 Research
 Open Access
Experimental evaluation of the WiMAX downlink physical layer in highmobility scenarios
 Pedro SuárezCasal^{1}Email author,
 José RodríguezPiñeiro^{1},
 José A GarcíaNaya^{1} and
 Luis Castedo^{1}
https://doi.org/10.1186/s1363801503399
© SuárezCasal et al.; licensee Springer. 2015
Received: 14 December 2014
Accepted: 24 March 2015
Published: 18 April 2015
Abstract
The experimental evaluation of the WiMAX downlink physical layer in highmobility scenarios is extremely difficult to carry out because it requires the realization of measurement campaigns with expensive fastmoving vehicles. In this work, however, we succeeded in doing such an experimental evaluation with a lowmobility vehicle. The key idea is the enlargement of the symbol period prior to its transmission over the air. Such enlargement reduces the frequency spacing between the orthogonal frequency division multiplexing (OFDM) subcarriers in WiMAX transmissions and hence induces significant intercarrier interference (ICI) on the received signals. The performance impact of such ICI in terms of error vector magnitude (EVM) and throughput is analyzed under different conditions like the use of multiple antennas, the placement of the receive antennas, or the accuracy of the channel information to adapt the transmission rate.
Keywords
 ICI
 Highspeed
 WiMAX
 MIMO
1 Introduction
Existing mobile communication networks are primarily designed for low user speeds below 15 km/h. Nowadays, however, there is an increased number of wireless terminals mounted on highspeed vehicles such as cars, trains, buses, subways, or airplanes. Highmobility communication networks (HMCN) aim at interconnecting such terminals to convey information not only for human users but also for machines to send all sorts of command, control, and safety information.
WiMAX is a communication standard suitable for the provision of wireless broadband connectivity. WiMAX is a term coined by the WiMAX Forum to promote the interoperability between the IEEE 802.16 family of wireless communication standards. WiMAX is the first commercially available and deployed technology for delivering mobile fourth generation (4G) services. IEEE 802.16 standardization activities started in 1999 and were the first ones to address broadband for wireless metropolitan area networks. Although longterm evolution (LTE) is currently being more widely used by 4G mobile network operators, specially in Europe and the United States, there is still a significant amount of network operators based on WiMAX, specially in Asia [1]. More specifically, WiMAX is being used in highmobility scenarios like the metropolitan transportation system of Japan [1]. In addition, the WiMAX Forum has recently instituted an Aviation Working Group within its organizational structure to collaborate on the adaptation of WiMAX to the specific needs of the aviation community [2].
The physical layer (PHY) of the WiMAX radio interface uses orthogonal frequency division multiplexing (OFDM) as the modulation scheme. OFDM is particularly suitable to carry data over broadband frequencyselective channels because it allows for a lowcost channel equalization. OFDM waveforms, on the contrary, are rather sensitive when transmitting over timeselective channels such as those encountered in highmobility scenarios. Timeselectivity causes the OFDM subcarriers to loose their orthogonality property and produces intercarrier interference (ICI). Timeselectivity is particularly harmful when considering complex scenarios with multiple antennas and/or multiple users.
Different techniques have been proposed in the literature to mitigate the effects of ICI. More specifically, issues like pilot pattern design [3], window design [4], channel estimation [5], and channel equalization have been thoroughly studied.
Basis expansion models (BEMs) have been extensively used to capture the double selectivity of the wireless channel and to implement costeffective channel estimation and equalization schemes [6]. The rationale under BEM is the efficient representation of either the impulse or the frequency response of a channel by means of a linear combination of some basis vectors. Different BEMs have been proposed to be used under doublyselective channels, such as the complex exponential BEM (CEBEM) [7], polynomial BEM (PBEM) [8,9], discrete prolate spheroidal BEM (DPSBEM) [10], and KarhunenLoève BEM (KLBEM) [11,12]. Regarding pilot patterns for OFDM under ICI conditions, the Kronecker delta model, where guard subcarriers are allocated around pilot subcarriers, has been proposed as an optimal design to minimize channel estimation errors [13,14]. A detailed review of the algorithms and techniques to combat doubleselectivity of the wireless channel can be found in [15].
Communication standards such as WiMAX or LTE define the pilot structure to be used by the elements in the network. Such pilot structures typically do not correspond to those assumed in many ICI estimation and equalization techniques proposed in the literature. In this work, we focus on specific algorithms which do not depend on such assumptions [12,16,17]. These methods typically estimate the ICI from a set of consecutive frequency response estimations and exploit time variations per subcarrier to estimate its spreading.
