A Hybrid Optimization-Based Technique for Channel Estimation in OFDM System: A Parametric Approach

The performance of OFDM system depends on the how efficient channel estimation is, which directly impactsits bit error rate (BER). In this research work, the novel technique of pseudo-pilot aided OFDM system is usedits channel estimation with least square and Recursive Least Squares approach. In proposed technique, for channel estimation Pseudo-pilot technique is used in OFDM system over AWGN fading channels. A hybridization technique has been implemented to alleviate the error rate in the signals. The hybrid optimization approach is implemented on the estimated output for the optimum solution. The combination of the two optimization techniques is particle swarm optimization (PSO) and Moth Flame Optimization (MFO) method that is used for optimization of the performance rate. To get more optimum solutions (reducing BER) OFDM system is trained with neural network. The performance of the proposed techniques and the weighted scheme are compared and verified using computerized simulation carried out using Matrix Laboratory software. The hybrid approach is able to achieve low BER of the network. The BER of PSO is 1.004 × 10−7 whereas for hybrid optimization (PSO + MFO) BER is 6.275 × 10−8 at 18 dB SNR. After training of the whole system is done with BPNN which will further reduce the BER while increasing SNR. By using BPNN, BER will further reduce to 1.243 × 10−8 at 18 dB SNR. It means hybrid optimization is done to optimize the performance of the channel and reduce the BERs, which will help to increase the channel estimation process as well as channel capacity and further reduces the losses while transmission from one end to the other end.


Introduction
OFDM is a multicarrier signal modulation scheme that is more advanced in wavelength selective switching (WSs) of IEEE standards, digital communication, audio-broadcasting, and fourth-generation mobile transmissions [1][2][3][4][5]. The interpretation of the pilot assisted wireless schemes often contains various glitches, these are pilot design, time, frequency domain, pilot placement, and definite channel estimation algorithms. Going through literature review, it is observed that much work has been done in the area of OFDM systems. However, more work is to be done in the area of channel estimation & equalization to enhance the system performance. Figure 1 gives the system architecture for the proposed system.
In proposed method, the work is done for the improvement of the signal for the given channel and to reduce the BER which further improves the channel estimation and its capability with hybrid optimization [6][7][8][9]. It also reduces error losses during the transmission from one end to the other end. In OFDM framework for pseudo pilot by utilizing hybridization of two advancement strategies for example hybridization utilizing molecule swarm improvement and moth fire advancement with a wellness capacity to get best outcome. Additionally, a neural network (NN) for OFDM framework with pseudo-pilot which is prepared with half breed streamlining to appraise channel boundaries is likewise proposed.
Optimization is related to the method of searching the best probable output for the specific problem.All optimization problems are not solved by single optimization method. Many optimization methods have been developed for problems in recent years. Complex engineering problems are solved by modern optimization methods like as GA, PSO algorithm, neural networks, artificial immune systems, ant colony optimization and fuzzy optimization.

