1. Practical Energy-aware Real- Time Systems 2012.10.24 Koo 1

2. Energy-aware Real-Time Systems There will be three main types of power management techniques. 1.DVFS (Dynamic Voltage & Frequency Scaling) 2.DMS (Dynamic Modulation Scaling) 3.Network Coding 2

3. Energy-aware Real-Time Systems Contents Part I. Introduction of Practical Real-Time systems Part II. Energy-aware Real-time scheduling (DVFS) Part III. Introduction of Network Coding 3

4. Part I. Introduction of Practical Real-Time systems Contents 1.Block diagram 2.Function applications 3.Case study by worse case 4.Power consumption by L-04A (NTT DoCoMo) 5.Conclusion 4

5. 1. Block Diagram Application processor with modem RF ICs [LTE, WCDMA, GSM, GPS, etc.] Memory (DRAM, eMMC) PMIC I/O devices 5

6. 1. Block Diagram (RF; e.g. Qualcomm RTR8600 ) RF IC GSM WCDMA LTE GPS GSM PA WCDMA dual PA RX SAW * PA : Power Amplifier * SAW : Surface Acoustic Wave * RX: Receiving * TX : Transmitting * SP6T: Single Pole 6 Throws SP 6T Coupler Duplexer TX SAW Coupler Duplexer  There are various types of RF ICs such as Blue Tooth, WLAN, and RFID. 6

7. 1. Block Diagram (Baseband ; e.g. Qualcomm MSM8960 ) Application processor with modem Memory (DRAM, MMC) PMIC I/O devices LCD Camera Speaker Vibrator USB Keypad Battery Adapter LPDDR2 SDRAM eMMC *LPDDR : Low Power Double Data Rate * eMMC : embedded Multi Media Card 7

8. 2. Function application (LCD operation) MSM 8960 MDP (Mobile Display Processor) HDMI (High Definition Multimedia Interface) MIPI (Mobile Industry Processor Interface) LCD Camera LED backlight  When a LCD module operates, it needs a LED backlight. A LCD module displays an image by LED backlighting. The backlight is one of major power consumptions in a cell phone. Its amount depend on brightness of LCD. PMIC 8

9. 2. Function application ( wireless communication ) Wireless communications Bluetooth GSM, WCDMA, HSPA, LTE GPS RFID IrDA (Infrared Data Association) Wireless LAN Modes of power amplifier Stand-by mode Low mode Middle mode High mode  Max. power owing to lack of connectivity 9

10. Example(from previous slide) A brief flow chart of power scheduling 10 Computation System(off-line operation) Idle mode Sleep modeActive mode Communication System (on-line operation) Medium power mode High power mode Low power mode Communication off / on Power-onStart-up Low freq. & low vtg. High freq. & low vtg. *For DVS, there are low & high frequency clocks. DC converter and LDO provide various types of voltages. High freq. & high vtg. *Low /medium/high power mode is decided by antenna condition. Three types of call modes

11. Video call & high power mode on WCDMA : This case needs a LCD and a camera with the max. power of power amplifier at WCDMA communication. RFIC 3. Case study (worse case) Application processor with modem Memory (DRAM, MMC) PMICs WCDMA WCDMA dual PA Coupler Duplexer TX SAW LCD LED backlight Camera 11

12. 4. Power consumption by L-04A (NTT DoCoMo) The most power consumption 5 & 5 No.Condition1Condition 2TX power Current consumption [Vin= 3.8V] 1 W2100VT23 dBm732mA 2 W800VT23 dBm721mA 3 W800Talk23 dBm520mA 4 W2100VT12 dBm487mA 5 W800VT12 dBm480mA No.Condition1Condition 2Current consumption [Vin= 3.8V] 1Camera VGA CAMERA 250mA 2Camera2M CAMERA213mA 3MP3Speaker92mA 4MP3Ear-mic57.5mA 5Stand-by 1mA 12

13. 5. Conclusion Multimedia partsMultimedia parts (ARM core, power regulators, LCD, camera, etc.) are the major part of power consumption when a wireless embedded system does not work for communication. RX power amplifier, RF moduleRX power amplifier, RF module will also critical when the system work for wireless communication. 13

14. Part II. Energy-aware Real-time scheduling Contents 1.PMIC 2.System Block Diagram 3.Case Study (DVFS) 4.General methods of Power Management 5.Future Works 14

15. 1. PMIC (Power Management IC) Why do we need PMIC? -. From a single battery to various types of inside modules and I/O devices, PMIC controls their power. What are its main tasks? -. Input power (battery, charger, USB) -. Output power (SMPS, LDO, charge pump) -. IC interface (PA control, GPIO) -. General housekeeping by internal CLKs -. User Interface (LED, LCD, Vibrator, Headset, Speaker) 15

