Cognitive Wireless Networking in the TV Bands

Cognitive Wireless Networking in the TV Bands

Networking Devices over White Spaces Ranveer Chandra Collaborators: Thomas Moscibroda, Rohan Murty, Victor Bahl, Srihari Narlanka Wi-Fis Success Story Wi-Fi is extremely popular (billion $$ business) Enterprise/campus LANs, Home networks, Hotspots Why is Wi-Fi successful Wireless connectivity: no wires, increased reach Broadband speeds: 54 Mbps (11a/g), 200 Mbps (11n) Free: operates in unlicensed bands, in contrast to cellular Problems with Wi-Fi Poor performance:

Contention with Wi-Fi devices Interference from other devices in 2.4 GHz, such as Bluetooth, Zigbee, microwave ovens, Low range: Can only get to a few 100 meters in 2.4 GHz Range decreases with transmission rate Overcoming Wi-Fis Problems Poor performance: Fix Wi-Fi protocol several research efforts (11n, MIMO, interference cancellation, ) Obtain new spectrum? Low range: Operate at lower frequencies? Higher Frequency

Analog TV Digital TV USA (2009) Japan (2011) Canada (2011) Broadcast TV UK (2012) China (2015) . Wi-Fi (ISM) . .. 5 What are White Spaces? Wireless Mic

TV 0 54-88 170-216 470 700 MHz50 TV Channels -60 ISM (WiFi) 2400 2500 5180 5300 White spaces 7000 MHz

Each channel is 6 MHz wide dbm FCC Regulations* TV Stations in America Sense TV-100 stations and Mics Frequency 700 MHz 470 MHz Portable devices on channels 21 - 51 White Spaces are Unoccupied TV Channels 6 Why should we care

about White Spaces? 7 The Promise of White Spaces TV 0 54-90 174-216 470 MHz Wireless Mic ISM (WiFi) 2400 2500 700

5180 5300 7000 MHz Up to 3x of 802.11g More Spectrum Longer Range } Potential Applications

at least 3 - 4x of Wi-Fi Rural wireless broadband City-wide mesh .. .. 8 Goal: Deploy Wireless Network Base Station (BS) Good throughput for all nodes Avoid interfering with incumbents 9 Why not reuse Wi-Fi

based solutions, as is? 10 White Spaces Spectrum Availability Fraction of Spectrum Segments 0.8 Urban 0.7 Differences from ISM(Wi-Fi) 0.6 Suburban

0.5 Rural Fragmentation Variable channel widths 0.4 0.3 0.2 1 2 0.13 4 5 0 1 1 2 3 4 5 2

3 4 5 6 >6 # Contiguous Channels Each TV Channel is 6 MHz wide Spectrum Use is Fragmented multiple channels for more bandwidth

11 White Spaces Spectrum Availability Differences from ISM(Wi-Fi) Fragmentation Variable channel widths Spatial Variation Cannot assume same channel free everywhere 1 2 3 4 5 1 2 3 4 5 TV Tower

Location impacts spectrum availability Spectrum exhibits spatial variation 12 White Spaces Spectrum Availability Differences from ISM(Wi-Fi) Fragmentation Variable channel widths Spatial Variation Cannot assume same channel free everywhere 1 2 3 4 5 1 2 3 4 5 Temporal Variation

Same Channel will not always be free Any connection can be disrupted any time Incumbents appear/disappear over time Must reconfigure after disconnection 13 Cognitive (Smart) Radios Frequency Signal Strength Signal Strength 1. Dynamically identify currently unused portions of spectrum 2. Configure radio to operate in available spectrum band

take smart decisions how to share the spectrum Frequency Networking Challenges The KNOWS Project (Cogntive Radio Networking) How should nodes connect? Which spectrum-band should two cognitive radios use for transmission? 1. Frequency? 2. Channel Width? 3. Duration? How should they discover one another? Need analysis tools to

reason about capacity & overall spectrum utilization Which protocols should we use? MSR KNOWS Program Prototypes Version 1: Ad hoc networking in white spaces Capable of sensing TV signals, limited hardware functionality, analysis of design through simulations Version 2: Infrastructure based networking (WhiteFi) Capable of sensing TV signals & microphones, deployed in lab Version 3: Campus-wide backbone network (WhiteFi + Geolocation) Deployed on campus, and provide coverage in MS Shuttles

