Baiocchi, Cuomo, and Bolognesi

1. IP QoS Delivery in a Broadband Wireless Local Loop: MAC Protocol Definition and Performance Evaluation Baiocchi, Cuomo, and Bolognesi IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 18, NO. 9, SEPTEMBER 2000

2. Abstract • In this paper, a complete broadband wireless local loop (WLL) network is presented. • The proposal is based on the OFDM-CDMA technique, to which an added dynamic reservation/request MAC protocol is proposed. • Central to our proposal is the support of different QoS profiles. • As a case study, the explicit presentation of the IETF integrated services (IntServ) support over our WLL system is addressed. • We prove that our scheme achieves high utilization efficiency, as well as a fair share of the available radio capacity.

3. I. INTRODUCTION • FWA (fixed wireless access) Architecture • A centralized radio node (RN) • A group of fixed radio terminals (RT) • a customer premises network/equipment (CPN/CPE)

4. I. INTRODUCTION -2 • FWA exploits the OFDM-CDMA[1][2] (orthogonal frequency division multiplexing - code division multiple access) technique which provides • protection against fading, • peak-average power ratio reduction capabilities, • and high flexibility in band-width assignment. • Duplexing can be managed dynamicallyto provide tight tracking of traffic asymmetry, by sharing the available pool of codesbetween uplink and downlink(code division duplex).

5. II. SYSTEM ARCHITECTURE • NSL = network service layer (e.g. IP) • Three layers:(i) Adaptation layer(AL); (ii) MAC layer; (iii) Physical layer.

6. II. SYSTEM ARCHITECTURE -2 • NSL • corresponds to • classical network functions • addressing, • routing • traffic handling functions • packet flow description and classification, • admission control, • traffic policing and/or shaping • Examples of NSL • IP layer enhanced with QoS handling capabilities (e.g. IntServ or DiffServ) • ATM traffic control (e.g. CBR, VBR, ABR, UBR)

7. II. SYSTEM ARCHITECTURE -3 • AL • maps NSL traffic classes into MAC service classes • Two service types in the MAC layer • Guaranteed bandwidth (GB) • Best effort (BE) • AL flow mapping table • mapping NSL traffic classes into MAC service classes • Updated by NSL when a new flow is admitted • Segmenting and reassembling (SAR)

8. II. SYSTEM ARCHITECTURE -4 • MAC layer • Capacity assignment • Sharing radio capacity among flows. • Performed at the RN, by a centralized functional entity named MAC scheduler controller (MAC-SC). • Two service types in the MAC layer • Guaranteed bandwidth (GB) • Best effort (BE) • Physical layer • Coding and transmitting/receiving signals according to OFDM/CDMA.

9. II. SYSTEM ARCHITECTURE -5

10. III. PHYSICAL LAYER • A. Modulation Technique • OFDM • a multi-carrier technique • Advantages of OFDM • Immune to channel dispersion compared to a single carrier technique; • equalizers require much less computational effort than for single carrier systems; • intercarrier and intersymbol interference can be eliminated by introducing a guard time interval and a cyclic symbol extension between successive symbols. • Disadvantages of OFDM • More sensitive to local oscillator phase noise and to carrier frequency offsets.

11. III. PHYSICAL LAYER -2 • The OFDM modulation can support multiple access by means of • OFDM-TDMA • Each symbol interval (SI) is used for the transmission of K data symbols of the same user on the K OFDM subcarriers • Delay caused by collecting K data symbols from a user • OFDM-CDMA • One SI can be used for the transmission of data symbols belonging to K different users; (K = 512 commonly) • The contemporary transmission is obtained by multiplying each user data symbol by an orthogonal spreading code • OFDMA • Allows an intermediate type of multiplexing by permitting each user to transmit x data symbols on a set of subcarriers per SI .(1≦ x ≦ K) • Or combination

12. III. PHYSICAL LAYER -3

13. III. PHYSICAL LAYER -4

14. III. PHYSICAL LAYER -5 • Advantages of OFDM/CDMA and OFDMA • Low packetization delay • Flexibility in bandwidth assignment • As the granularity gets finer (i.e., x gets lower), the benefits of multicarrier transmission tend to disappear for OFDMA, because a smaller and smaller subset of the available subcarriers is actually used by each user. Instead, they are intact in case of OFDM-CDMA.

