JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 32, NO. 11, JUNE 1,

Transcription

1 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 32, NO. 11, JUNE 1, Demonstration of High-Speed 2 2 Non-Imaging MIMO Nyquist Single Carrier Visible Light Communication With Frequency Domain Equalization Yuanquan Wang and Nan Chi Abstract We experimentally demonstrate a high-speed 2 2 non-imaging multiple-input multiple-output nyquist single carrier visible light communication system. Two commercially available blue light emitting diodes with 3 db electrical bandwidth of 10 MHz and two avalanche photo diodes with 3 db electrical bandwidth of 100 MHz are used as transmitters and receivers, respectively. A frequency domain equalization method based on two pairs of time-multiplexed training symbols is proposed, which allows demultiplexing and post-equalization simultaneously processing in one step. Frequency domain averaging and time-domain averaging are also implemented and analyzed in this paper. In this demonstration, the overall data rate is 500 Mb/s. The measured bit error rates for two receivers are both below the 7% pre-forward-errorcorrection threshold of after 40 cm free-air transmission. Index Terms Light emitting diode (LED), multiple input multiple output (MIMO), single carrier frequency domain equalization (SC-FDE), visible light communication (VLC). I. INTRODUCTION RECENTLY, there has been constantly gaining interest in visible light communication (VLC) motivated by the dramatic development of Light emitting diode (LED) technologies and increasingly scarce spectrum resources [1] [6]. Widespread used, cost effective, high brightness, larger bandwidth compared with other typical radio frequency (RF) based devices make it the most promising candidate for simultaneous illumination and communications, especially in some specific areas like hospitals, aircrafts, underwater and high security requirement environment. However, the relatively low modulation bandwidth is the main technical challenge in VLC system. Many research efforts have been dedicated to overcome this limitation such as digital signal processing (DSP) [2] [4], high-order modulation [5] and equalizations [6]. But most of these researches are Manuscript received November 22, 2013; revised March 5, 2014 and April 22, 2014; accepted April 22, Date of publication April 24, 2014; date of current version May 21, Y. Wang is with the Department of Communication Science and engineering, China ( speedboy_yq@163.com). N. Chi is with the School of Information Science and Engineering, Fudan University China ( nanchi@fudan.edu.cn). Color versions of one or more of the figures in this paper are available online at Digital Object Identifier /JLT focused on single input single output (SISO) systems, and there have been few reports on parallel transmissions which can offer a linear capacity gain with the number of channels in an ideal crosstalk-free system. Multiple-input multiple-output (MIMO) has been widely used in radio communication and multimode fiber communication [7]. In VLC, the availability of a large number of LEDs in a single room makes MIMO a promising candidate for achieving high data rates. In [8], a comparison of non-imaging MIMO and imaging MIMO is made. The channel matrix of imaging MIMO is always full-rank and diagonal matrix. But it needs precise alignment to make each LED image onto a dedicated detector. As for non-imaging MIMO, the precise alignment is not needed, and the tolerance to the misalignment become larger, but at the symmetry position, the channel matrix becomes less well conditioned. In [9], the characterization of a diffusetransmission MIMO is described. These examples of parallel free-space transmission are all discussed via simulation. Reference [10] experimentally demonstrate a 4 9 MIMO-OFDM VLC system with the data rate of each LED at 250 Mb/s, but the MIMO processing algorithm complexity is very high due to the large number of receivers (RXs). IN this paper, the configurationof RXs is simplified but with the same data rate of each LED at 250 Mb/s, compared to the results achieved in [10]. A 2 2 non-imaging MIMO 4- ary quadrature amplitude modulation (4-QAM) VLC system based on nyquist single carrier [11], [12] is demonstrated. This N-SC-FDE scheme has the similarity of spectral efficiency performance to the multicarrier modulation scheme, i.e., OFDM, and with a reduced calculation complexity compared with traditional SC scheme based on time domain equalization. Moreover, Single carrier frequency domain equalization (SC-FDE) can outperform OFDM in terms of PAPR, which is critical to VLC system due to the high nonlinearity of LED. The demultiplexing and post-equalization are simultaneously realized in the same step at frequency domain based on time-multiplexed training symbols. Frequency domain averaging (FDA) and time domain averaging (TDA) are also implemented and analyzed in this paper. The measured bit error rates (BERs) for two RXs are both below the 7% pre-forward-error-correction (pre-fec) threshold of after 40 cm free-air transmission. This paper is organized as follows. In Section II, we describe the non-imaging system, together with theoretical model and demultiplexing algorithm. The experiment results and discussions IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See standards/publications/rights/index.html for more information.

