Flashback: A New Control Plane for Wireless Networks

1. Flashback: A New Control Plane for Wireless Networks AsafCidon (Stanford), KanthiNagaraj (UCLA), PramodViswanath (UIUC), SachinKatti (Stanford) Stanford University

2. Agenda • Motivation and Overview • Wi-Fi PHY Primer • Design of Flashback • Experiment Results • Higher Layer Applications

3. Wireless Control Channels • Wireless networks require control channels for synchronization and coordination across multiple clients • Example: LTE • Dedicated frequencies for control and coordination • Used for resource allocation, QoS, scheduling, power level information, etc.

4. Unlicensed Networks Are Out of Control • Unlicensed networks do not have an explicit control channel - they use implicit coordination • RTS/CTS • Collision prevention and backoff mechanisms (CSMA/CA) • 802.11e QoS queues • Problems of implicit control mechanisms • Overhead on data channel • Do not scale with number of nodes, congested networks • Limited central control (lack of fairness, starvation)

5. The Holy Grail: Control Channel for Wi-Fi • Our goal: a control channel for Wi-Fi • Centrally Managed: AP provides coordination and QoS through control channel • Independence: Data and control independent • Simplicity: Throw away RTS/CTS, CSMA/CA • Constraint: low-overhead • Backwards compatibility • No big hardware changes

6. Wi-Fi PHY Primer: OFDM • OFDM widely used in wireless networks • Key idea: multiple narrowband sub-carriers at a low symbol rate • Main advantage: cope with severe channel conditions (frequency-selective fading) without complex equalization filters

7. Wi-Fi PHY Primer: Bit Rates and Channel Codes

8. Flashback Intuition • Wi-Fi channel codes have robust SNR margins (~3db) • Insight: even if we lose a couple of bits here and there, channel codes will prevent data loss • Key idea: erase subcarrier instead of treating it as an error • Gives us an even higher SNR margin

9. Flashback in a Nutshell • Control signaling using ‘flashes’ • High power single sub-carrier flash sent on top of data transmission • Receiver can detect flashes independently of on-going data transmission • If flash detected, erase the sub-carrier from data packet • Flashes are not modulated (i.e. they are binary) • Flashes provide a near-zero overhead separate PHY control channel • Backwards compatible • No synchronization required

10. Flashback Receiver Design 64 FFT Equalizer ADC Sync Flash Eraser Demodul-ator Flash Detector Viterbi Decoder Flash Demodulator Control Message Data Packet

11. Implementation • Implementation using NI PXIe-8130 RTOS Dual-Core Controller • NI PXIe-7965R FlexRIO, NI 5781 BB Transceiver • Setup • 1 data transmitter, 1 flash transmitter, 1 receiver • ~300 runs for each data point • Flashes sent at 8-10 db relative to data transmission

12. 5000

14. Improving Flash Detection • Flash detection is not perfect: flashing node is not synchronized to transmitter node • Flashes can be ‘smeared’ over 2 symbols in time • Solution: • Run additional FFT to detect if flash is smeared over 2 symbols

16. Applications • Given control channel PHY, we can use Flashback to improve the MAC: • Get rid of overhead in RTS/CTS • Implement QoS scheduling • Use flashes for estimating SNIRs between networks and improving spatial reuse • Use flashes to indicate power/sleep modes

17. Example 1: Don’t RTS, Just Flash • AP assigns flash subcarriers during association • Clients maintain overall flash rate by estimating number of nodes • Flash instead of RTS • Wait until AP is listening • Benefits • No RTS = no contention period = no overhead! • AP can do smart scheduling by estimating SNRs of nodes using flashes

19. Example 2: QoS

20. Example 3: Estimate SNIRs • Clients flash at constant power  receivers can estimate link SNR  estimate SNIR () • APs can know SNIR of all the links in the network • Use flashes to communicate between APs • Maximize spatial reuse

21. Thank You! Stanford University