238_1_lecture9 - Lecture 8: Advanced topics in multimedia...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Lecture 8: Advanced topics in multimedia communications: Delay-sensitive Cognitive Radio Networking and Multimedia transmission over multi-hop wireless networks 1 Cognitive radio networks Wireless users are able to sense and learn the environment and correspondingly adapt their transmission strategies. Identify available spectrum resources (wireless sensing to identify the existence of primary users) Spectrum sharing (resource management) (information exchange to identify the actions of other competing secondary users) Spectrum sharing is the main challenge for users in cognitive radio networks and ISM band 2 Motivation – why cognitive radios are important for media streaming? Increasing demand for video applications over wireless networks Emerging cognitive radio networks – standards, e.g. IEEE 802.22 Problems of existing work in cognitive radio networks Dynamic networks Delay-sensitive applications such as multimedia streaming Multi-user environment 3 Network Settings Secondary users SU 1, SU 2 ,..., SU N channels Autonomous competition F1 PU 1 F2 PU 2 F3 F4 PU 1 PU 3 PU 4 “spectrum holes” PU 4 time Frequency channels: F = {F1,..., FM } Primary users: PU = {PU 1,..., PU M } Autonomous secondary users: SU = {SU 1,..., SU N } 4 Main Challenges Decentralized nature of cognitive radio networking environment Autonomous secondary users facing heterogeneous environmental conditions Multimedia characteristics need to be considered Loss-tolerance, delay-sensitivity, and priority Coupled decision making due to the wireless interference 5 Problem formulation for autonomous users a i = [ai1, ai 2 ,..., aiM ] ∈ {0,1}M Actions: Questions: = [si1, i 2 ,..., siM ] ∈ is suitable Strategies: si generalsutility function [0,1]M 1. What for video applications? Utility (video quality) functions: ui (ai , a−i ) 2. What information does an autonomous user need to dynamically evaluate such Expected utility functions with dynamic an utility function? adaptation sopt (s−i ) = arg maxU i (si , s−i ) i si U i (si , s−i ) = E(si ,s−i )[ui (a i , a −i )] 6 Utility model for video applications Video characteristics: Expected quality impact: λk (λ2 ≥ ... ≥ λK ) Average packet length: Lk Challenges: Number of packets one GOP duration: N k 1. Coupling between users for channel selection Delay deadline: Dk (DK ≥ ... ≥ D2 ) 2. Different priorities among users (traffic classes) 3. Users Successful probability: may experience different channel conditions ≠ 1 and C ≺ C , if Pksucc ⎧0 ⎪ k' k ' Pksucc = ⎪ ⎨ ⎪ (1 − PSolution: kPrioritized virtual queue enables users k ) = E [I (d ≤ Dk )], otherwise, ⎪ ⎩ to autonomously determine the expected utility sopt = arg max λk Lk N k Pksucc (si , different Problem considered:which they cansiobtain when deploying s−i ) i channel selection C k ∈Vi strategies ∑ Priority classes: C = {C 1,...,C K } Primary users: C 1 Secondary users (quality layers): {C 2 ,...,C K } 7 Physical queues and prioritized virtual queues PU 1 PU 2 PU M Physical queues at the secondary users SU 1 V1 s1 j to F1 to F2 a11 Cognitive Radio Network a12 aij = 1 aij = 0 to FM SU 2 V2 s2 j to F1 to F2 a1M a21 a22 a 2M aN 1 aN 2 aNM F1 F2 FM to FM SU N VN sNj to F1 to F2 to FM Virtual queues for different frequency channels More details regarding the MAC problem can be found in [Shiang, van der Schaar, Trans. on MM, Aug 2008] 8 Evaluating successful probabilities Given: Loading of primary users (1) ρ j , j = 1,..., M the first two moments of the 2 (2) TS−i , TSi = [Ci , Bi , Li , Xi , Xi ] service time of the secondary users Lk 2 (1 + pij ) Lk 2 E [X jk ] = , E [X jk ] = 2 , if C k ∈ Vi 2 Tij (1 − pij ) Tij (1 − pij ) Problem considered: maximize si ∑ C k ∈Vi M λk Lk N k ∑ sij (1 − Pjk (si , s−i )) j =1 9 Proposed Solution Priority-Scheduling-based Channel Selection (PSCS) Action information exchange ai Secondary user SU i ∈ C k Priority virtual queue Interface a−i ai Strategy modeling(si , s−i ) Channel