2 2088 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 32, NO. 11, JUNE 1, 2014 Fig. 1. Indoor network using non-imaging MIMO VLC. are provided in Section III. Concluding remarks are presented in Section IV. II. PRINCIPLE OF NON-IMAGING MIMO A. Configuration of Non-Imaging MIMO Multi-services with a variety of different standards should be hosted in the converged indoor network. Fig. 1 illustrates the configuration of such a network based on VLC, and different terminals are hooked up through visible light. As the visible light ranging from 380 to 780 nm cannot be able to penetrate walls, they are totally confined to a single room without interfering from adjacent rooms. Each single room can be regarded as a picocell, which can provide high capacity per user. The interfacing of the indoor network with the access network, such as optical fiber network, coax cable network, twisted pair network and 3G, 4G wireless network, can take place via the residential gateway (RG) proposed in [13], which contains media converters and possibly additional intelligence for signal conversion, local data storage, etc. In order to maintain the typical illumination level, there would be of course many LEDs in a real room, which provide the natural setup for MIMO transmission. There are two types of MIMO system: imaging MIMO and non-imaging MIMO. The former one requires each LED array imaging onto a detector array, while the latter one only needs non-imaging concentrators before each of the RXs. In this paper, the latter type is discussed and experimentally demonstrated as proof of concept. As shown in Fig. 1, different services passing through RG can be modulated on different LEDs via LED driver module, and transmitted in the free-space environment and detected by different RXs. After signal recovery, they can be sent to different terminals. B. Experimental Setup Fig. 2 shows a block diagram of this proposed MIMO SC- FDE system. The random binary data is generated in MATLAB and would be first split into two parallel streams, one for each transmitter (TX) channel. In each channel, the bit stream is mapped into M-ary quadrature amplitude modulation (M-QAM) format and then the training sequences (TSs) are inserted into the signals. The TSs in this experiment will be detailed in next section. After adding cyclic prefix (CP), and up-sampling by a factor of 10, filtering by a rectangular filter with roll factor of 0 is employed. Assuming the obtained signals at this procedure be X 0 (f) at frequency domain, and X 0 (t) at time domain, the up-converted process can be expressed as: X(t) =X real (t) cos(2πf n t) X imag (t) sin(2πf n t) (1) where the X real (t) and X imag (t) are the real component and imaginary component, respectively. f n denotes the center frequency of subcarrier. X(t) represents the up-converted signal in time domain and its frequency forms can be written as X(f) =X real (f) π(δ(w + w n )+δ(w w n )) X imag (f) iπ(δ(w + w n ) δ(w w n )) = X 0 (f f n )+(X 0 ( (f + fn))). (2) From Eq. (2) it can be seen that double sideband signals can be obtained after up-conversion. The up-conversion used here cannot only provide flexible frequency allocation, but also offer RF for I/Q modulation. Compared to baseband system, it can avoid low-frequency noise for the signals are up-converted to RF frequency. After up-conversion, the SC-FDE waveform is then loaded into an arbitrary-waveform generator (Tektronix 7122C) with the maximum sample rate of 24 GS/s and the maximum bandwidth of 6 GHz. The output signals of AWG are combined with a low-pass filter (LPF), which is used to remove out-ofband radiation. Subsequently, amplified by electrical amplifier (EA) (Mini-circuits ZHL-6A+, 25-dB gain), combined withdirect current (DC)-bias via bias tee. Assuming the bias voltage be V d, the output signals from bias tee X b (t) can be expressed

