CSI-SF: Estimating Wireless Channel State Using CSI Sampling & Fusion Riccardo Crepaldi∗, Jeongkeun Lee†, Raul Etkin†, Sung-Ju Lee†, Robin Kravets ∗University of Illinois at Urbana-Champaign †Hewlett-Packard Laboratories Improving Performance in High Speed Wireless Channel Quality Metrics How do systems describe the channel quality? Pros Finer granularity compared to SNR Fading and phase shift information for each subcarrier Tighter bound with channel performance [Halperin’10] Cons CSI is significantly larger than SNR CSI Describes a specific MIMO configuration CSI for multiple MIMO configurations must be measured for a complete channel characterization PHY Layer enhancements: MIMO technology OFDM Channel Bonding Adaptive protocols for: Rate adaptation Channel selection Optimal antenna scheme selection Best AP selection Choices based on: Channel quality RF AGC FFT Chan Estimation MIMO Decoding Demap [N Input streams] [M spatial streams] RF AGC FFT Chan Estimation Demap RF AGC FFT Chan Estimation Signal-to-Noise Ratio (SNR): One scalar value per packet Accurate only for narrow-band channel Channel State Information (CSI): M x N x W complex matrix W= Number of OFDM subcarriers Current 802.11n devices: M=3, N=3 W=56 (20MHz) or W=114 (40MHz) TX: M parallel data streams RX: N receivers What MIMO configuration should I choose? The Channel quality is… Why CSI? CSI Matrix Combining CSI-SF enables estimation of uninvestigated channel configurations Z(t) Bitstream Encoding Symbol Mapping X(t) Spatial Mapping (Q) RF Channel (H) RX Y(t) Application of CSI-SF A real networking problem: Rate Adaptation Metrics: SNR, effective SNR, estimated effective SNR What’s a good metric? Small or no gaps Monotonic behavior CSI Ex: Estimate of a 2x3 configuration using two 1x3 received packets 2x3 – 20MHz Large gaps Non-monotonic beavior Packet 1 (Tx: A, Rx: ABC) Packet 2 (Tx: B, Rx: ABC) Q1† 3 2 56 Q2 1 CSI1 CSI2 Estimated CSI Power scale factor 2x3 – 40MHz 3x3 – 20MHz eSNR (effective SNR based on real samples) e2SNR (effective SNR based on CSI-SF estimates) Quality of CSI-SF Estimate SNR is not a good metric, eSNR and e2SNR are better Compared to standard eSNR, e2SNR uses estimates instead of real samples thus limiting the overhead Example of CSI-SF combining two 1x3 packets in a 2x3 packets SNR [db] OFDM subcarrier index Packet 1: 1x3 – Ant A Packet 1: 1x3 – Ant B CSI-SF Reduces Overhead Hypoyetical channel variation Overhead required to rate control to converge using different metrics Packet 3: 2x3 – Ant AB Real CSI Estimated CSI Hewlett Packard Laboratories University of Illinois @ Urbana-Champaign, Department of Computer Science – Mobicom 2011