Estimation of CFO and Channels in Phase-Shift Orthogonal Pilot-Aided OFDM Systems with Transmitter Diversity

We present a CFO estimation algorithm and an associated channel estimation method for broadband OFDM systems with transmitter diversity. The CFO estimation algorithm explores the TD structure of the transmitted symbols carrying pilots and data, relying solely on the data component present on the symbols to estimate the CFO, thus avoiding additional overhead like training symbols or null subcarriers. An intermediate output of the CFO algorithm provides an easy-to-get initial CIR estimate that will be improved with the utilization of a TD LMMSE ﬁlter. The feasibility of the investigated methods is substantiated by system simulation using indoor and outdoor broadband wireless channel models. Simulation results show that the joint algorithms provide a near optimal system’s performance.


Introduction
Future mobile broadband applications will require reliable high data-rate wireless communication systems.In recent years, multiple-input multiple-output orthogonal frequency division multiplexing (MIMO-OFDM) transmission systems [1][2][3][4] emerged as the scheme with the potential to fulfill these conditions, with bandwidth efficiency and robustness to frequency selective channels, common in mobile personal communication systems.
The estimation and removal of the CFO has been the focus of a considerable number of works published in recent years.The algorithms can be categorized as blind or dataaided.The first category explores the properties of the received symbols (commonly the cyclic prefix (CP)) [10][11][12].The data-aided algorithms use dedicated training symbols [13,14] or exploit the presence of null subcarriers [15,16].
The accurate extraction of the channel state information is crucial to realize the full potential of the MIMO-OFDM system.The performance of the channel estimator is vital for diversity combining, coherent detection and decoding, and resource allocation operations.The cochannel interference inherent to the system, where the received signal is the superposition of the signals transmitted simultaneous from the different antennas, puts an additional challenge on the design of the channel estimation algorithm.
A decision-directed channel estimation scheme that attempted to minimize the cochannel interference was published in [17].The proposed algorithm exhibits a high computational load.A simplified and enhanced algorithm, introducing a data-aided scheme for the data transmission mode, is presented in [18].The topic attracted a significant attention and has been the focus of investigation in multiple publications [19][20][21] and references therein.
The design of training symbols and pilot sequences with the ability to decouple the cochannel interference and minimize the channel estimation mean square error (MSE) EURASIP Journal on Wireless Communications and Networking for MIMO-OFDM was addressed in several publications [18,22,23].
Most publications on the topic of training-signal or pilotaided channel estimation use the frequency-domain (FD) least squares (LS) estimates as the starting point for the analysis of the estimation algorithm or the design of the training sequence.It was established in [24] that in singleinput single-output (SISO) OFDM a time-domain (TD) equivalent to LS estimate could be obtained using a simple linear operation on the received signal, if the used pilot sequence fulfills certain conditions.This paper contains a proposal for a CFO estimation algorithm and associated channel estimation method for OFDM systems with transmitter diversity that exploits a standardized transmission format, where FD pilot symbols are regularly spread in the OFDM symbols.To minimize the pilot overhead, the pilot subcarriers are shared among all transmit antennas.To mitigate the resulting cochannel interference, the system adopts phase-shifted pilot sequences per transmit antenna [18].By exploring the TD properties of the received symbols, the proposed algorithms are able to estimate and remove the CFO, separate each of the CIRs, and generate the final channel estimate, without requiring any additional overhead (training symbols or null subcarriers).By performing most of the operations on the TD received symbols and sharing operations, the overall computational load required to implement both algorithms is affordable for real-time implementations.
The paper is organized as follows.Section 2 gives a brief introduction to the wireless multipath channel and the OFDM baseband model.In Section 3, the investigated CFO and channel estimation algorithms are developed.The feasibility of the developed method is substantiated by simulation results presented in Section 4. Finally, conclusions are drawn in Section 5.

OFDM in Mobile Wireless Channels
Before introducing the investigated method, we will briefly overview the mobile wireless multipath channel and the considered OFDM baseband model.
Throughout the text, the notation (∼) is used for TD vectors and elements, and its absence denotes frequencydomain (FD) vectors and elements.The index n denotes TD elements and k FD elements.Unless stated otherwise, the vectors involved in the transmission/reception process are column vectors with N C complex elements.The superscripts (•) T and (•) H denote transpose and Hermitian transpose, respectively.

