 Research
 Open Access
Analysis and modeling of spatial characteristics in urban microscenario of heterogeneous network
 Jianhua Zhang^{1}Email author,
 Nan Sheng^{1},
 Fenghua Zhang^{1},
 Lei Tian^{1},
 Guangyi Liu^{2},
 Weihui Dong^{2},
 Ping Zhang^{1} and
 ChiaChin Chong^{3}
https://doi.org/10.1186/168714992011187
© Zhang et al; licensee Springer. 2011
 Received: 2 March 2011
 Accepted: 28 November 2011
 Published: 28 November 2011
Abstract
Heterogeneous network (HetNet) is a typical deployment scenario for the IMTAdvanced system whereby the macro enhanced node B (eNB) provides the wide coverage while the lower power nodes such as micro, pico, femto, and relay nodes extend the coverage/capacity for coverage hole or hotspot. This literature addresses the spatial propagation modeling for urban micro (UMi) scenario of HetNet. Due to users distributed in canyon streets, the multipath with high power is not always coming from the lineofsight (LoS) direction in UMi scenario. Moreover, considering the impact of the directional antenna pattern, the current IMTAdvanced UMi channel model may lead to inaccurate interference modeling. To verify this, multipleinput multipleoutput (MIMO) field channel measurement is conducted in downtown Beijing for typical UMi. Based on the measurement data analysis, the multipath's angular offset from the LoS direction is clearly observed. In order to capture such spatial characteristic into the existing IMTAdvanced UMi channel model, the angular offset models are proposed for both LoS and nonLoS (NLoS) cases. Finally, the interference and capacity simulation prove that it is necessary to capture the angle offset model into the MIMO channel model in UMi scenario.
Keywords
 IMTAdvanced
 multipleinput multipleoutput (MIMO)
 channel model
 interference
 HetNet
1. Introduction
With the expansion of the mobile data market, the mobile operators have more and more pressure to expand the cellular capacity by cell splitting or carrier aggregation [1]. In order to make full use of the expensive spectrum, the cellular technology is required to improve the spectrum efficiency as much as possible. As reported in [2], the mobile data market will increase more than 50 times from 2010 to 2015. In order to meet the requirements of the future data market, 3rd Generation Partnership Project (3GPP) has started the research and standardization of the next generation cellular network technology, which is called as LTEAdvanced [3].
The cellular system is usually planned as hierarchical coverage. The macro enhanced node B (eNB) with high transmit power and high antenna height is deployed to provide wide coverage as the basic layer, whereas some low power nodes such as micro, pico, femto, and relay [4] nodes are deployed for the coverage/capacity expansion as the secondary layer. In order to alleviate the complexity of the network planning and optimization in hierarchical cellular deployment like Global System for Mobile (GSM) and Universal Mobile Telecommunication System, the macro eNB and micro/pico nodes are allocated with different carrier frequencies, and thus the interference between different coverage layers can be ignored.
According to the prediction from International Telecommunication UnionRadio communication sector (ITUR) [5], the required spectrum for IMTAdvanced is above 1 GHz, while the spectrum allocated for the IMTAdvanced by ITUR is less than 500 MHz now. In order to fill the spectrum gap between the required and the available, more aggressive spectrum usage strategies have to be considered for IMTAdvanced. In 3GPP, the hierarchical network with the same spectrum allocated for both basic and secondary layers is defined as heterogeneous network (HetNet) [6]. Compared to the homogeneous interference among the macro eNBs, the heterogeneous interference between macro eNB and lower power nodes becomes more serious. In order to deal with the serious intercell interference, the enhanced intercell interference coordination (eICIC) [6] and heterogeneous coordinated multiple point transmission and reception (CoMP) [7, 8] are proposed in 3GPP.
To facilitate the corresponding performance evaluation for HetNet, 3GPP has defined the evaluation methodology for HetNet and eICIC [6]. However, only the path loss and shadow fading are explicitly defined based on the existing models, such as the IMTAdvanced model [9] and ITUR M.1225 [10], whereas the fast fading is not defined explicitly. For the performance evaluation of eICIC in time domain, the path loss and shadowing may be sufficient; however, for the eICIC and CoMP in the spatial and frequency domains, the fast fading is necessary to show a reliable performance. Therefore, both the fast fading and slow fading of the MIMO channel should be captured in performance evaluation methodology of HetNet.
