Jan M. Rabaey BWRC University of Berkeley ISLPED 2001, Huntington Beach

Transcription

1 Wireless Beyond the Third Generation Facing The Energy Challenge Jan M. Rabaey BWRC University of Berkeley ISLPED 2001, Huntington Beach

2 It s all about Laws Moore s Law Shannon s Law Metcalfe s Law Gene s Law Maxwell s Laws Laws of Physics

3 The Fibonacci Law on Wireless Growth The number of world-wide wireless subscribers (in tens of millions) grows as a Fibonacci series? Source: Goldman-Sachs

4 The Shift to Wireless Data The New Internet The 1990s: Conquering the world The network revolution Gigabit Ethernet Massive Cluster Clusters The 2000s: Extending toward the Small Enabled by integration (Moore s Law at Work!) and wireless connectivity Pre-1990: Client-Server Systems Courtesy: R. Katz, UCB

5 (Projected) Growth in WLAN Units, k Mbps 10-11Mbps 20Mbps+ 54 Mbps Total Source: Cahners In-Stat 2001

6 The Evolving Wireless Scene More bit/($ nj nj) More bit/sec Data Rate 100Mb 10Mb 1Mb 100Kb 10Kb 1Kb 802.1a (LAN) Bluetooth (PAN) Sensor networks Metropolitan 3G Cellular 2.5 G Cellular Cellular (WAN) 1m 10m 100m 1km 10km Range

7 More bit/($ nj nj) More bit/sec Data Rate 100Mb 10Mb 1Mb 100Kb 10Kb 1Kb 802.1a (LAN) Bluetooth (PAN) Sensor networks Metropolitan 3G Cellular 2.5 G Cellular Cellular (WAN) 1m 10m 100m 1km 10km Range

8 How to Get More bits/sec in a Band-Limited Environment? The Shannon Bound: In an AWGN channel, the best bandwidth-efficiency (in bits/sec/hz) that can be achieved with arbitrarily low bit-error rate is given by 9 Bandwidth efficiency (bits/sec/hz) Bandwidth Efficient Energy Efficient bweff log 2 (1 + bweff * E b /N 0 ) Energy/bit over Noise Spectral Density E b /N 0

9 The Cost of Approaching Shannon s Bound The Bliss and Challenge of Error Coding Relative Complexity /2 LDPC, N=10 7, 1100 iterations 2/3 Capacity Bound 1/2 Capacity Bound 1/2 Turbo, ν=4, N=64k 1,2, and 3 iterations 8/9 Capacity Bound 2/3 Turbo, ν=4, N=64k 1,2, and 3 iterations 8/9 LDPC, N=4k 1,3, and 5 iterations 8/9 Turbo, ν=4, N=4k SNR (db) 1/2 Conv. Code, ν=4, N=64k for BER of /3 Conv. Code, ν=4, N=64k 8/9 Conv. Code, ν=3, N=4k Courtesy Engling Yeo, UCB

10 Dealing with Non-ideal Channels (e.g., fading) 1 st path, α 1 = 1 Array Processing Array Processing x(t) 2 nd path, α 2 = 0.6 Multi-antenna approach exploits multi-path fading by sending data along good channels Results in large theoretical improvements in bandwidth efficiency for fading channels But computationally hungry Capacity (bits/s/hz) y(t) (4,4) With Feedback (4,4) No Feedback (4,1) Orthogonal Design (1,1) Baseline SNR (db)

11 The Cost of Dealing with Non-ideal Channels MIPS OFDM + Multi-Antenna OFDM + Multi- Antenna + Coding Matched Filter Multi-User Detector OFDM + Coding 0 Data rate per user Spectral efficiency 0.8 Mb/s 0.9 b/s/hz 1.6 Mb/s 1.8 b/s/hz 1.9 Mb/s 2.1 b/s/hz 3.8 Mb/s 4.2 b/s/hz 5.6 Mb/s 6.3 b/s/hz Performance * Assume 25 MHz bandwidth and 28 users Source: Ning Zhang, UCB

12 Compelling Wireless Implementation Issues A Ferocious Quest for Performance Driven by the hunger for bits/sec Outstripping the technology evolution With a Premium on Reduction in Energy Consumption The compelling argument behind wireless is its untethered nature Power consumption key impediment to penetration of new services Energy sources on slow evolutionary path (5%/year)

