Interference Coordination and Cancellation for 4G Networks

4,348 views

Published on

Published in: Technology
  • Be the first to comment

Interference Coordination and Cancellation for 4G Networks

  1. 1. LTE PART II: 3GPP RELEASE 8 Interference Coordination and Cancellation for 4G Networks Gary Boudreau, John Panicker, Ning Guo, Rui Chang, Neng Wang, and Sophie Vrzic, Nortel ABSTRACT with varying degrees of success in terrestrial mobile networks for more than 20 years. Tradi- This article provides an overview of contem- tional approaches to interference mitigation porary and forward looking inter-cell interfer- between transmitted signals have focused on ence coordination techniques for 4G OFDM either ensuring orthogonality between transmit- systems with a specific emphasis on implementa- ted signals in time, frequency as well as spatially, tions for LTE. Viable approaches include the or by actively removing and canceling interfering use of power control, opportunistic spectrum signals from the desired signal if orthogonality access, intra and inter-base station interference between the desired signal and potential inter- cancellation, adaptive fractional frequency reuse, ferers cannot be achieved. In early 2G cellular spatial antenna techniques such as MIMO and systems such orthogonality was achieved primari- SDMA, and adaptive beamforming, as well as ly through static pre-planned allocations of radio recent innovations in decoding algorithms. The resources. 3G systems introduced interference applicability, complexity, and performance gains cancellation techniques based mostly on a com- possible with each of these techniques based on bination of blind information gathering at a base simulations and empirical measurements will be station such as spectrum usage monitoring and highlighted for specific cellular topologies rele- coarse exchange of interference indicators such vant to LTE macro, pico, and femto deploy- as the Rise over Thermal (RoT) indicator ments for both standalone and overlay networks. employed in the 3GPP2 1xEV-DO standard. Typically interfering signals have been estimated INTRODUCTION using blind detection and their estimates sub- tracted from the desired signals. Good overviews The growing demands on mobile networks to of historical approaches can be found in [1] and support data applications at higher throughputs the references contained therein. and spectral efficiencies has driven the need to From a link perspective the downlink (DL) develop Orthogonal Frequency Division Multi- allows for a more tractable analysis since if the plexing (OFDM) based 4th generation (4G) net- desired mobile terminal location is known, the works, including WiMAX and 3GPP Long Term distances to all potential interfering base stations Evolution (LTE). A key objective with respect to can be easily determined based on the network deployment of OFDM 4G networks is to utilize geometry, and hence a probabilistic based esti- a frequency re-use of one (denoted by N = 1) or mate of the signal-to-interference-plus-noise as close to N = 1 re-use as is practical. A fre- ratio (SINR) can be calculated based on the quency re-use of N = 1 implies that the base sta- channel fading conditions for the desired signal tions in cells transmit on all available and the interfering signals. In addition to time-frequency resource blocks (RBs) simultane- AWGN, both the desired signal and interfering ously. Due to transmit power limitations in signals will experience shadowing, which typically mobile terminals, the constraint on the uplink is log-normally distributed. Analysis of the uplink link budget will necessitate the need for smaller (UL) interference requires knowledge of not cell sizes than is typically deployed for present only the location of the desired mobile terminal 2nd generation (2G) and 3rd generation (3G) under consideration, but also the relative loca- cellular systems. This requirement is driven by tions of all potential interfering mobile termi- the need to meet targeted higher data rate nals, for which both the locations of the throughputs for users not only near the base sta- interfering terminals, the number of potential tion, but also for cell edge users. The resulting terminals, and their spatial velocity will be ran- interference limited system for N = 1 deploy- dom variables. ment will not achieve the full potential capacity LTE offers the capability to provide a flexible that the LTE standard can support without the dynamic inter-base station approach to interfer- implementation at the base station and mobile ence coordination through the use of inter-base terminal of one or more viable interference miti- station signaling capabilities, including the use of gation and/or cancellation techniques. UL reactive overload indicators (OIs) and proac- Interference cancellation and mitigation tive high interference indicators (HIIs) that pro- techniques have been investigated and deployed vide bit maps of interference conditions on a per 74 0163-6804/09/$25.00 © 2009 IEEE IEEE Communications Magazine • April 2009
