Opportunities in Communication

Transcription

1 Opportunities in Communication J OHN M. CIOFFI Hitachi Professor Emeritus of Engineering

2 Communication uses and applications April 20,

3 Basic Communication (digital) Communication is fundamental to all we do (information is the natural resource) Up there with water, electricity, food, Sender chooses a message to send (can be from large set) Channel will distort this message (maybe a lot or maybe only slightly) Receiver attempts to decide what the message was (mathematically detection ) Talking, texting, sending/viewing video, searching on internet, facial recognition, radar, lidar,. April 20, 2018 EE 102 3

4 Traditional Internet Service Provider Comm Channels Messages Internet Text video, audio Sensor/camera images Channel Ø Fiber, copper Ø Wireless Ø Many or single Ø Mesh Network edge OSI =? Open Systems Interconnect How much money spent on connection Subscriptions by consumers globally? $1.3T April 20, 2018 EE 102 4

5 Traditional Mobile and multi-path This could be a reflection Also (building, mountain, etc) Any one spot a problem? Delayed signal could be 180 degrees out of phase at some frequencies What happens then? Solution: Exploit the dimensions use/leverage other frequencies that reinforce (frequency dimensions) Look only in one direction (spatial dimensions) April 20, 2018 EE 102 L1: 5

6 Not so traditional Search Engine Message set Layer 7 = Layer 1? Channel output Receiver Estimated message Yes, a search engine is also receiver/detector decides which pages to display ( page rank ) Ø Searches among many possible messages The channel output is the typed phrase Ø That channel-output phrase for this user has a message (page or group of pages) detected for it Dimensions are the words, phrases, context, etc (along with some history and user knowledge or cookies ) The ultimate receiver is the woman herself she decides to click on one of the displayed pages iterative decoding the system may refine the decision with additional passes (more dimensions added) April 20, 2018 EE 102 L1: 6

7 Self-Driving Cars Messages to be estimated Ø Relative position Ø Chan out = Lidar reflection Ø delay = distance = position Ø Route Ø Channel out = destination Ø Location: current, anticipated Ø Chan out = GPS coordinates, recent locations Ø Message is the position Ø Driver ID Ø Chan out facial image, finger prints (cameras in or out) Ø Time Ø Chan out = electronic ref, GPS, base station signal Ø Info to/from other vehicles Ø Chan out = Wi-Fi, Bluetooth, other So self-driving cars have many communication channels and receivers Ø Multi-dimensional in many ways April 20, 2018 EE 102 L1: 7

8 Drones/Satelites and Communication Messages Ø Ø Ø Ø Ø s/web pages Tests/chats images Video Audio Channel Ø air Ø Moving fast Ø attenuates Ø noise Dimensions are in time, possibly space (multiple antennas) April 20, 2018 EE 102 L1: 8

9 3D Cameras (images and videos) Messages = Ø Ø Ø Ø Sporting event Traffic conditions Security/intruders People/faces CHANNEL OUTPUTS Visible spectrum Obstacles (opaque) affect this Multiple paths 2D arrays of pixels Multiple places Who is in the photo? April 20, 2018 EE 102 9

10 Manufacturing/Factories Messages Ø Status of widget Ø Test results Ø Position of machines Ø Position of widget Ø faults Channel Ø Wired and/or wireless Ø Interference from others Ø Sensor and other low-power limitations April 20, 2018 EE

11 Health Care Messages Ø Diagnostics Sensors around the body Sensors in the body Channel.. could be Ø Ø Ø Ø Half-way around world From internal organ to cellphone Unusual propagation Very high frequencies within body Ø Or very low frequencies (pills with radios) April 20, 2018 EE

12 Agriculture Feed the world better Message Ø Plant status Ø Animal status Ø Water, fertilizer levels Ø Machine position Channel Ø wireless Ø To drone, satellite, local AP April 20, 2018 EE

13 Smart Cities and/or Smart Homes Messages appliance status, typical internet messages, Channel wireless and wires, many paths April 20, 2018 EE