The Wireless Multipath Channel.
Let us consider that the system transmits over multipath Rayleigh fading wireless channels modeled by the discrete-time channel impulse response (CIR): where L p is the number of channel paths, α l and τ l are the complex value and delay of path l, respectively.The paths are assumed to be statistically independent, with normalized average power, , where σ 2 h [l] is the average power of path l.The channel is time variant due to the motion of the mobile terminal (MT), but we will assume that the CIR is constant during one OFDM symbol.The time dependence of the CIR is not present in the notation for simplicity.Assuming that the insertion of a long enough cyclic prefix (CP) in the transmitter assures that the orthogonality of the subcarriers is maintained after transmission, the channel frequency response (CFR) can be expressed as where N C is the total number of subcarriers of the OFDM system.

OFDM Baseband Model.
Consider the OFDM baseband system with n S transmit antennas depicted in Figure 1.The n S vectors d s hold the M-ary PSK or QAM coded data to be transmitted.
To assist in the channel estimation process, pilot symbols are added in each transmit antenna path.The n S vectors p s hold the pilot values for each path.The pilots are transmitted in dedicated subcarries (vectors p s and d s contain nonzero values in disjoint positions).The resulting FD signal transmitted by antenna s is s s =d s +p s .All transmit antennas use the common set of subcarriers ℘ to convey the overlapping pilot sequences.The pilots are regularly spread every N f subcarriers.The pilot separation can range from 1 (particular case where all subcarriers in the OFDM symbol are dedicated to transmit pilots-training symbol) to N C , fulfilling the condition N C /N f n S ∈ N.
The system uses distinct phase-shifted pilot sequences in each transmit antenna to allow the separation of the sequences in the receiver.The kth element of the vector p s is defined by where N t = N C /N f , and k ini ∈ {0, . . ., N f −1} is the first pilot subcarrier.
The inverse discrete Fourier transform (DFT) block present in each antenna path transforms the input vector into the TD vector s s , using an efficient N C -points inverse fast Fourier transform (FFT) algorithm.
An L samples long guard interval, in the form of CP, is prefixed to vector s s , resulting in the TD transmitted vector where The nth received signal sample of the ith symbol can be expressed as where n i [n] is a sample of independent and identically distributed (iid) zero mean additive white Gaussian noise (AWGN) with variance σ 2 n .Collecting the (N C + L) samples of the symbol, where the vector n i collects the noise samples that affect the ith symbol, the vector z i represents the intersymbol interference (ISI) caused by the channel dispersion, and the matrix H lin s,i is the (N C +L)×(N C +L) lower triangular Toeplitz channel convolution matrix with first column h s,i (column (N C + L)-vector with the discrete-time CIR (its elements are defined by ( 1)) padded with zeros).The (N C + L) × (N C + L) diagonal matrix that holds the phase rotation that affects each symbol sample is The receiver starts by removing the CP from the received symbol.Dropping the symbol index, the resulting vector is where is the matrix that removes the CP with 0 (NC×L) representing the (N C × L) null matrix, n = R CP n is the resulting TD noise column N C -vector, and is the common phase that affects all samples of the ith symbol.The last step in (8) was possible considering the structure of the matrices involved With the assumption that the length of the CP is larger than the duration of CIR, the ISI is completely removed, and (8) can be written as where circulant matrix with circulant vector h s and, due to the properties of the DFT, , with the elements of h s defined by (2).
The CFO and channel estimation block is responsible for estimating both the CFO that affects the received samples and the n S channels that disturbed the transmission process.Both estimation algorithms will be introduced in the next section.Moreover, this block is also responsible for reducing the CFO, using the estimated CFO value w o .This operation can be described by and the CFO is completely removed.As it will be demonstrated in the next section, the CFO ambiguity remaining after this block is an integer multiple of the pilot subcarrier separation N f Δ f , where Δ f is the subcarrier separation.This acquisition range should be sufficient for current OFDM systems; however coarse CFO estimation techniques [25] can be used to tackle this limitation, if proven necessary.
The DFT block transforms the vector r to FD with an efficient FFT operation.Assuming that the CFO is completely eliminated, the resulting FD column N C -vector can be expressed as where n = FC H NC ( w o ) n is the resulting FD noise vector.The remaining phase-rotation θ ini is naturally removed in the channel estimation process, assuming that the pilot-aided scheme calculates the LS estimates (back-rotated received signal).
The deframing block separates the signals in the subcarriers conveying pilots and data symbols.The values in the data subcarriers are collected in vector d and fed to the decoding block.Together with the channels' estimate h s , this block is now able to decode the received symbols, according to some decision rule, and generate the estimate of the transmitted data b.