  2. 2. RB basis. DL inter-cell interference coordina- tion (ICIC) is supported through the use of DL Histogram and CDF of SNR over drive route (100% load on all cells) relative narrowband transmit power (RNTP) bit 450 120.00% Average SNR is 7 dB maps providing a coarse power indication on a 400 per RB basis. 100.00% 350 The main body of the article is comprised of the next section providing an overview and anal- 300 80.00% Frequency ysis of existing interference mitigation and can- 250 cellation techniques that have been studied in Frequency 60.00% the open literature and/or implemented in exist- 200 Cumulative % ing cellular systems. We then provide a compari- 150 40.00% son of the identified ICIC approaches for LTE 100 along with implementation recommendations, 20.00% followed by the conclusions in the final section. 50 0 .00% -15 -13 -11 -9 -7 -5 -3 -1 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 More OVERVIEW OF ICIC TECHNIQUES Bin The challenge with deploying a static N = 1 fre- quency re-use OFDM system in an interference Figure 1. Typical measured SINR distribution for a deployed LTE system in limited environment is that for a fully loaded an urban environment. deployment, significant regions of coverage will experience negative SINR levels, resulting in gaps in the deployed coverage, irrespective of to compensate for a fraction of the path loss, as the inter-cell distance. Figure 1 illustrates a typi- opposed to the full path loss for cell edge mobile cal SINR distribution for 100 percent loading, as terminals ([2] and the references therein). Use measured in an urban LTE trial deployment of fractional power control in interference limit- with nine cells. It can be seen that on the order ed environments has been shown to provide of 15 percent of users will experience negative improvements in aggregate sector throughput on SINR, with some users experiencing negative the order of 20 percent over traditional power SINR levels of –10 to –15 dB. It should be noted control for scenarios employing inter-cell dis- that in a fully loaded interference limited cellu- tances of 500 m to 1 km and a bandwidth of 10 lar deployment the severity of the SINR degra- MHz. For deployment scenarios with a small dation will be highly dependent on the average inter-cell distance (ISD) on the order of 500m, path loss exponent. For a cellular deployment fractional power control also shows an increase with a fixed inter-cell distance, high path loss in 5 percentile cell edge throughput on the order propagation environments with path loss expo- of 10–15 percent over traditional power control. nents up to a 5th or 6th order will experience However for larger macro cells with ISDs on the less overall interference than deployments with order of 2 km, the gain in aggregate throughput lower path loss exponents, since potential inter- is traded off against a small loss in cell edge user fering signals from neighboring cells will be throughput on the order of 20–30 percent [2]. It more greatly attenuated in the former case. should be noted that although theoretically opti- Even though there will be significant SINR vari- mal power control approaches such as water fill- ation depending on the propagation environ- ing can provide higher aggregate throughput ment, in order to robustly deploy an LTE OFDM gains, this is typically achieved at the expense of system one will have to mitigate the inevitable fair cell edge throughput. negative SINR coverage regions that will exist. Moving beyond the intra-eNodeB power con- The following sections provide an overview of trol algorithms described above, power transmit- a number of techniques to address the potential ted to or from a mobile can be optimized from a SINR coverage holes, including power control, network inter-base station or macro diversity both static fractional frequency reuse (FFR) and perspective. This approach can be applied in adaptive fractional frequency reuse (AFFR), combination with relay networks. It is possible to AFFR with tiering, intelligent SINR based define a number of distributed power control scheduling, opportunistic scheduling and access scheduling algorithms optimized for cell edge techniques, multiple-input multiple-output user throughputs, aggregate network through- (MIMO), and space division multiple access puts, and max-min fairness, as well as intermedi- (SDMA) antenna techniques, as well as beam- ate variants such as proportional fair share forming. Note that these techniques can be (PFS) or harmonic mean fairness [3]. The PFS based on both intra and inter-base station com- and min-max algorithms have shown potential munication approaches. gains of 30-80 percent. Note that unless other- wise specified, performance results quoted in POWER CONTROL this article are based on the 3GPP simulation Historically power control has been employed in methodology typically with ISDs of 0.5 to 2 km cellular systems to minimize near-far dynamic and log-normal shadowing with a standard devi- range effects by constraining the UL power to be ation of 7–8 dB. received with a constant power level at the base station. Such an approach, while not optimal STATIC FRACTIONAL FREQUENCY REUSE from an aggregate throughput or spectral effi- Fractional frequency reuse (FFR) involves parti- ciency perspective, assures fairness to cell edge tioning the usable spectrum into a number of users. Recent investigations for 4G propose a subbands and assigning a given subband to a cell compromise approach by controlling the power in a coordinated manner that minimizes inter- IEEE Communications Magazine • April 2009 75
