Maximizing Throughput with Ultra-Compact Diversity Antennas

Transcription

1 IEEE VTC 2003 FALL 1 Maximizing Throughput with Ultra-Compact Diversity Antennas David T. Auckland, William Klimczak, and Gregory D. Durgin *Georgia Institute of Technology Etenna Corporation 511 Van Leer Building 6100-C Frost Place Atlanta, GA Laurel, MD Abstract Portable wireless devices currently use antenna diversity designs that waste much of the physical and electromagnetic space allotted to their antenna elements. With case studies of an integrated IEEE b laptop radio, we demonstrate how unconventional, ultra-compact antenna diversity maintains data throughput in wireless devices that are notoriously ill-suited for space diversity. We overcome much of the measurement and computational difficulties of evaluating diversity designs by using a 3-step evaluation procedure: 1) in-situ antenna range measurement (measuring the antennas mounted onto the device housing), 2) multipath modeling, and 3) throughput mapping. I. INTRODUCTION Though antenna diversity in portable wireless devices leads to performance gains, design of the antenna element, layout, and mounting often becomes intractable. There currently exists no clear, quantifiable method for trading-off the four most important design factors in antenna diversity: cost, form factor, mounting requirements, and link performance. This paper clarifies these trade-offs by presenting a systematic approach for designing and evaluating ultra-compact diversity antennas in portable wireless devices. In the process, we demonstrate how to exceed the current commercial limits of antenna diversity technology in terms of size, cost, and performance. By presenting case-study results using ultra-compact IEEE b diversity antennas in laptops, we show that throughput can be maintained despite the extraordinarily small size and compactness of the diversity antennas. The practice of using multiple, space-stingy diversity antennas on a wireless device is an important source of gain for the ever-increasing bit rates and carrier frequencies of the family of air interfaces. A critical difficulty in choosing and evaluating antenna diversity designs is the rampant coupling between each antenna element and the device casing. Conventional rules for designing space diversity antennas prohibit placing the elements closer than 0.5λ to avoid both coupling and envelope correlation effects [1]. However, Ebine, et. al. showed that when dipoles are brought close together, their coupling actually suppresses the detrimental envelope correlation between elements [2], [3]. Furthermore, the signal losses due to parasitic element coupling are, in many cases, minimal particularly when evaluated in terms of throughput. We illustrate these design principles using a 3-step analysis procedure: 1) insitu range measurement of antenna elements, 2) realistic multipath modeling, and 3) extrapolating physical layer performance to observable throughput. II. ANTENNAS This section discusses the antenna element type, layout, and measurement used in the diversity analysis. A. Antenna Elements There are two types of antenna elements used in this study, both radiating in the 2400 MHz ISM band. The baseline antennas are large planar inverted-f (PIFA) type antennas, each occupying a volume of 900 mm 3.We should also note that the two PIFAs industry standards in many laptops were different designs (for mounting purposes), thereby requiring two different part numbers. The second type of antenna is a tabletop-shaped element (the Etenna EA2400 AccuWave TM ). The tabletop antenna is significantly smaller than the PIFA element, occupying a volume of only 350 mm 3. Figure 1 shows the tabletop antenna. Figure 2 shows the in-situ range measurement of each type of diversity antenna. B. Antenna Placement Our diversity experiment tested many different antenna configurations. This paper reports results of 5 basic

2 IEEE VTC 2003 FALL 2 Return Loss Results on 25 Samples Return Loss (db) Frequency (GHz) Frequency: Return Loss: Efficiency: Dimensions: Mass: Fig. 1. x y 2.4 to 2.48 GHz < -10 db >70% (45mm sq board) 14x10x2.5mm grams Small tabletop antenna (Etenna s EA2400 AccuWave). SATIMO arch Laptop z Compact diversity pair #1 #2 Baseline diversity pair #1 #2 Fig. 2. In-situ pattern measurement of two types of diversity antennas, both mounted in the laptop s screen hinge. configurations, shown in Figure 3. These configurations include: 1) a pair of baseline PIFAs mounted in the laptop s screen hinge, 2) an ultra-compact pair of Etenna tabletop antennas in the hinge, 3) another pair of tabletops mounted in the hinge, 4) a pair of tabletops mounted along the screen edge, and 5) four tabletops mounted in the hinge. In addition to the geometrical layout of the antenna elements, the measured configurations included a laptop screen in both open and closed positions. C. Measurement In-situ antenna pattern measurements were made at 2400 MHz with each antenna in every configuration, both with open and closed laptop screens. The measurements were performed rapidly using Etenna s