The impact of LED transfer function nonlinearity on high- speed optical wireless communications based on discrete- multitone modulation B. Inan, 1 S. C. J. Lee, 1 S. Randel, 2 I. Neokosmidis, 3 A. M. J. Koonen, 1 and J. W. Walewski 2 (1) COBRA Research Institute, Technical University of Eindhoven, P.O. Box 513, 5600 MB, Eindhoven, The Netherlands. (2) Siemens AG, Corporate Technology, Information & Communications, Otto-Hahn-Ring 6, 81739, Munich, Germany. (3) Department of Informatics and Telecommunications, University of Athens, Greece. E-mail: joachim.walewski@siemens.com Abstract: The nonlinear dependence of the optical power from white LEDs on the applied driving current and its impact on discrete-multitone modulation was investigated by use of numerical simulations for the case of optical wireless communications. ©2009 Optical Society of America OCIS codes: (060.2605) Free-space optical communication; (060.4080) modulation 1. Introduction There has been a steadily growing interest concerning optical wireless communications by use of light in recent years. This technology is frequently referred to as visible-light communications (VLC). Research on VLC was pioneered by the group of Nakagawa at the Keio University, Japan, and the maturity of this technology recently triggered global standardization efforts within IEEE 802.15 [1]. High-speed data transmission by use of white LEDs gained interest after the advent of high-power white LEDs [2]-[4], and 100 Mbit/s transmission over short distances was recently demonstrated [5]. The modulation bandwidth of white LEDs is limited to 20 MHz and less [4], so that such high data rates can only be achieved by equalization [6], multiplexing, multi-level modulation [4], or a combination of all three. Furthermore, LEDs are notoriously nonlinear, a fact that affects the performance of such systems. The objective of the work presented here was to assess the impact of LED transfer function nonlinearity on VLC for quadrature amplitude modulation (QAM) on discrete-multitone (DMT) modulation by use of numerical simulations. 2. Measurements Static LED transfer functions were measured with a straight-forward measurement consisting of a DC supply, a current meter, a single-chip LED, an amplified photo detector, PDA10A), and a voltage meter. The photo-detector output voltage was measured for a wide range of driving currents. Due to the lack of LED models describing the measured concave LED transfer function, we resorted to a parameter-free model in form of a polynomial. Very good fits were achieved for a polynomial order of five; however, we found that over 70% of the power contributable to the transfer function nonlinearity was contributed by the second-order term. For the sake of simplicity we therefore used a second-order polynomial model for our simulations. Due to relying on a static transfer function, our model is only valid for modulation frequencies well below the LED 3-dB bandwidth. 3. Simulations The numerical model consists of data source, transmitter, free-space channel, receiver, and data sink. Random data from the source is mapped to QAM symbols, and preamble block are added for the purpose of channel equalization. The data stream is parallelized into N subcarriers together with one subcarrier that represents the DC value and only contains zeros. This data is streamed into a DMT module consisting of an inverse discrete 2N+2 Fourier transform, which also introduces an oversampling of factor two. The output of the DMT module is serialized, the signal digitally clipped, and then piped through a DAC. For all simulations a clipping factor of 10 dB was chosen [7]. After the DAC, the analogue signal is two-times oversampled, which increases the clipping-noise bandwidth and reduces its interference with the signal. A DC current corresponding to the working point of the LED is added to the DAC output and the resulting current drives the LED. The latter can be represented by a linear or a nonlinear transfer model. The LED signal is transmitted through a free-space channel and detected by a photo detector composed of a photodiode and a trans-impedance amplifier. Both channel and photodetector have a flat frequency response. Optionally, white noise can be added to the received signal. A low-pass filter is used to filter out unwanted noise outside the clipping-noise bandwidth. Afterwards, the signal is digitized, piped through a DMT demodulator, and the a2143_1.pdf JThA52.pdf © 2009 OSA/OFC/NFOEC 2009 JThA52.pdf