CFO and Channel Estimations by Exploring the TD Properties of Phase-Shifted Pilot Sequences
The algorithms implemented in this block estimate both the n S channels over which the transmission occurred and the CFO that affects the received signal.The inputs to the CFO estimation algorithm are the TD symbols carrying both pilots and data, according to the model defined in the previous section.The channel estimation algorithm reuses an intermediate output of the previous operation to attain an initial CIR estimate with minimal computational load.
3.1.Analysis of the TD Symbol's Structure.From (10), each element of the TD received symbols (carrying pilots and data), after CP extraction, can be expressed by where the elements of the TD data vector d s are where d s [k] is the kth element of d s (complex data symbol conveyed by the kth subcarrier of the s transmit antenna path), and the elements of the TD pilot vector p s are Replacing ( 14) and ( 15) in (13), where r d and r p hold the data-dependent and pilotdependent components in r, respectively.By expanding the pilot-dependent vector it becomes clear that it is made up of N f frequency-shifted and scaled replicas of each of the n S CIR.Moreover, the replicas of each CIR are separated by N t samples and transmit antenna s CIR replicas are time-shifted (s/n S )N t samples from the reference position mN t , m ∈ {0, . . ., N f − 1}.

CFO Estimation.
The CFO estimation method introduced in the following uses the pilot structures, introduced primarily for channel estimation purposes, to estimate the CFO present in the received samples.Therefore, it is absolutely bandwidth efficient, as it does not require any additional specific overhead.The algorithm exhibits a fast acquisition, being able to output an estimate with low deviation from a single OFDM frame.It proves adequate for burst mode transmission, where the frequency offset varies from frame to frame.The algorithm requires a search within the acquisition range to find the minimum value of the cost function.An initial candidate angular frequency offset w is applied to the input signal r i , together with the TD equivalent of the FD multiband filter that selects the pilot subcarriers [24] (phaseshifted sum of the samples in the same relative position in all N f segments of N t samples).This operation can be described by where the ( , the column N t -vectors g d and g p hold the data-dependent and pilot-dependent components in g, respectively, and v is the resulting noise vector.The elements of g p can be expressed by If channel s maximum delay τ s (normalized to the system's sampling interval Δt) is short enough so that the adjacent CIR replicas in (19) to not overlap (fulfils the sampling theorem), (19) can be further simplified to that clearly shows that the pilot-dependent samples are limited to the n S sets of samples {L p } (with L p elements), where the corresponding phase-shifted CIRs have energy.
The remaining samples will depend only on the transmitted data and noise.
The elements of g d can be expressed by where where f is the initial candidate frequency offset.The elements of the noise vector v can be expressed by Figure 2 depicts an example of the constitution of vector g for a system with 2 transmission antennas, N C = 1024 subcarriers and pilot separation N f = 4(N t = 256).The plots put in evidence that the CIRs energy is limited to 2 sets of samples and the data-dependent component spans the entire symbol duration.
A careful inspection of ( 23) reveals that the factor Ψ (and the data-dependent component) is zero for independently of the considered sample.Keeping in mind that k / ∈ ℘ and (k − k ini ) / = mN f , m ∈ Z, the solution for ( 25) is where Δ f is the subcarrier separation.It should be noted that the solution in (26) presents a periodicity N f Δ f and includes the condition when the CFO is completely eliminated ( f o − f = 0).A similar analysis reveals that the factor Ψ has maximum magnitude for We can conclude that Ψ has minimum values spread N f Δ f Hz, with (N f − 1) maximum magnitude values in between, separated by Δ f Hz.Let