Cognitive Radio: a (biased) overview

Transcription

1 Dept. of ECE, IISc Apr. 10th, 2008

2 Outline Introduction Definition Features & Classification Some Fun 1 Introduction to Cognitive Radio What is CR? The Cognition Cycle On a Lighter Note 2 Cognitive Radar 3 Info. Theoretic Aspects System Design 4 and Future Directions

3 A Simple Introduction Definition Features & Classification Some Fun Radio-frequency spectrum, like the acoustic spectrum, is a natural resource, but its use is regulated by governments via licensing agreements However: Some frequency bands are unoccupied most of the time Some are only partially occupied Others are very heavily used If a frequency band is unused now, it is gone forever - so think in terms of detecting and utilizing spectral holes Cognitive Radios attempt to improve spectral utilization by Radio scene analysis Dynamic spectrum management Transmit power control

4 The Case for Cognitive Radio Definition Features & Classification Some Fun If you look at the entire RF frequency up to 100GHz, and take a snapshot at any given time, you ll see that only 5 to 10 percent of it is being used. So there s 90GHz of bandwidth available. - Ed Thomas, former chief engineer at the FCC. So a Cognitive Radio that can intelligently use spectrum licensed to other users when they aren t using it offers a significant benefit.

5 Definition Features & Classification Some Fun Some Definitions of a Cognitive Radio Wikipedia: Cognitive radio is a paradigm for wireless communication in which either a network or a wireless node changes its transmission or reception parameters to communicate efficiently avoiding interference with licensed or unlicensed users. This alteration of parameters is based on the active monitoring of several factors in the external and internal radio environment, such as radio frequency spectrum, user behavior and network state. FCC: A Cognitive Radio is a radio that can change its transmitter parameters based on interaction with the environment in which it operates.

6 Definition Features & Classification Some Fun Cognitive Radio Tasks/Features : detect unused spectrum and use it without causing interference to other users Spectrum Management: matching the available spectrum to user requirements Spectrum Mobility: maintaining seamless connection while frequently(!) changing over to better frequency(!) bands Spectrum Sharing: fair spectrum scheduling methods

7 The Basic Ingredients Definition Features & Classification Some Fun Source: S. Haykin, Cognitive radio: brain-empowered wireless communications, IEEE JSAC, Feb. 2005

8 Terminology Introduction Definition Features & Classification Some Fun Two basic kinds: The Mitola CR: every possible parameter is taken into account and chosen in a context-aware manner Spectrum-sensing CR: only the radio frequency spectrum is considered Some say that a lot of CR is already implemented, e.g., power control, adaptive modulation and coding! In terms of their spectrum use: Licensed Band CR: IEEE Work Group (WRAN using licensed TV bands) Unlicensed Band CR: IEEE Task Group 2 (coexistence of IEEE and bluetooth)

9 Terminology Introduction Definition Features & Classification Some Fun Two basic kinds: The Mitola CR: every possible parameter is taken into account and chosen in a context-aware manner Spectrum-sensing CR: only the radio frequency spectrum is considered Some say that a lot of CR is already implemented, e.g., power control, adaptive modulation and coding! In terms of their spectrum use: Licensed Band CR: IEEE Work Group (WRAN using licensed TV bands) Unlicensed Band CR: IEEE Task Group 2 (coexistence of IEEE and bluetooth)

10 Standardization Efforts Definition Features & Classification Some Fun Dynamic Frequency Selection (DFS) in IEEE a Now DFS refers more generally to automatically selecting a frequency band to minimize or avoid interference to a primary transmitter-receiver. Transmit Power Control (TPC) in IEEE a Now TPC refers more generally to automatically setting the transmit power based on the spectrum used and the regulatory requirement in the current environment (interference temperature) Also in IEEE h (DFS and TPC for WLAN sharing), IEEE P1900 (standards for advanced spectrum management), IEEE (WRANs in unused TV bands)

11 Spotlight on CR Introduction Definition Features & Classification Some Fun CR workshop/tutorial in every major communications and signal processing conference IEEE JSAC special issue: April 2007 CR Workshop here in IISc: Apr. 19, 2007! IEEE Comm. Mag.: May 2007 IEEE Wireless Comm. Mag.: June 2007 SpringerLink special issue on Cognitive Radio: May 2008 IEEE JSAC, JSTSP 2008 Conferences: CrownCom, CogNet, CWNet, DySPAN

12 Definition Features & Classification Some Fun Companies Interested/Working on CR

13 Definition Features & Classification Some Fun Source: N. Holmes and J.H. Snider, The cartoon guide to federal spectrum policy, available online.

14 Definition Features & Classification Some Fun Source: N. Holmes and J.H. Snider, The cartoon guide to federal spectrum policy, available online.

15 Definition Features & Classification Some Fun Source: N. Holmes and J.H. Snider, The cartoon guide to federal spectrum policy, available online.