In this literature, the spatial models of the typical UMi scenario of HetNet are addressed. Regarding the existing IMTAdvanced UMi channel model, due to the impact of the directional antenna pattern of eNB transmitter and the phenomenon as described above, the intrasite interference from the neighboring sectors of the same micro eNB may be underestimated and thus, leads to overestimation of the single point MIMO system performance. Dedicated MIMO field channel measurements were conducted in downtown Beijing for typical UMi scenario. The angular offset of the multipath is observed from the measurement results, and a modified MIMO channel model is proposed to capture such spatial characteristics into the UMi channel model, where a random angular offset is captured into the fast fading.
To verify the influence of the proposed model, the theoretical channel capacity based on the measured data is analyzed, and system level simulations of Time Division LTEAdvanced (TDLTEAdvanced) system [11] are performed. The numerical results show that the intrasite interference has been underestimated by the original IMTAdvanced UMi model, while the proposed model provide better CoMP gain due to taking into account the impact of the directional antenna pattern and the angular offset on the multipath.
The rest of the article is organized as follows. The limitations of the existing channel models are discussed in Section 2. The field channel measurement is described in Section 3. The proposed channel model for UMi is presented in Section 4. The theoretical analysis and system level simulation results are given in Sections 5 and 6, respectively. Finally, the conclusions are drawn in Section 7.
2. Limitations of existing channel model
Wireless channel consists of many propagation paths, which diffuse in the spatial domain at both the transmitter and the receiver. The performance of MIMO system is greatly affected by the extent of the angular dispersion of angle of departure (AoD) and angle of arrival (AoA), which is described by the angular spread (AS) in the existing channel models. Moreover, due to the application of sectorized antennas, the spatial characteristics of the channel between the eNB and a certain UE may be influenced by the antenna pattern of the eNB.
However, it might not be the case in UMi scenario, especially in the downtown of dense urban like Beijing. The canyonlike streets environment in such scenario may lead to peculiar spatial characteristics of the propagation channel, e.g., the center of the PAS has some offset from the LoS direction. Considering the impact of the sectored antenna pattern, it may influence the interference modeling much and thus influence the network capacity.
where θ_{3 dB} is the mainlobe's 3 dB beam width, and A_{ m } is the maximum attenuation. Typically, θ_{3 dB} = 70° and A_{ m } = 25 dB.
As illustrated in Figure 2, the green line denotes the interference departing at θ_{1} from the LoS direction, whereas the red line represents the interference departing at θ_{2}. And the channel gain affected by the eNB transmission antenna at θ_{1} and θ_{2} can be calculated as G(θ_{1}) = A_{eNB}(θ_{1})·P(θ_{1}) and G(θ_{2}) = A_{eNB}(θ_{2})·P(θ_{2}), respectively. In conventional channel model, P(θ_{1}) ≤ P(θ_{2}), thus the values of P(θ_{1}) and P(θ_{2}) determine the relative magnitude between G(θ_{1}) and G(θ_{2}) if A_{eNB}(θ_{1}) and A_{eNB}(θ_{2}) are not taken into account. However, according to the antenna pattern gain (the blue curve in Figure 2) depending on the departure angle of each path, it is found that A_{eNB}(θ_{2}) is much larger than A_{eNB}(θ_{1}). So it is very possible that G(θ_{2}) could be in the same order of magnitude with G(θ_{1}). If G(θ_{2}) is ignored in the interference calculation from the channel modeling, the interference experienced by the UE will be inaccurate.
3. Field MIMO channel measurement
In order to verify our hypothesis, the dedicated field MIMO channel measurements are carried out in downtown Beijing for typical UMi scenario. This section introduces the details of the environment, equipment, and procedures of the field channel measurement.
3.1. Measurement equipment
Measurement parameters
Items  Settings 

Carrier frequency (GHz)  2.35 
Bandwidth (MHz)  50 
Code length (chips)  255 
Transmitting power (dBm)  26 
Types of antennas  ODA 
Number of eNB antenna (N_{BS})  16 
Number of MS antenna (N_{MS})  32 
Height of eNB antenna (m)  7 
Height of MS antenna (m)  1.8 
3.1.1 Environment and measurement procedures
The measured data are stored in the memory of the Rx and the channel characteristics are extracted by accurate data post processing. LoS and NLoS cases are processed separately to analyze the possible differences.
4. Data processing and proposed angle offset model
In this section, the postdata processing for the field channel measurement results are introduced. The random angle offset from the LoS direction is observed from the extracted spatial parameters cyclebycycle. Based on the statistics of the angle offset values from an amount of measurement cycles, the empirical model is regressed and proposed to capture such spatial characteristic into the GBSM channel model.