Experimental performance evaluations of 4G technologies in highmobility situations are scarce in the literature due to the huge difficulties of carrying out experiments in such scenarios. In this work, we follow an approach where the timeselective wireless channels of highmobility situations are recreated from experiments carried out at low speeds and hence are more costefficient to implement [18]. The approach consists of timeinterpolating OFDM symbols prior to its transmission overtheair (OTA), which leads to a reduction of the bandwidth of the whole OFDM signal. This produces OFDM waveforms which convey exactly the same information as the original one but with a reduced subcarrier spacing, hence artificially increasing the waveform sensitivity to ICI. At reception, and prior to its demodulation, the interpolation operation is inverted via a simple decimation operation. The resulting OFDM symbols are affected by ICI similarly as if they were transmitted over a highmobility wireless channel. In fact, interpolating the original signal by a factor I will affect the transmitted signal similar to what would happen if it were transmitted at I times the original speed.

Evaluation of ICI estimation and cancellation algorithms with experimental measurements of multipleinput multipleoutput (MIMO) transmissions under timeselective conditions. These results are compared to simulations with comparable model parameters.

Experimental study of the impact of ICI and feedback delay on performance metrics such as throughput.

Improved measurement methodology with respect to that used in previous works ([18,19]). Instead of carrying out a dedicated measurement for each emulated speed value, the signals corresponding to all emulated speeds (time interpolation factors) are transmitted sequentially right after each other, thus making possible their latter comparison under similar conditions. Additionally, two receivers are used simultaneously, and therefore outdoortooutdoor as well as outdoortoindoor measurements are recorded under similar conditions.
This work focuses on OFDM waveforms based on the WiMAX physical layer. Howerver, note that the experimental methodology also applies to other OFDMbased transmissions, such as those used in LTE. Although the results are obtained for a particular standard, this work provide hints on the behavior of other OFDMbased transmissions on highmobility conditions and on the performance of the ICI cancellation techniques in different scenarios.
2 Signal model
We consider the transmission of MIMOOFDM symbols synthesized according to the PHY layer specifications of the IEEE 802.16e (Mobile WiMAX) standard. We assume OFDM symbols with N subcarriers and a cyclic prefix of N _{ g } samples are transmitted. In total, each OFDM symbol occupies N _{ t }=N+N _{ g } samples. OFDM symbols are grouped in frames of K consecutive symbols, including a symbol reserved as a preamble. We also assume multiple antennas at transmission and reception, i.e., MIMOOFDM transmissions.The number of transmit and receive antennas is M _{ T } and M _{ R }, respectively. OFDM symbols are spatially multiplexed over the M _{ T } transmit antennas except the preamble which is transmitted only through the first antenna.
Let \({\mathbf {s}}_{k}^{(m)} \in \mathbb {C}^{N \times 1}\), k=1,…,K, m=1,…,M _{ T } be the column vector that represents the N complexvalued information symbols transmitted in the kth OFDM symbol over the mth antenna. Such vectors contain data, pilot, and guard symbols. Similarly, \({\mathbf {x}}_{k}^{(m)} \in \mathbb {C}^{N_{t} \times 1}\) contains the N _{ t } samples corresponding to the kth OFDM symbol transmitted over the mth antenna.
where \({\mathbf {s}}_{k} = \left [ {{\mathbf {s}}_{k}^{(1)}}^{T},{{\mathbf {s}}_{k}^{(2)}}^{T},\cdots,{{\mathbf {s}}_{k}^{(M_{T})}}^{T} \right ]^{T}\) is the N M _{ T }×1 column vector containing the information, pilot, and guard symbols transmitted in the kth MIMOOFDM symbol; F is the standard N×N discrete Fourier transform (DFT) matrix; G _{1} is a N _{ t }×N matrix which appends the N _{ g } samples of the cyclic prefix; ⊗ denotes the Kronecker product; \({\mathbf {I}}_{M_{T}}\) is the M _{ T }×M _{ T } identity matrix; and \({\mathbf {x}}_{k} = \left [ {{\mathbf {x}}_{k}^{(1)}}^{T},{{\mathbf {x}}_{k}^{(2)}}^{T},\cdots,{{\mathbf {x}}_{k}^{(M_{T})}}^{T} \right ]^{T}\) is the N _{ t } M _{ T }×1 column vector with the samples of the kth MIMOOFDM symbol. We assume the samples in \({\mathbf {x}}_{k}^{(m)}\) are serially transmitted over the mth antenna at a sampling rate F _{ s }=1/T _{ s } where T _{ s } is the sampling period.
