Building Cognitive Radio Networks Prof. Joseph B. Evans Prof. Gary J. Minden The University of Kansas Information and Telecommunications Technology Center 2335 Irving Hill Road Lawrence, KS 66045 <evans@ittc.ku.edu> Building Cognitive Radios • • • • Introduction and Motivation Example implementation – KU Agile Radio “Cognitive” Radio Rethinking design… (-100dBm) AD8347 I-Q DEMODULATOR 800 MHz - 2.7 GHz MGA-86576 Gain = +19dB@6.0 GHz P1dB = +4.3dBm@6.0 GHz NF =1.8dB@6.0 GHz +8dBi HMC488MS8G CLoss = 10 dB P1dB = +8 dBm LO Drive 0dBm 5.250-5.850 GHz (-77dBm) -3dB LNA -4dB R -10dB -5dB 0-30dB RX Attenuation Control Active RX Antenna Module BASEBAND I 0 90 I L +19dB -5dB 1.850-2.450 GHz (-81dBm) (-66dBm) LNA +19dB GAIN CONTROL 3.4 GHz I DET DPB Input +3.3 AD8347 AD8349 ADF4360 ADF4113 SMV3300A FOX801BE-160 BGA2031 MGA-83563 MGA-82563 MGA-545P8 MGA-86576 HMC488MSG8 ERA-1SM 4 @60mA +5 80 150 MGA-82563 Gain = +10dB@4.0 GHz P1dB =+17dBm@4. 0 GHz 240 15 20 2 @25mA 2 @50mA 3 dB POWER DIVDER -2dB 50 100 40 770mA 440mA ADF4360-1 2.150-2.450 GHz RX IF GAIN CONTROL AD5601 6 BIT DAC RX LO 1 FOX801BE-160 TCXO LO 3 (+3.5dBm) 5 70 200 100 140 Q DET Jumper Select SPI Bus BASEBAND Q ADF4360-2 1.850-2.150 GHz MC68HC08 Microcontroller I²C BW=30 MHz 16.0 MHz REF CLK OUT ADF4113 +9dB (+3.5dBm) -5dB SMV3300A (+5dBm) TX LO 2 Microstrip Lumped (+25dBm) +8dBi 5.250-5.850 GHz (+17dBm) (+21dBm) (+15dBm) (+9dBm) TX IF GAIN CONTROL ERA-1SM Gain = +6dB@6.0 GHz P1dB = +12dBm (-2dBm) (-8dBm) 3.4 GHz R +17dB ADF4360-1 2.150-2.450 GHz BASEBAND I (+7dBm) (+3dBm) L -4dB AD8349 I-Q MODULATOR 700 MHz - 2.7 GHz 1.850-2.450 GHz (-3dBm) PA -5dB OPTIONAL MGA-545P8 Gain = +11.5dB@5.8 GHz P1dB = +21dBm@5.8 GHz PSAT = +22dBm@5.8 GHz ADF4360-2 1.850-2.150 GHz AD5601 6 BIT DAC Active TX Antenna Module I -10dB +9dB 0 90 BW=30 MHz -5dB BASEBAND Q For use w ith passi ve anten na MGA-83563 Gain = +17dB@6.0 GHz P1dB = +15dBm@6.0 GHz PSAT = +18dBm@6.0 GHz BGA2031 Gain = +23dB@1.9 GHz (2.7 V CT RL) G = 56dB@1.9 GHz P1dB = +13dBm@1.9 GHz 5 GHz RF PCB Block Diagram 1 D. DePardo 1 1 27 JULY 05 KU Agile Radio KU Agile Radio Concept Digital Board and Control Processor Power Supply RF Transceiver 7” H x 3” W x 6” D KUAR Power Supply • Provide 1.8 VDC, 2.5 VDC, 3.3 VDC and 5 VDC power to the radio, separate supplies for the digital and RF sections • External power from battery, vehicle, or mains KUAR Control Processor • Five functions: radio control; signal processing; configuration management; adaptive algorithms; and interface with wired networks. • Intel Pentium-M; 1.4 GHz; 1 GB of RAM; 8 GB micro-disk; 100 Mbps Ethernet; USB; VGA; Floating Point • GPS • Linux OS (Kernel 2.6); Full TCP/IP protocol stack; SSH/SSL; Web Server; NFS; Samba • KUAR CP fully participates in a wired network with standard IP services KUAR Digital Board • Xilinx Vertex II Pro V30; 2 PPC 405 cores; 31K logic cells; 350 MHz operation • Analog Devices AD9777 DAC; I & Q; 160 Msps; 16-bit • Linear Technologies LTC2284 ADC; I & Q; 105 Msps; 14-bit • 4 MB (1 M x 36-bit) SRAM DRAM 1 GB SRAM 4 MB Disk 8 GB To/From RF Board 32 2x14 Ethernet USB Video GPS Intel Pentium-M 1.4 GHz 32 Vertix-II Pro XC2VP30 2x16 LTC LTC2284 105 Msps Rx_I DAC AD9777 160Msps Tx_I Rx_Q Tx_Q KUAR 5 GHz RF Transceiver (-100dBm) +8dBi 5.250-5.825 GHz (-77dBm) HMC488MS8G CLoss = 10 dB P1dB = +8 dBm LO Drive 0dBm -3dB LNA -4dB AD8347 I-Q DEMODULATOR 800 MHz - 2.7 GHz MGA-86576 Gain = +19dB@6.0 GHz P1dB = +4.3dBm@6.0 GHz NF =1.8dB@6.0 GHz LNA +19dB R SHIFT REGISTER +3.3 4 @60mA I DET MGA-82563 Gain = +8.8dB@6.0 G Hz P1dB =+ 16. 