NEURAL NETWORK BASED ROBUST ADAPTIVE BEAMFORMING FOR SMART ANTENNA SYETEM

1. NEURAL NETWORK BASED ROBUST ADAPTIVE BEAMFORMING FOR SMART ANTENNA SYETEM Presented by : PARAMANAND SHARMA Roll No. : 207EE104 Under the Guidance of : Prof. Susmita Das

2. Contents • Smart Antenna • Beamforming • Problem Formulation • Simulation Results • Conclusion • Future Work • References

3. Smart Antenna Introduction • Antennas are themselves not smart. It is rather whole antenna system that makes a smart antenna . • It is an antenna array with digital signalprocessing capability to transmit and receive in an adaptive and spatiallysensitive manner . Types of Smart Antenna • Switched Beam antenna • Adaptive array antenna

4. Goals of Smart Antenna : • Features • Signal gain • Interference rejection • Power efficiency • Benefits • Better range/coverage • Increased capacity • Multipath rejection Drawbacks of Smart Antenna : • Complex • More expensive

5. Beamforming “A general signal processing technique used to control the directionality of the reception or transmission of a signal” • It creates the radiation pattern of the antenna array by adding the phases of the signals in the desired direction and by nulling the pattern in the unwanted direction • phases and amplitudes are adjusted to optimize the received signal

6. Adaptive Beamforming • A technique in which an array of antennas is exploited to achieve maximum reception in a specified direction by estimating the signal arrival from a desired direction while signals of the same frequency from other directions are rejected. • Resolution and interference rejection capability when the array steering vector is precisely known. • It optimize the array weights

7. Block diagram of Adaptive Beamforming Antenna-1 W1* Antenna-2 W2* ∑ Antenna-M WM* ∑ Adaptive Algorithm Output : Error :

8. Problem Formulation • Mathematical Model Consider a uniform linear array (ULA) with M omnidirectional sensors spaced by the distance d and D narrow-band incoherent plane waves, impinging from directions {θ0 θ1 …..θD-1} The observation vector is where s0(k)is signal waveform a is signal steering vector i(k) and n(k) are interference and noise components.

9. The output of a narrowband beamformer is where w is the complex vector of beamformer weight. The Signal to interference plus noise ratio has following form – Where are signal and interference-plus-noise covariance matrices of order .

10. Adaptive beamformer weight is computed in order to optimize certain criteria. Here we consider output SINR criterion which is written as where is the signal power. The following optimization problem to maximize the SINR min wHRi+nw subject to wHa=1 So , the optimal weight vector is and optimal SINR is

11. a) Sample Matrix Inversion (SMI) Algorithm • The sample matrix is a time average estimate of array correlation matrix using N-time samples. • The sample covariance matrix for SMI algorithm is where N is number of snapshots. SMI weights can be defined as where is a normalization constant. • Very sensitive to the mismatch between presumed and actual signal vector.

12. b) Loaded sample matrix inversion(LSMI) algorithm: This algorithm attempts to improve the robustness of the SMI technique against an arbitrary spatial signature mismatch by means of diagonal loading of the sample covariance matrix. It replace the covariance matrix by diagonally loaded covariance matrix where ξ is a diagonal loading factor . The LSMI weight vector is in the following form – • Improves the performance of SMI algorithm • Proper choice of is a serious problem

13. Neural Network based Robust Adaptive Beamforming Robust Adaptive Beamforming • It accounts for desired signal array response mismatch • Based on explicit modeling of uncertainty in desired signal array response Here actual signal steering vector be where is assumed steering vector . is norm of steering vector that bounded by some known constant as

14. The optimization problem can be formulated as subject to The weight vector for RAB be where λ is a Lagrange multiplier . • wRAB is a non-linear function of sample covariance matrix • wRAB is not suitable for real time implementation.

15. Radial Basis Function Neural Network Ω x1 w1 w2 Ω x2 y ∑ w3 x3 Ω wM Ω Hidden Layer Output Layer Input Layer The network output be Where is a Gaussian function. Where is center , is variance and is a set of weights.