The information symbol vectors are constructed as \({{\mathbf {s}}_{k}^{(m)}={\mathbf {P}}_{k}^{(m)}{\mathbf {p}}_{k}^{(m)}+{\mathbf {D}}_{k}^{(m)}{\mathbf {d}}_{k}^{(m)}}\) where \({\mathbf {p}}_{k}^{(m)}\) is a P×1 vector containing the pilot symbols in the kth OFDM symbol transmitted over the mth transmit antenna, whereas \({\mathbf {p}}_{k}^{(m)}\) is the N×P matrix that defines the positions of the pilots in such a symbol. Similarly, \({\mathbf {d}}_{k}^{(m)}\) is the D×1 vector containing the data symbols, and \({\mathbf {d}}_{k}^{(m)}\) is the N×D matrix that defines their positions. Data symbols are the output of a quadrature amplitude modulation (QAM) constellation mapper whose inputs are the channel encoded source bits. Note that P+D<N, with N−(P+D) being the number of guard subcarriers. Matrices \({\mathbf {p}}_{k}^{(m)}\) and \({\mathbf {d}}_{k}^{(m)}\) consist of ones and zeros only and are designed so that data and pilots are assigned to different subcarriers. According to the Mobile WiMAX standard, pilot subcarriers allocated in the mth antenna are set to 0 in all other antennas.
Note that \({\mathbf {H}}_{k}^{i,m} {\mathbf {x}}_{k}^{(m)}\) represents the timeconvolution between the signal transmitted over the mth antenna and h ^{ i,m }(t,τ).
where \(\bar {{\mathbf {G}}}_{k}\) is a block matrix with the main diagonals of the submatrices of G _{ k } and \({{\mathbf {z}}_{k} = ({\mathbf {G}}_{k}\bar {{\mathbf {G}}}_{k}){\mathbf {s}}_{k}}\) represents the ICI in the received signal. Note that in multiantenna systems, ICI occurs not only among subcarriers but also among different transmit antennas.
with f _{ c } the carrier frequency, c the speed of light, and v ^{ I }=I v the emulated speed as a result of an actual measurement speed v and a timeinterpolation factor I. Consequently, enlarging the symbol length T ^{ I } allows for the emulation of a velocity v ^{ I }, which is I times higher than the actual speed of the receiver, namely v.
It should be noticed that interpolation does not allow for a perfect recreation of highmobility channels because the signals overtheair have a reduced bandwidth and are less sensitive to the channel frequency selectivity.
Nevertheless, note that in this work, we are mostly interested in conducting experiments to test the performance of ICI cancellation methods in WiMAX receivers over realworld channels rather than channel equalization methods which can be tested in static experiments. Time interpolation does not reproduce the exact conditions of high mobility scenarios but provide a costefficient approximation to them.
3 Receiver structure
The ensuing subsections present a more detailed description of each processing block represented in Figure 2.
3.1 Frame detection and synchronization
Frame detection is carried out using the correlation properties of the preamble symbol. WiMAX defines a preamble symbol with pilot subcarriers generated from a pseudonoise sequence modulated as binary phaseshift keying (BPSK), with a spacing of two guard subcarriers between them. This structure leads to a threefold repetition in the time domain, which can be exploited for frame detection and time and fractional carrier frequency offset (CFO) estimation. Integer frequency shifts are corrected after the DFT by performing a crosscorrelation between the differential sequences of the transmitted and received preamble pilot sequences [20,21].
3.2 ICI estimation
The frequency response matrix G _{ k } in the received MIMOOFDM signal model given by Equation 5 needs to be estimated. We start determining a noisy frequency response estimation as a least squares (LS) estimation on the pilot subcarriers, affected by the interferences z _{ k } and w _{ k }, followed by a linear minimum mean squared error (LMMSE) interpolation of the channel coefficients on the data subcarriers. Recall that pilot subcarriers are transmitted over disjoint sets of subcarriers at different antennas. Assuming spatially uncorrelated channels, this estimation can be done independently for each transmitreceive antenna pair. For simplicity reasons, antenna indices will be dropped in the following expressions.