Version 2: WhiteFi System Prototype Hardware Platform Base Stations and Clients Algorithms and Implementation Discovery Spectrum Assignment Handling Disconnections Evaluation Deployment of prototype nodes Simulations 17 Hardware Design Send high data rate signals in TV bands

Wi-Fi card + UHF translator Operate in vacant TV bands Detect TV transmissions using a scanner Avoid hidden terminal problem Detect TV transmission much below decode threshold Signal should fit in TV band (6 MHz) Modify Wi-Fi driver to generate 5 MHz signals Utilize fragments of different widths Modify Wi-Fi driver to generate 5-10-20-40 MHz signals KNOWS Platform: Salient Features Can dynamically adjust channel-width and center-frequency. Low time overhead for switching

can change at fine-grained time-scale Transceiver can tune to contiguous spectrum bands only! Frequency Changing Channel Widths Scheme 1: Turn off certain subcarriers ~ OFDMA 10 20 MHz Issues: Guard band? Pilot tones? Modulation scheme? Changing Channel Widths Scheme 2: reduce subcarrier spacing and width! Increase symbol interval

10 20 MHz Properties: same # of subcarriers, same modulation Adaptive Channel-Width Why is this a good thing? 1. Fragmentation 5Mhz 20Mhz Frequency White spaces may have different sizes Make use of narrow white spaces if necessary

2. Opportunistic, load-aware channel allocation Few nodes: Give them wider bands! Many nodes: Partition the spectrum in narrower bands KNOWS White Spaces Platform Windows PC TV/MIC detection Scanner (SDR) FFT Net Stack FPGA

UHF RX Daughterboard Whitespace Radio Connection Manager Atheros Device Driver Wi-Fi Card UHF Translator Variable Channel Width Support 25

WhiteFi System Challenges Fragmentation Spatial Variation Temporal Variation Impact Discovery Spectrum Assignment Disconnection 26

Discovering a Base Station Discovery Problem 1 2 3 Goal 4 5 Quickly find channels 1 BS is 2 using 3 4 5 Discovery Time = (B x W) B x W) Fragmentation Try center channel

and widths How does thedifferent new client discover BS and Clients must use same channels Can we optimize this discovery

time? channels used by the BS? 27 Whitespaces Platform: Adding SIFT PC TV/MIC detection Net Stack Scanner (SDR) FFT Temporal Analysis (SIFT)

FPGA UHF RX Daughterboard Whitespace Radios Connection Manager Atheros Device Driver Wi-Fi Card UHF Translator SIFT: Signal Interpretation before Fourier Transform

28 SIFT, by example 10 5 MHz MHz SIFT SIFT Does not decode packets Pattern match in time domain Amplitude ADC

BeaconData Beacon ACK SIFS Time 29 BS Discovery: Optimizing with SIFT 1 2 3 4 5 1 2 3 4 5 Amplitude 18 MHz

Matched against 18 MHz packet signature Time SIFT enables faster discovery algorithms 30 BS Discovery: Optimizing with SIFT Linear SIFT (L-SIFT) 1 2 3 4 5 Jump SIFT (J-SIFT) 1 2 3 4 5 6 7 8 31 D is co ve ry T im e R a ti o (co m p a re d to b a s e lin e )

Discovery: Comparison to Baseline Baseline =(B x W) B x W) L-SIFT = (B x W) B/W) J-SIFT = (B x W) B/W) 1 0.9 Linear-SIFT 2X reduction 0.8 Jump-SIFT 0.7 0.6