us define the column (N t − n S L p )-vector j that collects the samples of g with no CIRs energy (only datadependent and noise; example depicted in Figure 2) and the cost function J( w) as the energy in j: The definition of the cost function guarantees that, if within the acquisition range, its minimum value will converge to the true estimate as the number of elements in j increases (and the noise term tends to a floor in J( w)).The elements in j may be obtained from one OFDM symbol or a set of symbols (with data and pilots) if higher accuracy on the estimate is required.From the previous analysis of the factor Ψ, it is clear that the acquisition range of our cost function is The CFO estimate can be found by a line search within the acquisition range to find the minimum value of the cost function: where w o is the estimated CFO value.The exhaustive line search is computationally demanding, depending on the search's granularity.Hence, there is a tradeoff between complexity and estimate's variance.
The cost function has a closed form expression, and its behavior is perfectly described.In the acquisition range, there are N f maximum values; in the interval limited by the maximum values that surround the perfect estimate, J( w) presents a smooth shape with a single minimum.Using the knowledge we possess of the cost function, we propose a 2step approach to find its minimum value.The initial step performs a coarse line search to locate the global minimum interval.Testing N f candidate CFO values evenly spaced by Δ f Hz should suffice.The candidate CFO will be the one with the lowest cost.If the number of elements in j is small and SNR is very low, the probability of wrong identification may not be negligible and the number of candidate CFO values can be increased thus decreasing the wrong identification probability.In the final step, we use the gradient descent method [26] to find the global minimum.Figure 3 depicts an example of the cost function for a 2 × 1 Alamouti OFDM system with N C = 1024 subcarriers, sampling interval Δt = 10 nanoseconds and pilot separation N f = 4.The values in the plot were acquired using an SNR = 20 dB.In Figure 3(a), the separation of ≈390 kHz between consecutives minimum values is visible.Figure 3(b) shows in detail the interval around (w o − w) = 0.It is clear that it has a unique global minimum that is easy to find (no problem with local minimum values).In Figure 3(c), the N f − 1 maximum values between consecutive minimum values are clearly visible.It also shows in detail that the separation of the maximum values around (w o − w) = 0 is ≈195 kHz.
Figure 4 shows the evolution of the estimated CFO standard deviation with the number of samples used in estimation algorithm (elements of j).The plots depict the standard deviation when the number of samples goes from 200 to 2000, in steps of 200 samples.

Channel Estimation.
Assuming that the CFO is completely eliminated, the output of the initial operation of the CFO algorithm is made up of the pilot-dependent component and noise g = g p + v.The data-dependent component was eliminated from this vector, opening way to easily obtain an initial CIR estimate.
The channel estimation algorithm starts by isolating each of the n S phase-shifted CIRs from g and removing the modulating exponential factor.The elements of the resulting vectors h s,LS can be expressed by (30)  In [24], it was demonstrated that for a single transmitting antenna OFDM system with perfect synchronization, (30) is the TD counterpart of FD LS estimate.By using phase-shifted pilot sequences that allow the separation of the different CIRs, the same result holds in the present model.
This initial estimate can be significantly improved by incorporating a TD linear minimum MSE (LMMSE) filter W s to reduce the estimate's error, taking advantage of the CIR energy concentration.The improvements provided by this filter are especially significant for low values of SNR.
The LMMSE filter can be expressed by [27] , 0, . . ., 0 . (31) The resulting CIR and CFR estimates are, respectively, h s = W s h s,LS and h s = F h s .

