Choir: Empowering Low-Power Wide-Area Networks in Urban Settings - PowerPoint PPT Presentation

Choir: Empowering Low-Power Wide-Area Networks in Urban Settings Rashad Eletreby Diana Zhang, Swarun Kumar and Osman Yağan http://www.witechlab.com/LoRa/ChOIR.html 1

Imagine a world where every single object is connected to the Internet… Few kbps Several miles Simple and cheap 10 year battery away RF interface 2

Smart Infrastructure Smart Homes Smart Vehicles 3

Low-Power Wide-Area Networking (LP-WAN)

Low-Power Wide-Area Networking (LP-WAN) Long Range Low Data rate Low Cost Low Power • Up to 10 • Order of • < $5 • Up to 10 KMs in rural kilobits per years of areas second battery life Initiatives from Industry (LoRa, SIGFOX) and standardization bodies (3GPP LTEM, NBIoT)

Key Challenges 6

Key Challenges Interference Range Collisions emerge from the LPWAN ranges drop by 10x in sheer density of nodes and urban areas due to excessive the simplicity of the current multipath, shadowing, etc. MAC protocols (e.g., transmit as soon as wakeup) 7

Past work Wireless WiFi/Cellular sensor LPWANs networks LoRaWAN MegaMIMO Glossy SAM Sigfox ACR …. ZigZag …. …. 8

Choir Preserving Scalability Range simplicity • Decodes 10’s • Extends the • Fully of collided range of implemented transmissions teams of at a single- cooperating antenna base nodes station base station over an area of 10 Km 2 in Pittsburgh 9

: Chirps Chirp in T.D. Chirp on a spectrogram Data encoding 10

: 1-bit encoding ‘0’ ‘1’ In general, 𝒐 bits -> divide the BW to 𝟑 𝒐 initial frequencies 11

: 2-bit encoding 12

: 2-bit encoding 13

Choir in action Interference Range 14

Collision of chirps Different data + 15

Collision of chirps Same data 600 500 500 400 400 Abs. FFT Abs. FFT + 300 300 200 200 100 100 0 0 80 100 120 140 160 180 0 100 200 300 400 500 600 FFT Bin FFT Bin 16

Hardware imperfections 𝑔 + 𝜀𝑔 & 𝑔 𝑔 + 𝜀𝑔 ' Local oscillator mismatch 17

Hardware imperfections Packet 1 Packet 2 Sub-symbol timing offsets 18

Frequency Chirps are signals whose frequency increases linearly with time Time TO 19

Frequency An offset in time maps to an offset in FO frequency! Time TO 20

Frequency Two chirps with an offset in FO frequency! Time 21

Collision of chirps Same data 500 400 Abs. FFT 300 + 200 100 0 80 100 120 140 160 180 FFT Bin Hardware offsets! 22

23

Decoding data Symbol 1 U1 data: U2 data: 500 125 130 400 Abs. FFT 300 200 U1 data + U1 hardware offsets = 125 100 U2 data + U2 hardware offsets = 130 ! 0 90 100 110 120 130 140 150 160 170 FFT Bin 24

Decoding data Preamble Sym. 1 Sym. 2 … Sym. n Preamble Sym. 1 Sym. 2 Sym. n … Hardware offsets Peak locations are remain constant across used to estimate the packet hardware offsets U1 data + U1 hardware offsets = 125 Symbol 1: U2 data + U2 hardware offsets = 130 25

Decoding data Preamble Sym. 1 Sym. 2 … Sym. n Preamble Sym. 1 Sym. 2 Sym. n … Hardware offsets Peak locations are remain constant across used to estimate the packet hardware offsets How to measure accurate hardware offsets across the preamble? 26

