Hybrid Networks: Cellular-Relay Architecture Harish Viswanathan, Sayandev Mukherjee and Ram Ramjee Bell Labs, Lucent Technologies Murray Hill, NJ
Network Architecture for 4G � Macro-cell architecture may not have enough capacity and coverage for ubiquitous high data rates � Alternatives – Pico-cellular, Hierarchical, Ad Hoc, etc. – Integrated cellular and WLAN � Focus: Can macro-cellular capacity be significantly enhanced through deployment of cheap “pseudo base stations” or relays ? – Multiple hops to and from base station through the relays – Intermediate step towards multihop through terminals 2 Lucent Technologies 7/4/2005
Cellular Networks with Relays Relays decode received packets, then store and forward them on the same/different frequency to terminals Relays can improve coverage and capacity Mobile relays to cover hot spots Adaptation to traffic imbalance Additional delay due to multihop 3 Lucent Technologies 7/4/2005
Cellular-WLAN Architecture Intermittent connectivity/low data-rate Opportunistic Base Station Relaying using high data-rate dual-mode Poor WAN Relay terminals Channel Proxy Good WAN 802.11b Channel � Proxy discovery and maintenance � Incentives � Routing – WAN operator – Node mobility – Proxy & relay – 3G channel dynamics 4 Lucent Technologies 7/4/2005
Cellular-Relay Architecture • Service provider deployed relays Base to cover hot spots or enhance Station coverage (eg. in-building) • Same spectrum is used for base to relay and relay to user transmission • Low cost base stations with no backhaul � Routing and Scheduling Wireless Base Stations � Signaling � Location of Relays 5 Lucent Technologies 7/4/2005
Ev-DO Based System Medium Access Features Downlink Frame Structure Base TX Slot 1 Slot 2 Slot 3 • Time division multiplex transmission with spreading • Base and relay transmissions identified by unique PN Pilot Preambles Data scrambling (allows simultaneous transmission) Relay Tx/Rx • Rate adaptation based on link quality • CDMA reverse link Pilot Preamble • Link quality feedback on the uplink (as in 1X EV-DO) – helps deal with mobility across relays 6 Lucent Technologies 7/4/2005
Why do relays provide a gain? Multiple simultaneous transmissions possible With increasing number of relays the size of each relay region decreases - transmit power can be reduced With power control, one transmission for every k relays even for a large number of relays 7 Lucent Technologies 7/4/2005
Large System Performance Gain N Number of simultaneous transmissions ~ k Number of Hops ~ N Throughput Gain ~ N Assumes no peak rate limit from base station 8 Lucent Technologies 7/4/2005
First Ring Bottleneck In the multihop system, data has to be first transmitted to the first ring of relays before the reuse efficiency kicks in In the limit of large number of hops transmission from the base to the first ring can be the bottleneck since base has to transmit one at a time to each of the receivers in the first ring - peak rate limit Multiple antennas at the base station can alleviate the problem ≤ R peak Gain R avg 9 Lucent Technologies 7/4/2005
First ring upper bound with peak rate limit First Ring Upper Bound 6.5 Rate from base 35 dBm Cell Radius = 2 Km 6 40 dBm to first ring of 45 dBm 5.5 relays is limited by peak SNR 5 achievable 4.5 Gain One hop 4 performance is 3.5 sensitive to the 3 base transmit power while p-hop 2.5 performance bound 2 is not 1.5 12 14 16 18 20 22 24 26 28 30 Peak SNR (dB) Higher gains can be achieved with multiple antenna transmission in the first hop 10 Lucent Technologies 7/4/2005
Routing and Scheduling with relays � Which terminals to schedule at each time step? � Constraints: – Relays cannot transmit and receive in same frame – Queues at base and each relay for every user; packets transmitted by relay must have been queued from base to relay first – Transmission rates dictated by interference power � Optimization: given instantaneous feedback about transmission rates on all links, determine the optimal set of active links on the next frame so as to achieve throughput-optimality : stable queues for each user for largest possible set of arrival rates 11 Lucent Technologies 7/4/2005
Throughput-Optimal Scheduling (Derived from Tassiulas & Ephremides 1992) − ∑ i index of user max D r ij j ∈ − r R ( ) t j index of the link i j , − q j ( ) source node of link = − ( ) ( ) D Q t Q t ij iq j ( ) ih j ( ) − h j ( ) destination node of link R(t) is the feasible set of link rates that depends on power allocation and interference Algorithm determines optimal reuse based on fading conditions 12 Lucent Technologies 7/4/2005
Simulation Setup � Cell has three 120º sectors; 20 users/sector and 0, 1, 2, 3 or 4 relays/sector � Relays are at a distance of half the cell radius from the center, on lines that divide the central angle of the sector equally � In each slot, packets arrive independently for each user with equal probability. The size of the packets is exponential with a chosen mean � Cell radius is 2 km; we assume full interference from two rings of cells around the cell in question � Path loss model is COST231-Hata at 1900 MHz, with base height 30m, Relay height 3m, and user terminal height 1.5m 13 Lucent Technologies 7/4/2005
Impact of number of relays Effect of number of relays 3 Mean aggregate throughput in cell (nats/symbol) 2.5 2 1.5 1 Base only (40 dBm) Base + 1 Relay (37 dBm) Base + 2 Relays (37 dBm each) 0.5 Base + 3 Relays (37 dBm each) Base + 4 Relays (37 dBm each) 0 0 0.5 1 1.5 2 2.5 3 Mean aggregate load in cell (nats/symbol) 14 Lucent Technologies 7/4/2005
Impact of Relay Transmit Power Effect of relay power 2.5 Gain saturates with Mean aggregate throughput in cell (nats/symbol) increasing relay power 2 1.5 1 Base only (40 dBm) 0.5 Base + 4 Relays (20 dBm each) Base + 4 Relays (30 dBm each) Base + 4 Relays (37 dBm each) 0 0 0.5 1 1.5 2 2.5 Mean aggregate load in cell (nats/symbol) 15 Lucent Technologies 7/4/2005
Results with Constant Total Cell Power Performance of throughput-optimal scheduling policy with constant total cell power 2.5 Mean aggregate throughput in cell (nats/symbol) 2 Distributing the power is better 1.5 1 Base only (40 dBm) 0.5 Base (37 dBm) + 1 Relay (37 dBm) Base (37 dBm) + 2 Relays (34 dBm each) Base (37 dBm) + 3 Relays (32 dBm each) Base (37 dBm) + 4 Relays (31 dBm each) 0 0 0.5 1 1.5 2 2.5 3 Mean aggregate load in cell (bits/symbol) 16 Lucent Technologies 7/4/2005
Summary � Analysis for large number of relays shows gains increasing linearly with number of hops – First ring of relays can be a bottleneck � Simulation results show – About 60% gains for 3 relays in uniform traffic distribution – Increasing gains with with increasing relays when total cell power is held constant – Diversity gains from fading � Potentially large gains for non-uniform traffic with optimal placement of relays or with multi-hop through WLANs � Coverage enhancement 17 Lucent Technologies 7/4/2005
Other Results � Cellular-WLAN Architecture – Proxy assignment and Routing protocols – ~50% throughput improvement � Connectivity – Significant coverage improvement – Energy savings on the reverse link � Multicasting – Significant gains from SINR improvement to cell edge users – Use of Network Coding 18 Lucent Technologies 7/4/2005
Recommend
More recommend