SATIMO near-field antenna chamber. This antenna range measures dual-polarized, complex antenna patterns as a function of azimuth and elevation angles. III. CHANNEL MODELING Once the antennas are measured in the antenna range, we must extrapolate their pattern characteristics to link behavior in a fading channel. This section discusses in detail the procedure used for modeling performance in a multipath fading channel. A. Creating Channel Ensembles Whenever invoking a stochastic channel model, it is essential to define that model s channel ensemble (the collection of all possible channels), both with rigid mathematics and physical meaning. It is a bad habit of radio engineers to construct an elegant stochastic model for the wireless channel and then proceed to calculate a physically meaningless result because the ensemble did not reflect the correct application. To avoid this error, we will use the rigorously define stochastic local area channel model with independent phases (called the I- SLAC model in [4]). The narrowband I-SLAC model assumes that all multipath waves within a local area may be modeled as homogeneous plane waves. The mathematical form of this model as a function of vector position r is h( r) = N i=1 ( [ Pi exp j Φ i ]) k i r where P i and k i are the power and wavevector, respectively, of the ith multipath wave. The phases {Φ i } are independent random variables, uniformly distributed over the interval [0, 2π). The channel ensemble of the I-SLAC model has several equivalent meanings: 1) The multipath amplitudes of a specified local area are known, but the phases are unknown; the I- SLAC ensemble represents the collection of all possible phase combinations with uniform likelihood. 2) The complete complex channel is known, but the exact position of a receiver antenna within the spatial channel is unknown. 3) The complete complex channel is known, but the exact position of the carrier frequency (perhaps in a channelized air interface) is unknown. Essentially, the I-SLAC model allows us to generate a plausible ensemble of wireless channels given minimal information, such as a multipath angle spectrum. (1)

3 IEEE VTC 2003 FALL cm 1.8 cm 15.1 cm 27.9 cm 1.8 cm 15.1 cm 1.8 cm Configuration 1: PIFA Antennas Configuration 2: Compact Tabletops Configuration 3: Separated Tabletops Configuration 4: Side-mount Tabletops Configuration 5: Four Tabletops Fig. 3. Configurations of diversity antenna elements under test. Antennas are measured with the screen opened and closed. B. Polarization Extension Since we are studying the physical interface between an antenna and its multipath environment, let us extend the I-SLAC model to include antenna gain and polarization effects. In a range, we measure the complex antenna pattern, ã(θ, ϕ), which records the amplitude and phase response of an antenna as a function of 3D radio wave incidence (azimuth angle, θ, and elevation angle, ϕ). This complex antenna pattern is related to the familiar gain pattern by the following relationship: G(θ, ϕ) = ã(θ, ϕ) 2 (2) Furthermore, there are two separate responses of the antenna corresponding to polarization in the ˆϕ ( Eplane ) and the ˆθ ( H-plane ) directions. We denote these two polarization responses as ã ϕ and ã θ. If the antenna response has two different polarizations, then the multipath characterization must have two polarizations as well. We use P i,ϕ and P i,θ to denote the power in each polarization contributed by the ith multipath wave. Each multipath polarization must also be assigned corresponding phases (Φ i,ϕ and Φ i,θ ) which are assumed to be independent. Thus, the complete I- SLAC model becomes h( r) = N [ã ϕ (θ i,ϕ i ) P i,ϕ exp(jφ i,ϕ ) i=1 }{{} ϕ polarization +ã θ (θ i,ϕ i ) P i,θ exp(jφ i,θ ) } {{ } θ polarization ]exp ( j ) (3) k i r Equation (3) is the basis for the channel simulations in this paper. C. Choosing a Multipath Angle Spectrum An I-SLAC model can be represented in terms of an angle spectrum, p(θ, ϕ), that records arriving multipath power as a function of incident angles. For discrete multipath propagation, the multipath angle spectrum is given by p(θ, ϕ) = N i=1 P i δ(ϕ ϕ i )δ(θ θ i ) cos ϕ i (4) In computer simulation, it is most convenient to represent a discrete angle spectrum by uniformly sampling power with respect to elevation and azimuth. If there are M samples in elevation and N samples in azimuth, we can specify each multipath wave with a pair of integer coordinates [m,n], which correspond to the original index i by i = m N + n ϕ i = mπ M π θ i =2π n 2 N (5) With this uniform sampling, Equation (4) becomes p(θ, ϕ) = M 1 N 1 m=0 n=0 P m N+n δ ( ϕ+ π 2 mπ M cos ( mπ M π 2 ) ( δ θ 2π n ) (6) Our next step in analysis is to choose a realistic, analytical angle spectrum and back-solve for the discrete values of power, P i, based on Equation (6). The simplest choice for an indoor, realistic propagation environment is the isotropic angle spectrum, which assumes that equal amounts of multipath power arrive at the receiver from all directions. Recent measurements suggest that the isotropic angle spectrum is a fairly good approximation of indoor angle-of-arrival [5]. Such propagation is diffuse and has the following angle spectrum: p(θ, ϕ) =1 P m N+n =cos ) ( mπ M π 2 ) N (7) Notice that, although the power spectrum is uniform, the actual samples {P i } taper as the elevation angle increases. This is an artifact of the uniform sampling, which requires reducing the contribution of highelevation multipath for the same reason that latitudelongitude crossings become much denser near the poles of a globe.

