Presentation is loading. Please wait.

Presentation is loading. Please wait.

Analog Network Coding Sachin Katti Shyamnath Gollakota and Dina Katabi.

Similar presentations


Presentation on theme: "Analog Network Coding Sachin Katti Shyamnath Gollakota and Dina Katabi."— Presentation transcript:

1 Analog Network Coding Sachin Katti Shyamnath Gollakota and Dina Katabi

2 Current Wireless Router C

3 Current Wireless Traditional Routing requires 4 time slots C

4 Router Current Wireless Traditional Routing requires 4 time slots C

5 XOR = Router Traditional Routing requires 4 time slots Last Year  Network Coding  COPE C

6 Router Traditional Routing requires 4 time slots C

7 Last Year  Network Coding  COPE XOR = = Router Traditional Routing requires 4 time slots COPE requires 3 time slots  Higher throughput Can we do it in 2 time slots? C

8 Analog Network Coding (ANC) Instead of router mixing packets… Exploit that the wireless channel naturally mixes signals

9 Analog Network Coding Router C

10 Analog Network Coding 1)Dina and Jon transmit simultaneously Interference C

11 Router Analog Network Coding 1)Dina and Jon transmit simultaneously 2)Router amplifies and broadcasts interfered signal C

12 Router Analog Network Coding 1)Dina and Jon transmit simultaneously 2)Router amplifies and broadcasts interfered signal 3)Dina subtracts known signal from interfered signal C

13 Router Analog Network Coding 1)Dina and Robert transmit simultaneously 2)Router amplifies and broadcasts interfered signal 3)Dina subtracts known signal from interfered signal Analog Network Coding requires 2 time slots  Higher throughput Analog Network Coding requires 2 time slots  Higher throughput C

14 It Is More Than Going From 3 To 2! Philosophical shift in dealing with interference Strategically exploit interference instead of avoiding it Promises new ways of dealing with hidden terminals

15 C C CC Hidden Terminal Scenario R1 R2 Src Dst

16 P1 Hidden Terminal Scenario C C CC R1 R2 Src Dst

17 P2 Hidden Terminal Scenario P1 1)Src and R 2 transmit simultaneously C C CC R1 R2 Src Dst

18 Hidden Terminal Scenario 1)Src and R 2 transmit simultaneously 2)R 1 subtracts P 1, which he relayed earlier to recover P 2 that he wants P1 P2 C C CC R1 R2 Src Dst

19 Hidden Terminal Scenario R2 and Src are hidden terminals Today : Simultaneous transmission  Collision ANC : Simultaneous transmission  Success! P1 P2 C C CC R1 R2 Src Dst

20 Hidden Terminal Scenario Other Benefits of ANC: First step toward addressing hidden terminals ANC extends network coding to new scenarios Other Benefits of ANC: First step toward addressing hidden terminals ANC extends network coding to new scenarios C C CC R1 R2 Src Dst

21 How do we make it work?

22 Practical Challenges Interfered signal is not exactly the sum Channel distorts signals Two signals are never synchronized It is not s D (t) + s J (t) but f 1 (s D (t)) + f2(s J (t-T)) Prior work assumes full synchronization and ignores channel distortion Not Practical! C

23 Key Idea: Exploit Asynchrony!

24 Time Signal No Interference Dina uses interference-free parts to estimate channel and timing Dina compensates for her interfering signal Jon runs the same algorithm backwards! Exploit asynchrony to make it practical

25 Cross layer realization of our idea

26 Router senses idle medium and broadcasts a trigger to Dina and Jon Dina and Jon jitter their start times randomly and transmit Router amplifies and forwards interfered signal Dina and Jon receive and decode Protocol How do they decode?

27 Primer on Modulation Nodes transmit vectors on channel Focus on MSK (Minimum Shift Keying) modulation D2 lags D 1 by 90 degrees  Bit “0” D2 D1 D2 D1 D2 leads D 1 by 90 degrees  Bit “ 1 ”

28 Primer on Channel Effects Attenuation D2 and D1 are attenuated by the same amount Channel D2 D1 D2 D1

29 Primer on Channel Effects Attenuation Rotation Channel D2 D1 D2 D1

30 Primer on Channel Effects Attenuation Rotation Angle between vectors is preserved Channel D2 D1 D2 D1 To decode, receiver computes angle between received vectors Angle (D2, D 1) = 90 degrees  Bit “ 1 ” was transmitted