Related Work
In the last two decades, the research on OFDM is progressing at a fast pace. In recent times, OFDM was designated as a high-performance LAN transmission technique. However, in this thesis, the proposed work is confine to channel estimation using a hybrid optimization technique with moth flame optimization & particle swarm optimization trained with a neural network. Gorbunov et al. [13] addressed the LTE fading channels as having an appropriate BER of QPSK effective multi-carrier signal. Simulation outcome shows that SD equal modulator may manage a similar BER presentation as OFDM bandwidth savings of 25%. Xie et al. [14] considered a new virtual pilot assist channel approximation approach for multiple inputs and one output carrier frequency domain (CFD) equalizations. Wang et al. [18] introduced research on various utilizations of the speculative strategy and significant philosophies of the WDM conspire through reasonable OFDM. Recreation result shows that model Q of WDM channel was at 12.8 Tb/s and finished a value of 15.5 dB correspondence for the 1000 km-length in a typical specific mode fiber with no visual scattering and non-linear instances. Li et al. [16] proposed research on OFDM frame infrastructure for deduction of the system complexity and recompense carrier frequency offset (CFO). A simulation outcome shows that the planned method described better MSE and BER performance as a comparable method. The experimental outcome shows that this method has the best performance rate. It was more realistic as compared to CS-based channel approximation. Munshi et al. [10] discovered the efficiency of compressive for channel approximation in single input and single output and multiple input and multiple output OFDM scheme. They established a hybrid LS-CS approximation method. The other significant observation was that the channel was sparser as compared to the hybrid LS-CS approximation method produces greater results.
The Table 1 will give the different Techniques used in OFDM for channel estimation. Optimization is related to the method of searching the best probable output or solution for the specific problem. Over the last few decades, the complexity issue rises and the requirement for novel methods becomes more obvious than before. In obtaining the synchronization and estimation of the channel is done if the insertion of pilots will be there with the payload of symbols. With each pilot the pilot burst is increased independently in the time domain which will cause the overhead problem in the system. So, it becomes difficult for us to select the data for the pilot, means planning for the pilot data and planning for the channel estimator with its low unpredictability & high channel capacity. The performance of the system will be restricted by the pilot overhead and it became essential to overwhelm the load in that system which is made by the pilots. For increasing in the system performance, the overhead problem due to pilots the novel pseudo-pilot is used instead of pilots. Then the channel impulse response is optimized using hybrid optimization which is then trained with a neural network for better system performance and high transmission rate.

Proposed Work
In the proposed approach, a hybrid model combining the functionalities of LS and RLS to moderate overhead and increment the OFDM execution with low intricacy, postpone time and high transmission rate. Further, this exploration attempted to streamline the crossover assessed OFDM framework for pseudo pilot by utilizing hybridization of two advancement strategies for example hybridization utilizing molecule swarm improvement and moth fire advancement with a wellness capacity to get best outcome. Additionally, a neural system for OFDM framework with pseudo-pilot which is prepared with half breed streamlining to appraise channel boundaries is likewise proposed. At long last, the presentation of the NN identifies with LS and RLS calculation utilizing BER and MMSE versus SNR situation and calculation was finished utilizing MATLAB. The performance will be evaluated on the basis of SNR, MSE, BEER, PDR, Mean Square Error and End to End delay. Figure 2 will explain the research methodology for the proposed system.
Assume the signal used for transmission to be denoted as x and the signal for the reception to be denoted as y. The communicated signals are occupied from one of the multiamplitude signal groups. Equation 1 is used for the calculation of CIR value in the proposed system model.
From Eq. 1, T s is sampling interval, amplitude is represented as r and delay is represented as r .
Then after transmitting acknowledged signal is: where DFT matrix is represented as y and x contains the diagonal elements. The value of Q, m, f and x is given below

Channel Estimation Using LS
Least square CE is determined by following equation where Form the Eqs. 7 and 8, it is derived that, z LS and w LS are the column vectors.

Channel Estimation Using RLS
The channel approximation for the given RLS is given as follows: where (N) = 1 + (N−1) , (0) = 0 and is the forgetting factor,

Channel Estimation Through LS + RLS
The squared error estimation will be done on the basis of LS and RLS estimation: LS + RLS channel estimation is calculated using equation and P is the reference length and L is the guard period with the bipolar elementsm i € {− 1, + 1}. In Eq. 11, () −1 is inverse matrices and () z denoted the Hermitian matrices.

Equation 12
is more simple solution for the autocorrelation function (ACF). It will give the simple relationship between the training sequence and received signal.
Step 2: Evaluation function In Eq. 13, reference channel model is representing as Z LS [12].

Identify the Best Channel Using PSO + Moth
The Hybrid optimization is used for best channelindentification. In this section, equations proposed in chapter 3 are being. The process is explained briefly: Step 1: Initialize the algorithm parameters such as month number n, number of iterations T, dimension d and so on.
Step 2: Initialize the population according to the Eq. 4.5.
Step 3: Calculate the fitness function of the moths and artificial flame in the population and categories them according to their fitness function.
Step 4: Update the flame according to the best solution so far.
Step 5: Update the number of flames according to this equation Step 6: Update the dynamic weight according to the where is the average fitness value of the first optimization is processed and b(j) is the fitness of the first month j and it represents the current number of iterations.
Step 7: Calculate the best moth position Step 8: Optimal solution is obtained if the termination condition is satisfied, else return to step 3.

