Timing-Offset-Tolerant Universal- Filtered Multicarrier Passive Optical Network for Asynchronous Multiservices-Over-Fiber

Kang et al. VOL. 8, NO. 4/APRIL 2016/J. OPT. COMMUN. NETW. 229 Timing-Offset-Tolerant Universal- Filtered Multicarrier Passive Optical Network for Asynchronous Multiservices-Over-Fiber Soo-Min Kang, Chang-Hun Kim, Sang-Min Jung, and Sang-Kook Han Abstract We propose a universal filtered multicarrier (UFMC) passive optical network for asynchronous multiservices-over-fiber to overcome the limitations of a conventional discrete multitone (DMT). The channel performance in terms of error vector magnitude was compared by adding a timing offset to conventional DMT and UFMC signals. In addition, both signals were adaptively modulated at the most asynchronous timing-offset case. Our experimental results demonstrated improvement in UFMC performance, which was 32% and 32.5% higher in terms of bit rate and spectral efficiency, respectively, compared with conventional DMT in the most asynchronous case. Index Terms Asynchronous PON system; DMT; Multiservices-over-fiber; UFMC. I. INTRODUCTION I n a passive optical network (PON), high data traffic demands have been required because of the heavy increase in multiservices such as fixed mobile convergence, the Internet of Things (IoT), and 5G [1 3]. To provide these multiservices, a PON system should have high spectral efficiency (SE) as well as stable performance in an asynchronous transmission environment [4]. To satisfy these demands, a discrete multitone (DMT), which is a real-valued type of orthogonal frequency division multiplexing (OFDM), is widely used in a cost-effective intensity modulation, direct detection (IM/DD) PON system [5 8]. Multicarrier systems such as DMT and OFDM are able to transmit by dividing an entire band into several subbands per frame. Subband is defined as a subset of a subcarrier. To support stable performance of multiservices in this system, two key conditions must be satisfied. First, orthogonality among the subcarriers of a DMT signal should be maintained. Second, perfect synchronization must to be kept to prevent intersymbol interference (ISI) and interchannel interference (ICI) in multiservice transmissions [7 10]. Manuscript received November 5, 2015; revised February 11, 2016; accepted February 24, 2016; published March 22, 2016 (Doc. ID 253365). The authors are with the Department of Electrical and Electronic Engineering, Yonsei University, 134 Shinchon-dong, Seodaemun-gu, Seoul 120-749, South Korea (e-mail: skhan@yonsei.ac.kr). http://dx.doi.org/10.1364/jocn.8.000229 However, when multiservices are provided in subbands, there are some timing differences between them because of the asynchronous transmission properties of each service. Thus, multiservices are generated at different times [11,12]. Consequently, these asynchronous transmissions can cause a loss of orthogonality and lead to ICI. To prevent these problems, conventional DMT (or OFDM) uses a guard band (GB) in the frequency domain or a cyclic prefix (CP) in the time domain [9,13]. However, too many GBs or CPs degrade system performances in terms of throughput and SE [14,15]. In this paper, a timing-offset-tolerant universal-filtered multicarrier (UFMC) PON was experimentally verified as a proof-of-concept for asynchronous multiservices-over-fiber. UFMC applies subband filtering using a specific filter function to suppress the sidelobes of each subband [16,17]. By adoption of subband filtering in UFMC, interference between signal groups could be mitigated because of the sidelobe suppression effect. In our study, we focused on the timing-offset effect in the asynchronous situation. Timing offset was realized by signal sample delay between the signal groups of different services. First, the channel state was estimated by the degree of timing offset. Second, adaptively modulated UFMC and conventional DMT systems were transmitted under the same timing-offset environment, and system throughputs such as the bit rate and the SE of each system were evaluated. System throughputs were improved by more than 30% compared with conventional DMT in the most asynchronous case. Consequently, UFMC PON was more robust for asynchronous multiservicesover-fiber than conventional DMT. In Section II, we discuss the basic concept of PON system for multiservices-overfiber, the block diagram of UFMC and DMT system, and the filter properties of UFMC system. In Section III, we describe the experimental setup. In Section IV we experimentally verify the performance comparison of UFMC compared to DMT in multiservices-overfiber. Finally, in Section V, we present our conclusions. II. THEORETICAL APPROACH In this section, the basic concept of a PON system for multiservices-over-fiber is described, followed by 1943-0620/16/040229-09 2016 Optical Society of America

230 J. OPT. COMMUN. NETW./VOL. 8, NO. 4/APRIL 2016 Kang et al. presentation and discussion of block diagrams of UFMC and conventional DMT for asynchronous multiservicesover-fiber in PON system. Finally, the filter properties of UFMC are analyzed theoretically. A. Concept of Multiservices-Over-Fiber System Figure 1 illustrates the basic concept of a PON system for multiservices-over-fiber. In Fig. 1(a), there is one optical line terminal (OLT) and N number of optical network units (ONUs) separated by a 1 N passive remote node. To support multiservices to one ONU (such as ONU 1), various service signals such as broadcast, wired, mobile wireless, and IoT services should be tightly allocated over a fixed bandwidth. Because the allocated bandwidth of each ONU is limited, a multicarrier system such as an OFDM (or DMT), which provides a high data-transmission rate, has to be used. In addition, these service signals are independently modulated and demodulated. Therefore, we focus on a multicarrier-system-based PON system for various and independent multiservices-over-fiber. For multiservices-over-fiber, there are two possible situations in the time domain. That is, both synchronous and asynchronous situations could occur, as shown in Fig. 1(b). At an OLT, differences in signal-generation time among each service signal can cause a timing offset. Therefore, perfect synchronization in both situations is needed to recover the original signals at an ONU. For example, i number of services having different signal-generation times are depicted in Fig. 1(b), and these service signals have different timing offsets designated Δτ i 1.In Fig. 1(b), (i) denotes service 1, which is a fixed standard in the time domain; (ii), (iii), and (iv) denote services 2, 3, and i, respectively. Synchronous situation means that the timing offset among different services is zero; this is the ideal situation. In Fig. 1(b), the timing offset in (i) and (ii) (services 1 and 2) is zero (Δτ 1 0); therefore, this is a synchronous situation. Consequently, two service signals can be perfectly synchronized in both time and frequency domains. When the region marked X in Fig. 1(b) is assumed to be a symbol duration, the rectangular window of both service signals can be maintained in this region. Thus, sinc-shaped service bands composed of orthogonal subcarriers are zero-crossed, and their zero-crossed intervals are equal. This maintains orthogonality between two inter-service bands, as shown in Fig. 1(c). In addition, the orthogonality of intra-service bands is maintained in each service band. Therefore, there is no interference between each service band. However, an asynchronous situation would be generated in practice because each service signal is generated at a different time. Therefore, different timing offsets exist between each service. To protect a service signal from a different timing offset, a conventional multicarrier system uses CP in the time domain, as mentioned in Section I. If the timing offset is smaller than the CP length [(i) and (iii) in Fig. 1(b)], perfect synchronization could be achieved and the orthogonality of inter- and intra-service bands can be maintained [equal to the case of (i) and (ii) in Fig. 1(b)]. However, if the timing offset is larger than the CP length [(i) and (iv) in Fig. 1(b)], Δτ i 1 could affect the frequency domain, as depicted in Fig. 1(c). Because the next frame signal in-flows in the current frame region, a perfect rectangular window in the X region is impossible. Fig. 1. