angle plane with approximately beamwidth with overlap at the 3 db point. This is equivalent to approximately beam separation. A set of down-tilted beams and a set of up-tilted beams are used to form a 5x3 beam array. Also, back-to-back planar phased array antennas are used to cover the front and the back of the UE to provide coverage all around the UE. UE-mB association is a two stage process. In the first stage, the received power from all mbs antenna beams is measured at the UE location with a quasi-omni antenna pattern in the case where the UE antenna is directional, or an ideal-omni antenna if no directional antenna is used. The quasi-omni antenna pattern is just the pattern of a single element patch antenna which effectively constrains the possible pointing direction of the array. The mb array and corresponding beam providing the largest Rx power is selected. In the second stage (only for directional receive antennas), all possible Rx beams at the UE are varied to determine the best receive antenna pattern. C. Self Blockage It is assumed that the signal coming from the back side of the UE is attenuated by db. For an average blocker 175 cm tall and cm wide holding the UE cm away, the blocking angle is a azimuth angle centered at the back of the UE and elevation angles between 62.8 and. D. SINR and Throughput Calculation After calculating the instantaneous SINR for the scheduled UEs, the SINR is mapped to its corresponding throughput. Shannon capacity formula provides an upper bound on performance where R = Blog 2 (1+SINR), B is the bandwidth and R is the achieved rate. A second method is to map SINR to throughput via a lookup table. We use tables based on LTE (QPSK, 16QAM and 64-QAM) [7] and also based on adaptive modulation and coding schemes from 2.11ad [8]. E. Default Simulation Parameters We consider UE drops, each comprising samples (TTIs). The total number of UEs per drop is which corresponds to a UE density of 7 UE/km 2. A TTI duration is assumed to be 1 msec. The carrier frequency is GHz. Following 2.11ad, a total of 26 MHz bandwidth is used and divided into 512 sub-carriers of which 336 are used for data. The UE noise figure is 6 db. Unless otherwise noted, Round Robin scheduling is used to distribute the resources 4147

3 equally among the users and ideal omni-directional Rx Antennas are used. The mb antenna s down-tilt angle is and mbs have an EIRP of dbm. The mb s height is 4 m and the UE s height is 1.5 m. III. UNIVERSITY CAMPUS PERFORMANCE EVALUATION The studied campus shown in Figure 2 is an abstracted version of a large, public university that contains trees and buildings of various heights. Two proposed mb placement schemes are considered. The first relies on 4 mbs placed at the corners of the college campus with the intent that they provide coverage to the main quadrangle. The other has an additional 5th mb placed at the center of the quadrangle itself. Access link system performance is measured by the metrics of user throughput and total network throughput. A. Multiple mb Configurations Two mb deployment schemes are tested. The number of azimuth beams per array is varied from 5 to 9. The default tilt angle is set to. For the 9 beams/array case, two tilt angles were applied to the antenna arrays, and 15. The EIRP is held constant for each mb antenna type so the improvement seen as one moves to narrower beams is not due to increased transmit power. Figures 3 and 4 show the user and the total cell throughput, respectively. The additional mb at the center of campus improves performance where the added interference does not overcome the SNR improvement and the frequency reuse gain. As expected, generally, narrower beams (7 vs. 5 beams/array) provide better performance since there is less likelihood of inter-mb interference, however when the beams are very narrow (as for the 9 beams case), the results are sensitive to the tilt angle. This is seen where the performance for the 7 beams/array scenario outperforms the 9 beams/array case. For the 9 beams/array case, the tilt angle is superior to the 15 tilt angle but there are cases in which having a down-tilt is helpful such as when the UE is very close to the mb, shown in figures 3 and 4 as (T). This is seen by noting the significant improvement when the best of the two tilt angles is chosen for each UE, the 18 beams/array in figures 3 and 4. B. Interference Interference is a matter of concern in all wireless systems. To determine the effect of interference, we calculate a hypothetical value which is the SNR throughput. The SNR throughput is the achieved throughput assuming zero interference. Comparing this value to the achieved SINR throughput, one can determine the effect of the interference on the system under study. Figure 5 shows the SNR and SINR UE throughput CDF. The mbs in this test are assumed to have 5 beams/array. Two schedulers have been tested, the Round Robin and the proportional fair schedule (PF) [9] schedulers. A PF is considered to balance between delay and fairness constraints and maximizing the network throughput. The results show that 5 beams/array 5mB 7 beams/array 4mB 9 beams/array 4mB 5 beams/array 5mBs 7 beams/array 5mBs 9 beams/array 5mBs 9 beams/array 5mBs (T) 18 beams/array 5mBs Fig. 3: User throughput for various mbs deployments 5 beams/array 5mB 7 beams/array 4mB 9 beams/array 4mB 5 beams/array 5mBs 7 beams/array 5mBs 9 beams/array 5mBs 9 beams/array 5mBs (T) 18 beams/array 5mBs Total Cell Throughput (Gbps) Fig. 4: Total cell throughput for various mbs deployments even for the Round Robin scheduler, the SNR throughput and the SINR