Identify the Best Channel Trained with BPNN
A three layer BPNN was designed for channel estimation and compensation as shown below. BPNN structural design has three layers. In Fig. 3, all the three layers are Input, Hidden and output layer are shown. Input layer used for the processing of the data. In the hidden layer the output of each unit is represented as: where I i is the input data for ith unit and w ij is the weight between the input and hidden layer. Then the activation function for the hidden layer is as: where A h j is the output of the hidden layer, f is the activation function. The output is expressed as: Basic structural design of BPNN [35,36] where w jk is the weight between hidden and output layer. Now the output layer is given as: In this O o k is output layer of kth unit. The BPNN algorithm is based on the gradient descent. To reduce the error and the error function is the main motive which is defined as: where R k is the required output. For reducing error weight is updated with given equation In this is learning rate, which tells that how much weight is changed in each step. Therefore, weight update through hidden to output layer is given as: And the weight update from input to hidden layer is given as: When the learning is done the received signal is sent to the neural network for the compensation.

Parameters of Consideration
Before discussing the result of the proposed technique, various parameters are needed to be taken under consideration while simulating the proposed technique on computer software. Table 2 will give the parameters to be used for the proposed techniques. Filter length 32 7 Step size 0.08 8 Carrier frequency 2407 MHz 9 Sampling frequency 37 kHz 10 FFT size 32

Experimental Results and Comparison
In this section, the analyses and observation of the proposed hybrid technique is discussed here. This analysis will be carried out with respect to various parameters of evaluation. The hybrid optimization (PSO + MFO) is applied on the output of the hybrid (LS + RLS) channel estimator, then the optimized output is trained in terms of various iterations in the back propagation phase to reduce the number of error rates with a reduction in the loss functions to achieve less training error. Figure 4 gives the effects of OFDM signal in the AWGN channel.
In this simulation design, 32-QAM technique is used. Convolution encoding is used in this proposed technique to encode the random bits. During transmission of signal through free space, AWGN noise is added into the original signal. After receiving, the effect of noise on signal is shown in Fig. 5.  Figure 5 shows the BER comparison with the optimization. The analyses of the PSO and Hybrid (PSO and firefly optimization) optimization is performed. Which will reduce the error rate and its losses. With hybrid optimization BER is reduce with respect to SNR.
The Table 3 gives the BER values for the different signal to noise ratio (E b /No). According to the table values which are derived from the above graph shows that the hybrid optimization works better than a swarm optimization.  In the last, a neural network has been added in the system for better training of the estimated output of the hybrid estimator. Figure 6 gives the Neural Network Training tool in MATLAB.
Below figure shows the BER performance in terms of LS, LS + RLS, PSO, Hybrid optimization (PSO + MFO) and Hybrid optimization + neural. It can be noticed from the below diagram that the proposed technique (hybrid + neural) is achieving bit error rate performance and reducing the low error rates to increase the efficiency of the system to achieve highly efficient communication from transmission to the reception. Figure 7 will gives the BER comparison for different techniques used in Proposed system. Table 4 gives the different values for the different techniques used in the proposed work.

Conclusions
Optimization design in engineering tends to be a challenging approach because of the complexity, high non-linearity of the problem and programming in engineering. Optimization is the technique that is implemented selectively by comparing the different solutions unless the optimum solution is found. Optimization plays an essential role in the computer aided technology. In this research proposal the hybrid optimization on the estimated output has been used. Hybrid approach is able to achieve the low BER and further it can increase the throughput. For further reduction in the BER the system is trained with BPNN. Eventually the comparison is made based on all the techniques that the proposed approach using neural is trained in terms of the back propagation network to achieve low error rates for low latency and high throughput in the OFDM systems which is one of the optimal approaches for better communication systems.

Author Contributions
All authors contributed to the study conception and design.All authors read and approved the final manuscript.
Funding The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.

Data Availability
The datasets generated during and/or analysed during the current study are not publicly available as the signals are randomly generated but are available from the corresponding author on reasonable request.

Conflict of interest
The authors have no relevant financial or non-financial interests to disclose.