15. III. PHYSICAL LAYER -7 • B. The FWA Physical Layer • This paper assumes an OFDM/CDMA with FDD (frequency division duplex) technique. • This paper consider millimeter wave region of the radio spectrum because of the availability of larger bandwidth blocks. • The number of OFDM subcarriers is chosen to be 512. • Tradeoff: increasing number of subcarriers ＋ improves the multipath robustness, ＋ reduces the guard interval overhead, ＋ and increases the flexibility in bandwidth assignment; － increases the phase noise sensitivity, － makes base-band processing (i.e., FFT) more complex.

16. III. PHYSICAL LAYER -6

17. IV. MAC PROTOCOL • A. MAC Service Classes • Two service types in the MAC layer • Guaranteed bandwidth (GB) • Used for services with stringent requirements for delay and delay jitter, i.e. real time services (e.g. video and audio) • Traffic descriptors (TDs) are required (e.g. peak bit rate) for each information flow • Relevant admission control and flow parameter compliance checks must be defined in network layer • Best effort (BE) • To provide for economic use and efficient use • Capacity left from more demanding flows can be filled with traffic with loose requirements • To accommodate the existing Internet application traffic

18. IV. MAC PROTOCOL -2 • B. MAC Signaling • The radio capacity with OFDM-CDMA is structured as • Korthogonal codes that can be used simultaneously. • Each code is used in a TDMA fashion; a time slot carries a MAC_PDU (TSLOT = TMAC_PDU). • [RT ID, Other Info., Data Load] • Time is structured into frames (TFRAME) lasting N time slots by K * N * MAC_PDUs. • The structure is referred to as the TC-matrix (time slots-code matrix). • See [Fig. 6] for N=3

19. IV. MAC PROTOCOL -3 • The capacity assignment is performed frame by frame. • Each RT can transmit (uplink) on several time slot-code pairs (TC-pairs) without restrictions. • See the gray slots in Fig. 6

20. IV. MAC PROTOCOL -4 • The basic MAC signaling consists of • the request channel (ReqCh) • an UL ( Uplink Logical) channel to make capacity requests • the allocation channel (AlCh) • a DL (Downlink Logical) channel to answer the requests • The ReqCh and AlCh is structured in minislots. • A ReqCh-AlCh minislot pair is dedicated to each RT.

21. IV. MAC PROTOCOL -5 • The ReqCh in UL is structured in minislots: • Each minislot contains the bandwidth request. • [RT ID, Request GB Class, Request BE Class] • The request issued in the kth frame by each RT is just the number of MAC_PDUs of each service class found in the RT at the beginning of the kth frame for which there is no pending request.

22. IV. MAC PROTOCOL -6 • The AlCh in DL has the same minislot structure as the ReqCh. • A ReqCh-AlCh minislot pair is dedicated to each RT. • Each minislot contains the allocation reply. • [Starting Code, Starting Offset, No. of TC-pairs] • The RN uses the AlCh to signal to each RT • the number of assigned TC-pairs, • the starting code (the row of the TC-matrix) • the starting offset in the code row. • Detailed format and dimensioning of ReqCh and AlCh are re-ported in [14].