3 WANG AND CHI: DEMONSTRATION OF HIGH-SPEED 2 2 NON-IMAGING MIMO NYQUIST SINGLE CARRIER 2089 Fig. 2. The architecture for the proposed 2 2 non-imaging MIMO system, Inset (a) detailed process of TX, inset (b) detailed process of offline processing. Fig. 3. Time-multiplexed training symbols for MIMO de-multiplexing and post-equalization (TS: training sequence, TX: transmitter). as X b (t) =X(t)+V d. (3) By adjusting the bias voltage, the bipolar signals can become positive signals, and then applied to blue LED chips acting the optical TX. Passing through free-space transmission, and lens (50-mm diameter, 18 mm focus length), the signals are detected by two avalanche photo diodes (APDs). Subsequently, the received signals from each of the 2 RXs are routed to a high speed oscilloscope (Lecroy) with the maximum sample rate of 40 GS/s and the maximum bandwidth of 16 GHz and are acquired for further DSP. After synchronization, down-converting to baseband and removing CP, the received data streams are processed in a MIMO de-multiplexer at frequency domain. The final streams are then passed through the QAM decoder to recover the original binary stream. It should be noted that in this demonstration, each of the RXs will capture signals from both TXs, only with the different coefficients. C. Theoretical Model and De-Multiplexing In our system, the free-space link can be regarded as MIMO model, which can be expressed as ( Y1 Y 2 ) ( H11 H = 12 H 21 H 22 ) ( ) ( ) X1 N1 + X 2 N 2 (4) where (Y 1 Y 2 ) T represent two received SC-FDE signals after free-space transmission, and (X 1 X 2 ) T is the two independent SC-FDE signals at the TX, while (N 1 N 2 ) T denote the system noise. The channel matrix elements H i,j (i =1, 2; j =1, 2) represent the gain from jth TX to ith RX. All the signals are processed at frequency domain. The de-multiplexing can be easily realized once the channel matrix H is known [14]. Unlike the method that an extra MI- MOtraining run used in [11], we adopt the time-multiplexed TS based frequency domain equalization as shown in Fig. 3. Two pairs of TSs are transmitted in the front of signal to obtain the matrix for channel estimation and they can be expressed as ( ) ( ) TS1 0 T 1 =,T 0 2 = (5) TS 2 of which TS 1 and TS 2 are TSs which are made up with binary phase shift keying signals inserted in front of two independent streams. Zero-forcing (ZF) and minimum-mean-square-error can be applied for MIMO processing. In this demonstration, ZF is adopted due to the low algorithm complexity and the system noise can be ignored. The obtained channel matrix can be expressed as ( ) ( ) H11 H H = 12 Y1,1 /TS = 1 Y 1,2 /TS 2 (6) H 21 H 22 Y 2,1 /TS 1 Y 2,2 /TS 2 where Y 1,1 and Y 2,1 represent the received TS of the first symbol of RX1 and RX2, while Y 1,2 and Y 2,2 represent the received TS of the second symbol of RX1 and RX2, respectively. After obtaining the channel matrix H, the transmitted signal would

4 2090 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 32, NO. 11, JUNE 1, 2014 Fig. 4. RXs. Experimental setup for the proposed non-imaging MIMO system, the picture in the dashed box is the schematic diagram of the arrangement of TXs and be recovered by using (7) and (8) X 1 =(H 22 Y 1 H 12 Y 2 )/(H 22 H 11 H 12 H 21 ) (7) X 2 =(H 11 Y 2 H 21 Y 1 )/(H 22 H 11 H 12 H 21 ). (8) By using this method, the de-multiplexing and post-equalization can be simultaneously realized, and no extra phase recovery process of 4-QAM SC-FDE signals is needed. D. Time and Frequency-Domain Averaging The channel matrix obtained in (6) is critical to the recovery of the received signals. In order to improve the accuracy of channel estimation in the presence of noise, the time-domain averaging and frequency-domain averaging are needed [15], [16]. In this system, channel transfer function is usually highly correlated for adjacent frequencies. After obtaining channel matrixh at frequency domain in (6), frequency-domain averaging process can be applied. The frequency response at w k can be smoothed by averaging the estimated for itself and its multiple adjacent frequencies in the ith TS pair. Typically, for w k,the averaging can be performed over w k and its m left neighbors and/or m right neighbors, or totally up to (2m + 1) adjacent frequencies. It should be noted that the window size has to be narrowed when w k is near two edges of H. The improved channel matrix at frequency w k after the FDA process can be expressed as H i (w k )= 1 2m +1 k+m n=k m H i (w n ). (9) As the indoor channel is a