  3. 3. Power Power Gains due to SDMA Power and MIMO Cell #1 Cell #1 implementations in Cell #1 F1 Frequency F1 Frequency OFDMA systems Power Power f1 Frequency Power have been Cell #2 Cell #2 extensively Cell #2 researched and F1 F1 Frequency Power Frequency Power f2 Frequency a full suite of MIMO Power and SDMA tech- Cell #3 Cell #3 niques are currently F1 Cell #3 F1 Frequency Frequency defined in the F3 Frequency LTE standard. a) N=1 b) N=1/3 c) PBP for N=2/3 Figure 2. Adaptive FFR modes: a) N = 1; b) N = 1/3 and; c) PBP for N = 2/3. cell interference. For example, for an N = 1/3 maintain a record of throughput levels, coverage FFR re-use pattern, each cell would be assigned and or interference problems through use of a a subband spanning 1/3 of the available spec- channel quality indicator (CQI) and handover trum. Basic N = 1/3 FFR results in an improve- feedback from the mobiles it is serving. If a base ment in SINR due to reduced interference levels; station in mode 1 detects a problem (for exam- however, due to a reduction in available BW, the ple, the number of users at the cell edge with aggregate throughput is typically about 75–80 data to transmit/receive exceeds a certain thresh- percent of the aggregate throughput of an equiv- old) then the scheduler for that base station can alent system employing N = 1 reuse. However, transition to employing mode 2 (i.e., N = 2/3 due to the improvement in SINR, any coverage reuse) and send a message to neighboring base gaps for cell edge users can typically be mitigat- stations to transition to use of mode 2. Note that ed, although total cell edge user spectral effi- for an LTE deployment this signaling can be ciency compared to N = 1 performance does not accomplished over the X2 interface between typically improve significantly. It should also be base stations (i.e., eNodeB’s). If a neighbouring noted that use of FFR in an LTE OFDM system base station is already using mode 2 or higher no will reduce the available number of RBs that can further action by that base station is necessary. be scheduled and hence the peak throughput, Each base station will keep track of which neigh- potentially impacting the possible quality of ser- boring base station made the request and the vice for high bandwidth applications. requested mode of operation. ADAPTIVE FRACTIONAL FREQUENCY REUSE SPATIAL TECHNIQUES Further improvements in SINR can be achieved Gains due to SDMA and MIMO implementa- by adapting the FFR assignments based on inter- tions in OFDMA systems have been extensively ference levels, as well as scheduling users with a researched, and a full suite of MIMO and given subband based on channel quality mea- SDMA techniques are currently defined in the surements (CQI) fed back from the mobiles [4]. LTE standard, including spatial multiplexing, An AFFR implementation can be based on the cyclic delay diversity (CCD), and space frequen- notion of Power Bandwidth Profiles (PBPs). Fig- cy block coding (SFBC) on the DL and multi- ure 2a–c illustrates allocations of bandwidths for user MIMO (MU-MIMO) on the UL. The gains N = 1, static N = 1/3, and PBPs for an N = 2/3 of these techniques have been well established FFR scheme, respectively. As illustrated in Fig. through simulation [5] as well as trial implemen- 2c, the difference in power spectral density with- tations. Throughput gains from MIMO in actual in a given PBP is defined as the power split implementations have been shown to approach (PS). For the example considered here the power the theoretical maximums for uncorrelated spa- split is 3 dB. tial paths (i.e., a 2 times throughput gain for 2 × Table 1 provides a summary comparing the 2 MIMO and a 4 times throughput gain for 4 × 4 throughput of 10 MHz LTE deployments MIMO). employing N = 1 re-use, N = 1/3 re-use, and AFFR with N = 2/3. It can be seen that the sec- NETWORK MIMO tor throughput of AFFR is close to N = 1 and The notion of network MIMO involves transmis- both outperform N = 1/3 by close to 30 percent. sion of multiple spatial paths to or from a mobile AFFR also improves cell-edge user throughput from multiple base stations. On the DL, multiple over N = 1. base stations can transmit one or more MIMO The process of adapting the FFR operation paths to a mobile, whereas on the UL, the trans- comprises a number of steps based on moving mission by a mobile can be received by one or between the N = 1, 1/3, and 2/3 reuse modes as more base stations. In its simplest form com- defined in Fig. 2. Each base station would begin prised of a single path from each respective base in, for example, mode 1 ( i.e., N = 1 re-use) and station to a mobile, network MIMO takes on the 76 IEEE Communications Magazine • April 2009