where \({\mathbf {A}} = {\mathbf {C}}_{hh_{p}}({\mathbf {C}}_{h_{p}h_{p}} + {\sigma ^{2}_{I}}{\mathbf {I}}_{P})^{1}\) is a N×P matrix with the LMMSE interpolation coefficients, \({\mathbf {C}}_{hh_{p}} = {\mathbf {P}}_{k}^{T}\mathrm {E}\{ {\mathbf {g}}_{k} {\mathbf {g}}_{k}^{T} \}\) and \({\mathbf {C}}_{h_{p}h_{p}}={\mathbf {P}}_{k}^{T}\mathrm {E}\{{\mathbf {g}}_{k} {\mathbf {g}}_{k}^{T}\}{\mathbf {P}}_{k}\) are submatrices of the covariance matrix of the channel frequency response, and \(\hat {{\mathbf {g}}}_{k}\) is a N×1 vector with the estimated channel coefficients per subcarrier. The secondorder statistics of the channel are estimated by using all the received pilot subcarriers in a frame. Also, the power \({\sigma ^{2}_{z}}\) of the ICI arising from z _{ k } in Equation 6 is estimated according to [22], and it is considered as part of \({\sigma ^{2}_{I}} = {\sigma ^{2}_{w}}+{\sigma ^{2}_{z}}\). As explained below, these frequency response estimations play a fundamental role in the estimation of the ICI.
where b _{ r,q } is the qth column of B _{ r }, G _{1} and G _{2} are the matrices which append and remove the cyclic prefix in Equation 5, and \(\hat {{\mathbf {c}}}_{q}\) is a N×1 vector with the columns of the estimate \(\hat {{\mathbf {C}}}^{T}\).
Finally, the estimates \(\hat {{\mathbf {c}}}_{q}\) in Equation 9 are obtained from the columns of \(\hat {{\mathbf {C}}}_{r}^{T}\), which provide the coefficients for the channel matrices G _{ r } of the group.
The results in this work are all obtained by using the discrete prolate spheroidal BEM (DPSBEM) [10]. Such a BEM is built on the Slepian sequences arising from time sequences whose energy is localized on a given frequency interval. For the purpose of ICI estimation, the frequency interval is the one corresponding to the Doppler spectrum, whose domain is bounded by the maximum Doppler frequency. Recall that the use of DPSBEMs is equivalent to assuming a timeselective channel with a flat Doppler spectrum. Consequently, to determine the specific DPSBEM appropriate for a given scenario, it is necessary to know the mobile speed, and more specifically, the emulated speed after interpolation [10].
3.3 ICI cancellation and equalization
Once the full frequency response matrices for all antenna pairs are obtained, both ICI cancellation and equalization of the received signal are done to obtain estimates of the information symbols s _{ k }. In the literature, both block interference cancellation (BIC) and sequential interference cancellation (SIC) schemes have been proposed [15,23]. In this work, a LMMSE SIC receiver of seven taps is used to remove the ICI and equalize the channel. Such a method has been chosen due to its good tradeoff between computational cost and performance.
 1.
Obtain the matrices \(\hat {{\mathbf {C}}}_{r}\) for the K−R+1 groups in a frame, as explained in the previous subsection.
 2.Remove ICI from each OFDM symbol as follows:
 (a)
For symbols k∈[1,R/2], ICI is suppressed with the coefficients obtained from \(\hat {{\mathbf {C}}}_{1}\).
 (b)
For symbols k∈[R/2+1,K−R/2], ICI is suppressed with the coefficients obtained from \(\hat {{\mathbf {C}}}_{kR/2+1}\).
 (c)
For symbols k∈[K−R/2+1,K], ICI is suppressed with the coefficients obtained from \(\hat {{\mathbf {C}}}_{KR+1}\).
 (a)
 3.
Obtain a new frequency response estimate from the ICIreduced received signal and return to step 1.
As can be seen, except for the first and last symbols of the frame, ICI is estimated and cancelled for the central OFDM symbol of each set. The final equalization of the ICIreduced signal is done by zero forcing. Detected information symbols are demapped and sent to a Viterbi decoder to obtain the information bits.
4 Experimental setup
A single board is used in continuous transmitonly mode for implementing the base station transmitter for the downlink. The base station is equipped with two MiniCircuits TVA11422 highpower amplifiers [27], two Interline SECTOR ISG14F2425A120V vertically polarized transmit antennas [28], and an Ubiquity AM2G15120 cross polarized transmit antenna [29]. Notice that a single vertically polarized antenna was used for the singleinput singleoutput (SISO) transmissions, while the cross polarized antenna was employed for measuring the 2×2 MIMO ones.
The remaining two USRP B210 boards are used for implementing two different mobile receivers, both mounted on a car. The first one is connected to a couple of eRize ERZA24O09MBR omnidirectional 9 dBigain antennas placed outdoors, on the roof the car. The second mobile receiver is connected to another two omnidirectional antennas placed inside the car, between the two front seats. Using both mobile receivers allows us to capture, at the same time, the signals transmitted by the base station to both outdoor and indoor receivers.