throughput are very close to each other. This is because of the mmw characteristics (high path loss, material absorption, limited diffraction and the Oxygen absorption) in addition to the use of the transmitter narrow beams. This result shows that a mmw system is not an interference limited system and pushing more power in the system will increase system capacity. C. Modulation We consider various upper limits on modulation schemes when mapping SINR to throughput. QPSK, 16-QAM and 64-QAM tables based on LTE from [7] are used to map the calculated SINR to the achieved throughput. We compare these results with the Shannon capacity SINR to throughput mapping. Modulation and coding schemes based on IEEE 2.11ad [8] are tested as well, one is based on single carrier (MCS indices 1-12) and the other is based on OFDM (MCS indices 13-24). Figure 6 shows the total cell throughput CDF. Because of the availability of bandwidth, a simple modulation scheme can be used where this would be enough to provide a high quality 4148

4 Ideal omni dir planar array 5x1 dir planar array 5x3 2 dir planar array 5x1 2 dir planar array 5x3 SINR throughput RR SNR throughput RR SINR throughput PF, B=1 SNR throughput PF, B=1 Fig. 5: SINR throughput vs SNR throughput Fig. 7: User throughput under various Rx antennas Ideal omni dir planar array 5x1 dir planar array 5x3 2 dir planar array 5x1 2 dir planar array 5x3 SC 2.11ad MCS OFDM 2.11ad MCS Shannon Capacity QPSK 16 QAM 64 QAM Total Cell Throughput (Gbps) Total Cell Throughput (Gbps) Fig. 6: Total cell throughput for various modulation schemes Fig. 8: Total cell throughput under various Rx antennas of service and at the same time reduces receiver complexity. This can be shown from the results in the figure where even when limiting the modulation scheme to QPSK, the mmw system provide performance that exceed any known deployed technology. The average total network throughput for QPSK is 29.8 Gbps. The area of the college campus is.76 km 2 and with that average network throughput, a throughput of 392 Gbps/km 2 can be achieved. D. Receive Antenna Pattern Both one- and two-sided uniform rectanguar and uniform linear beam directional antennas are used at the receiver. Based on the total received power, the receiver selects the beam pattern that maximizes the total received power. We compare 5 antennas at the UEs: 1) a db uniform gain ideal omnidirectional antenna 2) a planar phased array that can generate 5x1 beams (azimuth steering only), 3) a planar array antenna that can generate 5x3 beams (azimuth and elevation steering), 4) two back-to-back planar phased array antennas each can generate 5x1 beams (azimuth steering only), and 5) two backto-back planar phased array antennas each can generate 5x3 beams (azimuth and elevation steering). Figures 7 and 8 show the user throughput and the total cell throughput, respectively. A general observation is that the use of directional antennas can benefit the system compared to an ideal omni antenna. An omni directional antenna outperforms the 5x1 beam antenna since the range that this directional antenna covers is limited compared to the omni-directional antenna. Adding the two back-to-back antennas was beneficial where this expands the azimuth angle over which the UE can receive signal to almost 3. The down-tilted and the uptilted beams are very important because of the directivity of the receive beam patterns and allows the possibility of the UE to be oriented in any direction. This effect should be clearer with higher modulation schemes where the extra gain provided by the antenna in case of QPSK might not be used due to its limited spectral efficiency. E. mmw System Threshold A threshold for the average UE total received power is used to determine whether the UE should be served with mmw service or not. This guarantees a high throughput and a good use of of the mmw resources. The UEs that are not served with mmw are still covered by the cellular underlay. 4149

5 mmw RX thr = Inf mmw RX thr = mmw RX thr = 85 mmw RX thr = mmw RX thr = 75 mmw RX thr = (a) 7 mbs/16 arrays Fig. 11: Urban deployment (b) 9 mbs/25 arrays Fig. 9: mmw UE throughput CDF Percentage of UEs served with mmw Inf mmw RX power service threshold Fig. : mmw penetration percentage Fig. 12: User throughput, Urban Scenario We test various received power thresholds by calculating the percentage of mmw penetration and the performance of the total system in terms of throughput. Figure 9 shows the mmw UE throughput CDF and Figure shows the percentage of UEs that are served with mmw with respect to all UEs in the system. Note that there is a trade-off between the mmw penetration percentage and throughput. Using a high threshold guarantees a high service quality but at the same time will limit the number of UEs that are served. In the system under study, - dbm is considered a good threshold that guarantees % mmw penetration and around Mbps tenth percentile UE throughput. IV. URBAN DEPLOYMENT Identical metrics have been generated for a small urban morphology which is loosely based on a portion of London, England in which a series of uniform buildings of similar dimensions are laid out in a uniform pattern. This deployment includes a high preponderance of urban canyons. Figure 11a shows a 2-D view of this scenario for 7 mbs and 16 arrays while Figure 11b shows the identical scenario with 9 mbs and 25 arrays. In both cases, adequate SNR coverage can be provided down the urban canyons but the denser deployment provides increased network throughput due to higher signal strength overcoming the effects of additional