13 What Technology Offers Us Gene (Frantz) s Law mw/mips 1, Gene s Law DSP Power Energy Efficiency of DSPs Doubles Every 18 Months Source: Gene Frantz (TI) Year

14 Shannon beats Moore beats Chemists G Algorithmic Complexity (Shannon s Law) 3G Processor Performance (~Moore s Law) G Battery Capacity Courtesy: Ravi Subramanian (Morphics)

15 The Need for Flexibility Cost-of-design issues point towards component reuse Design complexity impacts time-to-market Physical effects increase verification costs and design risk NRE of new designs is increasing significantly (mask making, fab cost) Multi-standard has become a must in the diverse wireless landscape Adaptive solutions lead to better spectral utilization A wide variety of unpredictable services

16 An Attractive Option: Multi-Processor System-on on-a-chip The New NAND gate Copyright Tensilica, Inc 2001 Courtesy: Chris Rowen, Tensilica

17 Flexibility Comes at a Huge Cost Flexibility Prog Mem µp Prog Mem µp 500 pj/op Prog Mem µp Satellite Processor MAC Unit Addr Gen General Purpose µp 0.5 pj/op Dedicated Logic Direct Mapped Hardware Satellite Satellite Processor Processor Hardware Reconfigurable Processor Software Programmable DSP 3 orders of magnitude! Inefficiency

18 The Opportunity of Configurable Architectures Power (mw) 250 Area (mm^2) Execution Memory Control Clocking 0 DSP (TI C54) DSP Extension Configurable Dedicated Hardware DSP (TI C54) DSP Extension Configurable Dedicated Hardware * Based on the implementations of a multi-user detector Source: N. Zhang, UCB

19 The Opportunity of Reconfigurable When Does it Work? J/Transform Lower limit Function-specific reconfigurable hardware Data-path reconfigurable processor FPGA Low-power DSP High-performance DSP Transforms/sec/mm 2 2 ) Function-specific reconfigurable hardware Data-path reconfigurable processor FPGA Low-power DSP High-performance DSP FFT size FFT size Energy and Area Efficiency of Various FFT Implementations Source: N. Zhang, UCB * All results are scaled to 0.18µm

20 The Ideal Radio-on on-a-chip Platform Combines performance, flexibility and energy-efficiency Reconfigurable State Machines FPGA Dedicated DSP Embedded up + DSPs Reconfigurable DataPath Heterogeneous Matches the computational model Provides flexibility only where needed and desirable and at the right granularity Supports massive concurrency Operates at minimum supply voltage and clock frequency

21 An Orthogonal Approach to Bit/sec Ultra-Wide Band Radio Traditional Sinusoidal, Narrowband Time Frequency Impulse, Ultra-Wideband Time Frequency Splurge on Bandwidth (> 5 GHz) and Punt on Bit/sec/Hz Possible advantages: easy co-existence, low-power, simple

22 Digital Pulse-Based Radio Simple Digital Architecture: V - + V - + A/D Transmit Only Narrow Pulses (No Carrier Frequency) Spread Energy Over Existing Noise Floor The Architectural Challenge: Providing accurate timing resolution without high-frequency clocks! Predicted performance: 100 < 10-4 bit/sec/hz and ~ 10 nj/bit

23 More bit/($ nj nj) More bit/sec Data Rate 100Mb 10Mb 1Mb 100Kb 10Kb 1Kb 802.1a (LAN) Bluetooth (PAN) Sensor networks Metropolitan 3G Cellular 2.5 G Cellular Cellular (WAN) 1m 10m 100m 1km 10km Range

24 More Bits/(nJ nj $ mm 3 ): Wireless Sensor Networks Pushing the Bounds in Ultra [Small, Cheap, Low-Power] Berkeley PicoRadio s Meso-scale low-cost radio s for ubiquitous wireless data acquisition in sensor/actuator networks that are fully integrated and consume less than 100 µw enabling energy scavenging The Smart Building Integrated Sensor/Actuator/ Control System Improves quality-of-living Saves energy Provides security Helps localizing items Extends building-human interface

25 The Energy-Scavenging Opportunity µwatts 10 Lithium Zinc-Air NiMH Li (rechargeable) Solar Vibrations Alkaline Years Battery size: 0.5 cm 3 Vibration: 1 cm 2 piezo-electric Solar: 1 cm 2 single-crystal Courtesy: S. Roundy (UCB)

26 Opportunity: Metcalfe s Law The true value of a network increases as the square of the number of users on the network A Variant: Jan s Law The power efficiency of a wireless sensor node increases as the square of the number of nodes in the network (or is proportional to the node density).