With respect to the software, we use a multithread receiver implemented in C++ with Boost and using the Ettus USRP Hardware Driver (UHD) software. The main thread of the receiver is responsible for retrieving the samples coming from the USRP through the USB 3.0 bus and store them in a set of buffers in the main memory of the host laptop. The second thread reads the samples from such buffers and saves them persistently in a dedicated solidstate drive. Finally, there is a lowpriority thread for logging important information for documenting the measurement campaign. On the other hand, the transmitter is a singlethread process since the same signals are cyclically transmitted overtheair. Therefore, the signals to be radiated are first stored in a temporary buffer and next transmitted in a loop to the USRP through the USB 3.0. The rest of the software was implemented in MATLAB.
Parameter values of the WiMAX profile used in the experiments
Parameter  Value 

Sampling frequency  8 MHz 
Useful bandwidth  7 MHz 
Number of subcarriers  1,024 
Number of data subcarriers  720 
Number of pilot subcarriers  120 
Subcarrier spacing  7.81 kHz 
Cyclic prefix length  128 samples 
Permutation zone  PUSC 
Once the transmit signals have been generated, the base station is notified and the overtheair transmission process starts. First, the signals are read from the corresponding source file and transferred to the USRP, where they are again interpolated in the FPGA before reaching the (DAC). Note that this interpolation stage is needed for adapting the signal sampling rate to the sampling frequency of the DAC, thus not affecting the signal bandwidth. Next, the signals are upconverted to the central frequency f _{ c }=2.6 GHz, preamplified inside the USRP (configured with a gain value of 60 dB out of 89.5 dB), amplified by the two MiniCircuits TVA11422 highpower amplifiers (one per transmit antenna) at their maximum gain of 40 dB, and finally radiated by the antennas (we use the vertically polarized transmit antennas for SISO transmissions, while the cross polarized ones are employed for 2×2 MIMO ones). The mean transmit power value measured at each antenna input is +17.5 dBm when a single transmit antenna is used. In case both transmit antennas are employed, the transmit gain is reduced to 3 dB per antenna to keep the total transmit power equal despite of the number of transmit antennas.
Figure 3 shows the base station placed outdoors, on the second floor of the CITIC building located in the Campus de Elviña at the University of A Coruña. It also shows the power amplifiers, the antennas (vertically or cross polarized), the USRP B210, and the laptop running the software for the base station. Looking at the picture of the crosspolarized antennas in Figure 3, one can also see part of the road traversed by the car during the measurements.
The two receivers are also built around the USRP B210 and the UHD with the software installed on two laptops, one per receiver. Notice that during the measurements, the acquired signals are persistently stored in dedicated solid state drives attached to laptop receivers, but they are not processed using the WiMAX receiver at that moment. As shown in Figure 4, the two outdoor receive antennas are sticked on the roof of the car used for measuring and they are connected directly to the USRP B210, which is powered by its corresponding laptop. The receive gain of the USRP is set to 35 dB (out of 73 dB), ensuring a linear operation. The second receiver is completely installed inside the car used for the measurements, with the antennas between the two front seats of the car. We use another laptop for running the receive software and for persistently storing the signals acquired by the indoor receiver. Unlike the outdoor receiver, the receive gain of the indoor receiver is set to 45 dB to better accommodate the amplitude of the acquired signals to the range of the ADC, while ensuring that the receive signal is not distorted by the amplifiers.
4.1 Physical layer configuration
OFDM symbol structure of the transmitted signal follows the downlink of mobile WiMAX standard regarding subcarrier allocation, symbol mapping, and channel encoding. The frames consists of K=25 symbols, with the first one reserved for the preamble signal transmitted only by the first antenna. The other symbols carry the information of six bursts with the modulation and channel coding profiles: 4QAM 1/2, 4QAM 3/4, 16QAM 1/2, 16QAM 3/4, 64QAM 1/2, and 64QAM 3/4. Each burst spans during four consecutive OFDM symbols along all the available data subcarriers. The permutation matrices \({\mathbf {p}}_{k}^{(m)}\) and \({\mathbf {d}}_{k}^{(m)}\) are generated according to the partial usage of subcarriers (PUSC) zone with the corresponding modifications for MIMO as specified in the standard and pilot subcarriers on \({\mathbf {p}}_{k}^{(m)}\) generated as a boosted BPSK sequence obtained by mapping the output of a linear feedback shift register. Channel coding is performed by the convolutional encoder defined in the standard, which features tailbiting to terminate coding blocks.
