PHY-Level Security Protection Month Year doc.: IEEE 802.11-yy/xxxxr0 July 2017 PHY-Level Security Protection Date: 2017-04-20 Authors: Li, Jiang, Segev, Abramovsky, et al, Intel Abramovsky, Ghosh, Segev & Li, Intel
July 2017 Abstract Previously in [1][2], we identified a threat model with two types of adversaries: Type A 1ms response time Type B 1us response time In this submission, we present a scheme to detect and suppress Type B adversary attacks at PHY level [1] Doc.: IEEE 802.11-17/0120r2 Intel Secured Location Threat Model [2] Doc.: IEEE 802.11-17/0801r1 Intel Discussion on FTM Protection – follow up Li, Jiang, Segev, Abramovsky, et al, Intel
Outline PHY-level technique to protect measurement symbols: July 2017 Outline PHY-level technique to protect measurement symbols: Prevention of wrong sense of distance through detection of adversary attack: Discarding contaminated measurements ensures security Suppression of adversary attack: Suppressing adversary attacks enhances robustness Li, Jiang, Segev, Abramovsky, et al, Intel
Needs for High Security July 2017 Needs for High Security Some applications require high security Door lock, PC lock, ATM Spoofed measurement should be discarded to prevent property loss Li, Jiang, Segev, Abramovsky, et al, Intel
HW Impersonation/Data Integrity – How to Spoof Legacy Sounding Month Year doc.: IEEE 802.11-yy/xxxxr0 July 2017 HW Impersonation/Data Integrity – How to Spoof Legacy Sounding L-STF & L-LTF give the timing reference to the VHT-LTF, which could be spoofed by the adversary RSTA (AP) Transmission NAV is set by 11a RTS CTS Beamforming makes DATA not accessible to third parties Add a backup slide for random sequence in 11mc Note: Quotation of Slide 11 in [1] Li, Jiang, Segev, Abramovsky, et al, Intel Abramovsky, Ghosh, Segev & Li, Intel
MAC protection is insufficient July 2017 MAC protection is insufficient Although transmissions of time stamps i.e. t1, t2, t3, t4 can be encrypted, the measurements of t2 and t4 themselves are still vulnerable AP Adversary STA t1 t2 Spoofed 1st tap arrives before the true one t3 t4' RTT is perceived smaller because t4'-t1 < t4-t1 t4 Li, Jiang, Segev, Abramovsky, et al, Intel
Goals Detecting adversary attack ensures security July 2017 Goals Detecting adversary attack ensures security Once adversary attack is detected, spoofed measurement can be discarded and further damage is prevented Suppressing attack signals enhances resilience Processing gain of random sounding sequence suppresses spoofing signal Li, Jiang, Segev, Abramovsky, et al, Intel
July 2017 Adversary Detection Conduct two sounding measurements within channel coherence time Shift 2nd sounding symbols (i.e. HE-LTF or VHT-LTF) by a random CSD unknown to spoofer Check consistency across two channel measurements CSD e.g. 170 ns applied to HE-LTF TF 1 UL NPD 1 NDP-A 1 DL NDP 1 TF 2 UL NPD 2 NDP-A 2 DL NDP 2 Channel measurement 1 Channel measurement 2 Li, Jiang, Segev, Abramovsky, et al, Intel
Procedures Transmitter: Receiver: July 2017 Transmit two sounding signals within channel coherence time e.g. 1ms Apply CSD to 2nd sounding signal, where CSD value is known to the receiver over encrypted message so that spoofer can’t adapt to the CSD Receiver: Remove the CSD from each measurement, and compare the channel estimates of two adjacent measurements Channel estimates should be consistent unless spoofing occurred Channel estimates from 1st measurement Channel estimates from 2nd measurement Inconsistent Due to spoofer Due to user Li, Jiang, Segev, Abramovsky, et al, Intel
July 2017 Discussions Spoofing detection by CSD requires almost no implementation changes CSD is currently used in legacy transmitter and receiver. For example, CSD is compensated before channel interpolation in 11n/ac/ax Adversary attack can be detected but can’t be suppressed by random CSD Li, Jiang, Segev, Abramovsky, et al, Intel
Suppression of Adversary Attack — Random sounding symbols July 2017 Suppression of Adversary Attack — Random sounding symbols Replace existing sounding signal (i.e. LTF binary sequence) by a random binary sequence unknown to spoofer Sequence generation key is exchanged and encrypted before measurement L-STF, L-LTF, L-SIG, RL-SIG, HE-SIG-A HE-STF Random BPSK sequence +1, -1,+1, +1, +1, -1, -1, … Li, Jiang, Segev, Abramovsky, et al, Intel
Suppressed Spoofing Impact July 2017 Suppressed Spoofing Impact Spoofed 1st tap True 1st tap Noise level With Legacy LTF symbols With random sounding symbols Concentrated, high power spoofed taps Spread, low power spoofed taps Li, Jiang, Segev, Abramovsky, et al, Intel
July 2017 20 dB Suppression Suppress spoofed 1st tap by about 20 dB for 80 MHz sounding True 1st tap Spoofed 1st tap Li, Jiang, Segev, Abramovsky, et al, Intel