14 Consumer /User/Thing Happiness? a state of uncontrollable fury or violent anger induced by the delayed or interrupted enjoyment of streaming video content from over-the-top (OTT) services. Over 50% of under-35 internet consumers experience it ~ daily Despite all our communication algorithms/knowledge Tech Mahendra 2016 April 20, 2018 EE

15 Fifth-Generation Communcations: Application trends (or verticals ) Self-Driving Auto Rage coming? ~ Patient Rage even worse? Currently Telecom Services is $2T/year Global industry Security as service, privacy as service? What happens if these have outage? April 20, 2018 EE

16 Basic Communication April 20,

17 3 Basic Problems to Solve CHANNEL IDENTIFICATION what kind of limits does it have, does it vary, vary with what? CODING - What is a good (best) set X for a given channel? DETECTION What is a good (best) receiver for deciding X? The same 3 problems occur in Machine Learning April 20, 2018 EE 102 L1: 17

18 Simple Additive White Gassian Noise Channel X Detection Problem First n y 16QAM - 3d 2 3d 2 4QAM d 2 - d 2 - d 2 d 2 3d 2 + y ˆX - 3d 2 Pick closest point ForAWGN Maximum Likelihood Uniform input distn 4 messages or 16 messages QAM à 2 dimensional Equally Likely Cos / Sine Add noise Zero mean Variance Pick closest point is maybe obvious, But also mathematically optimum April 20, 2018 EE 102 L1: 18

19 The Intersymbol Interference (ISI) Channel The ISI channel with additive white Gaussian Noise (AWGN) n t ( ) x k p(t) + y( t ) Every T seconds, a symbol is transmitted the channel stretches them and causes intersymbol interference (ISI) Most detection designed for no band limitations, but how about this bandlimited ISI Channel? Channel sums message with.9 x message delayed T sec 1 2T Equalization?? 1 2T Enhances the noise April 20, 2018 EE 102 L1: 19

20 Multiple carriers/dimensions in frequency water-filling How do we learn and adjust Ø Dynamically Some of very first AI methods in com (from Stanford) Use bit-swapping AI/ML Method SU patent #3 in Engineering (#7 overall) April 20, 2018 EE

21 2 x 2 Antenna System (MIMO) x 1 Wireless Channel P y 1 x 2 y 2 σ 2 =.1 There is crosstalk between dimensions (kind of like ISI, just different dimensions) Can actually double the data rate if right signal processing used (and antennas are not too close) April 20, 2018 EE

22 Multiple directions in Space Best Energy will also be water-fill over The singular vectors of the channel Essentially matrix form of machine learning From earlier April 20, 2018 EE 102 L1: 22

23 MU-MIMO (multi-user) involves more learning/factoring Virtually all communication problems can be cast into the Same theory of finding the critical modes of channel April 20, 2018 EE 102 L1: 23

24 An Example from the field April 20,

25 A Network State Machine Net may be in a state Mbps clean Mbps interference GYR Ø More detail Mbps clean 5-12 Mbps Weak signal Mbps interference 5-12 Mbps Weak and Int Relates to more traditional models 2-6 Mbps Low signal 2-6 Mbps Bad int 2-6 Mbps both 1-2 Mbps ~ nothing 1-2 Mbps Real Bad 1-2 Mbps Ugly April 20, 2018 EE

26 Wi-Fi and QoE (the desired response) Strong correlation between WiFi problems and Complaint Rate 48% of APs have WiFi issues Customers identified as having poor Wi-Fi performance are 5 times more likely to complain 20.00% 18.00% 16.00% 14.00% 12.00% 10.00% 8.00% 6.00% 4.00% 2.00% 0.00% Customer Complaint Rate No Problem Stable WiFi Needs Attention Selling more mesh points to ISP often increases instability percentage WiFi problems not quite unstable GYR 26April 20, 2018 EE 102

27 Root Causes of Wi-Fi QoE Degradation 2.4GHz Interference Coverage Noise Top Three Wi-Fi Problem Drivers Ø Correlated with the main User QoE April 20, 2018 EE