16. is applied as input of RBF neural network and wRAB produced as output. RBFNN is designed to perform an input-output mapping • Here for training we use supervised selection of centers as the learning strategy. • The centers of the radial basis functions and all other free parameters of the network undergo a supervised learning process.

17. Simulation Results Array Factor Plots with variation of number of array elements with different spacing between element • Parameters The angle of arrival of desired user = 30 degree The angle of arrival of interferer = -60 degree Diagonal loading factor Element spacing λ/2, λ/4 and λ/8

18. 1. SMI algorithm Array Factor plot for d=0.5λ Array Factor plot for d=0.25λ

19. Array Factor plot for d=0.125λ of SMI algorithm

20. 2. LSMI algorithm Array Factor plot for d=0.5λ Array Factor plot for d=0.25λ

21. Array Factor plot for d=0.125λ of LSMI algorithm

22. 3.RAB algorithm with RBFNN Array Factor plot for d=0.5λ Array Factor plot for d=0.25λ

23. Array Factor plot for d=0.125λ of RAB with RBFNN

24. 2.Comparison of Array Beampatterns No. of omnidirectional sensors M=10 , spacing between element =0.5λ DOA of interfering sources 30° and 50° DOA of presumed signal spatial signature = 0° DOA of actual signal spatial signature = 2° SNR =10 dB (a) for no mismatch

25. (b) Comparison of Beampattern for 2° mismatch The SMI algorithm is very sensitive to mismatch.

26. 3. Comparison of Performance for known signal steering vector Output SINR versus N plot for fixed SNR = 10dB Output SINR versus SNR plot for fixed N = 500

27. 4. Comparison of Performance for Signal look direction mismatch(3° ) Output SINR versus N plot for fixed SNR = 10dB Output SINR versus SNR plot for fixed N = 500

28. Conclusion • RAB Algorithm based on RBFNN is useful in tracking the desired signal while nulling the interference sources. • RAB Algorithm based on RBFNN is much less sensitive to signal steering vector mismatches . • RAB Algorithm based on RBFNN achieves excellent performance , it achieves the values of SINR that are close to the optimal one in a wide range of the SNR and N but values of SMI and LSMI algorithm did not achieve to the optimal one.

29. Future Work • Neural network like Recurrent Neural Network (RNN) with reduced structural complexity can be incorporated for adaptive beamforming. • Adaptive Neuro-Fuzzy Inference System (ANFIS) may be considered better robustness to the beamforming algorithms

30. References • Frank B. Gross, “Smart Antenna For Wireless Communication” Mcgraw-hill, September 14, 2005. • A. H. El Zooghby, C. G. Christodoulou, and M. Georgiopoulos, “Performance of radia basis function networks for direction of arrival estimation with antenna arrays,” IEEE Trans. Antennas Propagat., vol.45, pp. 1611-1617, Nov. 1997 • V.V. Zaharov, F.S. Csco, O.A. Amin “ The tracking Algorithm and Processor for Smart Antenna Cellular Communication” IEEE –Russian Conference pp. 175-181 , Oct 2001. • Li J., Stoica P. and Wang Z., “On Robust Capon Beamforming and Diagonal Loading”, IEEE Trans. Signal Processing, 51, pp. 1702-1715, 2003. • H. Cox, R. M. Zeskind, and M. H. Owen, “Robust adaptive beamforming,” IEEE Trans. Acoust., Speech, Signal Processing, vol. 35, pp. 1365-1376, Oct. 1987.

31. Xin Song, Jinkuan Wang, and Yinghua Han, “Robust Capon Beamforming in the Presence of Mismatches”, Proceedings of ISCIT 2005,pp.135-138, July2005. • Xin Song Jinkuan Wang Xuefen Niu “Robust Adaptive Beamforming Algorithm Based on Neural Network” IEEE International Conference on Automation and Logistics , pp. 1844-1849 , Sep 2008. • Simon Haykin, “Neural Networks”, Second edition, Pearson education asia, Fourth Indian reprint, 2004.

32. THANK YOU