4.1. Data processing
In data post processing, firstly, the CIRs are converted from the raw data by sliding correlating the received signals with a synchronized copy of the sequence. Then SpaceAlternating Generalized Expectation maximization (SAGE) algorithm [14–17], which has widely been used for channel parameter estimation, is applied in order to extract the channel parameters from the CIRs. As an extension of MaximumLikelihood (ML) method, the SAGE algorithm provides a joint estimation of the parameter set with no constrains on the response of antenna array. τ_{ n } , υ_{ n } , ϕ_{ n } , φ_{ n } , A _{ n }, P_{ n } denote the propagation delay, Doppler shift, AoD, AoA, polarization matrix, and the power of the n th propagation path, respectively.
where N(θ)is the number of paths of which the ϕ_{ n } = θ, and the P_{n, θ}is the power of the corresponding path.
4.2. Data analysis and angular offset modeling
It is reasonable to assume that the mean angle of all existent paths locates approximately in the middle of the two groups, which means that the center of the PAS distribution in this environment is not in accordance with the LoS direction. It can be explained by the fact that in the crossroads environment, most of the propagation paths come along the street and are at the same side of the LoS line. Thus, the offset between the mean angle of PAS distribution and LoS direction appears and it determines the amount of signals deflected to nearby sectors. In typical UMi environment, since the eNB is deployed at the roof top, where lots of buildings are higher than the eNB, offset values can easily be observed for the canyon propagation in the street.
However, in the existing channel models such as IMTAdvanced channel model, it assumes that the PAS follows wrapped Gaussian or Laplacion distribution with its center along the LoS direction. The offset between the mean angle of PAS distribution and LoS direction is ignored.
where ϕ_{ n } is the AoD, the minimization over Δϕ is to eliminate the additional angle spread introduced by different selection of reference zero angle. r(·) convert angles to (180, 180) degree. ϕ_{mean} is the power weighted mean angle which is derived from ${\varphi}_{\mathsf{\text{mean}}}={\sum}_{n}^{N}r\left({\varphi}_{n}+\text{\Delta}\varphi \right){P}_{n}/{\sum}_{n}^{N}{P}_{n}$.
For the PAS that follows wrapped Gaussian or Laplacian distribution, ϕ_{rms} will reach its minimum value when Δϕ is at the center of the distribution. Thus, during the data processing of the measurement data, we can obtain the Δϕ that minimizes ϕ_{rms}, which is also the center angle of the PAS distribution.
To find the mean angle denoted in Figure 6, i.e., the center of all paths in one cycle, the coordinate system in which the angle spread is minimized need to be determined. Offset values are obtained by subtracting the mean angles of all paths from LoS direction in the coordinate system.
According to the measurement result, it is found that the distribution center is not in accordance with the LoS direction (referred as 0 degree) and an angular offset is observed. In current channel models, the ASA and ASD are calculated with Equation 6 from measured data, and when we use the channel model to generate ϕ_{ n } , the center of the Gaussian or Laplacian distribution is assumed to be the LoS direction, which is in contradiction with our observation.
where μ = 1.524 and σ = 1.048 are extracted from the measured results. The α can be set as positive and negative values in equal probability.
where μ = 66.461 and σ = 18.325. The α can be set as positive and negative values in equal probability.
It is shown that under LoS environment, the absolute values of the offset angles are best fitted with loglogistic distribution, while there are best fitted with logistic distribution under NLoS environment. So, such spatial characteristic of typical UMi scenario can be captured into the IMTAdvanced channel model with the proposed distribution.
where α is randomly generated according to the distributions described by Equation 9 for LoS or Equation 10 for NLoS. A random variable Y with uniform distribution in the discrete set of {1, 1} is multiplied to assign positive or negative sign to the offset angles. Since the offset is an overall shift of the PAS in one drop, it should be noted that the offset angle α is generated for each link and all the ϕ_{ n, m } in the same link should be shifted with the same offset angle (all the AoD of the paths have the same offset). The generation of all the other parameters can be the same as the original IMTAdvanced channel model.
Proposed parameters
Items  LoS  NLoS  

Angular offset distribution  Loglogistic  Logistic  
Parameters (°)  μ  1.524  66.461 
σ  1.048  18.325 
To verify the necessity of this offset modeling to the channel modeling and the corresponding network performance modeling, theoretical capacity and network capacity analysis are conducted; the results are presented in the following section.