Simulation Results
A simulation scenario was implemented using an Alamouti 2 × 1 OFDM system with Nc = 1024 modulated subcarriers, sampling interval Δt = 50 nanoseconds and a CP with 100 samples.The transmitted OFDM symbols carried pilots and data, with a pilot separation N f = 8.The OFDM frame consists of 16 symbols.The CFO value was randomly generated in each frame with a value inside the acquisition range ] − N f Δ f /2, N f Δ f /2[.The CFO estimation and removal was performed on a frame basis.Two channel models with exponentially decaying power delay profile (PDP) were used to simulate indoor (50 nanoseconds rms delay spread) and outdoor environments (250 nanoseconds rms delay spread).To validate the proposed method, several simulations were performed using Eb/N0 values in the range of 0 dB to 20 dB.CFO.The solid lines represent the standard deviation of the CFO estimate.In our method, the gradient descent was stopped for a step of 10 HZ.For the method in [16], one null subcarrier was added to each OFDM symbol, and an exhaustive search was performed with a 10 HZ step.The results show that our method is unbiased for the all range of Eb/N0 values.The algorithm generates estimates with small deviation from the true value using a limited number of symbols (16).The estimate deviation is ∼1% of that of [16] for low values of SNR.The method in [16] requires a large number of symbols and/or more receive antennas to generate accurate estimates.This result shows that the investigated method is quite adequate for burst systems.Figure 6 depicts the MSE of the channel estimation algorithm (blue plots) when the system has perfect synchronization and the MSE of the joint CFO and channel estimation process (red plots), when using QPSK, 16-QAM, and 64-QAM modulations.Figure 6(a) presents the results for the indoor channel and Figure 6(b) for the outdoor channel.
Figure 7 depicts the system BER for 3 scenarios: the ideal situation where the receiver has perfect channel state information (CSI) (with no pilot overhead) and the CFO is absent (green plots), the situation when the receiver must estimate the channel from the received samples (blue plots), and the more realistic scenario where the receiver needs to estimate both the CFO and channel (red plots).Simulations results were obtained also for QPSK, 16-QAM, and 64-QAM modulations as identified in the figures.Figure 7(a) presents the results for the indoor channel and Figure 7(b) for the outdoor channel.
The channel estimation MSE improvement that can be observed for higher-order modulations is due to the fact that the ratio between the powers in the pilot symbols and data symbols is kept constant in all simulations.The large increase of delay spread between both channels is the origin of the ∼3 dB MSE degradation when moving from the indoor channel to the outdoor channel plots.This acceptable degradation shows the ability of the estimator to deal with the increasing channel delay spread by always weighing the energy of channel taps versus noise variance.The channel estimation BER plots present a degradation of ∼1,2 dB that can be largely attributed to the 12.5% pilot overhead.
The joint CFO and channel estimation MSE is an effective measure of the degradation caused by both algorithms.In these plots, the estimated channel was compared against the true channel affected by the same CFO that distorted the received signal (according to (10)).The results plotted in Figures 6 and 7 show that the performance degradation of the joint process is marginal when compared with channel estimation only, substantiating the performance of the proposed algorithms.

Conclusions
We have investigated a CFO estimation algorithm and an associated channel estimation block for OFDM with transmitter diversity that explores the TD structure of transmitted symbols carrying pilots and data.The CFO algorithm relies solely on the data component present on the symbols to estimate the CFO, avoiding additional overhead like training symbols or null subcarriers.Simulation results show that the residual CFO has a minimal impact in the system's performance, confirming that the CFO estimates have minimal deviation from the true value.The definition and shape of the cost function determine a very low-complexity scheme.An intermediate output of the CFO algorithm provides an easy to get initial CIR estimate minimizing the overall complexity.By incorporating a TD LMMSE filter, the initial CIR estimate is significantly improved.Simulation results of the joint algorithms confirm a reduced degradation of the system's performance when compared with the ideal scenario.
matrix, and A CP = [IN C , L I NC ] T is the matrix that adds the CP, with I NC denoting the N C × N C identity matrix and I NC,L denoting the last L columns of I NC .The TD vectors d s and p s collect, respectively, the components of the data symbols and pilot symbols present in s s .The n S vectors s s are simultaneously transmitted to the receiver's antenna.Let w o = 2π f o Δt be the normalized angular CFO, where f o is the frequency offset due to the frequency mismatch of the oscillators of the transmitter and the receiver, and Δt is the sampling interval.

6 EURASIPFigure 2 :
Figure 2: Example of the constitution of vector g.

Figure 4 :
Figure 4: Estimated CFO standard deviation versus the number of samples in J(w).

Figure 5
shows the remaining CFO at the output of the CFO mitigate block when using the indoor channel model.The dashed lines represent the average remaining EURASIP Journal on Wireless Communications and Networking