8dBm@6. 0 GHz 240 15 20 Q DET 3 dB POWER DIVDER -2dB 50 100 40 770mA 440mA ADF4360-1 2.150-2.425 GHz RX IF G AIN CONTROL AD5601 6 BI T DAC RX LO 1 FOX801BE-160 TCXO LO 3 (+3. 5dBm) BASEBAND Q ADF4360-2 1.850-2. 150 GHz Jumper Select +5 80 150 BW=30 MHz GAIN CONTROL 3.4 GHz 5 70 200 100 140 2 @25mA 2 @50mA 0 90 I -10dB-5dB 0-30 dB RF ATTENUATION CONTROL AD8347 AD8349 ADF4360 ADF4113 SMV3300A FOX801BE-160 BGA2031 MGA-83563 MGA-82563 MGA-545P8 MGA-86576 HMC488MSG8 ERA-1SM BASEBAND I L +19dB -5dB Act ive RX Antenna Module 1.850-2.425 GHz (-81dBm) (-66dBm) 16.0 MHz REF CLK OUT ADF4113 +9dB (+3. 5dBm) -5dB SMV3300A SPI CONTROL (+5dBm) TX LO 2 Microst rip Lumped (+25dBm) +8dBi 5.250-5.825 GHz (+17dB m) (+21dBm) (+15dBm) (+9dBm) TX IF GAIN CONTROL ERA-1SM Gain = +6dB@6.0 GHz P1dB = +12dBm (-2dBm) (-8dBm) 3.4 GHz (+7dBm) (+3dBm) L -5dB O PTIONAL MGA-545P8 Gain = +11. 5dB@5.8 GHz P1dB = +21dBm@5.8 GHz PSAT = +22dBm@5.8 GHz ADF4360-1 2.150-2.425 GHz BASEBAND I R +17dB AD8349 I-Q MODULATOR 700 MHz - 2.7 GHz 1.850-2.425 GHz (-3dBm) PA -4dB ADF4360-2 1.850-2.150 GHz AD5601 6 BIT DAC Active TX Antenna Module I -10dB +9dB 0 90 BW=30 MHz -5dB BASEBAND Q For use w ith passive anten na MGA-83563 Gain = +17dB@6.0 GHz P1dB = +15dBm@6.0 GHz PSAT = +18dBm@6.0 GHz BGA2031 Gain = +23dB@1.9 GHz (2.7 V CT RL) G = 56dB@1.9 GHz P1dB = +13dBm@1.9 GHz Agile Radio 2.0 5 GHz RF PCB Block Diagram Revision: 1 D. DePardo Page: 1 of 1 Date:1 JUNE 05 Tx/Rx Antenna Patch 5 GHz wideband integrated LNA receiver patch 5 GHz wideband (1 GHz) integrated patch computed pattern Preliminary performance measurements KU Agile Radio Version 3.0 • Complete package • Version 3.0 digital board with CP and RF boards KU Agile Radio Enables • Rapid service definition and deployment • Bring new services to the public • Dynamic service access • Rapidly find and access available radio services • Dynamic spectrum access • Improve utilization of spectrum resource • Spectrum commons/markets • Devolve spectrum management to local regions Cognitive Radios Cognitive Radio Learning Structure Hours Milli-Seconds ~Minutes/Hours “Cognitive” Challenges • Mission Oriented Radio Configuration • Develop techniques to select appropriate communications modules to accomplish defined mission • Self Configuring Radios • Software should automatically determine capabilities of hardware and use those capabilities • Adaptation • Change radio operation based on current environment • ElectroSpace resource models • Policy Adherence • Software Architecture Cognitive Radio Learning Structure Hours Milli-Seconds ~Minutes/Hours Mission Oriented Properties • Low probability of detection and interception (LPD/LPI) • Interference avoidance and rejection • Multipath channel mitigation/exploitation • Information assurance (jam resistance, security enhancement, etc.) • Communication range (e.g. foliage penetration) • QoS requirements • Communications capacity • Power/energy efficiency Consider Natural Disaster Communications • Initial deployment • • • • Robust communications messages must get through; minimize first responder stress