interference. Figures 12 and 13 show the user and total throughput CDF. The main result is that adding more mbs and antenna arrays provides significant performance improvement. The median total cell throughput was improved by over % by increasing the number of mbs from 7 to 9. A. Inter-site Distance A single street (urban canyon) was created with mbs positioned at either end m apart. mbs use x URA antenna arrays. Additional mbs are then equally spaced between the end points to determine the performance gains for a higher base station density and to determine if there becomes a point of diminishing returns in which further reducing inter-site distance (ISD) provides little benefit. The median total throughput was plotted for a varying number of mbs, Figure 14a. Continuing to add mbs improves performance as the received signal power increases with the reduced mb-ue distance. In Figure 14b the median UE throughput is plotted for the cases in which the throughput is mapped to the SINR as well as to the SNR. The latter is a purely hypothetical exercise designed to estimate the impact of interference on user throughput. Without interference, the throughput improves linearly with the increasing mbs (decreasing ISD) and when interference is considered, the throughput improvement begins to level off. Taken together, 41

6 were chosen with some coordination (e.g. 22% improvement (a) A random plaza deployment (b) mb drops total throughput Fig. 15: mb placement sensitivity Fig. 13: Total cell throughput, Urban Scenario between the min and max cases for 16 mbs). V. CONCLUSION Throughput (Gbps) Number of mbs (a) Median network throughput Throughput (Mbps) SINR SNR Fig. 14: Single street Number of mbs (b) Median UE throughput the results show that an ISD of 1 m yields good performance but further reducing ISD provides diminishing returns. B. Effects of mb Placement An outdoor shopping plaza was created to accommodate 32 potential storefronts, Figure 15a. Obviously not every store will supply enhanced WiFi service but some will as shown by the red dots. As a result, the number of storefronts providing mmw service was treated as a simulation variable where the locations supplying mmw service are then randomly selected. For several combinations of active mbs, numerous random mb âăijdropsâăi are done to determine system wide performance sensitivity to the positioning of the mbs. Figure 15b shows the total throughput CDFs for 4 (blue), 8 (green), and 16 (red) active mbs. It is clear that network performance improves when more mbs are active; however, the absolute spread between the best and worst drops increases with the number of active mbs. The median throughput improves significantly with the number of active mbs while the relative spread âăş measured as the ratio of the median throughputâăźs standard deviation to its mean âăş is largest with only 4 active mbs. This means that there is greater sensitivity to mb placement when there are few mbs. In general, performance improves with the increasing number of mbs so it is reasonable that uncoordinated mmw deployments are viable though performance would be better if the active mbs A system level simulation for a network of a small mmw base stations that are overlaid on a cellular network is carried out for a university campus and an urban area. Results show that a goal of around Gbps/km 2 total network throughput can be easily achieved even with simple modulation and coding schemes. The characteristic of the GHz carrier frequency and the use of transmit antennas with narrower beamwidths result in less inter-mb interference and, therefore, better overall performance. It is shown that placing mbs 1 m apart is adequate as long as there are clear lines of sight between possible UE locations and nearby mbs. Double sided directional receive antennas are necessary to at least provide the same performance as that of an ideal omni. Uncoordinated placement of mbs can be supported since adding mbs always improves overall performance and the increase in intercell interference is more than offset by the reduced UE-mB distance. REFERENCES [1] W. Jing, R. Prasad, and I. Niemegeers, Analyzing GHz radio links for indoor communications, IEEE Transactions on Consumer Electronics, vol. 55, no. 4, pp , Nov. 9. [2] J. Wang, R. Prasad, P. Pawelczak, and I. Niemegeers, A link stability model for indoor GHz radio wireless networks, in Vehicular Technology Conference Fall (VTC 9-Fall), 9 IEEE th, sept. 9, pp [3] H. Lee, J. Bok, B. G. Jo, G. H. Baek, and H. G. Ryu, Indoor WPAN communication system using 2-dimensional array antenna in GHz frequency band, in Computing, Communications and Applications Conference (ComComAp), 12, jan. 12, pp [4] X. Zhang, L. Lu, R. Funada, C. S. Sum, and H. Harada, Physical layer design and performance analysis on multi-gbps millimeter-wave wlan system, in Communication Systems (ICCS), IEEE International Conference on, nov., pp [5] Z. Pi and F. Khan, System design and network architecture for a millimeter-wave mobile broadband (mmb) system, in Sarnoff Symposium, 11 34th IEEE, may 11, pp [6] AWE Communications, WinProp software suite, [7] IEEE 2.16m evaluation methodology document (EMD), IEEE 2.16 Broadband Wireless Access Working Group, Jan 9. [8] IEEEdraft standard for local and metropolitan area networks-specific requirements-part 11: Wireless LAN medium access control (MAC) and physical layer (PHY) specifications. amendment 3: Enhancements for very high throughput in the GHz band, IEEE2.11ad, Mar 12. [9] D. Tse and P. Viswanath, Fundamentals of Wireless Communication. Cambridge University Press,