time slowly varying channel, the adjacent training symbols can be considered to experience the same channel effect, therefore, TDA can be performed by combining different pairs of TSs. A more accurate estimation of the channel matrix can be obtained through H(w k )= 1 N N H i (w k ) (10) i=1 where H i (w k ) is the channel response at w k estimated by using the ith pair of TSs calculated in (9). After the two averaging strategies both in time and frequency domain, the channel matrix is much more precise. III. EXPERIMENTAL RESULTS AND DISCUSSIONS Fig. 4 illustrates the experimental setup of the proposed SC- FDE VLC system. In this demonstration, two commercial available blue LED chips (Cree PLCC6, blue: 470 nm) are used as the TXs and two APD (Hamamatsu APD C5331, about 0.05 A/W sensitivity at 470 nm at the gain of 1) are used as the RXs. The dynamic range of the overall system is about 4 to 18dBm.The block FFT size is 128, and the CP length is 1/16 of symbol length in this experiment. The up-sampling factor is 10, and the sample rates of AWG and OSC are 1.25 and 2.5 GS/s, respectively. Thus, the occupied electrical bandwidth of each LED chip is 125 MHz ranging from to MHz. The center frequency is located at MHz. The distance between two TXs and two RXs are 5 and 10 cm, respectively, and the offset of the center positions of two sides is 2.5 cm, which will break the ill condition. The transmission distance is varied from 20 to 50 cm. The illuminance at 40 cm is about 3.5l which is far smaller than the indoor illuminance standard. Once multiple LEDs are implemented, the illuminance can become larger, thus to enhance the transmission distance. Fig. 5 shows the captured time signals of RX1 and RX2 in the case of only one TX work and both TXs work. The yellow colors (upper) represent the signals of RX1, while red colors (under) for RX2. From this figure, we can find that the crosstalk between these two TXs is very high, which requires MIMO processing in the DSP. Fig. 6 illustrates the measured electrical spectra of the two RXs in blue color LED chips. The attenuation of the highest frequency is about 30 db larger than the lowest frequency component in this system. And the signal-to-noise in the case of both TXs work is higher than only one TX works, which can be conclude by comparing Fig. 6(a), (b) and Fig. 6(c), (d). Fig. 7 shows the amplitudes of channel matrix coefficients estimated without and with the frequency-domain equalization process as a function of the frequency, under 40 cm free-space transmission of blue LED chip. The original elements of the

5 WANG AND CHI: DEMONSTRATION OF HIGH-SPEED 2 2 NON-IMAGING MIMO NYQUIST SINGLE CARRIER 2091 Fig. 5. Captured signals of two RXs in the case of (a) only TX1 transmitted, (b) only TX2 transmitted, (c) TX1 and TX2 transmitted at the same time. Fig. 6. Measured electrical spectra of (a) RX1, (b) RX2 in the case of two TXs work at the same time and (c) RX2 (only TX1 transmitted), (d) RX2 (only TX2 transmitted). Fig. 7. Channel estimation of frequency matrix with and without FDA (a) H11, (b) H12, (c) H21, (d) H22. estimated channel matrix are displayed in Fig. 7 as blue line. The estimated channel coefficient without the FDA exhibits high-frequency fluctuations due to the presence of the optical noise, and with FDA, the high-frequency fluctuations can be removed shown as red line. In this paper, the FDA algorithm is applied in order to make the estimated channel response much smoother. One of the samples of the channel matrix measured in this experiment at 40 cm is: ( ) ( ) H11 H H = =. (11) H 21 H We can find that the channel matrix is a full-rank matrix, which means the ill condition is broken owe to the offset alignment between TXs and RXs. And the eigen values of this matrix is 13 and 2. Fig. 8 shows the BER performance versus different averaging window size from 4 to 24. We can find the BER performance will be improved with the increase of averaging window size, but when it is larger than 18, the BER performance will degrade. So an optimal averaging window size should be set to 18. A larger window size will reduce the frequency resolution and a smaller one cannot remove the frequency fluctuation.