  4. 4. form of macro diversity. With respect to the UL, FFR FFR the notion of network MIMO can also be com- N=1 N = 1/3 AFFR PS = 3 dB PS = 6 dB bined with the notion of MU-MIMO and inter- ference cancellation to allow mobiles in adjacent Sector Tput (Mb/s) 8.01 7.92 7.79 6.11 7.89 cells to be assigned the same RBs, thus improv- ing overall system spectral efficiency. Receiver 5%-ile Cell-Edge algorithms such as dirty paper coding can also 286 313 352 292 312 User Tput (kb/s) be employed in combination with network MIMO to improve overall efficiency. One of the 10%-ile Cell-Edge challenges with network MIMO implementations 311 362 395 368 357 User Tput (kb/s) is the latency for exchange of information between base stations. In the current Release 8 Table 1. Comparison of N = 1, N = 1/3, and AFFR throughputs in a of the LTE standard, the minimum X2 latency 10 MHz LTE system. for exchange of information between base sta- tions is 20 msec, whereas RBs are typically assigned on a 1 msec subframe basis, making real time processing of interference cancellation and inter-cell interference. For effective cancel- data from adjacent base stations unfeasible. lation of inter-cell interference, inter-cell com- munication is essential. The current version of INTERFERENCE REGENERATION AND the LTE standard supports non-real time IC CANCELLATION TECHNIQUES through the X2 interface, which provides inter- base station communication. IC techniques can The notion of interference cancellation (IC) is also be used for canceling interference from well known and was seriously proposed more MU-MIMO (also referred to as Virtual-MIMO) than 20 years ago for commercial implementa- configurations in the LTE uplink, for which two tions. The basic concept is to regenerate the or more users are scheduled with the same set interfering signals and subsequently subtract or overlapping set of physical RBs for uplink them from the desired signal. Typically this transmissions. The simultaneous transmissions involves storing the received signal in a sample could interfere with each other, resulting in buffer. From an implementation standpoint, it is degraded SINR performance, and consequently, simplest to cancel signals whose modulation less than the expected spectral efficiency symbols are a priori known at the base station improvement for MU-MIMO. (e.g., reference symbols) since the IC process The high-level architecture of a typical inter- can avoid data demodulation and decoding. The ference canceling LTE UL receiver at a base sta- most advanced forms of IC with the largest tion with two receive antenna is shown in Fig. 3. capacity gains require cancellation of interfer- UE1 is the desired user and UE2 to UEn are the ence from data signals as well. interfering users. Note that user equipment It is feasible to implement IC in OFDMA (UE) is the LTE equivalent term for a mobile systems such as LTE. Although IC techniques and the subscript refers to the mobile number. can be applied to both downlink and uplink of Interfering users could arise from within the cell LTE, due to complexity considerations, IC is (intra-cell interference) or from outside the cell considered mainly as a technique for the UL and (inter-cell interference). Intra-cell interference implemented in the base station receiver. IC scenarios include UEs purposely scheduled on techniques can be used to cancel both intra-cell the same RBs within the cell (MU-MIMO), or eNodeB receiver Freq domain UE1 TX channel estimation Ant1 pre- and processing equalization Optimal antenna combining and o/p UE2 TX and Reference symbol/data Data Ant2 pre- symbols demodulation processing interference cancellation UEn TX UE2 estimates UEn estimates Figure 3. High level view of interference canceling receiver at eNodeB. IEEE Communications Magazine • April 2009 77
  5. 5. 2-branch MMSE-IC is about 9.5 dB at a BLER 100 level of 1 percent, whereas the gain from a 4- No IC 2-branch Rx IC branch MMSE-IC reaches 13.5 dB. 4-branch Rx IC INTELLIGENT SCHEDULING In [6, 7] the performance of OFDMA frequency re-use schemes is analyzed with a particular emphasis on LTE. The simulated results of [7] 10-1 claim that a frequency re-use scheme of N = 1 provides the best overall throughput for both the BLER (afec) UL and DL; however, cell edge users are inter- ference limited in the N = 1 reuse case resulting in approximately equal per user throughput at the cell edge. In [6] similar results are provided; 10-2 however, cell edge throughputs are claimed to be maximized through a partial frequency re-use scheme in which different subbands within a cell have different transmit power limits. Gains of up to 20 percent in aggregate throughput are shown for partial frequency reuse, but with a potential penalty of up to 20 percent for cell edge users. It 10-3 should be noted that the analysis does not take -14 -12 -10 -8 -6 -4 -2 0 2 4 6 8 CIR (dB) into account the effects of lognormal shadowing. In order to mitigate cell edge SINR degradation Figure 4. : Interference cancellation performance with one major co-channel it is recommended that a partial frequency re- interferer (CCI). use scheme be employed. OPPORTUNISTIC AND having UEs within a cell allocated with the same ORGANIZED INTER-BASE STATION ACCESS RB during an overload situation. Inter-cell inter- ference scenarios typically involve UEs in neigh- Opportunistic or organized access can be imple- boring cells being