BLIND IDENTIFICATION OF CHANNEL TAP NUMBERS IN WIRELESS COMMUNICATION

Abstract

The present disclosure relates to a method for blind identification of channel tap numbers in wireless communication by using deep neural networks (DNN). In the proposed method, it is possible to train a DNN using only the transmitted and received signals of a wireless system in order to obtain the number of channel taps. We propose a robust and efficient sparse representation technique for the identification of wireless channels. We estimate the number of channel taps which is considered as one of the sparse features of the hannel The blind estimation performed in the proposed system, enhances the spectral efficiency of the used wireless communication system since the employed DNN does not require to transmit extra signals for identifying the channel taps. In our identification method, physical insights are not available or used.

Claims

1. A method for blind identification of the number of channel taps in wireless communication comprising the steps of: a. importing channel samples from the real-world channel datasets, b. modifying an existing DNN and analysing its performance in terms of training, validation loss and accuracy c. selecting the basic structure comprising; i. using the feed-forward design that has fully-connected layers with residual connections and batch normalization ii. including a final ‘softmax’ layer in order to provide normalized probabilities of input signals belonging to classes and iii. incorporating Dropout layers to the deep neural network d. training the employed DNN with an optimizer e. sending the transmitted signals through different wideband frequency selective channels with the generated CIR of length corresponding to the number of multipath components f. using the transmitted and received signals with their corresponding number of channel taps as training dataset.

2. A method for blind identification of the number of channel taps in wireless communication according to claim 1, wherein said real-world channel datasets are generated using a simulator.

3. A method for blind identification of the number of channel taps in wireless communication according to claim 1, further comprising the step of applying regularization.

Description

DETAILED DESCRIPTION

[0011] The invention proposes a method for blind identification of the number of channel taps in wireless communication comprising the steps of: [0012] a. importing channel samples from the real-world channel datasets, [0013] b. modifying an existing DNN and analysing its performance in terms of training, validation loss and accuracy [0014] c. selecting the basic structure, stated in Xin B. et al. (2016), comprising; [0015] i. using the feed-forward design that has fully-connected layers with residual connections and batch normalization [0016] ii. including a final ‘softmax’ layer in order to provide normalized probabilities of input signals belonging to classes and [0017] iii. incorporating Dropout layers to the deep neural network stated in Xin B. et al. (2016), [0018] d. training the employed DNN with an optimizer [0019] e. sending the transmitted signals through different wideband frequency selective channels with the generated CIR of length corresponding to the number of multipath components [0020] f. using the transmitted and received signals with their corresponding number of channel taps as training dataset.

[0021] In a preferred embodiment, said real-world channel datasets are generated using a simulator. For example, MATLAB-based NYUSIM simulator (2016) could be used as also suggested by Samimi, M. K. & Rappaport T. S. (2016) and Sun S. et al. (2016). As Sun S. et al. (2016) suggest, the preference found in NYUSIM is that it can support broad set of frequencies, beamwidths, bandwidths, wireless channel scenarios, etc. The simulation parameters used in our implementation are similar to that of the default simulation setup described in NYUSIM simulator (2016) for spatial channel model using NYUSIM.

[0022] In the example implementation, we select the basic structure proposed by Xin B. et al. (2016) due to its memory and computation efficiency and the ability to employ it on various heterogeneous systems.

[0023] Last but not least, in order to avoid overfitting, regularization is applied in a preferred embodiment.

REFERENCES

[0024] [1] New York University, NYUSIM, 2016. [Online]. Available: http://wireless.engineering.nyu.edu/5gmillimeter-wave-channel-modeling-software/.

[0025] [2] M. K. Samimi and T. S. Rappaport, “3-D Millimeter-Wave Statistical Channel Model for 5G Wireless System Design,” in IEEE Transactions on Microwave Theory and Techniques, vol. 64, no. 7, pp. 2207-2225, July 2016.

[0026] [3] S. Sun et al., “Investigation of Prediction Accuracy, Sensitivity, and Parameter Stability of Large-Scale Propagation Path Loss Models for 5G Wireless Communications,” in IEEE Transactions on Vehicular Technology, vol. 65, no. 5, pp. 2843-2860, May 2016.

[0027] [4] Bo Xin, Yizhou Wang, Wen Gao and David Wipf. Maximal sparsity with deep networks arXiv preprint arXiv: 1605.01636, 2016

[0028] [5] Z. Xiao, H. Wen, A. Markham, N. Trigoni, P. Blunsom, and J. Frolik, “Non-line-of-sight identification and mitigation using received signal strength,” IEEE Transactions on Wireless Communications, vol. 14, no. 3, pp. 1689-1702, 2014.

[0029] [6] A. Li, C. Dong, S. Tang, F. Wu, C. Tian, B. Tao, and H. Wang, “Demodulation-free protocol identification in heterogeneous wireless networks,” Computer Communications, vol. 55, pp. 102-111, 2015.

[0030] [7] Dengao Li, Gang Wu, Jumin Zhao, Wenhui Niu, and Qi Liu, “Wireless Channel Identification Algorithm Based on Feature Extraction and BP Neural Network,” Journal of Information Processing Systems, vol. 13, no. 1, pp. 141-151, 2017.