Regarding the receiver, a DPSBEM with Q=5 is used for ICI estimation, given its good properties to estimate ICI assuming that only the maximum Doppler frequency is known.
In order to ensure similar channel characteristics for all the interpolation factors in each measurement, a superframe consisting of four different parts is built. The first part is a small time period during which the base station is kept silent to allow for noise variance estimation at the receiver as it was already explained. The remaining three parts are built from the same mobile WiMAX frame consisting of the six burst plus the preamble that was described above. Consequently, the second part of the superframe corresponds to the aforementioned mobile WiMAX frame interpolated in time by the factor I=4. Next, the third part is the same as the second one but using I=12, while the fourth part employs I=20. Finally, the superframe is cyclically transmitted from the base station to both mobile receivers, one placed outside the car and the other installed inside.
5 Results
This section presents the results obtained from the measurements described above. We also carried out simulations to support the experimental results. Simulations were designed to create scenarios comparable to those obtained from the measurements. Parameters estimated from the realdata captured are the mean SNR of each frame, and the Kfactor of the wireless channel. The mean SNR is estimated taking as a reference the noise variance estimated during the silence preceding each frame; and the Kfactor is estimated and averaged for all frames. Regarding simulations, the estimated mean SNRs for each frame in the real scenario are applied to the simulated transmissions, all with the same average Kfactor. Finally, full characterization of the channel frequency responses or spatial covariance matrices have not been obtained, so simplified assumptions have been made while conducting the simulations. Basically, we assumed a frequencyflat spatially uncorrelated MIMO channel. The frequencyflat assumption arises from the fact that for the highest interpolation factors, the bandwidth of the transmitted signals will be rather narrow, and therefore the frequency selectivity observed by the receiver would be negligible. As for the MIMO channel, the coefficients for each pair of antennas are drawn from independent Rice distributions, although not identically distributed, picking random complex mean values for each MIMO channel matrix entry.
5.1 Error vector magnitude and SNR
being EVM_{ F } the mean EVM for the Fth frame. Note that the original transmitted symbols are used to normalize the (EVM). Notice that if only symbol estimates were available, EVM measures would be less accurate.
5.2 Throughput
In order to evaluate the impact of the ICI cancellation methods on the final user quality of experience (QoE), throughput estimations have been carried out using the information conveyed in the bursts. Two types of throughput estimation have been considered. The first one assumes that the transmitter knows, for each frame, the optimal modulation and coding profile to be used. This case was modeled by estimating the bit error ratio (BER) of the six received bursts per frame and assuming that the whole frame was transmitted with the profile which carries more information bits without suffering any bit error. The second method assumes that the transmitter estimates the most suitable burst profile, and consequently only transmits the corresponding profile during all the burst. Errors are measured in this case assuming only the burst picked by the transmitter is available, regardless whether this burst has errors or not. This method is intended to determine the impact of the channel fast timevariations on this type of decisions.
EVM to burstprofile mapping for throughput estimations
Profile  EVM [dB] 

4QAM 1/2  5 
4QAM 3/4  8 
16QAM 1/2  10.5 
16QAM 3/4  14 
64QAM 1/2  16 
64QAM 3/4  20 
Mean throughput in Mbit/s for the 1×1 scenario when the receive antennas are placed outside the car
Ideal throughput  Ideal throughput  Feedback throughput  Feedback throughput  

(with ICI cancellation)  (with ICI cancellation)  
I=4  14.32  14.73  13.01  13.93 
I=12  12.22  13.78  8.81  11.39 
I=20  9.76  11.37  5.48  6.51 
Mean throughput in Mbit/s for the 1×1 scenario when the receive antenna is placed inside the car
Ideal throughput  Ideal throughput  Feedback throughput  Feedback throughput  

(with ICI cancellation)  (with ICI cancellation)  
I=4  3.15  3.48  2.11  2.45 
I=12  2.90  3.26  1.96  2.23 
I=20  2.67  3.03  1.82  1.97 
Mean throughput in Mbit/s for 2×2 scenario when the receive antennas are placed outside the car
Ideal throughput  Ideal throughput  Feedback throughput  Feedback throughput  

(with ICI cancellation)  (with ICI cancellation)  
I=4  16.95  17.46  10.56  10.99 
I=12  15.74  17.48  9.20  11.01 
I=20  12.61  13.99  7.49  8.87 
6 Conclusions
We have experimentally evaluated the performance of the WiMAX downlink physical layer in highmobility scenarios. Both SISO and MIMO transmissions were considered, as well as placing the receive antennas outside and inside the car used for the experiments. We focused on the ICI caused by channel time variations in such scenarios. Costeffective measurement campaigns were carried out where highmobility scenarios are emulated with a vehicle moving at a low speed. The key idea is the enlargement of the symbol period prior to its transmission over the air. Such enlargement reduces the frequency spacing between the OFDM subcarriers in WiMAX transmissions and hence induces high ICI values on the received signals. Experiments illustrate the performance of WiMAX receivers with and without ICI cancellation in terms of EVM and throughput. ICI cancellation produces significant performance gains mainly when the received SNR is high. Otherwise, thermal noise dominates over the ICI, and gains are not so much appreciated. Furthermore, the spacing of subcarriers in this standard is sufficient to provide a robust behavior for moderate mobility environments, and the gain observed after ICI cancellation is not very significant for speeds up to 80 km/h. Note that the profile tested in this work is the WiMAX profile with the lower subcarrier spacing, so it is foreseeable that other profiles present better performance under these scenarios.