scheduled on the same RB. mented from a perspective of spectral, temporal The first step in the interference cancellation or spatial re-use of scheduled RB’s between receiver is the regeneration of the interference base stations. The notion of opportunistic spec- from UE2 to UEn, for which the signal and chan- tral access can be exploited both for hierarchal nel parameters of the interfering users need to overlay systems such as femto overlays on a be estimated. The base station receiver first macro system as well as for spectral re-use with- acquires the information on the reference sym- in a single macro cellular network. For LTE bol (RS) sequences employed by interfering macro networks, opportunistic access takes the users UE 2 to UE n and subsequently estimates form of dynamic channel allocation (DCA) at both the channel gain and phase as well as the the base station to enable the assignment of the time delay estimates of the interfering signals same RB’s to multiple mobiles, thus increasing from this RS information. With these estimates, overall spectral efficiency. If DCA decisions at the base station receiver can regenerate the the candidate base station are undertaken inde- interfering signals of UE 2 to UE n and subse- pendently of RB assignment decisions from quently, subtract them from the received signal. neighboring base stations, the candidate base If the channel estimates are accurate, the regen- station must have a means to measure the level erated signal will be an accurate estimate of the of interference in the candidate RBs. For time actual interfering signal and the subtraction pro- division duplex (TDD) systems, in which the cess can substantially cancel the interference channel is stationary across the frame of inter- from the signal of the desired user UE 1 . It is est, the received signal subframe can be therefore essential that accurate channel esti- employed as an accurate estimate of the chan- mates, frequency offset estimates, time delay nel conditions for the transmit portion of the estimates, and received power estimates are subframe. However, in frequency division duplex available under all fading scenarios. Minimum (FDD) implementations of LTE, a separate mean squared estimation (MMSE) and maxi- measurement of the transmit channel quality mum likelihood sequence estimation (MLSE) will need to be signaled through CQI feedback techniques are both viable channel estimation from the mobile. approaches. Note that the interference regenera- Opportunistic beamforming and scheduling tion and cancellation steps need to be performed are relatively recent areas of investigation to successively to realize the potential capacity and achieve throughput and spectral efficiency gains throughput advantages offered by this technique. in a cellular system [8]. For an OFDMA system Figure 4 shows the possible improvement in such as the DL of LTE, multiuser diversity in block error rate (BLER) from employing both the frequency domain can be exploited by intelli- 2-branch and 4-branch MMSE interference can- gently scheduling mobiles during the peaks of cellation in the presence of one major interferer. the channel fading. The potential large CQI The modulation and coding scheme employed is overhead to implement such a scheme can be QPSK with a coding rate of 1/2. The fading minimized by forming the beams randomly based channel model is Pedestrian B with a speed of 3 on clustered CQI feedback and scheduling a km/h. It can be seen that the improvement from user on the instantiation of a beam within a high 78 IEEE Communications Magazine • April 2009
  6. 6. SINR cluster. Simulation results from [8] show vector quantization. Recent research has shown gains in aggregate throughput of up to 50 per- that DPC combined with 2 × 2 MIMO can Although the poten- cent for opportunistic beamforming plus propor- potentially achieve spectral efficiency gains of tional fair share scheduling over a conventional 0.8 b/s/Hz over 2 × 2 MIMO alone and 1.35 tial performance equal gain beamformed system with round robin b/s/Hz for DPC over 4 × 4 MIMO [11]. DPC has improvements of scheduling. also been shown to outperform frequency re-use OBH are compelling, In [9] Hu et al. investigate the performance of schemes. One of the current challenges with organized beam hopping (OBH) between nodes implementing DPC is the complexity of the a number of out- in a multi-cellular environment. OBH is a tech- implementation. DPC requires an implementa- standing implemen- nique to pseudo-randomly hop beams in an tion of vector quantization at both the transmit- organized quasi-orthogonal pattern between ter and receiver as well as an iterative type tation issues remain, base stations to maximize throughput by RB re- decoder, such as a Viterbi, turbo, or a successive including contention use while minimizing inter-base station interfer- trellis decoder. It order to achieve the full theo- and collision resolu- ence. The performance of OBH is shown to be retical gain of DPC, the number of symbols per superior to opportunistic beamforming for all codeword has to be on the order of up to 10 4 , tion methods for levels of mobile populations in a cell. Gains in resulting in prohibitively