28 Wi-Fi Hot Spots and Problems Coverage (noise) & signal strength Crosstalk = who controls phy April 20, 2018 EE

29 Cocktail Party Effect (crosstalk) TALK LOUDER Sorry, can t hear, Talk louder OK, I ll SHOUT Solution: All speak politely at low volume (lower power) Ø All send more information (more power and/or higher data rate) This can be learned and optimized I NEED TO TALK VERY LOUD EG: Wi-Fi Box/Chips blasting at 1-10 Gbps! Ø Or worse yet install repeaters/mesh and have them all blast Ø Or use MU-MIMO when 3 people are collinear (BAD) April 20, 2018 EE

30 Channel-opt versus top-selling chip auto-select Chip X s Auto-Channel vs ML Cloud Optimization Auto Channel ML Optimization 14% 43% 9% 22% 12% 17% 54% After ML Optimization 5% 7% 17% 0-3M 3-5M 5-10M 10-20M Throughput 62% reduction in Wi-Fi below 3Mbps 45% reduction in Wi-Fi below 5Mbps Pre-Optimization Population % Post-Optimization population % %Gain 3M 12.3% 4.7% 61.5% 5M 20.9% 11.4% 45.5% 10 M 42.8% 28.6% 33.3% April 20, 2018 EE

31 Optimization, QoE, and Stability based on several hundred million daily internet users Consumer Survey Did your overall performance improve with Booster? management? 40% Yes No 53% Neutral / Not sure 8% Machine-Learned Stability United Kindom France USA Canada Hungary Germany Romania Czech Republic Unmanaged Internet connections Ø Globally ~20% of unmanaged fixed connections are unstable each month! Dynamic (optimized) Managed Connections Ø Ø Often see the above numbers drop by ½ or more From experience both in electronic measures and consumer feedback 0% 20% 40% 60% 80% Stability improvement April 20, 2018 EE

32 QoE Time Trend learned (300k IPTV customers) 100% 80% 60% 40% 20% 0% QoE Tracker 3/1/2016 4/1/2016 5/1/2016 6/1/2016 7/1/2016 8/1/2016 9/1/ /1/ /1/ /1/2016 1/1/2017 2/1/2017 Note the dynamic learning and adjusting ongoing Bad QoE Warning QoE Good QoE April 20, 2018 EE

33 Data Wars Game Uses FM loading on several cross-talking links/connections Signal from other user is treated as Gaussian noise x 1 Wireless Channel P y 1 x 2 y 2 If Link 2 could use lower energy, then Link 1 sees less noise à Link 1 loads to higher data rate But Link 2 might want Link 1 to save energy so it can transmit faster Who wins sometimes called the prisoner s dilemma in game theory April 9, 2018 EE 102 L4: 33

34 Example for 25 users all sharing same spatial channel 25 Crosstalking systems versus connection distance 25 users all sharing same spectrum Iter-W-Fill all-max rate Near-end ( echo ) as well grows with f 1.5 and far-end ( xtalk ) attenuates with same transfer but has additional f 2 coupling. Each effects the other 24 users All are of same distance between xmit/rcvr Speeds roughly double connection distance April 9, 2018 EE 102 L4: 34

35 Spectra after convergence 4000 The echo-crosstalk is strong see FDM effect between down and up UPLINK It was all learned by water-filling No clever designer No FCC / regulator Called cognitive radio by some DOWNLINK Not best, but pretty good! Can work with different topology, rates, etc How s that for Machine Learning, eh? April 9, 2018 EE 102 L4: 35

36 April 20, 2018 Virtualization Software Defined Network

37 Converged Virtual 5G networks (fixed & wireless lines) Virtual Network Operators Need functions (VNFs) Infrastructure Provider Software- Defined-Network (SDN) Single infrastructure shared by multiple operators April 20, 2018 EE

38 X-Haul basically allows for just ADC/DAC at antennas Increasingly everything in software Ø For most/all connections 5GPP X-Haul Group April 20, 2018 EE

39 The Opportunity 4B people using internet à hopefully all Ø They all need, and increasingly depend seriously, on the connection It affects all applications Ø The new hot ones of today Ø The old voice, video Ø The ones we don t know yet These networks and problems will always be there Ø And need help, design, improvement from EE s like those in this class! April 20, 2018 EE

40 Thank you J OHN M. CIOFFI Hitachi Professor Emeritus of Engineering