4 IEEE VTC 2003 FALL 4 The angle spectrum used in our analysis is actually a hemisphere model, which is uniform in the upper hemisphere of the angle spectrum and 0-valued below. This angle spectrum is designed to approximate the conditions of a laptop resting upon a desktop. The surfaces of the laptop and the desktop preclude the arrival of multipath power from negative elevation angles. We should also mention that there should be 2 angle spectra: one for each polarization. For simplicity, we will assume complete depolarization that multipath power is likely to be distributed in ϕ and θ polarizations regardless of the transmitter polarization. Thus, the same angle spectrum will be used to describe propagation in both polarizations. The dearth of indoor polarization measurements in the research literature affords only this assumption. IV. ESTIMATING DIVERSITY PERFORMANCE A. Average Power and Correlation We now have all of the channel-dependent variables for using the model of Equation (3): measured complex antenna patterns provide ã ϕ and ã θ and samples from a multipath angle spectrum model provide P i and k i.now we must generate many trials of channel samples for each configuration of antenna elements. The algorithm for performing the trials is given below: 1) Designate one antenna as the origin element. Assign this a position of r = 0. 2) Choose a set of independent random phases for all of the Φ-terms in Equation (3). 3) Apply the origin element s ϕ-pol and θ-pol pattern measurements to Equation (3). Record the resulting complex channel sample h. 4) Now calculate the complex channel value for the next antenna element using the same set of random phases that were used at the origin element in the previous step. Measure the 3D displacement distance of the next element with respect to the origin element and set this equal to r. Apply this element s ϕ-pol and θ-pol pattern measurements to Equation (3) and record the resulting complex channel sample h. Repeat this step for every remaining antenna element. 5) At this point, the current trial is complete; each antenna element should have a complex channel sample generated from a common set of phases. Go to step 2 and repeat the procedure until a satisfactory number of trials have been performed. Upon completion, a large data set is available for calculating channel statistics between diversity antennas. There are two primary channel statistics that we wish to calculate for each antenna element configuration. First is the average received power for each element (the average h 2 ). Second is the average envelope correlation between antenna elements [4]. With these two statistics, it is now possible to generate link performance distributions. B. SINR Distributions The signal-to-interference+noise ratio (SINR) for switched diversity is calculated from the SINR of each antenna element, selecting the branch with the highest value; this branch s SINR is the one used by the diversity receiver. By performing this calculation for each trial in a large data set, we can generate distributions of SINR for the diversity receivers operating in these Rayleighfading links. Now we could have performed this step as part of the algorithm in the previous section. However, computing the resulting channel h at the terminals of each antenna given their complex antenna patterns is a timeconsuming operation. It takes far fewer samples to calculate reliable mean power and envelope correlation statistics than it does to calculate a reliable distribution. Therefore, once we have enough channel trials to calculate mean and correlation, we can switch to a simpler technique to generate the performance distributions. We use the Ertel method for generating a large number of correlated Rayleigh-fading channel samples [6]. C. Throughput Modeling Modeling data throughput from physical layer parameters can be somewhat arbitrary due to many network and application factors that affect the final quantity of usable bits arriving at a wireless terminal. As an approximation, we will use the rules-of-thumb for mapping SINR to transport-layer throughput as developed by Henty, et. al. in [7]. Based on this work, throughput (in bits/second) as a function of SINR in an unloaded network follows an exponential curve: ( T = T 0 [1 exp SINR SINR 0 SINR 1 )] u (SINR SINR 0 ) (8) where T 0, SINR 0, and SINR 1 are constants (often devicedependent) and u( ) is the unit-step function. Typical values for IEEE links are T 0 = 2000 kbps, SINR 0 = 0 db, and SINR 1 = 3 db. All throughput results presented in this paper use the model in Equation (8).

