Performance Analysis of Beam Sweeping in Millimeter Wave Assuming Noise and Imperfect Antenna Patterns Vutha Va and Robert W. Heath Jr. Wireless Networking and Communications Group Department of Electrical and Computer Engineering The University of Texas at Austin http://www.profheath.org Funded by D-STOP Tier 1 University Transportation Center and a gift from TOYOTA InfoTechnology Center, U.S.A., Inc.
Millimeter wave for high data rate applications Bandwidth 54 Max. channel bonding 1 4 802.11n 802.11ac 802.11ad Large bandwidth at mmwave Low freq. antenna mmwave antenna Shrinking antenna aperture Beam alignment required Arrays needed for gain and aperture Accurate beam alignment is crucial in enabling high data rate mmwavelinks 2
Beam training for beam alignment Beam sweeping 1 3 2 Active sector Quasi-omni Wide beams at initial stages for efficient searching Low SNR due to low antenna gain Slides Robert W. Heath Jr. (2016) Antenna gain Fluctuations* Measurement error leading to beam misalignment Derive beam misalignment probability considering both measurement noise and antenna gain fluctuation * K. Hosoya et al., IEEE Transactions on Antennas and Propagation, vol. 63, no. 1, pp. 81 96, Jan. 2015. 3
Related work sector patterns Quasi-omni patterns Examples of codebook based beam sweeping: ª Tree-based search [1] ª Adaptive subspace sampling [2] Existing analysis ª Bound on misalignment probability due to measurement noise [2] ª Misalignment probability due to antenna gain fluctuations [3] Our analysis ª Considers both measurement noise and antenna gain fluctuations ª Can capture the severity of beam misalignment [1] J. Wang et al., IEEE Journal on Selected Areas in Communications, vol. 27, no. 8, pp. 1390 1399, Oct. 2009. [2] S. Hur et al., IEEE Transactions on Communications, vol. 61, no. 10, pp. 4391 4403, Oct. 2013. [3] K. Hosoya et al., IEEE Transactions on Antennas and Propagation, vol. 63, no. 1, pp. 81 96, Jan. 2015. 4
Receive power model Tx antenna gain Rx antenna gain Tx antenna gain Slides Robert W. Heath Jr. (2016) Rx antenna gain Rx power excluding antenna gain Noise Tx Rx This includes channel effect (path loss and fading), Tx power, and spreading gain Power due to this ray (excluding antenna gain) seen by the beam pair is modeled by Applicable to both narrowband and wideband models because only Rx power is used 5
Quasi-omni pattern gain fluctuation model Path A path loss -100 db Path B path loss -95 db Slides Robert W. Heath Jr. (2016) Antenna gain measurement from [1] 10 db gain fluctuation ULA 4 elements STA Weaker path A is chosen because of gain fluctuation Laplace distribution provides good fit Antenna gain in db Average antenna gain Fluctuation level [1] K. Hosoya et al., IEEE Transactions on Antennas and Propagation, vol. 63, no. 1, pp. 81 96, Jan. 2015. 6
SLS and 3c beam alignment methods IEEE 802.11ad (SLS method) (SLS: Sector Level Sweep) Active sector IEEE 802.15.3c (3c method) Quasi-omni Tx Quasi-omni Rx Further training involving narrower beams can be done in both methods Quasi-omni SLS suffers gain fluctuation only on one side while 3c method suffers both Our analysis involves only this initial stage with quasi-omni Training over all sectors with a quasi-omni Exhaustive search at each level of beam codebook 7
Power loss probability Definition of power loss If the i-th best beam is chosen, we say there is a power loss of (By definition, ) Probability of power loss Measured receive power is greater than all other beam pairs Power [db] Optimal beam pair 1 2 3 4 Beam pair index Number of beam training pairs depending on alignment method The distribution of is different between SLS and 3c method 8
Sketch of the derivation Assuming the gain fluctuations experienced by each beam pair are independent Rational for the assumption AoA/AoD of each path are independent and so are the gains of each path Measured receive power are independent Deriving PDF of the receive power For both SLS and 3c method, These events are independent when conditioned on Deconditioning on Can be computed when knowing PDF of constant Independent RVs whose PDF can be computed using convolution. 9
Simulation setting Larger gap between best and 2 nd best pair for LOS channels Consider 2 types of channels: for i 2 LOS channel: NLOS channel: Power [db] SNR after alignment are set to 5, 15, and 25 db Antenna gain setting: SLS Beamwidth Gain Beamwidth Gain Quasi-omni 120 60 Sector 15 13.8 15 13.8 3c 1 2 3 4 Beam pair index Narrower quasi-omni for 3c method so both have the same overhead Denotes Laplace dist. with mean 7.44 and spreading param. 1.5 10
Numerical results: SLS method Slides Robert W. Heath Jr. (2016) Still has error (small) due to gain fluctuation Lower alignment probability than in LOS Power loss [db]: Power loss [db]: Alignment fails mainly at low SNR LOS channel NLOS channel Effect of antenna gain fluctuation is more severe in NLOS channels 11
Numerical results: 3c method Almost no misalignment due to the narrower quasi-omni used in the 3c method Power loss [db]: Power loss [db]: LOS channel NLOS channel Generally follow similar trend as the SLS case but with smaller gain fluctuation 12
Some implications LOS dominant use cases Use cases where NLOS is common Wireless backhaul LOS dominant use cases can tolerate antenna gain fluctuation Use cases where NLOS is common require high fidelity in antenna pattern High enough SNR is needed for correct beam alignment Spreading factor should be set to provide the needed SNR If position info can be used, should adapt the spreading factor 13
Conclusions We derived probability of power loss for beam sweeping methods o Power loss can capture the severity of misalignment o Both thermal noise and antenna gain fluctuation are considered Key insights o Spreading is needed to reduce the impact of measurement noise o Effect of antenna gain fluctuation cannot be compensated by large spreading o Adaptive spreading factor will be useful if wide range of link distance is needed o LOS dominant use cases have low requirements on antenna gain fluctuation 14
Thank you! 15