March 14, 2007 Steve Ellingson & Mahmud Harun Bradley Dept. of Electrical and Computer Engineering Virginia Polytechnic Institute & State University http://www.ece.vt.edu/swe/ ellingson@vt.edu
About Virginia Tech About 26,000 students College of Engineering: Grants the 7 th largest number BS degrees in the US Ranked in the top 25 by US News & World Report and others The Bradley Department of Electrical & Computer Engineering One of the nation s largest ECE departments ~ 75 tenure-track faculty and 12 research faculty members ~ 1,100 undergraduate and ~ 570 graduate students Wireless communications is a principal focus area ~ 25 ECE faculty involved in various aspects of wireless communications research and teaching. Large fraction of ECE graduate students specialize in wireless telecommunications or related technologies. 2
30+ members of the VT faculty with a common interest in wireless research. http://wireless.vt.edu 3
Overview of My Research z z z z Applied Electromagnetics Antennas / Arrays Propagation Radio Astronomy Remote Sensing z z z ellingson@vt.edu Applied Signal Processing Multiple antenna systems Interference Mitigation Radio Geolocation z z Instrumentation & Systems High-Performance Transceivers FPGA-Based Design http://www.ece.vt.edu/swe/ 4
Outline of this Talk One-Slide MIMO Primer Matrix Channel Measurement System (MCMS) Lateral position dependence of the 2.4 GHz hallway wireless channel Motivation Propagation measurements & analysis Capacity results 5
One-Slide MIMO Primer Generalized Shannon Bound: Mean SNR per RX antenna Capacity [bps/hz] C = k i= 1 log + 2 1 ρ N T λ i Eigenvalues of HH, where H is the [ N R x N T ] matrix of channel coefficients N T =1 or N R =1 rank{hh }=1 C log 2 N N T >1 and N R >1 and rank{hh }>1 C N Up to k=min{n T,N R } independent MIMO subchannels, each with SNR the associated eigenvalues of HH 6
MCMS Matrix Channel Measurement System 24V Battery Pack MCR MCT MCMS stowed MCR set up for demo http://www.ece.vt.edu/swe/mcms 7 S.W. Ellingson, A Flexible 4 x 16 MIMO Testbed with 250 MHz 6 GHz Tuning Range, 2005 IEEE Int l Ant. & Prop. Symp., Vol. 2A, July 2005, pp. 309 12.
8
MCMS High-Level Block Diagram RFDC Dig IF RFDC Dig IF RFDC Dig IF Embedded PC cpci Multi- Channel Transmitter (MCT) Dig I/O QDUC RFUC Dig I/O QDUC RFUC Dig I/O QDUC RFUC Dig I/O QDUC RFUC Clock & LO Synthesis & Distribution Matrix Channel Under Test RFDC RFDC RFDC RFDC RFDC RFDC RFDC RFDC RFDC RFDC RFDC RFDC RFDC Dig IF Dig IF Dig IF Dig IF Dig IF Dig IF Dig IF Dig IF Dig IF Dig IF Dig IF Dig IF Dig IF Clock & LO Synthesis & Distribution Aggregation & Corner Turning Dig I/O cpci Quad DSP Embedded PC Multi- Channel Receiver (MCR) 9
MCMS MCT: Direct Digital IF Synthesis 200 MSPS sample clock From PC: I-Q (symbols) or arbitrary waveform or sinusoid parameters fc = 78 MHz Ch.1 RF UC Ch.2 RF UC Ch.3 RF UC Ch.4 RF UC Direct-Sequence Spread Spectrum signal with 12 MHz BW (RRC filtering), synthesized at 78 MHz and upconverted to 1250 MHz 4-channel quadrature digital upconverter (QDUC) board using the Analog Devices AD9857 10
MCMS RF Downconverter / Digitizer (RFDC) 11
Digital IF Board FS = 104 MSPS (Real) 78 MHz fc = BW = 40 MHz From RFDC LVDS RX LVDS RX FS = 52 MSPS (I/Q) 0 MHz fc = BW = 40 MHz Altera Stratix EP1S10 O I SHIFT E FS = 26 MSPS (I/Q) 0 MHz fc = BW = 20 MHz FS/4 Q FIR, 2 A FIR, 2 A 16K FIFO A A NCOM CIC+FIR, R LVDS RX LVDS TX Daisy Chain MCMS FS = var. (I/Q) fc = 0 MHz BW = var. Analog Devices AD6620 12
MCR Digital Chassis MCMS Digital IF Boards 1.248 Gb/s LVDS from RFDC Daisy Chain #1 (320 Mb/s LVDS Serial Bus) Control In, Data Out BW < 1.5 MHz: Streaming mode BW > 1.5 MHz: Burst mode Corner Turner Board MCR Digital Chassis Daisy Chain #2 Daisy Chain #3 Daisy Chain #4 32-bit High-Speed DIO Board 13