5. Impact of angular offset on channel capacity
To evaluate the impact of angular offset under different kinds of antenna configurations and different antenna patterns, the 50 paths extracted from SAGE algorithm, where the impact of the measurement antenna is excluded, are used to reconstruct the channel of desired antenna setup.
where β indicates the slant angle between the antenna element and the vertical direction. Finally, H(τ) is converted into frequency domain by applying Discrete Fourier Transform, i.e., H_{recon}(f).
where α_{ i } means the offset value of the cycle i. Thus, the comparison of the channel capacity with and without the offset angle can be considered as the comparison of the proposed model and the traditional model.
where H_{recon}(f) is acquired by transforming the reconstructed channel impulse response into frequency domain.
Besides, from the eigenvalues of the channel matrix, we can get a better understanding of the eigen multipath channels which determines the performance of beamforming. Therefore, the cumulative distributive function (CDF) of the ratio of the biggest eigenvalue and the smallest eigenvalues is also presented.
To verify the angular offset on the theoretical channel capacity, the MIMO channel statistics for two cases are compared. Case 1 is named as "measured data excluding offset", which is equivalent to the original IMTAdvanced UMi channel model, where the angle offset from the LoS direction is ignored. The case 2 is named as "measured data", which is equivalent to the modified IMTAdvanced UMI channel model, where the offset value from the LoS direction is taken into account for the fast fading modeling.
6. Impact of angular offset on network capacity
To verify the impact of angular offset on the network interference and capacity, the dedicated network simulation is conducted. The details of the simulation can be found in [19].
Simulation parameters
Parameter  ITUR UMi 

Site layout  3sectorized hexagonal grid with 7 cells and wraparound 
Operating bandwidth  10 MHz 
DL/UL ratio  2 DL/2 UL 
Special subframe  [10:2:2] for DwPTS, GP and UpPTS 
UpPTS Antenna boresight points toward flat side of cell 

Antenna pattern  Polarized antenna/horizontal antenna 
eNB transmission power  46 dBm 
UE receiver structure  Minimum mean square error 
UE number  10/sector in full queue 
Linktosystem interface for simulations  MIESM 
HARQ combining  Chase combining 
Penetration loss  20 dB 
To show the impact of the proposed angular offset model, the simulations based on the existing IMTAdvanced UMi channel models and the modified IMTAdvanced UMi channel models are performed independently.
Impact of the proposed angular offset model
Simulation case  IMTAdvanced UMi  Proposed UMi  Performance Loss  

SE  CSE  SE  CSE  SE  CSE  
UMi  EBB(bps/Hz)  1.81  0.054  1.45  0.016  20%  71% 
MUBF(bps/Hz)  2.91  0.088  2.21  0.022  24%  75%  
MUCoMP(bps/Hz)  4.38  0.135  4.31  0.053  2%  61% 
From Table 3, it is also found that the performance gain of MUCoMP over EBB is underestimated by 55 and 80%, respectively, in spectrum efficiency and cell edge spectrum efficiency by conventional IMTAdvanced UMi channel model. Whereas 44 and 87% performance gain of CoMP over MUBF has been underestimated by IMTAdvanced UMi channel model.
The conclusion based on the analysis and simulation results above quite align with our original hypothesis. The current GBSM without considering the angular offset underestimates the intrasite interference of the UMi scenario due to the special characteristics of the environment and the impact of the directional antenna pattern. The proposed angular offset is necessary for the IMTAdvanced UMi model to capture the spatial characteristic of the scenario better and improve the accuracy of the channel modeling for UMi of Hetnet.
7. Conclusions
For HetNet research and performance evaluation, accurate channel model is very important. However, the current standardization work only explicitly specifies the path loss and shadowing model. To evaluate the solutions in spatial and frequency domain and to investigate the interference effects in HetNet scenarios, the accurate fast fading model is vital. This literature addresses the spatial propagation modeling for UMi of HetNet. Based on the analysis on angular offset characteristics of the UMi scenario and the impact of eNB transmitter antenna pattern on the interference modeling, it is concluded that the existing GBSM model, e.g., IMTAdvanced UMi channel model, leads to underestimated intrasite interference. To verify this, field MIMO channel measurements are carried out in downtown Beijing for typical UMi scenario. From the measurements, the angular offset of the multipath is clearly observed. In the proposed model, the angular offset in LoS and NLoS cases are modeled as loglogistic and logistic distribution, respectively. The theoretical channel and the network capacities' analysis proves that it is necessary to capture the random angle offset modeling into the IMTAdvanced UMi channel model.
