Ad-hoc Networks with Smart Antennas – Medium Access Control

1. Ad-hoc Networks with Smart Antennas – Medium Access Control Karthikeyan Sundaresan GNAN Research Group

2. Outline • Ad-hoc or multi-hop wireless networks • Smart antennas • Smart antennas in multi-hop wireless networks • Gains from smart antennas that improve performance • Importance of tailored protocols • MAC for a specific type of smart antenna (MIMO links)

3. Ad-hoc Networks D S • Multi-hop wireless networks • Infrastructureless • Typically used in military applications (where there is no infrastructure), or disaster relief (where infrastructure has been destroyed) • A packet from a source reaches the destination by traversing multiple hops in between • Every node acts as a source for its own packets as well as router (forwards packets) for other nodes -multi-hop relay

4. Ad-hoc Networks (contd.) • Typical data rates (on a per-link basis) same as WLANs (~10Mbps) • End-to-end data rates can be significantly smaller (depending on network size, diameter of network, etc.) • Very different network environment (highly dynamic, routers also mobile!, etc.) • Smaller transmission ranges increase “spatial reuse” but also increase hop length • Multi-hop burden has significant impact on network capacity

5. MAC and Routing D S • Shared wireless medium requires access control • Lack of central coordinating entity requires a distributed channel access mechanism • Nodes take independent decisions during channel access • Multi-hop nature of flow requires route/path determination • Quality of route/path determines performance (throughput, delay, etc.) • MAC and routing are important aspects of ad-hoc networks

6. Medium Access Control (MAC) • When multiple stations share a common channel, the protocol that determines which station gets access to the shared channel • Key characteristics based on which MAC protocols are evaluated: utilization and fairness • Channel partitioned approaches • FDMA, TDMA, CDMA • Random multiple access schemes • ALOHA, slotted-ALOHA • CSMA • CSMA/CA

7. CSMA/CA RTS CTS DATA ACK A B C D • Carrier sense multiple access with collision avoidance • Four-way handshake between Tx-Rx for every packet transmission • RTS (request to send) and CTS (clear to send) control packets used for channel reservation; ACK for acknowledgement

8. IEEE 802.11 • IEEE 802.11 has 2 modes – PCF and DCF • PCF – point coordination function • Access point is central coordinating entity • Contention free protocol • DCF – distributed coordination function (CSMA/CA) • Ad-hoc mode – completely distributed operation • Contention based protocol • Ad-hoc networks use existing WLAN MAC protocol – CSMA/CA by default • Not necessarily efficient but simple distributed implementation

9. Routing • Routing is essential due to multi-hop nature • Lot of research into designing efficient routing protocols • Key aspects • Quick route detection • Efficient and low-overhead route maintenance • Quality of the routes (high throughput, low delay, low energy consumption, etc.) • Single and multi-path routing protocols • Pro-active and reactive routing protocols

10. Proactive vs. Reactive Routing Table-driven routing protocols Faster route discovery High overhead in dynamic networks Better performance in heavy traffic, small networks High overhead in large networks Eg. DSDV On-demand, source-initiated Slower route discovery Low overhead in dynamic networks Large overhead in heavy traffic, small networks Low overhead in large networks Eg. DSR

11. Outline • Ad-hoc or multi-hop wireless networks • Smart antennas • Smart antennas in multi-hop wireless networks • Gains from smart antennas that improve performance • Importance of tailored protocols • MAC for a specific type of smart antennas (MIMO links)

12. Smart Antennas LS Beamformer 1 X1(t) W1 Array output 2 X2(t) W2 y(t) N Xn(t) Wn Control algorithm Ref. Error - + Signal processor + Adaptive processor • Smart antennas represent sophisticated PHY layer antenna technology coupled with significant signal processing • Multiple element arrays (MEAs) that vary from simple switched beam to sophisticated digital adaptive arrays • They provide fundamental performance improvement over omni-directional antennas; also help overcome the impairments of the wireless channel