16. 2. System Block Diagram RF(Radio Frequency) vs. BB (Baseband) Application processor with modem RF ICs [LTE, WCDMA, GSM, GPS, etc.] Memory (DRAM, MMC) PMIC I/O devices 16

17. 2. System Block Diagram According various scenarios, it needs a good algorithm for PM. Game Task A State : Running Interrupt Interrupt Vector Table Interrupt Service Routine Message Task B State : Blocked Waiting Context Switch State : Ready Save context Message State : Ready State : Running 17

18. 3. DVFS (Dynamic Voltage and Frequency Scaling) DVFS -. According to scheduled modes, PMIC provides processors’ core with different types of power by scaling voltage and frequency. 18

19. 4. General Power Management Step 1) Analyze target task/application : min./max. power requirement Step 2) Seek leakage/unnecessary power e.g.) pending task after interrupt Step 3) Make up specific algorithm with possible scenarios e.g.) DVFS, FSM, etc. Step 4) Verify a side effect after a new PM algorithm 19

20. Part III. Introduction of Network Coding (COPE) Contents 1.Abstract of COPE architecture 2.COPE: basic idea 3.Three main parts of COPE 4.COPE: Opportunistic Coding Protocol 5.COPE implementation 20

21. 1. Abstract of COPE architecture 21

22. 2. COPE: basic idea Bob Alice Relay Alice’s packet Bob’s packet Alice’s packet 3 transmissions instead of 4  Saves bandwidth & power  33% throughput increase 3 transmissions instead of 4  Saves bandwidth & power  33% throughput increase XOR = 22

23. Opportunistic Listening  COPE tries to listen all packets by analyzing the headers. Opportunistic Coding  By XOR, COPE performs network coding based upon next-hop basic. Learning Neighbor State  In order to encode a packet, COPE needs know what packets a neighboring node needs, and the packets the node has received so fat. 3. Three main parts of COPE 23

24. A simple example of “Opportunistic Coding” 24

25. 4. COPE: Opportunistic Coding Protocol Alice Relay Bob Charlie Alice’s packet Charlie’s packet Bob’s packet Charlie’s packet Alice Bob Bob Charlie Charlie Alice XOR = 25

26. 20-node wireless testbed There are two floors which have 10 node each; it runs on 802.11a with a bit-rate of 6Mb/s. 5. COPE implementation

27. Software Nodes in the testbed run Linux; COPE is implemented using “Click Modular Router” toolkit like the under. The implementation sends and receives raw 802.111 frames from the wireless device using a lipcap-like inteface 5. COPE implementation

28. MIT’s Click “Push-Pull” semantics Single-threaded Network element database: 200+ elements Tight integration with Linux During Push (forwarding) During Pull (backwarding) 5. COPE implementation

29. Router [Routing protocol] Srcr (source-routes data packets); a state-of-the art routing protocol for wireless mesh network. [Algorithm] the protocol use Djikstra;s shortest path algorithm on a database of link weights based on ETT expected transmission time) metric. The router output queue is bounded at 100 packets. 5. COPE implementation

30. Srcr (Roofnet: An 802.11b Mesh Network) Srcr: DSR(Dynamic Source Routing) like protocol. -. Each link has metric. -. Data packets contain full source routes (robust aganist loops; metric may be dynamics.) -. Nodes keep database of link metrics. -. Run Dijkstra’s algorithm over data to compute source routes. 5. COPE implementation

31. Hardware Each node in the testbed is a PC equipted with an 802.11 wireless card attached to an omni- directional antenna. (The card are based on the NETGEAR 2.4 & % GHz 802.11 a/g chipset.) They transmit at 15 dBm power and operate in the 802.11 ad hoc mode with RTS/CTS disabled as in the default MAC. 5. COPE implementation

32. Traffic model They use an utility called “updgen” to generate UDP (User Datagram Protocol) traffic and “ttcp” to generate TCP (Transmission Control Protocol) traffic. 5. COPE implementation

33. Q&A Please give me a question which you are interested in or not clear. Thank you. 33

34. References Qinglong Liu, and Gang Feng “Optimization Based Queue Management for Opportunistic Network Coding”, 2011 6th International ICST Conference on Communications and Networking in China,2011, pp 1159-1164 S. Katti, H. Rahul, W. Hu, D. Katabi, M. M. Medard and J. Crowcroft, “XORs in the Air”: Practical Wireless Network Coding,” in Proc. of ACM SIGCOMM’06, Pisa, Italy, Sept. 2006. R. Ahlswede, N. Cai, S. Y. R. Li, and R. W. Yeung, “Network Information Flow”, IEEE Transactions on Information Theory, vol. 46, no. 4, July 2000, pp.1204-1216. P. Glatz, J. Loinig, C. Steger, and R. Weiss, “A first step towards energy management for network coding in wireless sensor networks,” in 9th IEEE Malaysia International Con-ference on Communications, dec. 2009, pp. 905 – 910.