MCMS MCR Signal Path Demo Time Domain 52 MSPS complex Output of Fs/4 Downconversion is 16-bits (+/-32K) Filter Specs: 63-tap FIR, 12-bit coeff., 12-bit in, 16-bit out, 20 MHz LP ~70 db -F S /4 Spectral Shift (104 / 4 = 26 MHz) +27 MHz shifts to +1MHz Frequency Domain Blue: 1 Yellow: 100, average 14
MCMS Acknowledgements S. Ellingson OSU, now VT PI / Systems Engineer G. Hampson OSU, now CSIRO Electronics W. Theunissen OSU Electronics B. Reynolds Aeroflex Aeroflex Program Manager P. Bohley Aeroflex Integration S. Fisher Aeroflex Electronics W. Koehler Aeroflex Software Students S. Horst RF Design W. Taylor RF Design M. Nuhfer Software National Science Foundation Grant No. ECS-0215990 15
- Motivation Universities Study areas often located within hallway-like spaces Courthouses Laptop/PDA users seated on benches Hospitals Wireless-enabled gurneys and monitoring equipment often parked in hallways Related but significantly different: Aircraft passenger cabins Ship corridors Tunnels Reverberant spaces 16
Hallways: MIMO Capacity Same trend seen for any link for which one end is in hallway. 17 D. Porrat, P. Kyritsi, and D.C. Cox, MIMO Capacity in Hallways and Adjacent Rooms, IEEE GLOBECOM 2002, Vol. 2, Nov.2002, pp. 1930 34.
Simplified Model for Hallway Propagation ε r moderate, lossy Horizontal Plane ε r moderate, lossy Vertical Plane σ, or ε r large σ, or ε r large 18
Simplified Model for Hallway Propagation ε r moderate, lossy Horizontal Plane ε r moderate, lossy Vertical Plane: Efficient Containment σ, or ε r large σ, or ε r large 19
Simplified Model for Hallway Propagation ε r moderate, lossy Horizontal Plane: Leakage into rooms, Loss Mode Order Rank collapse occurs because low-order modes propagate best Vertical Plane: Efficient Containment ε r moderate, lossy σ, or ε r large σ, or ε r large 20
Test Conditions Hallway 1.5 m wide x 2.7 m high Doors closed Range = 12 m 21
Antenna Array Identical broadside-pointing uniform linear arrays at both ends V-polarized ¼-wave monopoles ¼-wave spacing at 2.4 GHz Ground plane 25 cm x 2.5 cm -10 db max return loss over bandwidth -13 db max coupling 22
Measurement / Processing 4 CW signals used on transmit (1 per antenna) with frequency offset << coherence bandwidth used to discriminate between antennas 4 receive antennas coherently sampled with 40 MHz BW and recorded in 156 µs (<< coherence time) segments 4 complex channel coefficients (1 per transmit antenna) extracted from each of the 4 receive antennas signal captures (16 coefficients) System nose-to-nose self-response calibrated out, yields measurement matrix K Data checked to confirm: No noise-dominated eigenvalues (sufficient SNR) No corruption due to RFI (esp. from IEEE 802.11 WLAN) 2 N The usual SNR normalization: T HH' = KK' Tr { KK' } 23
Calibration / Sanity Check Eigenvalues of HH Wired Full Rank Channel Wired Keyhole Channel MCT MCR MCT MCR combiner splitter 24
K 4i, Horizontal Cut i = 1, 2, 3 (transmit antennas) Sample spacing ~ λ/10 Note this is not a traditional fast fading environment! R = 12 m 25
Horizontal Cut Propagation Model Measurement Model 2D (perfect vertical containment) model assuming infinitely thick walls with ε r = 4.4; LOS + 15 bounce modes (31 terms) R = 12 m 26
H-Cut Prop Model vs. Measurement Null regions near walls because reflected field partially cancels incident field Pretty good agreement on overall level Null regions closer to center are anticipated (but not accurately) R = 12 m 27
K 4i, Vertical Cut i = 1, 2, 3 (transmit antennas) Sample spacing ~ λ/10 No obvious structure; Consistent with vertical containment theory Starting to see some traditional fast fading R = 12 m 28
Similarity Metric: Horizontal Cut Tx Ant 1 Tx Ant 2 Tx Ant 3 Similarity [1 1 1 1] k i / 4 k i = 1 for LOS channel = 0 for an orthogonal channel Channel structure changes slowly, so must be simple! Similar trends (as expected, since array elements are following same path), but significant differences Note repeatibility (max, mean, min shown) 29