complex vector quantiz- inter-base station spectral efficiency of 1 b/s/Hz by OBH over both ers and decoders due to the exponential increase opportunistic beamforming and coherent beam- in complexity with codeword size. beam selection as forming were demonstrated in [9]. Results were well as the ability of obtained through simulation of a cellular system comprised of cells with a 1 km cell radius, each ICIC PERFORMANCE RESULTS the OBH algorithm having six beams per cell (i.e., 60 degree sec- Based on the overview of ICIC techniques pro- to adapt to varying tors). Although the potential performance vided in the previous section, Table 2 provides mobile traffic improvements of OBH are compelling, a num- a summary comparison of the various ICIC ber of outstanding implementation issues remain, techniques from a perspective of performance, patterns. including contention and collision resolution complexity, as well technology maturity and methods for inter-base station beam selection as risk. The applicability of the technique to the well as the ability of the OBH algorithm to adapt LTE DL, LTE UL, or both is indicated in the to varying mobile traffic patterns. leftmost column. Current mature approaches that should be considered as baseline features RECENT RECEIVER for initial UL and DL LTE deployments DECODING ALGORITHM INNOVATIONS include power control, some form of FFR, preferably with adaptation, as well as MIMO. Sphere and dirty paper decoding are two recent SDMA is also relatively mature and can pro- innovations in decoding algorithms that are vide significant aggregate capacity improve- potentially applicable to 4G systems. Sphere ments for a small increase in complexity. decoding is a technique of considerable interest Interference cancellation promises significant based on its potential to provide close to a maxi- gains but is likely limited to the LTE UL due mum likelihood (ML) decoding result at signifi- to processing complexity, and will require real- cantly lower complexity than a ML decoder. The time exchange of information between base basic concept is to search an N-dimensional stations on the order of every msec to maxi- hypersphere of some predefined radius R within mize the gain for an LTE system. Beamform- the code space. The choice of the size of radius ing technologies such as opportunistic spectrum to search over is a trade-off between exponen- access and organized beam hopping show sig- tially increasing search complexity as R increas- nificant theoretical promise; however, as prac- es, and a higher probability of an error in tical technologies for a deployed LTE system decoding as R decreases (i.e., the correct code they are still unproven. Newer decoding point may not be inside the search radius). approaches such as sphere decoding and dirty Recent analysis [10] has shown that the complex- paper coding show considerable promise in ity of a sphere decoding algorithm at high SINR terms of the gain they can provide; however, for 16- and 64-QAM modulations can be reason- the processing complexity of these approaches ably implemented with current processors. Fur- will likely limit their use to the UL for the near thermore, the performance of such an future. Network power control and network implementation is within 1 dB of a ML decoder MIMO are inter-base station techniques that assuming uncorrelated MIMO paths. If the seem viable as features for LTE ICIC evolu- MIMO spatial paths are correlated, the perfor- tion in the near future due to the potential mance of the sphere decoder can degrade up to performance gains and the feasibility of their 5 dB depending on the level of correlation implementation at the base station. between the paths. Dirty paper coding (DPC) is another tech- nique that has received considerable attention CONCLUSIONS recently as an approach to eliminate interference This article has provided an overview of tech- in systems for which joint decoding is not possi- niques that can be employed to mitigate interfer- ble. The essence of the approach is to precode ence in 4G OFDM systems with specific each transmission such that the desired signal is relevance to LTE. Viable existing and possible mapped into a known code space. The receiver future approaches have been reviewed, including will have knowledge of the precoding code space the use of fractional power control, static and which can be employed to successfully decode adaptive fractional frequency reuse, opportunis- the desired signal in the presence of interference tic spectrum access, intra and inter-base station using a trellis type decoder in combination with interference cancellation, MIMO, SDMA, adap- IEEE Communications Magazine • April 2009 79