13. Types of Smart antennas • Switched beam • Pre-determined beam pattern • Inefficient in multipath, require LOS • No flexible interference suppression • Adaptive array • Adaptive beam pattern • Degrees of freedom (DOFs) handle interference suppression • Single DOF for transmission • MIMO links • Multipath is exploited ! • Flexible usage of DOFs • Spatial multiplexing and diversity

14. Exploitable Gains • Directional / array gain (link) • Increases the mean SNR of transmission by focusing energy between communicating pairs • Used for extending communication range, increasing data rate, increasing reliability, minimizing transmit power • Diversity gain (link) • Reduces error probability on link due to fading by introducing redundancy in transmitted streams • Used for increasing communication reliability, range extension, minimizing transmit power • Spatial multiplexing gain (link) • Increases the capacity of the link directly by de-multiplexing independent data streams on different antenna elements and exploiting rich multipath scattering • Interference suppression gain (network) • Ability to form nulls towards interferers; allows multiple transmissions co-exist in same region • Referred to as “spatial reuse” gain

15. Relevant PHY properties Switched beam Adaptive array MIMO Links Open and closed loop communcation DOFs serve transmission & interference suppression Communication pattern Directional communication Fixed beam pattern Adaptive beam pattern DOFs determine interference suppression Array & diversity gains Large angular spreads impact array gain Array, diversity & spatial multiplexing gains Multipath is essential for multiplexing & diversity Directional/ antenna gain Scattering limits gain Potential gains DOFs decide # interfering streams Spatial correlation important Suppression along non-active beams Side lobes accumulate noise DOFs decide # interfering streams Spatial correlation important Interference suppression

16. Link Strategies • Rate • Increase in SNR or capacity used to increase transmission rate or send multiple packets simultaneously; range and error probability unchanged • Range • Gain in SNR used to increase communication range; rate and error probability unchanged • Reliability • Gain in SNR used directly to increase link reliability; rate and range unchanged • Power • Gain in SNR used directly to reduce transmission power

17. Illustration Interference suppression Omni Rate Directional gain Switched Beam Range Array gain Adaptive Arrays Multiplexing gain Reliability MIMO Links Diversity gain Power

18. Outline • Ad-hoc or multi-hop wireless networks • Smart antennas • Smart antennas in multi-hop wireless networks • Gains from smart antennas that improve performance • Importance of tailored protocols • MAC for a specific type of smart antennas (MIMO links)

19. MWNs with Smart Antennas • Related work • Protocol design for ad-hoc networks • MAC: [Tseng’02, Yang’02, Sadeghi’02 , Muqattash’03, Qiao’03] • Routing: [Haas’98, Johnson’99, Perkins’99, Li’01, Srinivas’03] • Protocols developed for omni-directional antennas • Smart antenna research • Extensive research at the physical layer of protocol stack: [Winters’99, Foschini’ 99, Catreux’00, Zheng’03, Gesbert’03] • Results extensible to single-hop wireless networks but cannot be applied to multi-hop wireless networks due to their varied characteristics • Advantages of smart antennas to higher layers of the MWN protocol stack is still an open problem • Objective: Identify the advantages of smart antennas to MAC and routing layers and design appropriate protocols to leverage their benefits

20. Need for New Protocols • Existing protocols (MAC and routing) do not exploit the gains of smart antennas • Eg. Directionality (beam pattern adaptation) increases spatial reuse

21. Relevance to Protocols • Two main advantages to networks • Different kinds of link gains (strategies) and interference suppression gain • MAC layer • The most important layer in helping realize the true potential of smart antenna gains • Need to identify optimization considerations to increase network utilization • Leverages link gains through rate/range/reliability increase or power decrease; and also leverages interference supp. gain for spatial reuse • Routing layer • Different link strategies (rate/range/rel/power) used to determine “quality” routes • Range extension can help avoid partitions • Power minimization can help in designing better energy-efficient routing protocols • Link strategies also improve routing layer mechanisms • As we go further higher up the stack, the impact of smart antennas diminishes!!