6 2092 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 32, NO. 11, JUNE 1, 2014 Fig. 8. BER versus different averaging window size. Fig. 10. Constellations of the RX1 and RX2 N-SC-FDE signals after 40- cm free-space transmission of blue LEDs at different procedures in the offline DSP, (a) RX1 before de-multiplexing and post-equalization, (b) RX2 before de-multiplexing and post-equalization, (c) RX1 after de-multiplexing and postequalization, (d) RX2 after de-multiplexing and post-equalization. Fig. 9. BER versus distances of two RXs using one TS and two TSs. After employing FDA, the TDA is discussed in this paper. Fig. 9 shows the BER performance versus distance with and without TDA. After using TDA with two pairs of TSs, the BER performance can be improved, but with an increase of overhead. Fig. 10 shows constellations of the RX1 and RX2 N-SC-FDE signals after 40-cm free-space transmission of blue LED chips at a data rate of 500 Mb/s in different procedures of the offline DSP. The transmitted SC-FDE signals in two different TXs both appear in RX1 and RX2, which causes the constellation to be distorted before de-multiplexing as shown in inset (a) and (b) of Fig. 10. The channel matrix obtained through the channel estimation with FDA and TDA is applied to realize the de-multiplexing, and after the de-multiplexing, the signals can be recovered. The constellations of recovered signals are depicted in inset (c) and (d) of Fig. 10. It should be noted that the estimation of channel matrix is imperfect due to the neglect of noise in ZF algorithm, which will lead to the slightly rotation of constellations. Next, we measured the BER performance versus different transmission distances after adopting time-domain averaging and frequency-domain averaging. The results are depicted in Fig. 11. The measured BER versus distances; inset (i)-(ii) shows the constellations of two different RXs. Fig. 11. We can find the performance of two RXs are almost the same, and the BER degrades with the increase of delivery distance. All BERs are below the 7% pre-fec limit of The constellations of two independent tributaries in this case are also inserted in Fig. 11. A comparison of the BER versus distances of a parallel MIMO transmission and SISO transmission is made. For the SISO case, the occupied bandwidth is also 125 MHz, but the modulation format is 16 QAM. Thus, the SISO transmission owns the same data rate of 500 Mb/s compared to MIMO. In

7 WANG AND CHI: DEMONSTRATION OF HIGH-SPEED 2 2 NON-IMAGING MIMO NYQUIST SINGLE CARRIER 2093 Fig. 12. channel. The measured BER versus distances of MIMO channel and SISO this comparison, the illuminance of SISO is half of MIMO. The measured results are depicted in Fig. 12. From this figure it can be seen that at the same transmission distance, the SISO case can support higher-order modulation format with slightly BER penalty. This is mainly caused by the crosstalk introduced by the multiple channel RXs. The constellation of the SISO transmission is also inserted in Fig. 12. IV. CONCLUSIONS In this paper, we have experimentally demonstrated a 2 2 non-imaging MIMO VLC system that is capable to deliver 500 Mb/s 4-QAM N-SC-FDE signals over 40-cm free-space transmission. The MIMO de-multiplexing and post-equalization are simultaneously employed in the same step at frequency domain. FDA and TDA are proposed to improve the performance. Moreover, the data rate can further improved by employing pre-fde and a larger bandwidth APD. In addition, the discrete optic and electric