  7. 7. Performance & Spectral Efficiency ICIC Option Description Implementation Complexity Maturity and technical risk Improvement Improvements of up to 40% in Regular power control Power Control Scales linearly with increase in Mature technology with aggregate throughputs and 100% for and fractional power (LTE UL & DL) number of RBs available known performance cell edge users over scenario with no control power control Aggregate throughput loss of up to Static FFR (LTE Fixed assignment of Minimal complexity. Fixed Mature technology with 25%. Small improvement in cell edge UL & DL) subbands to each cell assignment is preconfigured known performance user throughput Scales linearly with the number of FFR assignment is adap- mobiles served at the base New technology. Dynamics Up to 10 dB improvement in SINR Adaptive FFR tively updated based on station. Requires exchange of and stability in a loaded net- levels. Can translate into spectral (LTE UL & DL) the SINR in each subband SINR based metric between base work need to be validated efficiency gains of up to 1 b/s/Hz stations Use of multiple transmit Theory is well understood. Complexity scales linearly with Potentially 2–4 times spectral efficiency and receive antenna. Empirical trial results are MIMO (LTE DL) number of users. Exponential improvement with 4 × 4 MIMO for SFBC, STTD, CCD, and available confirming complexity with order of MIMO uncorrelated spatial paths MU-MIMO potential performance gains UL technique to assign Scales exponentially with number New technology. Scheduling Theoretical improvement in spectral MU-MIMO (LTE the same RB to multiple of mobiles simultaneously algorithms to select MU efficiency scales with order of MU- UL) mobiles in the same cell processed mobiles are still evolving MIMO. Typically less than 2 in practice. Mature technology that has Up to N times capacity improvement SDMA (LTE Fixed beam partitioning of Minimal additional processing is been deployed in 3G for N beams. Typically provides 2–4 UL&DL) the cell required at the scheduler systems times spectral efficiency improvement Scales exponentially with number Gain of up to 2 b/s/Hz in the presence Algorithms are mature and Interference Regenerate and subtract of mobiles served at the base sta- of a single strong interferer. Can understood. High cost of Cancellation (LTE interfering signal from tion. Requires exchange of informa- potentially improve channel perfor- implementation will likely UL) desired signal tion in real time between base mance to that of AWGN if accurate limit application to the UL stations. channel estimation is available High complexity. Exploits channel Opportunistic BTS assigns resources only sounding. Additional HW is New technology that has Gain of up to 50% in aggregate Spectrum Access to subchannels with low required in FDD systems. Suited to not been empirically channel throughput or spectral (LTE UL & DL) power spectral density mitigate interference levels and evaluated efficiency strong interferers New technology. Contention Organized Beams are pseudo-ran- Scheduling complexity will resolution and traffic pat- Improvements in spectral efficiency of Beamforming domly hopped in a increase exponentially relative to tern adaptability needs to be up to 1 b/s/Hz (LTE UL&DL) quasi-orthogonal manner the number of mobiles serviced addressed Sphere Based on a decoding Complexity increases exponential- High cost of processing Improvements of up to 1 b/s/Hz in Decoding search in an N-dimensional ly with the size of the search likely to limit application to fading environments (LTE UL) hyper-sphere space the UL Dirty Paper Precoding of signal Complexity increases exponentially High cost of processing Potential spectral efficiency gains of decoding mapped into a known with the size of the search space likely to limit application to 0.8 to 1.35 b/s/Hz (LTE UL) code space and codeword size. the UL Complexity increases linearly with Network Power New technology that has Power control coordinated the number of base stations. Potential gains in throughput or spec- coordination not been empirically between base stations Requires exchange of power levels tral efficiency of up to 80% (DL) evaluated between base stations Complexity increases exponentially MIMO implemented with the number of base stations New technology that has Gains are expected to be consistent Network MIMO across multiple base and number of links. Will require not been empirically with higher order diversity (DL) stations exchange of information between evaluated implementations base stations Table 2. Summary of LTE ICIC options. tive beamforming, network MIMO, sphere tion such as fractional power control and adap- decoding, and dirty paper coding. The applica- tive fractional frequency reuse based on schedul- bility, maturity, complexity, and performance ing in high SINR regions could form the basis of gains possible from each of these techniques a robust LTE ICIC strategy. Longer term gains have been summarized in this article. No single in ICIC performance could potentially be approach will in itself provide complete interfer- achieved through the use of inter-base station ence mitigation for an LTE implementation. network based algorithms, including network However, it is expected that a combination of MIMO, opportunistic and/or organized beam- these approaches can be employed to provide a forming, and distributed power control, as well robust N = 1 reuse capability in a heavily loaded as coding strategies such as sphere decoding or LTE deployment. In the short term a combina- dirty paper coding. 80 IEEE Communications Magazine • April 2009