5 IEEE VTC 2003 FALL 5 V. RESULTS Figure 4 shows an example of throughput distribution for the open laptop with ultra-compact tabletop antennas. Although the plots mark nicely the distribution of throughput in a Rayleigh fading channel, the most critical portion of the curves are their left-hand behavior and, in particular, their y-axis intercept; this is essentially the outage probability of the link (0 kbps). Note the huge throughput gains available for using two elements with pure selection. 1 TABLE I SUMMARY OF RESULTS FOR 5 DIVERSITY CONFIGURATIONS Configuration Gain 1 Gain 2 Corr. PS (db) (db) Outage 1 Open % Close % 2 Open % Close % 3 Open % Close % 4 Open % Close % 5 Open Left % Open Right % 4 Br. PS 0.3% Close Left % Close Right % 4 Br. PS 0.8% Fig. 4. Cumulative throughput distribution in a Rayleigh channel for open configuration 2. The results of the study are summarized in Table I, which records antenna correlation values, average element gain (with respect to an average baseline PIFA), and link outage probability (assuming a 10 db average SINR in a Rayleigh fading channel). There are several key points in Table I: 1) For the case of four antenna elements, results are tabulated for both sets of pairs as well as selection diversity using all four. 2) The envelope correlation of compact tabletop antennas are higher than other pairs with higher separation distances; however, these correlation values ( ) are far below the theoretical values without coupling. 3) The degradation resulting from ultra-compact tabletop antennas is slight, with outage rising from 1.2% to 2.8% for the open screen measurements. 4) Hinge-mounted antennas perform consistently better when the laptop screen is open instead of closed. Overall, the results demonstrate the diversity gains present, even for ultra-compact antenna elements. VI. CONCLUSIONS The modeling technique of this paper a useful procedure for studying antenna effects on a variety of wireless communication links illuminated basic principles in antenna diversity design for small wireless devices. In fact, the results of this study suggest that small wireless devices of the future should be covered with many cheap, high-efficiency diversity antenna elements. VII. ACKNOWLEDGEMENTS The authors wish to acknowledge the advice and assistance of Professor Mike Buehrer during the course of this investigation. REFERENCES [1] W.C. Jakes, Ed., Microwave Mobile Communications, IEEE Press, New York, [2] Y. Ebine, T. Takahashi, and Y. Yamada, A Study of Vertical Space Diversity for a Land Mobile Radio, Electronics and Communications in Japan, vol. 74, no. 10, pp , [3] T. Taga, Characteristics of Space-Diversity Branch Using Parallel Dipole Antennas in Mobile Radio Communications, Electronics and Communications in Japan, vol. 76, no. 9, pp , [4] G.D. Durgin, Space-Time Wireless Channels, Prentice Hall Inc., Upper Saddle River NJ, [5] G.D. Durgin, V. Kukshya, and T.S. Rappaport, Wideband Measurements of Angle and Delay Dispersion for Outdoor and Indoor Peer-to-Peer Radio Channels at 1920 MHz, IEEE Transactions on Antennas and Propagation, May [6] R.B. Ertel and J.H. Reed, Generation of Two Equal Power Correlated Rayleigh Fading Envelopes, IEEE Communications Letters, vol. 2, no. 10, pp , Oct [7] B.E. Henty and T.S. Rappaport, Throughput Measurements and Empirical Prediction Models for IEEE b Wireless LAN (WLAN) Installations, Tech. Rep. Master s Thesis, Virginia Tech, Aug 2001.