Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
1 The Performance of PPM using Neural Network and Symbol Decoding for Diffused Indoor Optical Wireless Links S. Rajbhandari, Z. Ghassemlooy, and M. Angelova † Optical Communications Research Group † Intelligent Modelling Lab School of Computing, Engineering and Information Sciences Northumbria University, Newcastle upon Tyne UK. ICTON 2007, Rome, Italy Why Optical Wireless? Licence free spectrum Secure links Free from electromagnetic interference Low cost transmitter/receiver Small cell size Most importantly potential unlimited bandwidth, 10 millions times that of RF (which could solve the problem of bandwidth congestion in mobile system for an foreseeable future) ICTON 2007, Rome, Italy Challenges in Indoor Optical Wireless Strict link set-up for direct line-of-sight links Shadowing effects Lack of mobility Power limitation : due to eye and skin safety Intersymbol interference due to multipath propagations in diffused links Intense ambient light noise Large area photo-detectors - limits the data rate. ICTON 2007, Rome, Italy Digital Modulation Schemes On-off Keying (OOK) Pulse position modulation (PPM) Subcarrier modulation Digital pulse interval modulation (DPIM) Dual-header pulse interval modulation (DHPIM) ICTON 2007, Rome, Italy Diffuse Links Rx Tx 1.2 Received signal for non-LOS Links 1 Amplitude 0.8 Multipath path ISI, which in indoor depends on the room design and size. Delay spread Drms is used to approximate the dispersion in optical channel. 0.6 0.4 0.2 0 -0.2 -0.4 0 2 4 6 8 10 Normalized Time ICTON 2007, Rome, Italy Techniques to Mitigate the ISI Optimal solution - Maximum likelihood sequence detector Sub-optimal solution - Linear or decision feedback equalizer based on the finite impulse response (FIR) filter - The impulse response of filter c(f)= 1/h(f), where h(f) is the frequency response of channel. - Need to have pre-knowledge of channel - Difficult to realize if channel is non-linear. - Sometime the inverse filter may not exist …. What is the alternative? ICTON 2007, Rome, Italy ANN Based Equalization Redefine the problem of equalization as geometric classification problem in a complex plane. Use artificial neural network (ANN) for classification. Optical Receiver Artificial Neural Network Threshold Detector Pattern Classification ICTON 2007, Rome, Italy Advantages of ANN Equalization Parallel processing Universal approximates No assumptions are made on the channel model or modulation techniques Adaptive processing Channel non-linearity: not a problem ICTON 2007, Rome, Italy ANN: Basics Fundamental unit : a neuron Based on biological neuron Capability to learn Output is function of weight inputs and a bias as given by f(.) n y f (b xk .wk ) k 1 ICTON 2007, Rome, Italy ANN: Basics Neuron layers 1 Input layer Input Layer Hidden Hidden Layer 1 Layer 2 Output 1 or more hidden layer(s) 1 output layer Learning Method Supervised or unsupervised ICTON 2007, Rome, Italy Proposed System n(t) M 0100 M 0010 PPM Encoder PPM Decoder Xj Optical Transmitter Decision Device X(t) Z(t) h(t) Yj Neural Network ∑ Optical Receiver Zj Zj-1 Zj Matched Filter . Zj-n Ts = M/LRb . A feedforward back propagation neural network . ANN is trained using a training sequence at the operating SNR. Trained AAN is used for equalization ICTON 2007, Rome, Italy Impulse Response of Equalized Channel Impulse response of unequalized channel impulse response of equalized channel • Pulse are spread to adjust pulse . • Equalized response in a delta function which is equivalent to a impulse response of the ideal channel • ISI depends on pulse spread ICTON 2007, Rome, Italy Results and Discussion (1) Slot error rate performance of 8- PPM in diffuse channel with Drms of 5ns at 50 Mbps Adaptive linear equalizer with least mean square (LMS) algorithm is used. The performance of ANN equalizer is almost identical to the linear equalizer. ICTON 2007, Rome, Italy Results and Discussion (2) Slot error rate performance of 8- PPM in diffuse channel with Drms of 5ns at 100 Mbps Unequalized performance at higher data rate is unacceptable at all SNR range Linear and neural equalization give almost identical performance. ICTON 2007, Rome, Italy Further Work Using ANN as decoder and equalizer to reduce system complexity. Practical implementation. ANN with wavelet transform ICTON 2007, Rome, Italy Conclusions ANN is an effective equalizer for indoor optical wireless environment. No need for a prior knowledge of the channel for equalization. Performance of ANN is identical to or better than the traditional equalizer. The advantage of ANN over the traditional equalizer is its adaptability . ICTON 2007, Rome, Italy Acknowledgment My PhD students, Sujan, Rob, Maryam, Popoola Northunmbria University for the research funding ICTON 2007, Rome, Italy Thank you! ICTON 2007, Rome, Italy Questions and suggestions ICTON 2007, Rome, Italy