  8. 8. REFERENCES engineering from the University of Erlangen-Nuremberg, Erlangen, Germany, in 1992. In 1995, he joined the Bell- [1] J. Andrews, W. Choi, and R. Heath, Jr., “Overcoming Northern Research (BNR) and Nortel Networks, where he is Interference in Spatial Multiplexing MIMO Cellular Net- now a system architect for wireless systems. Since then, he works,” IEEE Wireless Commun., vol. 14, no. 6, Dec. has been intensively involved in the development of 3G/4G 2007, pp. 95–104. wireless standards and products [2] C. Castellanos et al., “Performance of Uplink Fractional Power Control in UTRAN LTE,” VTC Spring 2008, Singa- JOHN PANICKER received B.Sc. (Eng.) and M.E. degrees from pore, pp. 2517–21. India in 1979 and 1985 respectively. He received M.Sc. and [3] P. Pischella and J.C. Belfiore, “Power Control in Dis- Ph.D. degrees in 1992 and 1996 respectively, from the Uni- tributed Cooperative OFDMA Cellular Networks,” IEEE versity of Saskatchewan where he was a Canadian Com- Trans. Wireless Commun., vol. 7, no. 5, May 2008, pp. monwealth Scholar. He worked in BPL India for one year, 1900–05. was an Assistant Professor in India for 10 years. He worked [4] Nortel 3GPP RAN1 Contribution R1-072761 “Uplink for six months in SED systems, Saskatoon before joining his Power Control with Fractional Frequency Reuse for current job at Nortel Networks in 1996. In Nortel he is EUTRA,”, Orlando, FL, June 2007. involved in the product performance aspects of 2G, 3G, [5] NGMN TE WP1, “Radio Performance Evaluation Phase 2 and 4G wireless access technologies IS-95, 1xRTT, 1xEV-DO, Report,” Feb. 2008. and LTE. His interests are in the area of communication [6] S.-E. Elayoubi, O. B. Haddada, and B. Fourestie, “Perfor- theory, with particular application to OFDMA and LTE tech- mance Evaluation of Frequency Planning Schemes in nologies. He has authored several journal and conference OFDMA-based Networks,” IEEE Trans. Wireless Com- publications and served as a reviewer for journal and con- mun., vol. 7, no. 5, May 2008, pp. 1623–33. ference publications. [7] A. Simonsson, “Frequency Reuse and Intercell Interfer- ence Co-ordination in E-UTRA,” VTC Spring 2007, RUI CHANG jointed Nortel in 1994 and is currently a senior Dublin, Ireland, pp. 3091–95. advisor in Core RF Engineering Department, specialized [8] P. Svedman et al., “Opportunistic Beamforming and in LTE RF performance enhancement, SON algorithms Scheduling for OFDMA Systems,” IEEE Trans. Commun., and 4G network design and deployment. Previously, he vol. 55, no. 5, May 2007, pp. 941–52. was deeply involved in the RF deployment and perfor- [9] H. Hu et al., “Radio Resource Management for Cooper- mance of various 2G and 3G systems including CDMA, ative Wireless Communication Systems with Organized 1xEV-DO, and WiMax. He received M.S. and Ph.D. Beam-Hopping Techniques,” IEEE Wireless Commun., degrees in electrical engineering from the University of vol. 15, no. 2, Apr. 2008, pp .100–09. Maryland in 1986 and 1989. [10] L. G. Barbero and J. S. Thompson, “Performance of the Complex Sphere Decoder in Spatially Correlated N ENG W ANG received the Ph.D. degree in Electrical Engi- MIMO Channels,” IET Commun., 2007, pp 112–30. neering from Queen’s University, Kingston, Ontario, Cana- [11] W. Choi and J. Andrews, “The Capacity Gain from da in 2005. In 2006, he joined Communications Research Base Station Cooperative Scheduling in a MIMO DPC Centre, Ottawa, Canada, as a Research Scientist. Since Cellular System,” ISIT 2006, Seattle, WA, pp. 1224–28. 2007, he has been with Nortel Networks in Richardson, TX, as a Wireless Access Standards Engineer, working on 3GPP BIOGRAPHIES LTE and 3GPP2 standardization. His research interests lie in general areas of wireless communications and signal pro- GARY BOUDREAU [M‘84] (boudreau@nortel.com) received a cessing. B.A.Sc. in Electrical Engineering from the University of Ottawa in 1983, an M.A.Sc. in Electrical Engineering from SOPHIE VRZIC received her M.Eng. degree in electrical engi- Queens University in 1984 and a Ph.D. in electrical engi- neering from the Carleton University, Ottawa, ON, Canada. neering from Carleton University in 1989. He has been with She received her B.Sc. degree in mathematics and comput- Nortel for over 15 years in a variety of management and er science from McMaster University, Hamilton, ON, Cana- system engineering roles responsible for wireless product da. Sophie is currently a Senior Systems Engineer in the development. He is currently responsible for LTE modem Wireless Technology Laboratories, Nortel Networks, architecture at Nortel. His interests include wireless com- Ottawa, Canada. Her areas of research include advanced munications and digital signal processing. wireless access technologies in the physical, MAC and upper protocol layers, and next generation broadband N I N G G U O received the B.Eng. and M.Eng. degrees in wireless systems and networks. She is a major contributor telecommunications from Beijing University of Posts and to 3rd generation and 4th generation wireless standards Telecommunications (BUPT), Beijing, China, in 1982 and including IEEE 802.16, 3GPP2, and 3GPP. She has over 80 1984, respectively, and the Dr.Eng. degree in electrical patents granted or pending. IEEE Communications Magazine • April 2009 81

×