Requirements for Random Sounding Signal July 2017 Requirements for Random Sounding Signal Strong security protection A large amount of sounding signals to choose Easy implementation BPSK modulation and minimum storage Support for long distance ranging Low PAPR Scalable protection Nice tradeoff between security and overhead Li, Jiang, Segev, Abramovsky, et al, Intel
Golay Sequences Golay sequences have low PAPR July 2017 Golay Sequences Golay sequences have low PAPR About the same as 11ax LTF Golay sequences can be easily generated Duplication, shift, and sign change 512 sequences can be generated for 80MHz 1x sounding The odd for the adversary is below 2 ×10-3 per sounding Li, Jiang, Segev, Abramovsky, et al, Intel
Loading Golay Sequence to Subcarriers July 2017 Loading Golay Sequence to Subcarriers Slight puncturing is applied to Golay sequence for accommodating guard subcarriers Golay sequence, a complementary pair Li, Jiang, Segev, Abramovsky, et al, Intel
Easy Generation Using Concatenation July 2017 Easy Generation Using Concatenation Long Golay sequence can be generated by concatenating two short sequences +1, +1 +1, -1 [ ] + +1, +1 +1, -1 - +1, +1 - [+1, -1] [+1, +1, +1, -1], [+1, +1 -1, +1] [+1, +1, +1, -1], [-1, -1, +1, -1] Li, Jiang, Segev, Abramovsky, et al, Intel
Easy Generation Using Interleaving and Reversion July 2017 Easy Generation Using Interleaving and Reversion Generate new long sequence by interleaving two short ones Generate new sequence by reversing the order of one Two short sequences a b c d A B C D Interleave a A b B c C d D a -A b -B c -C d -D Reversion -D d -C c -B b -A a D d C c B b A a Li, Jiang, Segev, Abramovsky, et al, Intel
Length 2K Golay Sequences July 2017 Length 2K Golay Sequences 2K+1 sequences with length 2K 512 sequences with length 256 Large distances among generated sequences Concatenation, interleaving, and reversion generate orthogonal sequences, respectively Cross correlation among sequences is either 0 or 1/4 Li, Jiang, Segev, Abramovsky, et al, Intel
Higher Security by 4x LTF July 2017 Higher Security by 4x LTF 4x LTF quadruples the number of sounding sequences 2048 for 80MHz at the cost of 10 us sounding time 4.0 μs 13.6 μs 1x LTF, 512 sequences 4x LTF, 2048 sequences Li, Jiang, Segev, Abramovsky, et al, Intel
Higher Security by Multiple Measurements July 2017 Higher Security by Multiple Measurements Each measurement independently choose a sounding sequence Sequence space increases exponentially with the number of measurements conducted within the channel coherence time The chance left for the adversary is below 4 ×10-6 for passing three contiguous measurements Independent sounding sequences TF 1 UL NPD 1 … TF 2 UL NPD 2 … TF 3 UL NPD 3 … Measurement 1 Measurement 2 Measurement 3 Li, Jiang, Segev, Abramovsky, et al, Intel
July 2017 PAPR for 80 MHz Sounding 0.5 dB better than 11ax LTF and 3.5 dB better than fully random BPSK sounding sequence Li, Jiang, Segev, Abramovsky, et al, Intel
July 2017 PAPR for 40 MHz Sounding 0.3 dB worse than 11ax LTF and 2.8 dB better than fully random BPSK sounding sequence Li, Jiang, Segev, Abramovsky, et al, Intel
July 2017 PAPR for 20 MHz Sounding 0.3 dB worse than 11ax LTF and 2.5 dB better than fully random binary sounding Li, Jiang, Segev, Abramovsky, et al, Intel
July 2017 Summary MAC protection is insufficient for preventing Type B spoofing and PHY protection is needed Type B spoofing can be detected by using CSD unknown to spoofer Type B spoofing can be suppressed by using randomized sounding signal Sounding sequences, whose PAPRs are comparable to 11ax LTF, can be easily generated Li, Jiang, Segev, Abramovsky, et al, Intel
July 2017 Backup Li, Jiang, Segev, Abramovsky, et al, Intel
Random Sounding Sequence for 11mc Month Year doc.: IEEE 802.11-yy/xxxxr0 July 2017 Random Sounding Sequence for 11mc Replace VHT-LTF by a new VHTm-LTF BPSK sounding sequence is replaced Only 11mc devices can read VHT-SIG-B and DATA Although legacy devices can’t read the VHT-SIG-B and DATA, it should be fine NAV is usually set by legacy PPDU e.g. 11a RTS/CTS Legacy devices don’t need to read the duration field in the DATA L-STF L-LTF L-SIG VHT-SIG-A VHT-STF VTHm-LTF VHT-SIG-B DATA Li, Jiang, Segev, Abramovsky, et al, Intel Abramovsky, Ghosh, Segev & Li, Intel
Additional Suppression to Adversary July 2017 Additional Suppression to Adversary Instead of 1x LTF symbol duration, 4x LTF symbol duration may be used 6 dB processing gain Instead of 1 OFDM symbol, the random sounding signal may spread over 8 OFDM symbols 9 dB processing gain Li, Jiang, Segev, Abramovsky, et al, Intel