0.5 0.4 0.3 0.2 0.1 0 0 30 60 90 120 150

180 White Space - Contiguous Width (MHz) 32 WhiteFi System Challenges Fragmentation Spatial Variation Temporal Variation Impact Discovery

Spectrum Assignment Disconnection 33 Channel Assignment in Wi-Fi 1 6 11 1 6

11 Fixed Width Channels Optimize which channel to use 34 Spectrum Assignment in WhiteFi Spectrum Assignment Problem Goal Maximize Throughput Include Spectrum at clients 1 2 3 4 5 Assign

1 2 3 4 5 Center Channel & Width Fragmentation Optimize for both, center channel and width Spatial Variation BS must use channel iff free at client 35 Accounting for Spatial Variation 1 2 3 4 5 1 2 3 4 5

1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 = 1 2 3 4 5 36 Intuition

Intuition Use widest possible channel BS But Limited by most busy channel 1 2 3 4 5 Carrier Sense Across All Channels All channels must be free BS(2 and 3 are free) = BS(2 is free) x BS(3 is free) Tradeoff between wider channel widths and opportunity to transmit on each channel 37

3.5 3 2.5 2 1.5 1 0.5 0 M Cham -value T h ro u g h p u t ( M b p s ) Multi Channel Airtime Metric (MCham) 20 Mhz 5 MHz

10 MHz W BS ( c ) MChamn (F, W) = n 5 Mhz c( F ,W ) 0 10 20 30 40 Background1 traffic 3 4 delay

5 (ms) 2 - Packet 50 Pick (F, W) that maximizes (N * MChamBS + nMCham ) 2.5 n 1 20 Mhz 10 MHz (c) = Approx. opportunity

node n will (2) n BS(2)5 MHz Free BS Air Time on Channel 2 2 Contention BS(2) = Max (Free Air Time on 2, 1/Contention) 1.5 get to transmitchannel

on channel c 1 0.5 0 0 5 10 15 20 25

30 35 40 45 50 Background traffic - Packet delay (ms) 38 WhiteFi Prototype Performance Throughput (Mbps) 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40

5 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 WhiteFi 0 15

30 45 60 75 OPT 90 105 120 135 150 165 180 195 210 225 240 Seconds 39 WhiteFi System Challenges Fragmentation

Spatial Variation Temporal Variation Impact Discovery Spectrum Assignment Disconnection 40 MSR KNOWS Program Prototypes Version 1: Ad hoc networking in white spaces

Capable of sensing TV signals, limited hardware functionality, analysis of design through simulations Version 2: Infrastructure based networking (WhiteFi) Capable of sensing TV signals & microphones, deployed in lab Version 3: Campus-wide backbone network (WhiteFi + Geolocation) Deployed on campus, and provide coverage in MS Shuttles Geo-location Service Shuttle Deployment Worlds first urban white space network! Goal: Provide free Wi-Fi Corpnet access in MS shuttles Use white spaces as backhaul, Wi-Fi inside shuttle Obtained FCC Experimental license for MS Campus Deployed antenna on rooftop, radio in building & shuttle

Protect TVs and mics using geo-location service & sensing Some Results Demo Summary & On-going Work White Spaces enable new networking scenarios KNOWS project researched networking problems: Spectrum assignment: MCham Spectrum efficiency: variable channel widths Network discovery: using SIFT Network Agility: Ability to handle disconnections

Ongoing work: MIC sensing, mesh networks, co-existence among white space networks, 45 Questions SIGCOMM 2008 Talk A Case for Adapting Channel Width in Wireless Networks Ranveer Chandra, Ratul Mahajan, Thomas Moscibroda, Victor Bahl Microsoft Research Ramya Raghavendra

University of California, Santa Barbara Adaptation in Wireless Networks Existing knobs: Transmit rate/Modulation: auto rate algorithms Adapt how tightly bits are packed in spectrum Transmit power: TPC algorithms Adapt tx power for connectivity, spectrum reuse This paper: Channel Width: how & why? 49 Channelization in IEEE 802.11 802.11 uses 20 MHz wide channels