Declarations
Acknowledgements
The authors thank Ismael Rozas Ramallal for his support in developing and testing the testbed and conducting the measurement campaigns. This work was supported in part by Xunta de Galicia, MINECO of Spain, and by FEDER funds of the E.U. under Grant 2012/287, Grant IPT20111034370000, and Grant TEC201347141C41R.
Authors’ Affiliations
References
 UQ WiMAX gaining the wireless broadband market (2013). http://www.wimaxforum.org/LiteratureRetrieve.aspx?ID=177978.
 M Ehammer, E Pschernig, T Graupl, in IEEE/AIAA 30th Digital Avionics Systems Conference (DASC). AeroMACS  an airport communications system (IEEESeattle, USA, 2011), pp. 4–114116. doi:10.1109/DASC.2011.6095903.Google Scholar
 I Barhumi, G Leus, M Moonen, Optimal training design for MIMO OFDM systems in mobile wireless channels. IEEE Trans. Signal Process. 51(6), 1615–1624 (2003). doi:10.1109/TSP.2003.811243.View ArticleGoogle Scholar
 S Das, P Schniter, MaxSINR ISI/ICIshaping multicarrier communication over the doubly dispersive channel. IEEE Trans. Signal Process. 55(12), 5782–5795 (2007). doi:10.1109/TSP.2007.901660.View ArticleMathSciNetGoogle Scholar
 Z Tang, RC Cannizzaro, G Leus, P Banelli, Pilotassisted timevarying channel estimation for OFDM systems. IEEE Trans. Signal Process. 55(5), 2226–2238 (2007). doi:10.1109/TSP.2007.893198.View ArticleMathSciNetGoogle Scholar
 GB Giannakis, C Tepedelenlioglu, Basis expansion models and diversity techniques for blind identification and equalization of timevarying channels. Proc. IEEE. 86(10), 1969–1986 (1998). doi:10.1109/5.720248.View ArticleGoogle Scholar
 G Leus, in Proc. of EUSIPCO. On the estimation of rapidly timevarying channels (EURASIPVienna, Austria, 2004), pp. 2227–2230.Google Scholar
 DK Borah, BT Hart, Frequencyselective fading channel estimation with a polynomial timevarying channel model. IEEE Trans. Commun. 47(6), 862–873 (1999). doi:10.1109/26.771343.View ArticleGoogle Scholar
 H Hijazi, L Ros, Polynomial estimation of timevarying multipath gains with intercarrier interference mitigation in OFDM systems. IEEE Trans. Vehicular Technol. 58(1), 140–151 (2009). doi:10.1109/TVT.2008.923653.View ArticleGoogle Scholar
 T Zemen, CF Mecklenbräuker, Timevariant channel estimation using discrete prolate spheroidal sequences. IEEE Trans. Signal Process. 53(9), 3597–3607 (2005). doi:10.1109/TSP.2005.853104.View ArticleMathSciNetGoogle Scholar
 KAD Teo, S Ohno, in IEEE Global Telecommunications Conference, 2005. GLOBECOM ’05, vol. 6. Optimal MMSE finite parameter model for doublyselective channels (IEEESt. Louis, USA, 2005), pp. 3503–3507. doi:10.1109/GLOCOM.2005.1578424.Google Scholar
 P SuárezCasal, JA GarcíaNaya, L Castedo, M Rupp, in IEEE 14th Workshop on Signal Processing Advances in Wireless Communications. SPAWC ’13. KLTbased estimation of rapidly timevarying channels in MIMOOFDM systems (IEEEDarmstadt, Germany, 2013).Google Scholar
 A Stamoulis, SN Diggavi, N AlDhahir, Intercarrier interference in MIMO OFDM. IEEE Trans. Signal Process. 50(10), 2451–2464 (2002). doi:10.1109/TSP.2002.803347.View ArticleGoogle Scholar