22. MAC with MIMO Links

23. Outline … • Characteristics/capabilities of MIMO links • Optimization considerations for the MAC protocol • SCMA (Stream Controlled Medium Access) protocols • Performance evaluation

24. Characteristics of MIMO links Spatial multiplexing gain Diversity gain • Multiple Input Multiple Output (MIMO) is an antenna technology that provides high spectral efficiencies • MIMO is the key to handle multipath efficiently! • Do not require LOS and can operate in rich multipath environments • Capable of diversity and spatial multiplexing gain • Spatial multiplexing provides a linear increase in asymptotic capacity while diversity gain provides diminishing returns with large antenna elements [1] • Objective : To exploit spatial multiplexing gain to increase the capacity of the system • Spatial multiplexing gain increases the link capacity • Independent streams are transmitted simultaneously • Diversity gain reduces error probability on link to increase reliability during fading • Introduces dependency amongst transmitted streams [1] D. Gesbert et al., “From Theory to Practice: An Overview of MIMO Space-Time Coded Wireless Systems”, IEEE JSAC, 2003.

25. Capabilities of MIMO links • Adaptive resource usage • Number of elements correspond to “degrees of freedom” (DOFs) or “resources” at a node • Data transmitted on the different elements is given the abstraction of “streams” • Resources can be used for spatial multiplexing or diversity • Capacity-Reliability-Range tradeoff • Diversity increases link reliability and also provides increased range • Spatial multiplexing increases system capacity • Flexible interference suppression • Can suppress as many interfering streams as the number of DOFs in uncorrelated fading

26. Outline • Characteristics/capabilities of MIMO links • Optimization considerations for the MAC protocol • SCMA (Stream Controlled Medium Access) protocol • Performance evaluation

27. Simple CSMA/CA extension • Carrier sense multiple access with collision avoidance is the MAC protocol used in IEEE 802.11b DCF mode • RTS-CTS-DATA-ACK exchange • Is there a simple extension to CSMA/CA that can exploit spatial multiplexing? • Yes, with appropriate tuning of timers and other constants • CSMA/CA that spatially multiplexes on “k” elements is referred to as CSMA/CA(k) • CSMA/CA(k) provides reasonable scalability in elements • Is this the best performance we can expect?

28. Optimization considerations (1) 0.9 0.78 0.61 0.48 • Stream control • Every data stream has a corresponding stream gain • Stream gains are disparate in low and moderate SNR • It pays to operate multiple links simultaneously on smaller subset of resources than to operate a single link on all resources • Consideration 1: Multiple interfering links operating simultaneously using stream control achieve overall better throughput performance

29. Optimization Considerations (2) Flexible interference suppression Number of independent interfering streams important Spatial correlation and power of the interfering streams also matter Efficient usage of resources for desired transmission Consideration 2: Flexible interference suppression in conjunction with stream control helps create additional resources and hence additional gain Avg = 4 streams/slot Avg = 6 streams/slot

30. Optimization Considerations (3) Passive receiver overloading In a CSMA/CA(k) scheme passive receivers belonging to multiple contention regions can be overloaded in resources to increase spatial reuse In pure stream control, receivers belonging to multiple contention regions can potentially degrade performance by reducing spatial reuse Consideration 3: Receivers belonging to multiple contention regions must not perform stream control Avg = 8 streams/slot Avg = 4 streams/slot

31. Design Elements • Stream controlled must be performed to leverage the knowledge of channel gains • Stream control also helps utilize the available resources efficiently, thereby making it possible to use the additional resources created due to weak interference • Determine links that alone should perform stream control to overcome “passive receiver overloading” problem • Links belonging to a single contention region (“white” links) must alone perform stream control to enable receiver overloading • Links belonging to multiple contention regions (“red” links) must operate on all available streams • SCMA (Stream-Controlled Medium Access) • Incorporates above elements to improve network utilization

32. Outline • Characteristics/capabilities of MIMO links • Optimization considerations for the MAC protocol • Stream control • Flexible interference suppression • Passive receiver overloading • SCMA (Stream Controlled Medium Access) protocol • Performance evaluation

33. SCMA Components (1) 6 1 2 3 4 5 e b a c d f a d f 7 b c e • Graph generations • Network topology is represented as a network graph • Contention between active links is captured in the flow contention graph • 3 essential components: clique identification, coloring and schedule Node graph Flow contention graph

