Neuron-inspired maintenance-free, distributed sensing Challenges - - PowerPoint PPT Presentation

Neuron-inspired maintenance-free, distributed sensing – Challenges and algorithms

Stephan Sigg

Department of Communications and Networking Aalto University, School of Electrical Engineering stephan.sigg@aalto.fi NII, 24.02.2017

Stephan Sigg February 24, 2017 2 / 36

Stephan Sigg February 24, 2017 2 / 36

Stephan Sigg February 24, 2017 2 / 36

Stephan Sigg February 24, 2017 3 / 36

Stephan Sigg February 24, 2017 3 / 36

Stephan Sigg February 24, 2017 4 / 36

Group members and recent related research RF-based activity recognition Maintenance-free, intelligent distributed sensing Sensor graphs for distributed mathematical operation Probabilistic superimposed mathematical operations Neuron-inspired communication between distributed nodes Artificial neural computation from implicit channel inputs Conclusion

Stephan Sigg February 24, 2017 5 / 36

Stephan Sigg February 24, 2017 6 / 36

Exploiting the RF-channel for environmental preception

◮ Multi-path propagation ◮ Signal superimposition ◮ Scattering ◮ Signal Phase ◮ Reflection ◮ Blocking of signal paths ◮ Doppler Shift ◮ Fresnel effects

Stephan Sigg February 24, 2017 7 / 36

RF-based activity recognition

Sensewaves Video

Stephan Sigg February 24, 2017 8 / 36

– Video –

Stephan Sigg February 24, 2017 9 / 36

RF-based device-free activity recognition

L y i n g empty S t a n d i n g Crawling W a l k i n g

Stephan Sigg February 24, 2017 9 / 36

RF-based device-free activity recognition

L y i n g empty S t a n d i n g Crawling W a l k i n g

Stephan Sigg February 24, 2017 10 / 36

Monitoring attention from RF

Stephan Sigg February 24, 2017 10 / 36

Monitoring attention from RF

Stephan Sigg February 24, 2017 11 / 36

Situation and gestures from passive RSSI-based DFAR

10cm 10cm

Towards Away Hold over Open/close Take up Swipe bottom Swipe top Swipe left Swipe right Wipe No gesture

Stephan Sigg February 24, 2017 11 / 36

Situation and gestures from passive RSSI-based DFAR

10cm 10cm

Towards Away Hold over Open/close Take up Swipe bottom Swipe top Swipe left Swipe right Wipe No gesture

Stephan Sigg February 24, 2017 12 / 36

Group members and recent related research RF-based activity recognition Maintenance-free, intelligent distributed sensing Sensor graphs for distributed mathematical operation Probabilistic superimposed mathematical operations Neuron-inspired communication between distributed nodes Artificial neural computation from implicit channel inputs Conclusion

Stephan Sigg February 24, 2017 13 / 36

Energy-harvesting from Ambient RF noise

Efficiency: DC-conversion possible at about 70% efficiency1 7cm·7cm rectenna : transmissions at 0.2Hz for 3.4ms each2 0.5m2 rectenna : RF-activity at 20Hz for 300µs each

3

1Doan et al. ’Design and Fabrication of Rectifying Antenna Circuit for Wireless Power Transmission System Operating At ISM Band.’ International Journal of Electrical and Computer Engineering, 2016 2Nishimoto et al. ’Prototype implementation of ambient RF energy harvesting wireless sensor networks.’ IEEE Sensors, 2010. 3Song et al. ’On the use of the intermodulation communication towards zero power sensor nodes.’ EuMC 2013

Stephan Sigg February 24, 2017 14 / 36

Maintenance-free intelligent distributed sensing

Stephan Sigg February 24, 2017 15 / 36

Group members and recent related research RF-based activity recognition Maintenance-free, intelligent distributed sensing Sensor graphs for distributed mathematical operation Probabilistic superimposed mathematical operations Neuron-inspired communication between distributed nodes Artificial neural computation from implicit channel inputs Conclusion

Stephan Sigg February 24, 2017 16 / 36

Calculation during transmission on the channel

Envisioned paradigm shift in mobile computing

Parasitic operation Communication comes virtually for free Miniaturisation Processing and storage capabilities limited (passive, parasitic, backscatter)

Stephan Sigg February 24, 2017 16 / 36

Calculation during transmission on the channel

Envisioned paradigm shift in mobile computing

Parasitic operation Communication comes virtually for free Miniaturisation Processing and storage capabilities limited (passive, parasitic, backscatter) Potential: Trade processing load for communication load

◮ Shift computation towards the wireless

communication channel

Stephan Sigg February 24, 2017 16 / 36

Calculation during transmission on the channel

Envisioned paradigm shift in mobile computing

Parasitic operation Communication comes virtually for free Miniaturisation Processing and storage capabilities limited (passive, parasitic, backscatter) Potential: Trade processing load for communication load

◮ Shift computation towards the wireless

communication channel

◮ Computation below computational complexity

possible?