 X Ma, GB Giannakis, S Ohno, Optimal training for block transmissions over doubly selective wireless fading channels. IEEE Trans. Signal Process. 51(5), 1351–1366 (2003). doi:10.1109/TSP.2003.810304.View ArticleMathSciNetGoogle Scholar
 F Hlawatsch, G Matz, Wireless Communications Over Rapidly TimeVarying Channels (Academic Press, USA, 2011).Google Scholar
 H Hijazi, L Ros, G Jourdain, in Proceedings European Wireless Conference. OFDM channel parameters estimation used for ICI reduction in timevarying multipath channels (IEEEParis, France, 2007).Google Scholar
 M Meidlinger, M Simko, Q Wang, M Rupp, in IEEE International Conference on Communications 2013. ICC 2013. Channel estimators for LTEA downlink fast fading channels (IEEEStuttgart, Germany, 2013).Google Scholar
 J RodríguezPiñeiro, P SuárezCasal, JA GarcíaNaya, L Castedo, C BrisoRodríguez, JI AlonsoMontes, in IEEE 8th Sensor Array and Multichannel Signal Processing Workshop (SAM 2014). Experimental validation of ICIaware OFDM receivers under timevarying conditions (IEEEA Coruña, Spain, 2014).Google Scholar
 PS Casal, J RodríguezPiñeiro, JA GarcíaNaya, L Castedo, in The Eleventh International Symposium on Wireless Communication Systems (ISWCS). Experimental assessment of WiMAX transmissions under highly timevarying channels (IEEEBarcelona, Spain, 2014).Google Scholar
 W Nam, YH Lee, Preamblebased cell identification for cellular OFDM systems. IEEE Trans. Wireless Commun. 7(12), 5263–5267 (2008). doi:10.1109/TWC.2008.071426.View ArticleGoogle Scholar
 A CarroLagoa, P SuarezCasal, J GarciaNaya, P FragaLamas, L Castedo, A MoralesMendez, Design and implementation of an OFDMATDD physical layer for WiMAX applications. EURASIP J. Wireless Commun. Netw. 2013(1), 243 (2013). doi:10.1186/168714992013243.View ArticleGoogle Scholar
 Y Li, LJ Cimini, Bounds on the interchannel interference of OFDM in timevarying impairments. IEEE Trans. Commun. 49(3), 401–404 (2001). doi:10.1109/26.911445.View ArticleMATHGoogle Scholar
 X Cai, GB Giannakis, Bounding performance and suppressing intercarrier interference in wireless mobile OFDM. IEEE Trans. Commun. 51(12), 2047–2056 (2003). doi:10.1109/TCOMM.2003.820752.View ArticleGoogle Scholar
 J RodríguezPiñeiro, JA GarciaNaya, A CarroLagoa, L Castedo, in Proc. of 16th Euromicro Conference on Digital System Design. A testbed for evaluating LTE in highspeed trains (IEEESantander, Spain, 2013).Google Scholar
 Ettus USRP B210 (2013). https://www.ettus.com/product/details/UB210KIT.
 Analog Devices AD9361 RFIC (2013). http://www.analog.com/en/rfifcomponents/rfiftransceivers/ad9361/products/product.html.
 MiniCircuits highpower amplifier TVA11422 (2013). http://www.minicircuits.com/pdfs/TVA11422.pdf.
 Interline SECTOR ISG14F2425A120V vertically polarized antenna (2014). http://interline.pl/antennas/SECTORV12014dBi120deg2.42.5GHz.
 Ubiquity AM2G15120 cross polarized antenna (2013). http://dl.ubnt.com/datasheets/airmaxsector/airMAX_Sector_Antennas_DS.pdf.
 M Matthaiou, DI Laurenson, CX Wang, in IEEE Wireless Communications and Networking Conference, 2008. WCNC 2008. Capacity study of vehicletoroadside MIMO channels with a lineofsight component (IEEELas Vegas, USA, 2008), pp. 775–779. doi:10.1109/WCNC.2008.142.Google Scholar
Copyright
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 credited.