34. SCMA Components (2) d d f f a a d d a a c c e e c b b c c c a(2) d(1) f a(3) d(2) f a f b e a a(3) a d(2) d d f f f b c(1) e b e b b b c(1) c c e e e • Clique identification, ranking and coloring • Maximal cliques in flow contention graph correspond to contention regions in the network • Ranking is done based on tuple (clique degree, max clique size) • Bottleneck links are colored red based on rank and non-bottleneck links are colored white

35. SCMA Components (3) a b c d e f Slot 1 0 0 4 0 0 0 Slot 3 2 2 2 2 0 0 0 0 2 2 2 2 Slot 2 0 4 0 4 0 0 Slot 4 a d(2) f b c(1) e • Dual-level scheduling • Red links are scheduled first based on their rank • White links are scheduled next and perform stream control Flow contention graph

36. Recap • Step1: Obtain the network graph and hence the flow contention graph • Step 2: Identify all maximal cliques in the flow contention graph and color bottleneck links as “red” recursively based on rank and non-bottleneck links as “white” • Step 3: Perform dual scheduling with white links alone exploiting stream control and partial interference suppression

37. Protocol Handshake Mechanism • Persistence is used instead of back-off for contention • Adapted based on link color • Follows the RTS-CTS-DATA-ACK handshake as in IEEE 802.11b DCF protocol • Stream control requires knowledge of channel gains • The initial control packets (RTS, CTS) lack information on channel conditions and are transmitted using OL-MIMO • The diversity instead of spatial multiplexing gain is used to provide increased link reliability/longer range for control packets to be decode-able in two hop region • The control packet exchanges along with training sequences are used to estimate the channel gains • DATA and ACK packets are sent using spatial multiplexing

38. Distributed SCMA Components • Distributed coloring • Nodes locally color their links using neighborhood information • Fair share adaptation • White links adapt to the fair share of allocation in their contention region • Co-ordinated scheduling • The first white link to be scheduled in a contention region co-ordinates the schedule of the remaining white links

39. Distributed Coloring • Whenever a link enters a network/region, it operates on all ‘K’ streams - initially all links use K streams • Every uncolored link observes the channel access pattern between its successive accesses • Colors “red” if multiple overlapped transmissions received in any slot: # streams received > K • Tx and Rx of the link exchange local channel access patterns in deciding red color • Colors “white” if no overlapped transmissions: ≤ K streams • Closely approximates the centralized coloring

40. Fair Share Adaptation • New link entering region first determines color • If “red” - operates on all K streams • If “white” - needs to identify number of streams to be used • After coloring white, continues to transmit on all K streams for one slot along with color information • Every white link locally keeps track of number of white links in the region and hence fair share • If additional resources available (flexible interference suppression, white link departure) fair share adapted locally

41. Coordinated Scheduling • Channel access decisions by white links are made locally/probabilistically • Does not ensure all white links in a region transmit simultaneously to exploit stream control to best extent • Achieved by co-ordinated scheduling • First white link to gain channel access, sends a “co-ordinate schedule” directive to other white links in the region through its RTS/CTS on diversity mode • Other white links use the directive to transmit DATA/ACK in parallel with the coordinating link • Persistence values are appropriately adjusted for the co-ordinating white link and other white links in the schedule

42. Performance evaluation • Simulation environment • Event driven network simulator • Toy topologies and random topologies using random way-point model in ‘setdest’ tool • 100 nodes in 750m * 750m grid • Backlogged sources with rate of 100 pkts/sec • CBR as application with a packet size of 1024 bytes, DSR as routing protocol • MAC Protocols considered: CSMA/CA, CSMA/CA(k), PFCR(k) and SCMA • Metrics studied • Aggregate Throughput and normalized standard deviation in throughput

43. Results Cent. SCMA Dist. SCMA CSMA/CA(k) CSMA/CA(k) PFCR(k) PFCR(k) Dist. SCMA CSMA/CA • Aggregate throughput scales better with number of elements in SCMA due to stream control • SCMA exhibits better fairness properties than CSMA/CA(k) and PFCR(k)