Decoding data ' ∗ = 𝑏𝑠𝑕𝑛𝑗𝑜 {3 9 ∈ 3 9 67,3 9 87 } 𝑧𝐷 6& − ℎ & 𝑓 ?'@3 4 A + ℎ ' 𝑓 ?'@3 9 A ∗ , 𝑔 𝑔 & ' 4 ∈ 3 4 67,3 4 87 ,3 𝑔 B -> initial frequency offset estimate of user i ℎ B -> channel estimate of user i Δ -> bin size of the FFT 𝐷 6& -> conjugate nominal chirp 𝑧 -> received symbol ∗ 𝑔 -> correct frequency offset of user i B 27

Which peak corresponds to which user? 28

Which peak corresponds to which user? Symbol 1 Symbol 2 600 600 500 500 352.2 189.6 107.6 27.2 400 400 Abs. FFT Abs. FFT 300 300 200 200 100 100 0 0 0 100 200 300 400 500 600 0 100 200 300 400 500 600 FFT Bin FFT Bin 29

Which peak corresponds to which user? Symbol 1 Symbol 2 600 600 500 500 352.2 189.6 107.6 27.2 400 400 Abs. FFT Abs. FFT 300 300 200 200 100 100 0 0 0 100 200 300 400 500 600 0 100 200 300 400 500 600 FFT Bin FFT Bin 30

Which peak corresponds to which user? Symbol 1 Symbol 2 600 600 500 500 352.2 107.6 189.6 27.2 400 400 Abs. FFT Abs. FFT 300 300 200 200 100 100 0 0 0 100 200 300 400 500 600 0 100 200 300 400 500 600 FFT Bin FFT Bin User 1 User 2 Integer part depends on Fractional part depends both data and hardware only on hardware offsets offsets 31

We generalize this solution to account for… Near-far effect 1 Inter-symbol interference 2 Handling a general number of collisions 3 32

Choir in action Interference Range 33

Range Extension 62 °F 65 °F Each node is out-of-range! 67 °F 34

Range Extension 62 °F 65 °F Each node is out-of-range! 63 °F 67 °F 65 °F Can we exploit data correlations to obtain a coarse- grained view of the sensed data? 35

25 25 20 20 64.5 °F Abs. FFT Abs. FFT 15 15 Noise floor 65 °F Noise floor 10 10 5 5 62 °F 67 °F 0 0 0 100 200 300 400 500 600 0 100 200 300 400 500 600 FFT Bin FFT Bin Objective 36

25 20 Abs. FFT 15 Noise floor 65 °F 10 25 5 62 °F 67 °F 20 64.5 °F 0 0 100 200 300 400 500 600 Abs. FFT FFT Bin 15 Noise floor 10 Choir 5 0 0 100 200 300 400 500 600 Receive filter FFT Bin Approach 37

25 20 Abs. FFT 15 Noise floor 65 °F 10 25 5 62 °F 67 °F 20 64.5 °F 0 0 100 200 300 400 500 600 Abs. FFT FFT Bin 15 Noise floor 10 Choir 5 0 0 100 200 300 400 500 600 Receive filter FFT Bin 38

Implementation 39

Evaluation 40

Hardware offsets 1 0.8 0.6 CDF 0.4 0.2 Observed Ideal 0 0 20 40 60 80 100 120 140 Observed CFO+TO (Hz) Hardware offsets are truly diverse across LPWAN radios 41

Resolving interference 140000 Network Thrpt (bits/sec) Ideal 120000 100000 80000 60000 40000 20000 2 3 4 5 6 7 8 9 10 # Users 42

Resolving interference 140000 Network Thrpt (bits/sec) Ideal ALOHA 120000 100000 80000 60000 40000 20000 0 2 3 4 5 6 7 8 9 10 # Users 43

Resolving interference 140000 Network Thrpt (bits/sec) Ideal ALOHA 120000 ChOIR 100000 80000 60000 29x 40000 20000 0 2 3 4 5 6 7 8 9 10 # Users 44

Extending range Number of Range collaborating nodes 1 1 Km 10 2.5 Km 2.65X 30 2.65 Km 45

Conclusion Objective Platform Preserving Scalability Range Results simplicity • Decodes 10’s • Extends the • Fully of collided range of teams implemented transmissions of cooperating at a single- nodes antenna base station 46