Stephan Sigg February 24, 2017 17 / 36

Calculation during transmission on the channel

Motivation: Computation during transmissiona

◮ Max. rate to compute & communicate functions ◮ Mention: Collisions might contain information

• aA. Giridhar and P

. Kumar, Toward a theory of in-network computation in wireless sensor networks, IEEE Comm. Mag., vol. 44, no 4, pp. 98-107, april 2006

Calculation of by means of post- and pre-processinga

◮ Requires accurate channel state information ◮ Requires identical absolute transmit power

• aM. Goldenbaum, S. Stanczak, and M. Kaliszan, On function computation via wireless sensor multiple-access

channels, IEEE Wireless Communications and Networking Conf., 2009

Stephan Sigg February 24, 2017 18 / 36

Calculation during transmission on the channel

Utilising Poisson-distributed burst-sequences

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

t K burst

superimposed received burst sequence transmit burst sequences time

Stephan Sigg February 24, 2017 18 / 36

Calculation during transmission on the channel

Utilising Poisson-distributed burst-sequences

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

t K burst

superimposed received burst sequence transmit burst sequences time

Basic operations Addition, subtraction, division and multiplication at the time of wireless data transmission via Poisson-distributed burst-sequences

Stephan Sigg February 24, 2017 18 / 36

Calculation during transmission on the channel

Utilising Poisson-distributed burst-sequences

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

t K burst

superimposed received burst sequence transmit burst sequences time

Addition Adding Poisson processes i with mean µi will result in a Poisson process with mean n

i=1 µi.

Stephan Sigg February 24, 2017 18 / 36

Calculation during transmission on the channel

Utilising Poisson-distributed burst-sequences

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

t K burst

superimposed received burst sequence transmit burst sequences time

Multiplication Applying logarithm laws allows multiplication

Stephan Sigg February 24, 2017 18 / 36

Calculation during transmission on the channel

Utilising Poisson-distributed burst-sequences

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

t K burst

superimposed received burst sequence transmit burst sequences time

Division From two nodes, one transmits the Numerator and

• ne the Denominator (fraction)

Stephan Sigg February 24, 2017 18 / 36

Calculation during transmission on the channel

Utilising Poisson-distributed burst-sequences

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

t K burst

superimposed received burst sequence transmit burst sequences time

Subtraction Combining division with logarithm laws allows subtraction (two nodes only)

Stephan Sigg February 24, 2017 19 / 36

Calculation during transmission on the channel

Errors for calculating during transmission on the wireless channel t = 106; κ = 103 10 nodes 20 nodes 30 nodes 40 nodes 50 nodes mean err .0322 .0466 .0609 .051 .0719 std-dev. .0232 .0368 .0536 .0336 .0438 max Ni 9 14 18.5 26 31 median T 2653.5 5161.5 7393 101816 124179 t = 107; κ = 103 10 nodes 20 nodes 30 nodes 40 nodes 50 nodes mean err .0049 .0176 .0402 .0475 .0781 std-dev. .0062 .0127 .0233 .0292 .0405 max Ni 12 18 23 27 31 median T 25708.5 52617.5 78502 101381 114348 t = 107; κ = 102 10 nodes 20 nodes 30 nodes 40 nodes 50 nodes mean err .0190 .1337 .2619 .4903 .6597 std-dev. .0107 .0358 .0591 .0708 .1129 max Ni 9.5 16 19 24 27 median T 24165 50037 71686.5 96829 114383

Stephan Sigg February 24, 2017 20 / 36

Calculation during transmission on the channel

Case study to compare the calculation accuracy

◮ Utilise data from the Intel Berkeley laboratory network

(here: temperature)4

◮ Transmission of data by simple sensor nodes

4http://db.csail.mit.edu/labdata/labdata.html

Stephan Sigg February 24, 2017 21 / 36

Calculation during transmission on the channel

Offline Online

Offline Online

Stephan Sigg February 24, 2017 22 / 36

Further mathematical operations

Utilising the mean of the minimum of a convolution

◮ Exploiting the CDF of the minimum of a distribution, further

• perations are possible

◮ √n ◮ dn ◮ . . .