16 Cognitive Radar 1 Introduction to Cognitive Radio What is CR? The Cognition Cycle On a Lighter Note 2 Cognitive Radar 3 Info. Theoretic Aspects System Design 4 and Future Directions

17 Introduction Cognitive Radar Objective: Designing high quality spectrum sensing devices and algorithms for exchanging spectrum sensing data between nodes to Reliably detect spectral holes for use by the CR Reliably detect when the primary transmitter comes on Two challenges: hidden transmitter, silent receiver Possible approaches: Matched filter Energy detector Feature detector Yet another classification: Centralized (or local) sensing Decentralized (or cooperative) sensing

18 Cognitive Radar Matched Filter-Based Optimal for signal detection (even demodulation) Requires carrier and timing synchronization with primary Pilots, preambles, spreading codes, etc can be used Examples: TV signals have narrowband pilots CDMA systems have pilot and paging channels OFDM systems have preamble words for timing acquisition Drawback: Need special receiver for each primary transmitter!

19 Cognitive Radar Energy Detector-Based Sub-optimal non-coherent detector Detect primary if measured energy in a band > threshold More general and versatile than the matched filter Drawbacks: Susceptible to changing noise levels and fading Does not work well for wideband frequency hopping or direct-sequence spread signals Very limited discriminatory ability between signals, noise and interference

20 Cognitive Radar Feature Detection-Based Modulated signals are always coupled with sine-wave carriers, pilot/preamble sequences, cyclic prefixes, etc with built-in periodicity This periodicity implies the process is cyclostationary, which results in spectral correlation The spectral correlation is used for feature detection Assumption: stationary noise and interference exhibit no spectral correlation Different types of modulated signals could have Identical power spectral density, but Very different spectral correlation functions

21 Cognitive Radar An Example: Distributed Work by J. Unnikrishnan and V. V. Veeravalli, UIUC Decentralized detection with identical sensors, but correlated measurements Source: J. Unnikrishnan and V. V. Veeravalli, IEEE J. Sel. Topics in SP, pp , Feb.2008.

22 Cognitive Radar Decentralized Detection Problem Let Y = vector of energy measurements from all sensors Wish to distinguish the two hypotheses H 0 : Y N (0, σ 2 0 I) H 1 : Y N (µ 1 1, Σ) µ 1, Σ known at the fusion center Individual sensors perform a Likelihood Ratio Test to meet the global false alarm probability on their own

23 Two Possible Approaches Cognitive Radar Counting Rule: Optimal for conditionally i.i.d. observations Simply compare the number of sensors that decide in favor of H 1 to a threshold Linear Quadratic Detector: Uses a decision metric T (X) = h T X + X T MX where X is the vector of LLRs of received bits (with means under H 0 subtracted) Optimize h and M to maximize the deflection criterion D T = [E 1(T (X)) E 0 (T (X))] 2 Var 0 (T (X))

24 Simulation Result Introduction Cognitive Radar

25 Cognitive Radar Introduction Cognitive Radar Learning from Bats: The bat uses sonar to figure out the location, size, velocity, etc of the target with an accuracy that would be the envy of any radar engineer Source: S. Haykin, Cognitive Radar: A Way to the Future, IEEE SP Mag, Jan. 2006

26 Cognitive Radar Cognitive Radar Requirements A cognitive radar should be able to: 1 Build on learning through interactions of the radar with the surrounding environment 2 Use feedback from the receiver to the transmitter to facilitate acquisition of intelligence 3 Learn from past information by recursively updating state-related information All of the above features are currently implemented in bats Difference from bats: while bats track one target at a time, radar systems have to deal with multiple targets

27 Cognitive Radar Basic Ingredients of a Cognitive Radar Information preservation: through soft signal processing. Must leave hard decisions till the point where a final decision is made Adaptive tracking of multiple targets: define the state of a receiver as the a posteriori probability that a target exists. Then, should adaptively track the receiver state Feedback: from the receiver to the transmitter based on the inputs received from the environment to facilitate efficient tracking of the state

28 Cognitive Radar Example of Adaptive Tracking: Radar Scene Analysis Ocean environment under surveillance by a coherent radar The information from radar returns (after processing by a peak filter) is modeled as Clutter statistics: F-distribution w/ (2, 2k) degrees of freedom, denoted F 2,2k (z) where z is the power spectrum measurement, k is the number of neighboring doppler bins used in averaging (for the peak filter) Target + clutter statistics: F -dist., (1/γ)F 2,2k (z/γ), where γ is the target-to-clutter power ratio

29 Bayesian Filtering System Cognitive Radar Closed loop feedback system: propagates the state vector of probabilities from one iteration to the next Source: S. Haykin, Cognitive Radar: A Way to the Future, IEEE SP Mag, Jan Important note: need to establish the right relation between radar measurements and statistical characteristics of clutter and target-plus-clutter for filter stability

30 Cognitive Radar Networks Cognitive Radar Several radars work together in a cooperative manner to track multiple targets. Employs a central base station to fuse information from different radars. Two options: Distributed cognition: all radars including the central base station are cognitive Centralized cognition: the radars are dumb and all the intelligence is confined to the central base station Cognitive radar networks offer a remote-sensing capability far in excess of what the radar components are capable of achieving individually