27 Addressing the Communication Cost 20 Transmit Power (in dbm) pj/bit Bluetooth f c = 2.4 GHz f c = 0.43 MHz Minimum required transmit power increases as d 4 due to ground wave and multi-path Increasing carrier frequency costs 20db/decade Distance (in m) (assuming d 4 path loss of and 10 kb/sec data rate)

28 Adding a Single Relay Point in a Wireless LAN Bit-rate: 6 Mb/sec Packet Error Rate: 1% Fixed Receiver Power: 100 mw Path Loss Exp.: 3.8 d1/d2 =1 Reduces energy/bit up to 4 times! Source: M. Kubisch and H. Karl, TU Berlin

29 Trading-off Latency for Energy: Self-configuring Multi-hop Networks controller actuator sensor 1 hop over 50 m 1.25 nj/bit 5 hops of 10 m each 5 2 pj/bit = 10 pj/bit Multi-hop reduces transmission energy by 125! node range But how to ensure fairness?

30 Energy-Conscious Networking Simulated Energy Dissipation in Sensor Networks (BWRC) Performance-oriented Power Consumption Source: R. Shah (UCB) Energy-Conscious

31 The Sensor-Node RF Challenge Receiver Power Consumption (mw) Carrier Frequency Increasing carrier frequency increases power dissipation Mostly due to higher speed active components (synthesizers, mixers, A/Ds) But enables higher integration Smaller sizes of passives ands antennas Rx power consumption versus carrier frequency for a number of low-data rate, small-distance RF implementations (all operate in Shannon s energy-efficient zone )

32 Eliminating most High-Speed Components Sub-sampling receiver with passive frontend RF Filter LNA RF Filter f base A D Image Rejection RF Filter RF Filter RF Filter Provides diversity Shapes LNA thermal noise Does not require any high-frequency active components Receive energy reduced by 1-2 orders of magnitude to ~ 5 nj/bit

33 Enabled by High-Q Q Integrated Filters Thin-Film Bulk Acoustic Resonators Q > 2 GHz (FBAR Agilent) - RF-MEMS: Poly Si-Ge Resonator

34 The Importance of Power Management Activity in sensor (data) networks is low and random (< 1%) Receiving a bit is computationally more expensive than transmitting one Most Media Access protocols assume that the receiver is always on and listening! Why not power transceiver up for real events only (incoming data, sensor event, network maintenance)?

35 Reactive Media-Access Control Truly Reactive Messaging at the Physical and Media- Access Level Power Down the Whole Data Radio Reduces Monitoring Energy Consumption by 10 3 Times Wakeup Radio Power s Up Data Radio for Data Reception Sleeping nodes Communicating nodes

36 The Wake-up (Reactive) Radio PA RF Filter RF Filter RF Filter Pulse Generator Transmit RF Filter LNA RF Filter A D RF Filter RF Filter RF Filter Receive Always running Super low power: 10-4 ~ 10-3 active mode power Data radio shut down when idle, and powered up by wake-up radio Receiver response time: < 10ms Energy Detector Pulse Position Detector Wakeup Tone Receive Session Wake-up

37 What it Ultimately Boils Down To Power-Profile Profile of PicoRadio (Projected) Clock Clock distribution distribution 14% 14% Reactive Reactive receiver receiver 1% 1% Radio Radio receiver receiver 12% 12% Radio Radio transmitter transmitter 5% 5% Baseband Baseband 13% 13% Embedded Embedded microprocessor microprocessor 30% 30% Reconfigurable Reconfigurable protocol protocol stack stack 25% 25%

38 Summary/Perspective Both the bits/sec/hz and the bits/nj quests create formidable energy challenges Keep your eyes open for innovative, orthogonal approaches that re-stack the cards There is a whole lot of unexplored land available > 10 GHz In the end, it are the laws of physics that provide the ultimate bounds