31 So, How Does Dina Decode? Time Signal No Interference Dina’s Signal Jon’s Signal

32 Time Signal Small uninterfered part at the start Decodes uninterfered part via standard MSK demodulation Once interference starts, Dina changes decoding algorithm Interference So, How Does Dina Decode? No Interference

33 What did Dina send? D1 D2

34 What did Dina send? What did Jon send? D1 D2 J1 J2

35 What is Interference  Vector addition J1 J2 X1 X2 D1 D2

36 D1 What does Dina know? J1 J2 X1 X2 D1 D2

37 D1 What does Dina know? Amplitude of her Vectors  α Amplitude of Jon’s Vectors  β No Interference X1 X2 α β

38 D1 D2 Dina finds solutions for X 1 and X2 What does Dina know? D1’ J1’ J1 D1 X1 X2

39 D1 D2 Dina finds solutions for X 1 and X2 What does Dina know? Two solutions for each interfered vector! D1’ J1’ J1 D1 X1 X2 J2 D2 D2’ J2’

40 D1 D2 Dina finds solutions for X 1 and X2 What does Dina know? Four possible angles! D1’ J1’ J1 D1 X1 X2 J2 D2 D2’ J2’

41 D1 D2 Dina finds solutions for X 1 and X2 What does Dina know? Four possible angles! D1’ J1’ J1 D1 X1 X2 J2 D2 D2’ J2’

42 D1 D2 Dina finds solutions for X 1 and X2 What does Dina know? Four possible angles! D1’ J1’ J1 D1 X1 X2 J2 D2 D2’ J2’

43 D1 D2 Dina finds solutions for X 1 and X2 What does Dina know? Four possible angles! D1’ J1’ J1 D1 X1 X2 J2 D2 D2’ J2’

44 D1 D2 Dina finds solutions for X 1 and X2 What does Dina know? Four possible angles! D1’ J1’ J1 D1 X1 X2 J2 D2 D2’ J2’

45 D1 D2 Dina finds solutions for X 1 and X2 What does Dina know? Pick the correct angle  90 degrees D1’ J1’ J1 D1 X1 X2 J2 D2 D2’ J2’

46 D1 D2 Dina finds solutions for X 1 and X2 What does Dina know? Pick the correct angle  +90 degrees D1’ J1’ J1 D1 X1 X2 J2 D2 D2’ J2’

47 D1 D2 Dina finds solutions for X 1 and X2 What does Dina know? Dictates solution for Jon’s vectors! D1’ J1’ J1 D1 X1 X2 J2 D2 D2’ J2’

48 D1 D2 What does Dina know? Dina finds angle between J 1 and J2 and decodes J1 D1 X1 X2 J2 D2

49 Decoding Algorithm – Decoding interference Time Signal Interference Decode rest of the interfered part using this algorithm Decode final uninterfered part from Jon via standard MSK demodulation

50 Performance

51 ANC Implementation Software – GNURadio codebase Hardware – USRP frontend GHz frequency range SNR of dB Canonical topologies in mesh networks

52 Dina and Jon Router ANC throughput gain over current 4/2 = 2 ANC throughput gain over COPE 3/2 = 1.5 C

53 Throughput gain for Dina-Jon scenario Throughput gain CDF Median Gain over Routing – 70% Gain over Routing

54 Throughput gain for Dina-Jon scenario Throughput gain CDF Gain over Routing Gain over COPE Median Gain over Routing – 70% Median Gain over COPE – 30% Median Gain over Routing – 70% Median Gain over COPE – 30%

55 X topology Router Capture! Interference C

56 X topology Router Capture! Interference C

57 X topology Router ANC throughput gain over current 4/2 = 2 ANC throughput gain over COPE 3/2 = 1.5 ANC decodes interference using overheard signals C

58 Throughput gain – X topology Throughput gain CDF Gain over Routing Median Gain over Routing – 65%

59 Throughput gain – X topology Throughput gain CDF Gain over Routing Gain over COPE Median Gain over Routing – 65% Median Gain over COPE – 28% Median Gain over Routing – 65% Median Gain over COPE – 28%

60 Chain topology ANC throughput gain over current 3/2 = 1.5 C C CC R1 R2 Src Dst

61 Throughput gain – Chain topology Throughput gain CDF Median Gain over Routing – 37%

62 Conclusion Shifts in the design of wireless networks to recognize wireless for what it is Embrace Broadcast Embrace Interference Implementation that yields large throughput gains


Download ppt "Analog Network Coding Sachin Katti Shyamnath Gollakota and Dina Katabi."

Similar presentations


Ads by Google