31 Cognitive Radar Challenges in Cognitive Radar Networks Development of statistical models to describe the information content of radar returns Multi-sensor information fusion (limited computing resources at the base station) Defining success metrics for cognitive radar networks

32 Info. Theoretic Aspects System Design 1 Introduction to Cognitive Radio What is CR? The Cognition Cycle On a Lighter Note 2 Cognitive Radar 3 Info. Theoretic Aspects System Design 4 and Future Directions

33 Info. Theoretic Aspects System Design Soft-Sensing and Optimal Power Control for CR Work by S. Srinivasa and S.A. Jafar, UCI Look at transmit power control at the secondary transmitter Try to maximize the average received SNR (or capacity) at the secondary receiver, subject to A peak power constraint on the secondary tx An average interference constraint on the primary tx Show that a binary power control strategy maximizes the average received SNR at the secondary receiver Derive the power control strategy to maximize the average capacity to the secondary receiver

34 Info. Theoretic Aspects System Design Cooperative Work by Z. Quan, S. Cui (Texas A&M) and A. H. Sayed (UCLA) Due to fading, the individual CRs may not be able to reliably detect the existence of the primary user Propose spectrum sensing based on linearly combining local test statistics from individual secondary users Two optimization schemes are proposed to control the combining weights: Optimize the CDF of the test statistic at the fusion center Maximize the global detection probability under the constraints on false alarm probability Significant cooperative gains are demonstrated

35 Info. Theoretic Aspects System Design using Cyclostationary Properties Work by H-S. Chen, W. Gao (Thomson Research) and D. G. Daut (Rutgers) Spectrum sensing in a very low SNR environment (-20dB) Sensing algorithms based on the measurement of the cyclic spectrum, or the spectrum correlation density (SCD) function of received signals Present three different SCD measurement methods and analyze the statistics of the SCD of stationary white Gaussian noise Present simulation results of the receiver operating characteristics of the three SCD methods

36 Capacity Limits of CR Info. Theoretic Aspects System Design Related work by A. Jovicic and P. Vishwanath and N. Devroye, P. Mitran and V. Tarokh System Model: Source: N. Devroye, P. Mitran and V. Tarokh, IEEE Comm. Mag. June 2006

37 Capacity Limits of CR Info. Theoretic Aspects System Design Source: N. Devroye, P. Mitran and V. Tarokh, IEEE Comm. Mag. June 2006

38 Spectral Efficiency of CRs Info. Theoretic Aspects System Design Work by M. Haddad, A. M. Hayar (Eurecom) and M. Dabbah (Supelec) Primary and cognitive users communicate with different receivers, with perfect spectrum sensing CR only transmits when the channel is idle Users successively transmit over available bands through water-filling Derive the total spectral efficiency of the CR system

39 Info. Theoretic Aspects System Design How Much Spectrum Sharing is Optimal? Work by S. Srinivasa and S.A. Jafar, UCI Look at the tradeoff between opportunistic access and licensed access in multi-user CR networks Goal: find the optimum number of secondary users to maximize the total throughput of the system Two cases: Perfect SS at secondary and zero interference tolerance at the primary and secondary receiver(s) Imperfect SS at secondary and non-zero interference tolerance at the primary receiver(s) Show that the optimal fraction of users lies b/w the extremes of fully opportunistic and fully licensed operation

40 Cognitive Medium Access Info. Theoretic Aspects System Design Work by S. Geirhofer and L. Tong, Cornell, and B. Sadler (Army research lab) Propose cognitive medium access (CMA) as a protocol for coexistence within independently evolving WLAN bands Formulate the problem of maximizing the throughput of the cognitive radio (subject to interference constraints) within the constrained Markov decision process framework Optimal control policy obtained via linear programming Also find structured solutions that provide more insight

41 Info. Theoretic Aspects System Design Can CR Work in a Frequency Planned Environment? Work by E. G. Larsson and M. J. Skoglund, KTH First order analysis of the impact on the SINR in a wireless network of CR users starting to transmit If cognitive devices are to be introduced, they need to: Be few in numbers (aggregate power scales with the number of CR users) Transmit with extremely low power (30dB below primary) Have very sensitive receivers (about 20dB more sensitive than the primary receivers in terms of C/(C + I)) Conclusion: introducing CR transmitters in a frequency planned cellular network without causing substantial interference is very challenging.

42 R & D Opportunities Fundamental research: better CR enabling algorithms, performance limits Cooperative & decentralized spectrum sensing algorithms Achievable rates/performance limits of cognitive radio Network capacity and scaling laws for CR networks Cognitive MAC protocols for secondary networks Implementation: building software-defined radios with CR capabilities Government: policy and regulation related research and recommendation Standardization activities

43 CR Related Resources Wikipedia entry: radio Joseph Mitola III s Thesis: jmitola/mitola Dissertation8 Integrated.pdf FCC: DARPA XG program: Europe s E2R project: CRT Wireless blog: Last, but not least, there s always:

44 The End Introduction Thank you very much!