70 MHz 2427 MHz 2402 MHz 2452 MHz 2472 MHz 2412 MHz 1 2 3 11

6 2407 MHz 20 MHz 50 Why Adapt Channel Widths? One Scenario More spectrum + more capacity (Shannons) higher idle power consumption (coming up) Challenge: 40 20 5 MHz

Dynamically determine app demand & adapt channel width For throughput intensive apps, wider for best data rate When idle, go narrow forgo least power consumption 51 Our Contributions

Demonstrate feasibility of dynamic channel width adaptation on off-the-shelf hardware Characterize properties of channel widths Throughput, range, energy consumption SampleWidth to dynamically select best channel width 52 Implementing Variable Widths Antenna Typical Wireless Card Baseband/MAC RF Component (coding/decoding,

timing, encryption) (PLLs, upconverters Power Amplifiers) REF CLOCK Modify driver to programmatically tune clock frequency Channel width proportional to clock frequency 53 Variable Channel Widths in OFDM In 802.11: 48 data subcarriers, 4 pilots Pilot tone Data Subcarriers 20 MHz Subcarrier Spacing: 0.3125 MHz

At 20 MHz: Guard Interval: 0.8 s Symbol Period = 1/0.3125 s + GI = 4 s 54 Variable Channel Widths in OFDM To reduce width to 10 MHz, halve the clock frequency Pilot tone Data Subcarriers 20 10 MHz Subcarrier Spacing: 0.3125/2 MHz At 10 MHz: Guard Interval: 0.8*2 s Symbol Period = (1/0.3125 s + GI)*2 = 8 s 55

Our Implementation Using Atheros cards on Windows Implemented 5, 10, 20, 40 MHz MAC parameters scale with clock e.g. SIFS: 20 s at 20 MHz, 40 s at 10 MHz We keep 802.11 slot time constant for interop 56 Properties of Channel Widths Impact on: Throughput Transmission Range Battery Power 57

Experimental Setup Conducted (clean) experiment Using attenuator & CMU emulator Indoor experiments at MSR & UCSB Outdoor experiments in large park 58 Throughput Throughput increases with channel width U D P T h r o u g h p u t (in M b p s ) (Shannons) Capacity = Bandwidth * log (1 + SNR) In practice, protocol overheads come into play Twice bandwidth has less than double throughput Actual Data Rate: 108 [email protected] MHz

54 [email protected] MHz 5 MHz 10 MHz 20 MHz 40 MHz 27 [email protected] MHz 13.5 [email protected] Modulation 59 Transmission Range Reducing channel width increases range

Loss Rate Narrow channel widths have same signal energy but lesser noise ~ 3 dB better SNR 5MHz 10MHz 20MHz 40MHz Attenuation (dB) 60 Impact of Guard Interval Loss Rate (%)

Reducing width increases guard interval more resilience to delay spread (more range) 5MHz 10MHz 20MHz 40MHz Delay Spread (in ns) 61 Need for Width Adaptation With auto rate: 40 MHz 20 MHz 10 MHz 5 MHz

There is no single best channel width! 62 Energy Consumption Lower channel widths consume less power Similar to CPU clock scaling Send Idle Receive 5MHz 1.92 1.00 1.01 10MHz

1.98 1.11 1.13 20MHz 2.05 1.25 1.27 40MHz 2.17 1.41 1.49 When idle, lowest channel width is best During send/receive, best energy/bit width depends on distance 63

Recap: Channel Width Properties When nodes are near, higher channel widths have more throughput Lower channel widths have more range Better SNR, resilience to delay spread Lower channel widths consume less power Lower widths increase range while consuming less power! 64 Application: Song Sharing Zune Social over Wi-Fi 1. Zunes advertise (periodically beacon) their song list 2. Interested Zunes download songs from peers Issues: throughput, power! Our Solution: Adapt channel width based on traffic