elements and large size of lens and detectors are not to be practical. Integrated detector arrays will be our future research topic. REFERENCES [1] G. Cossu, A. M. Khalid, P. Choudhury, R. Corsini, and E. Ciaramella, 3.4 Gbit/s visible optical wireless transmission based on RGB LED, Opt. Exp., vol. 20, no. 26, pp. B501 B506, [2] Y. Wang, N. Chi, Y. Wang, R. Li, X. Huang, C. Yang, and Z. Zhang, Highspeed quasi-balanced detection OFDM in visible light communication, Opt. Exp., vol. 21, no. 23, pp , Nov [3] R. Li, Y. Wang, C. Tang, Y. Wang, H. Shang, and N. Chi, Improving performance of 750-Mb/s visible light communication system using adaptive Nyquist windowing, Chin. Opt. Lett., vol. 11, no. 8, p , [4] Y. Wang, Y. Wang, N. Chi, J. Yu, and H. Shang, Demonstration of 575- Mb/s downlink and 225-Mb/s uplink bi-directional SCM-WDM visible light communication using RGB LED and phosphor-based LED, Opt. Exp., vol. 21, no. 1, pp , [5] F. M. Wu, C. T. Lin, C. C. Wei, C. W. Chen, Z. Y. Chen, and K. Huang, 3.22-Gb/s WDM visible light communication of a single RGB LED employing carrier-less amplitude and phase modulation, presented at the Optical Fiber Communication Conf., Anaheim, CA, USA, 2013, paper OTh1G.4. [6] Y. Wang, Y. Shao, H. Shang, X. Lu, Y. Wang, J. Yu, and N. Chi, 875-Mb/s Asynchronous Bi-directional 64QAM-OFDM SCM-WDM transmission over RGB-LED-based visible light communication system, presented at the Optical Fiber Communication Conf., Anaheim, CA, USA, 2013, paper OTh1G.3. [7] Y. Wang, N. Chi, R. Li, W. Fang, J. Zhang, L. Tao, and Y. Shao, Theoretical and simulation analysis of a novel multiple-input multiple-output scheme over multimode fiber links with dual restricted launch techniques, Opt. Eng., vol. 51, no. 6, pp , [8] L. Zeng, D. C. O Brien, H. L. Minh, G. E. Faulkner, K. Lee, D. Jung, Y. Oh, and E. T. Won, High data rate multiple input multiple output (MIMO) optical wireless communications using white led lighting, IEEE J. Sel. Areas Commun., vol. 27, no. 9, pp , Dec [9] S. Jivkova, B. A. Hristov, and M. Kavehrad, Power-efficient multi spot diffuse multiple-input-multiple-output approach to broad-band optical wireless communications, IEEE Trans. Veh. Technol., vol. 53, no. 3, pp , May [10] A. H. Azhar, T. Tran, and D. O Brien, A gigabit/s indoor wireless transmission using MIMO-OFDM visible-light communications, IEEE Photon. Technol. Lett., vol. 25, no. 2, pp , Jan. 15, [11] D. Falconer, S. L. Ariyavisitakul, A. Benyamin-Seeyar, and B. Eidson, Frequency domain equalization for single-carrier broadband wireless systems, IEEE Commun. Mag., vol. 40, no. 4, pp , Apr [12] J. Kim, K. Lee, and H. Park, Power efficient visible light communication systems under dimming constraint, in Proc. IEEE 23rd Int. Symp. Personal Indoor Mobile Radio Commun., Dec. 2012, pp [13] A. M. J. Koonen and M. G. Larrodé, Radio-over-MMF techniques part II: Microwave to millimeter-wave systems, J. Lightw. Technol., vol. 26, no. 15, pp , Aug [14] F. Li, Z. Cao, X. Li, Z. Dong, and L. Chen, Fiber-wireless transmission system of PDM-MIMO-OFDM at 100 GHz frequency, J. Lightw. Technol., vol. 31, no. 14, pp , Jul [15] Q. Yang, N. Kaneda, X. Liu, and W. Shieh, Demonstration of frequencydomain averaging based channel estimation for 40 Gb/s CO-OFDM with high PMD, IEEE Photon. Technol. Lett.,vol.21,no.20,pp , Oct [16] X. Liu and F. Buchali, Intra-symbol frequency-domain averaging based channel estimation for coherent optical OFDM, Opt. Exp.,vol.16,no.26, pp , Dec Authors biographies not available at the time of publication.