23. IV. MAC PROTOCOL –7 (1) UL Request Channel RN to RT (2) DL Allocation Channel RT to RN (N =3)

24. IV. MAC PROTOCOL -8 • C. MAC Fair Scheduling Algorithm (FSA) • Each RT stores arriving MAC_PDUs into its buffers by separating GB and BE packets: • BE traffic is queued up into a single FIFO buffer • GB traffic is split among a set of FIFO buffers • GB traffic has priority over BE ones. • The overall available capacity in each frame (K * N – H TC-pairs) is assigned to each RTs according to FSA. • FSA shares radio link capacity hierarchically among groups of users as to support decreasing QoS targets.

25. IV. MAC PROTOCOL -9 • The FSA is a practical realization of the fluid GPS [15] • which shares a fixed resource (capacity) among competing users, according to their actual load and to predefined weights. • A. K. Parekh and R. G. Gallager, “A generalized processor sharing approach to flow control in integrated service networks—The single node case,” in Proc. IEEE Infocom’92, 1992, pp. 915–924. • Here, the weight is related to the packet flow TDs and is passed to MAC layer by the NSL traffic control. • (1) the output of the FSA must be integer; • (2) tradeoff between bandwidth and complexity;

26. IV. MAC PROTOCOL -10 • The FSA divided the scheduling operation into two phases. • (1) The overall radio link capacity is shared among RTs, according to their overall requests and weights, by the RN MAC-SC; • (2) Each RT shares the bandwidth it obtained among the competingGB flows and, if possible, the BE traffic. • an RT can use a single FIFO buffer for GB traffic yielding the maximum capacity penalty; • individual queues can be handled per GB flow, resulting in the most efficient use of capacity although at the price of running a per-flow scheduling algorithm. • FSA is applied in each phase.

27. IV. MAC PROTOCOL -11

28. IV. MAC PROTOCOL -12 • Property 1: • If packet flows (with different TDs) requiring the same delay bound are FIFO multiplexed, the common delay bound can be met provided the output capacity of the FIFO mux is equal to that required by a GPS scheduler with the same input.

29. IV. MAC PROTOCOL -13 • The parameters used by the FSA are reported in (based on flows QoS requirements and TDs.)

30. IV. MAC PROTOCOL -14 • For the GB class (Step 1) • The overall capacity to be shared is S* = S - min{SBE,Σj Rqj,BE} • SBE : BW always left for BE traffic • Σj Rqj,BE : sum of all BE requests • FSA steps for GB • (1) Assign to each RT what is guaranteed : min{Rqj,GB, Wj,GB} • Rqj,GB: GB request of the jth RT • Wj,GB : GB weight, given by AC to ensure Σj Rqj,GB ≦ S – SBE (~ priority) • (2) If there are some requests still pending (i.e. requesting more than its guaranteed share), then redistributeresidual bandwidth to these RTs, according to their respective weights. • For the BE class (Step 2) • The overall capacity to be shared is S* = S -Σj Aj,GB • Aj,GB : BW assigned to the jth RT for the GB traffic (i.e. fairness for BE) (i.e. left from GB)

31. IV. MAC PROTOCOL -15 • ThecompleteFSAalgorithm

32. IV. MAC PROTOCOL -16 • The complete FSA algorithm • Step 1.1 • assigns up to the floor of the weight to each RT. • Step 1.2 • evaluates whether to assign one TC-pair for the fractional part of the weight . • Step 1.3 • evaluates whether to assign one more TC-pair on account of f_Wj to let small occasional bursts be transmitted even if • Step 2 • distributes residualbandwidth to RTs that have still some pending requests, according to a fair sharing (by a random round-robin). • Step 3 • updates the algorithm variables.

33. V. THE FWA SYSTEM AS AN ACCESS RSVP CLOUD • A case study resulting from the application of the FWA system within an IntServ enabled IP network. • We assume • The existence of an NSL • to provide (different profiles of) QoS, • to identify packet flows, • to possibly attribute them a “weight,” • expressing the amount of guaranteed capacity • expressing some priority criteria • The FWA AL and MAC only make use of these general (and minimal) capabilities.