(SampleWidth) 65 De tail SampleWidth for Throughput s+ pro Goal: Use minimum width that satisfies demand of in p ape r Algorithm: Start at minimum width best energy, range When interface queue is full, probe higher width

During song transfer Periodically probe adjacent (higher/lower) widths Return to minimum width when no traffic 66 SampleWidth Evaluation SampleWidth adapts to best throughput width 67 Reducing Power Consumption Start 20 MB 5MHz 20MHz 10MHz

40MHz Energy (Joule) file transfer @ 25 sec Seconds 68 Total Energy (Joules) SampleWidth for Energy ~ 25% savings 69

Application Scenarios 1. Throughput/energy-aware song sharing 2. Load aware spectrum allocation in WLANs 3. Improved capacity in 802.11 4. Cognitive (DSA-based) networking 70 Summary Channel width can be adapted On off-the-shelf hardware To improve application performance To design better, more efficient networks Future work Explore other channel width strategies e.g. modifying number of subcarriers

Communication across channel widths Nodes on different widths cannot communicate Build larger systems using adaptive channel widths 71 Questions? http://research.microsoft.com/netres/projects/spawn/ 72 73

Recently Viewed Presentations

  • How to Find the Square Root of a Non-Perfect Square

    How to Find the Square Root of a Non-Perfect Square

    - Use square root & cube root symbols to solve equations in the form x2 = p and x3 = p. - Evaluate roots of small perfect square. - Evaluate roots of small cubes. - Apply square roots & cube...
  • Request to Approve Agriculture Nutrient Tracking and ...

    Request to Approve Agriculture Nutrient Tracking and ...

    Neuse & Tar-Pam Estuaries. Exceeding Chl-a standard. Nutrient Management Strategies for each basin. Address Point & Nonpoint Sources. ... Cropland Nitrogen Accounting N Loss Estimation Worksheet (NLEW) Empirical Spreadsheet-based Model. Developed by DWQ, NRCS, NCSU, and others.
  • WAR - dlb.sa.edu.au

    WAR - dlb.sa.edu.au

    BIBLIOGRAPHY Use SACE Guidelines to referencing: Reference all sources including all illustrations ENGLISH COMMUNICATIONS TEXT RESPONSE POETRY ANALYSIS ORAL PRESENTATION Compare and Contrast two poems The Poems _____ by _____ and _____ by _____ The Subject Both Poems deal with...
  • Register Transfer and Microoperations

    Register Transfer and Microoperations

    Dr.Chao Tan, Carnegie Mellon University
  • Instructor Morteza Maleki, PhD Components of a Modern

    Instructor Morteza Maleki, PhD Components of a Modern

    Macro-environmental forces & trends are "non-controllable", which the company must monitor & to which it must respond. A series of challenges firms face are; The steep decline of the stock market, which affected savings, investment, & retirement funds, Increasing unemployment,...
  • Friday 3 February 2017 Who Likes Playing Games?

    Friday 3 February 2017 Who Likes Playing Games?

    Friday 3 February 2017. The assembly plans have been devised to help teachers explain why your school is taking part in NSPCC Number Day and how everyone can make a difference by having fun with numbers and using maths activities...
  • Greece in the Hellenistic Age - Mr. Thurmond's History Class

    Greece in the Hellenistic Age - Mr. Thurmond's History Class

    Greece in the Hellenistic Age . Athenian Philosophers . Socrates (469-399 BCE): Developed the Socratic Method . Believed in aretĂȘ (virtue) In 399 BCE found guilty of corrupting the youth of Athens and not believing in the gods . Did...
  • PRACTICE SOURCE QUESTIONS FOR CORE - 12 Ancient History

    PRACTICE SOURCE QUESTIONS FOR CORE - 12 Ancient History

    A Marcus Holconius Rufus, military leader elected by the people, duumir five times, twice duumvir five-year, a priest of Augustus Caesar, protector of the Colony. Statue of NoniusBalbus in Herculaneum. ... PRACTICE SOURCE QUESTIONS FOR CORE