Similarity Metric: Vertical Cut Tx Ant 1 Tx Ant 2 Tx Ant 3 Note LOS-like conditions over extended portions of the cut quite different from horizontal cut Dissimilar trends (as expected, since array elements are following different paths) consistent with simple hallway prop model Note repeatibility (max, mean, min shown) 30
Capacity Calculations (Reminder) Mean SNR per RX antenna Capacity [bps/hz] C = 3 i= 1 log 2 1 + ρ N T λ i Eigenvalues of HH, where H is the 4 x 3 matrix of channel coefficients HH' 2 N = T Normalization to make coefficients independent of mean SNR KK' Tr { KK' } (Will have more to say about this shortly!) 31
Brief Digression: Eigenvalues & Capacity Indoor: Cluttered laboratory, approx 5 m x 10 m About 2 meters between arrays Transmit Array: 4 λ/4 monopoles, V-pol, 0.25λ spacing Receive Array: same Eigenvalues of HH 4 x 4: Optical LOS Exists 4 x 4: Optical LOS Blocked using 1 m x 2 m metal plate 32
Brief Digression: Eigenvalues & Capacity Indoor: Cluttered laboratory, approx 5 m x 10 m About 2 meters between arrays Transmit Array: 4 λ/4 monopoles, V-pol, 0.25λ spacing Receive Array: same High mean SNR; Capacity here is better. Low mean SNR; Capacity here is worse. Eigenvalues of HH 4 x 4: Optical LOS Exists 4 x 4: Optical LOS Blocked using 1 m x 2 m metal plate 33
Raw Eigenvalues Vertical Cut Eigenvalues of KK Rank collapse is evident clear that MIMO potential is limited. 34
Raw Eigenvalues Horizontal Cut Eigenvalues of KK Rank collapse is evident clear that MIMO potential is limited. However, also that magnitude taper is quite pronounced effect on SNR normalization? 35
Tr{KK } (Total Power Transferred) Vertical Cut In terms of total power transfer, variation across horizontal cut is even more pronounced. Horizontal Cut Traditional SNR normalization HH' = KK' Tr will significantly overestimate performance in horizontal cut 2 N T { KK' } 36
Modified SNR Normalization HH' = KK' Tr 2 N T { KK' } Problem: This is varying from trial to trial; in fact it should be constant so as not to neutralize the SNR variation κ E { Tr{ KK' } vertical Use vertical cut, over which SNR appears to be approximately constant HH ' = KK' 2 NT κ Computed capacity now varies properly with SNR 37
Capacity in Horizontal Cut, Traditional Norm Ideal full-rank MIMO channel MIMO, measured Ideal rank-1 channel Best rank-1, measured Facing beams, measured ρ = 10 db 3 x 4 in all cases. Assuming perfect CSI. Ideal 1 x 1 is 3.5 bps/hz. 38
Capacity in Horizontal Cut, Modified Norm Ideal full-rank MIMO channel MIMO, measured Ideal rank-1 channel Best rank-1, measured Facing beams, measured ρ = 10 db Striking reduction in capacity near walls is now apparent 3 x 4 in all cases Assuming perfect CSI. Ideal 1 x 1 is 3.5 bps/hz. 39
Capacity in Vertical Cut, Traditional Norm Ideal full-rank MIMO ch. MIMO, measured Ideal rank-1 channel ρ = 10 db No taper apparent; somewhat more better overall result. (Modified normalization yields approximately same result) Best rank-1, measured Facing beams, measured 3 x 4 in all cases. Assuming perfect CSI. Ideal 1 x 1 is 3.5 bps/hz. 40
Concluding Remarks MIMO propagation and capacity measurements have been performed in a hallway @ 2.4 GHz, 12 m range, with vertical monopoles arranged horizontally 3 x 4 results (10 db SNR, ideally 11.5 bps/hz) Vertical cut: ~ 7.5 ± 1 bps/hz Horizontal cut: ~ 9 bps/hz near center, ~ 4-6 bps/hz near walls Significant rank reduction evident in both cases Optimal and suboptimal rank-1 schemes come much closer to ideal 5.4 bps/hz Scenarios involving hallways cannot be treated like other indoor scenarios Expect rank collapse, unusual fading behaviors Implications for antenna array design and placement 41