Low capacity - perhaps voice only, simple user devices Low radio density - long links Minimal power - low maintenance • Early follow-on • • • • Higher radio density - more time and resources to deploy additional radios Medium capacity - increase data services; use capacity to maintain and increase robustness (e.g. digital transmission and error correcting codes) Increased power Tie into wired infrastructure • Extended Support • • • • Extensive data services - voice, video, and data services interoperate with established infrastructure Radio density as needed High capacity Power from grid Mission Oriented Configuration • Establish trade-offs between multiple mission goals • Case-based reasoning • Establish a case library of possible scenarios • Match desired mission goals against case library • Select closest case from library and adjust to present mission goals • Genetic algorithms • Establish utility function for present mission goals • Establish a population of possible configurations • Select “good” configuration and “inter-mingle” to make a new population; repeat as configurations improve • Expert systems • Build a set of rules for defining configurations from present mission goals Cognitive Radio Learning Structure Hours Milli-Seconds ~Minutes/Hours Cognitive Radio Software Architecture • For adaptation… • Sense RF, network, and communications environment performance • Adjust radio components to current operating conditions for best performance • Based on trade-offs between alternative adjustments Cognitive Radio Software Architecture Topology Manager • Determine which radios should communicate • Based on… • Available ElectroSpace resources • Application load (network queues) • Adaptation (determining when to adjust) • A connection involves… • Allocation of ElectroSpace • Scheduling reception and transmission • Adding network routes Cognitive Radio Software Architecture Cognitive Parameters • General Radio Model • Every processing stage is programmable and controllable Compression Error Ctrl Encrypt Spreading Modulation Transmit Thermal Noise Information Source Mutual Interference Composite Channel Jamming Multipath Information Sink Other De-Compress • Transmission parameters (Knobs) • • • • • • Transmit power Modulation Code rate Symbol rate Frame length Environmental parameters (Dials) • • • • SNR Path loss Battery life Delay spread Error Ctrl Decrypt De-Spread De-Modulate Every processing stage is programmable and controllable. QoS Objectives ( Dials ) SNR Delay Profile Spectral Occupancy Battery Life Cognitive Adaptation Module Tx Power Frequency Coding Rate ( Knobs ) Bandwidth Frame Size Receive Goal Conflicts Reasoning/Control Approaches • Exact Methods • • Advantages: Exact optimal solution can be found Disadvantages: Typically requires at least first derivative of a complex equation; Time complexity (pure random) • Heuristic Methods • • Advantages: Lower complexity than exact methods; Increased flexibility with regards to changes in the fitness equation Disadvantages; Sub-optimal solutions • Simulated Annealing • • Advantages: Ease of implementation Disadvantages: Only works on single solution (Local optima problem) • Neural Networks • • Advantages: Low memory usage, fast output Disadvantages: Processing complexity, training needed, final output not traceable (traceability is needed) • Genetic Algorithms • • Advantages: Parallel processing, well suited for large problem spaces Disadvantages: Processing time Genetic Algorithms Average convergence over 50 runs 0.95 0.9 0.85 0.8 Fitness Characteristics • Evolves toward the better solution • Typically requires large amounts of processing power • Parameters are represented as strings of bits called chromosomes • Genetic Algorithm selects the best chromosomes and combines them in hopes of creating a better generation 0.75 0.7 0.65 0.6 0.5 Adaptive Genetic Algorithm • Normally the population of chromosomes is randomly initialized • If we assume a slow fading channel we can bias the initial population with chromosomes from a previous cycle • We have shown this to improve the GA convergence rate dramatically Average Maximum Minimum 0.55 0 100 200 300 400 500 600 Generations 700 800 900 1000 Parameter Sensitivity • How much influence does one parameter have on communications? • It is obvious that if we do not allow the cognitive engine to adapt the power parameter bad things happen • What about frame length or symbol rate? ReTarget: A Radio Design Framework Re-Targeting Radio Design Motivation • The JTRS Software Communications Architecture (SCA) describes interfaces between radio components We focus on the design of the programmable components • Radio hardware platforms will evolve quickly, approximately every 12-18 months, and be a combination of new hardware and programmable components We focus on re-targeting a radio design to new platforms Design once, use many. Re-Targeting Radio Design Approach • Use a specification language, Rosetta, to describe radio components and systems of components through composition • Rosetta is an IEEE standards project, P1699 • Translate Rosetta designs into intermediate forms • Similar to the organization of compliers, e.g. gcc • Manipulate the intermediate design forms • Optimize for power, space, specific implementation (e.g. hardware, software, or FPGA), ... • Generate required design description, e.g. VHDL, C Translate from what a component does to how a component is implemented. ReTarget Design Flow Rosetta Specifications Rosetta Specifications Re-Use Specifications Intermediate Forms Intermediate Forms Re-Use Intermediate Forms Target Implementations ASIC FPGA DSP GPU Radio #1 ReTarget Implementations ASIC FPGA DSP GPU Radio #2 ReTarget Tool Organization Future Radio • Innovate • Encourage new approaches to radio and service delivery • Collaborate • Work with research agencies and industry to invest in the future • Experiment • Try new radios and economic approaches • Think • Anticipate impact of emerging technology and economic concepts • Stewardship • Demonstrate care of the public resource KU Agile Radio Team Investigators Research Assistant Professor Gary J. Minden Joseph B. Evans Alexander M. Wyglinski Staff Dan DePardo Leon Searl Students A. Veeragandham Tim Newman* Jordan Guffy* Dragan Trajkov Ted Weidling* Qi Chen Dinesh Datla Rakesh Rajbanshi * Ryan Reed Travis Short Preeti Krishnan V. Rory Petty* Megan Lehnherr Brian Cordill Building Cognitive Radio Networks Prof. Joseph B. Evans Prof. Gary J. Minden The University of Kansas Information and Telecommunications Technology Center 2335 Irving Hill Road Lawrence, KS 66045 <evans@ittc.ku.edu>