Declarations
Acknowledgements
This study was supported in part by the China Important National Science and Technology Specific Projects under Grant No. 2009ZX0300700301 and by China 863 Program and Major Project under Grant No. 2009AA011502.
Authors’ Affiliations
References
 Iwamura M, Etemad K, MoHan F, Nory R, Love R: Carrier aggregation framework in 3GPP LTEadvanced. IEEE Commun Mag 2010, 48(8):6067.View ArticleGoogle Scholar
 Zhao Z, Wang J, Guan H, Mogensen P E, Liu G, Shen X: TDLTE Network Deployment Evolution in a Metropolitan Scenario. Proceedings of the IEEE Vehicular Technology Conference (VTCFall) USA 2011, 15.Google Scholar
 3GPP TR 36.806, Evolved universal terrestrial radio access (EUTRA); relay architectures for EUTRA (LTEAdvanced) 2010.Google Scholar
 Pabst R, Walke BH, Schultz DC, Herhold P, Yanikomeroglu H, Mukherjee S, Viswanathan H, Lott M, Zirwas W, Dohler M, Aghvami H, Falconer DD, Fettweis GP: Relaybased deployment concepts for wireless and mobile broadband radio. IEEE Commun Mag 2004, 42(9):8089. 10.1109/MCOM.2004.1336724View ArticleGoogle Scholar
 ITUR Report M.2133, Requirements, evaluation criteria and submission templates for the development of IMTAdvanced 2008.Google Scholar
 3GPP TR 36.814, Evolved universal terrestrial radio access (EUTRA); further advancements for EUTRA physical layer aspects 2010.Google Scholar
 Venkatesan S: Coordinating base stations for greater uplink spectral efficiency in a cellular network. Proc IEEE 18th International Symposium Personal, Indoor and Mobile Radio Communications PIMRC 2007, 2007: 15.Google Scholar
 3GPP RAN1 R1111086, CoMP Simulation assumptions, NTT DoCoMo 2011.Google Scholar
 ITUR report M.2135, Guidelines for evaluation of radio interface technologies for IMTAdvanced 2009.Google Scholar
 ITUR report M.1225, Guidelines for evaluation of radio transmission technologies for IMT2000 1997.Google Scholar
 ITUR WP5D Document 5D/580, Submission of TDLTEAdvanced for IMTAdvanced candidate technology, China 2010.Google Scholar
 Stucki A: PropSound system specifications document: concept and specifications. In Switzerland, Technical Report Edited by: Elektrobit AG. 2001.Google Scholar
 Elektrobit : Propsound multidimensional channel sounder.[http://www.propsim.com]
 Fleury BH, Tschudin M, Heddergott R, Dahlhaus D, Ingeman Pedersen K: Channel parameter estimation in mobile radio environments using the sage algorithm. IEEE J Sel Areas Commun 1999, 17(3):434450. 10.1109/49.753729View ArticleGoogle Scholar
 Fleury BH, Yin X, Rohbrandt KG, Jourdan P, Stucki A: Performance of a highresolution scheme for joint estimation of delay and bidirection dispersion in the radio channel. Proceedings of the IEEE Vehicular Technology Conference (VTCSpring) Korea 2002, 1: 522526.Google Scholar
 Fleury BH, Jourdan P, Stucki A: Highresolution channel parameter estimation for MIMO applications using the SAGE algorithm. Proceedings of International Zurich Seminar on Broadband Communications Switzerland 2002, 30: 19.Google Scholar
 Fleury BH, Yin X, Jourdan P, Stucki A: Highresolution channel parameter estimation for communication systems equipped with antenna arrays. Proceedings of the 13th IFAC Symposium on System Identification (SYSID), no. ISC379, Netherlands 2003.Google Scholar
 Roh W, Paulraj A: MIMO channel capacity for the distributed antenna. Proceedings of the IEEE Vehicular Technology Conference (VTCFall) Canada 2002, 2: 706709.View ArticleGoogle Scholar
 Wang Q, Jiang D, Liu G, Yan Z: Coordinated multiple points transmission for LTEAdvanced systems. Proceedings of the IEEE Int Conf Communications workshop on LTEAdvanced, Germany 2009, 14.Google Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.