Primary user estimationRij modeling Priority virtual queuing analysis SINR Channel sensing U i (si , s−i ) Dynamic Strategy Learning ai Frequency selection based on strategy Packet transmission sopt (s−i ) = arg maxU i (si , s−i ) i si TSi , TS−i Traffic specification exchange 10 Simulation Results “Coastguard” CIF format frame rate:30Hz Lk = 1000bytes, Dk = 500ms Scheme comparison: • Priority-Scheduling-based channel Selection (PSCS) • Static Allocation (SA) • Dynamic Least Interference (DLI) (N = 6, M = 10) 11 Simulation Results “Coastguard” CIF format frame rate:30Hz Lk = 1000bytes, Dk = 500ms Scheme comparison: • Priority-Scheduling-based channel Selection (PSCS) • Static Allocation (SA) 34 • Dynamic Least Interference (DLI) Average Y-PSNR (dB) 32 30 28 PSCS SA DLI 26 24 22 20 25 30 35 Number of secondary users (N) 40 (M = 10) 12 Preliminary conclusions • Prioritized scheduling solution provides an efficient model to rigorously quantify the impact of other users (either primary or secondary) on the autonomous user’s received video quality • Prioritized scheduling analysis provides a fast evaluation of the expected utility given the source and channel dynamics. The effect of dynamic channel selection can then be predicted/forecasted when some dynamics change • Such distortion/priority-driven solutions outperform existing channel selection schemes, since they do consider video characteristics, such as delay-sensitivity and priority, and – are not able to make forecasts about the resulting rewards associated with different dynamic channel selection strategies – 13 Resource management in multi-hop infrastructure Majority of the resource management research in cognitive radio networks focuses on a single-hop wireless infrastructure Main differences for multi-hop infrastructure Network resources include Vacant frequency channels (spectrum holes) Links to different relays Transmission strategies need to be adapted at both source nodes and relay nodes. Informationally-decentralized resource management is required “Value” of information Gathering information in a timely fashion across different hops 14 Network settings General topology graph G {M, N, E, F} Primary users: M = {m1,..., mM } Autonomous network nodes (i.e. secondary users, relays): N = {n1,..., nN } Network links: E = {e1,..., eL } Frequency channels: F = {f1,..., fM } Set of resources: R = {(e, f ) | e ∈ E, f ∈ F} Resource Matrix: Rn = [Rij ] ∈ {0,1}L×M ⎧1, if link ei is connected to the node n ⎪ ⎪ ⎪ Rij = ⎪ and the frequency channel f j is available. ⎨ ⎪ ⎪ 0, otherwise. ⎪ ⎪ ⎩ 15 Example of the multi-hop cognitive radio networks m1 F = {f1, f2 } f1 N = {n1, n2 , n 3 } n2 E = {e1, e2 , e3 } e1 m2 Interference Interference matrix of the primary users: f1 f2 e1 ⎡⎢ 0 1 ⎤⎥ ⎢ ⎥ I1 = e2 ⎢ 1 1 ⎥ ⎢ ⎥ e3 ⎢ 0 1 ⎥ ⎣ ⎦ n1 e3 n3 e2 Resource matrix at each node: f1 f2 f1 e1 ⎡⎢ 1 1 ⎤⎥ e1 ⎡⎢ 1 ⎢ ⎢ ⎥ R1 = e2 ⎢ 1 1 ⎥ R2 = e2 ⎢ 0 ⎢ ⎢ ⎥ e3 ⎢ 0 0 ⎥ e3 ⎢ 1 ⎣ ⎦ ⎣ f2 f1 1 ⎤⎥ e1 ⎡⎢ 0 ⎢ ⎥ 0 ⎥ R 3 = e2 ⎢ 1 ⎢ ⎥ e3 ⎢ 1 1⎥ ⎣ ⎦ f2 0 ⎤⎥ ⎥ 1⎥ ⎥ 1⎥ ⎦ 16 Interference characterization Available resource depends on Topology connectivity Interference of other traffic using the same frequency channel Prioritization of the traffic Primary users: C 1 Secondary users: {C 2 ,...,C K } Interference Matrix Interference from primary users Available Resource Matrix R(I ) = Rn ⊗ Ik −1 ⊗ ... ⊗ I1 nk Interference from competing secondary users in class C k I 1 = [I ij ] ∈ {0,1}L×M , where ′ ] ∈ {0,1}L×M , where Ik if= [I ijprimary user is occupying frequency channel f ⎧1, the j ⎪ ⎪ ⎪ the ⎪ link ei can interfere with the primary can I ij = ⎪ and ⎧1, if link ei using frequency channel f j user.be ⎨ ⎪ ⎪ ⎪ ⎪ 0, otherwise. ′ ⎪ I ij = ⎪ interfered by the traffic of priority class C k . ⎨ ⎪ ⎩ ⎪ ⎪ 0, otherwise. ⎪ ⎪ ⎩ 17 Delay evaluation Effective Transmission Time (ETT): Lk ˆ , An = (e, f ) ∈ An (k ) ETTnk (An ) = T (e, f ) × (1 − p(e, f )) Feasible action set: ˆ An (k ) = {A = (e, f ) | R(I ) = [Ref ]L×M , Ref = 1} nk Delay of a packet (data unit) in C k dij (σij (Ai )) = ∑ ETTnk (An ) n ∈σij Routes of a packet (data unit) σij = {(e, f ) | the jth packet of Vi flows through link e using frequency channel f } 18 Problem formulation Centralized approach with global information at the sources opt Ai = arg max ui (Ai ,Gi ) ˆ s.t. A ∈ An for all A ∈ Ai Utility for a delay-sensitive application u i (A i ,G i ) = qi ∑ w(λij ) ⋅ Prob{dij (σij (Ai ,Gi )) ≤ Dij } j =1 Dij = Dk and λij = λk if j ∈ C k Greedy algorithms that performs optimization sequentially from highest priority to lowest, i.e. for C k Aopt = arg min ik ∑ dij (σij (Aik ,Gi )) j ∈C k ˆ s.t. dij (σij (Aik ,Gi )) ≤ Dk and A ∈ An for all A ∈ Aik 19 Problem formulation Proposed distributed approach with local information at each node Application layer scheduler according to the impact factor λij Delay decomposition P dij (σij ) = dn (σij ) + E [dn (k, σij )] Distributed optimization problem Questions: opt An 1. What should be included in the = arg min E [dn (k, σij (An ,Ln ))] local information Ln ? 1. With whom should a node P s.t. E [dn (k, σij (An ,Ln ))] ≤ Dk − dn (σij ) − ρ, j ∈ C k exchange local information with? ˆ ,An ∈ An Store and exchange E [dn (k , An )] = ETTnk (An ) + E [dn '(An ) (k )] Constrained Bellman-Ford shortest-delay routing algorithm 20 Dynamic resource management with information constraints Local information gathered from neighbor node that is x hops away I n (x ) = {Ik (nx , Anx ), Anx , dnx | nx ∈ Nn } x Information cell Nn – network nodes within the h hops x Ln (h ) = {I n (x ) | x = 1,..., h } Dynamic resource management given the info. cell opt An (t ) = arg min E [dn (k, σij (An ,Ln (h, t - 1)))] subject to E [dn (k, σij (An ,Ln (h,t - 1)))] ≤ P ˆ Dk − dn (σij ) − ρ, j ∈ C k , An ∈ An (t − 1) 21 Tradeoff on the information cell Benefit of acquiring more information (less constrained) More choices to select relays Avoid “information exchange mismatch problem” Interference range of Information horizon n2 (b) (a) m1 n1 m1 n1 An1 , Ik (n1 ), E [dk (n1 )] An1 , Ik (n1 ), E[dk (n1 )] n2 An5 , Ik (n5 ), E[dk (n5 )] n5 n2 An4 , Ik (n4 ), E [dk (n4 )] An3 , Ik (n 3 ), E[dk (n3 )] An3 , Ik (n 3 ), E [dk (n 3 )] n5 n4 An4 , Ik (n 4 ), E [dk (n4 )] n4 n3 n3 An6 , Ik (n6 ), E[dk (n6 )] n6 Decision making dI (Ln (h )) Cost of gathering information dI (Ln (h )) n 6 Packet transmission dI (Ln (h )) tI (ν ) 22 Learning the feasible action set Adaptive fictitious play (AFP) Propensity rA (n ′, t ) = α × rA (n ′, t − 1) + I (An ′ (t ) = A) Estimated frequency selection strategy rA (n ′, t ) ∑ rA (n ′, t ) sA (n ′, t ) = A∈(En ′ ,Fn ′ ) Update the interference matrix e Ie = [I ij ] = k ∑ n ′ ∈−n (k ) Ik = [I ij | I ij Ik (n ′) = ∑ ∑ sA(n ′)Ik (n ′, A) n ′ ∈−n (k ) A e ⎧ 1, if I ij ≥ μ ⎪ =⎪ ⎨ e ⎪ 0, if I ij < μ ⎪ ⎩ Update the feasible action set R(I ) = Rn ⊗ Ik −1 ⊗ ... ⊗ I1 nk 23 Dynamic resource management algorithms : blocks that are not covered in this paper Periodic information exchange algorithm Zn Channel Channel sensing for primary users Ik (−n ), A−n Info. exchange among secondary users Minimum-delay route/channel selection algorithm Determining resource matrix using AFP R(I ) Priority scheduled packet buffer Ck nk Interference matrix d−n Select a feasible Minimum-delay action that minimizes Route/channel E [dn selection(k, An )] Delay vectors RTS/CTS coordination {Ik (n, An ), An , dn } RTS (An ) CTS (An ) RTS/CTS coordination An Packet transmission Information update 24 Conclusions Two types of interference in cognitive radio networks are considered to form the feasible action set: from channel sensing and from explicit information exchange. We described the information constraints in multihop cognitive radio networks – what should be exchanged and who should a node exchanges info. With Open problems: What happens is users experience dynamics? Cooperative and non-cooperative solutions 25 Thank you For information on our research Multimedia Communications and Systems Lab, please visit http://medianetlab.ee.ucla.edu 26 ...
View Full Document

Ask a homework question - tutors are online