Stephan Sigg February 24, 2017 23 / 36

Environmental perception with CRFs

Stephan Sigg February 24, 2017 24 / 36

Group members and recent related research RF-based activity recognition Maintenance-free, intelligent distributed sensing Sensor graphs for distributed mathematical operation Probabilistic superimposed mathematical operations Neuron-inspired communication between distributed nodes Artificial neural computation from implicit channel inputs Conclusion

Stephan Sigg February 24, 2017 25 / 36

Neural communication for sensor networks

Stephan Sigg February 24, 2017 25 / 36

Neural communication for sensor networks

Problem

◮ Communication in sensor networks is omnidirectional ◮ In neural networks, the missing of edges is vital for the

network’s computational power

Stephan Sigg February 24, 2017 25 / 36

Neural communication for sensor networks

Proposal

◮ Transmit beamforming to establish dedicated links

Stephan Sigg February 24, 2017 26 / 36

Example closed-loop carrier synchronization

Receiver Transmitter Receiver Transmitter Receiver Transmitter

Source Source Source Source Source

common master beacon to all source nodes Receive node broadcasts Receiver Transmitter Receive nodes bounce the beacon back on distinct CDMA channels phase offset of each node on these CDMA channels Receiver transmits the relative Synchronised nodes transmit as a distributed beamformer to the receiver

5 ◮ Too computationally expensive for parasitic operation

• 5Y. Tu and G. Pottie, Coherent Cooperative Transmission from Multiple Adjacent Antennas to a Distant

Stationary Antenna Through AWGN Channels, Proceedings of the IEEE VTC, 2002

Stephan Sigg February 24, 2017 27 / 36

Example open-loop carrier synchronization

◮ Too computationally expensive for parasitic operation

Stephan Sigg February 24, 2017 28 / 36

Feedback-based open-loop carrier synchron.

f2t + γ2 2 π f1t + γ1 2 π

i

fn 2 π

it+γn

i+1

fn 2 π

i+1

t+γn

Iteration i+1 F r e q u e n c y Time Mutation Iteration i

1 2 3 4 1

Superimposed received sum signal Receiver feedback

6, 7

• 6R. Mudumbai, G. Barriac and U. Madhow, On the feasibility of distributed beamforming in wireless networks,

IEEE Transactions on Wireless Communications, 2007 7Sigg, El Masri and Beigl, A sharp asymptotic bound for feedback based closed-loop distributed adaptive beamforming in wireless sensor networks, IEEE Transactions on Mobile Computing, 2013

Stephan Sigg February 24, 2017 29 / 36

Feedback-based open-loop carrier synchronizat.

◮ Weak multimodal fitness function ◮ Single local=global optimum

e

j(2π f t +γi)

i

cos( ) ϕ

i

i i

e

j ( +γi) 2π f t γ i 1 ϕ

i

−δ δi

i

ϕ

e

j ( 2π f +γ t ) j ϕ cos( ) G a i n

Stephan Sigg February 24, 2017 30 / 36

Feedback-based open-loop carrier synchron.

Stephan Sigg February 24, 2017 30 / 36

Feedback-based open-loop carrier synchron.

Stephan Sigg February 24, 2017 30 / 36

Feedback-based open-loop carrier synchron.

Stephan Sigg February 24, 2017 31 / 36

Stephan Sigg February 24, 2017 32 / 36

Group members and recent related research RF-based activity recognition Maintenance-free, intelligent distributed sensing Sensor graphs for distributed mathematical operation Probabilistic superimposed mathematical operations Neuron-inspired communication between distributed nodes Artificial neural computation from implicit channel inputs Conclusion

Stephan Sigg February 24, 2017 33 / 36

ANN computation from implicit channel inputs

Stephan Sigg February 24, 2017 33 / 36

ANN computation from implicit channel inputs

Stephan Sigg February 24, 2017 33 / 36

ANN computation from implicit channel inputs

Stephan Sigg February 24, 2017 33 / 36

ANN computation from implicit channel inputs

hk(− → x , − → w ) = f (3)

act

 

D2

• j=1

w(2)

jk f (2)

act

 

D1

• i=1

w(1)

ij

xi + w(1)

0j

  + w(2)

0k

 

Stephan Sigg February 24, 2017 34 / 36

Group members and recent related research RF-based activity recognition Maintenance-free, intelligent distributed sensing Sensor graphs for distributed mathematical operation Probabilistic superimposed mathematical operations Neuron-inspired communication between distributed nodes Artificial neural computation from implicit channel inputs Conclusion

Stephan Sigg February 24, 2017 35 / 36

Maintenance-free intelligent distributed sensing

Stephan Sigg February 24, 2017 36 / 36

Thank you!

Stephan Sigg stephan.sigg@aalto.fi