34. V. THE FWA SYSTEM AS AN ACCESS RSVP CLOUD -2 • A. Motivation for the IntServ Case Study in the FWA • Work on QoS-enabled IP networks has led to two distinct approaches: • the integrated services (IntServ)architecture • the differentiated services (DiffServ) architecture

35. V. THE FWA SYSTEM AS AN ACCESS RSVP CLOUD -3 • IntServ • enables hosts to request per-flow, quantifiable resources, along end-to-end data paths • enables hosts to obtain feedback regarding admissibility of these requests (by using the RSVP resource reservation protocol). • lack of scalability (since complexity grows as the number of multiplexed flows) • DiffServ • targeting per class aggregate flows • no RSVP • enables scalability across large networks,

36. V. THE FWA SYSTEM AS AN ACCESS RSVP CLOUD -4 • In this context, an architectural solution is to support • the DiffServ paradigm in the core network • while a set of edge devices allow the interworking with IntServ hosts in the access section of the network. • QoS is provided by applying the IntServ model end-to-end across a network containing one or more DiffServ domains.

37. V. THE FWA SYSTEM AS AN ACCESS RSVP CLOUD -5 • Creating such an architectural framework requires several parts: (i) an explicit setup mechanism to request resources in accordance to the IntServ paradigm; (ii) a per flow traffic control at the edge of the network; (iii) the configuration of internal nodes (nodes of the DiffServ domains) so that aggregate flows have a well-defined minimum serving rate; (iv) the conditioning of aggregate flows (via policing and shaping) so that their arrival rates at any internal node are always less than the allocated capacity at that node. • (i) & (ii) : IntServ(iii) : explicit forwarding per hop behavior(iv) : the network boundary traffic conditioners

38. V. THE FWA SYSTEM AS AN ACCESS RSVP CLOUD -6 • B. IntServ Support in the FWA System:Admission Control • Assuming that the GB traffic is regulated by means of Dual Leaky Buckets (DLBs) • A packet flow can be characterized by only four parameters: • the peak rate (P bit/s) • the token bucket rate (r bit/s) • the bucket depth (b bit) • the maximum datagram size (M bit)(where P≧r and M≦b) • The amount of information that can be offered by a flow in a time interval of duration, t, is limited by X(t) ≦min{Pt+M,rt+b}

39. V. THE FWA SYSTEM AS AN ACCESS RSVP CLOUD -7 The maximum delay in the MAC layer to access the TC-Matrix The bandwidth negociated by the ith flow

40. V. THE FWA SYSTEM AS AN ACCESS RSVP CLOUD -8 The required bandwidth in TC-pairs/frame for the ith flow The admission verifies that The weight for the jth RT

41. V. THE FWA SYSTEM AS AN ACCESS RSVP CLOUD -8 The bandwidth Ri to assign for the ith flow

42. V. THE FWA SYSTEM AS AN ACCESS RSVP CLOUD -9 • C. IntServ Support in the FWA System: Signaling

43. VI. PERFORMANCE ANALYSIS • The simulated FWA comprisesfour RTs and a single RN. • Three types of traffic sources: (1) measured MPEG coded traces, used to model real time multimedia GB traffic; (2) measured LAN IP packet traces, used to model the BE traffic; (3) artificial sources with ad hocsynthesized emission pro-files (e.g., CBR and ON–OFF).

44. VI. PERFORMANCE ANALYSIS -2

45. VI. PERFORMANCE ANALYSIS -3

46. VI. PERFORMANCE ANALYSIS -4

47. VI. PERFORMANCE ANALYSIS -5

48. VI. PERFORMANCE ANALYSIS -6

49. VI. PERFORMANCE ANALYSIS -7 • A. Numerical Results and Discussion • The actual maximum delay incurred by GB packets is sensitively less than the target value (32 ms). • Delay fairness is achieved, as it is shown by the almost equal values of the delays of different RTs.

50. VI. PERFORMANCE ANALYSIS -7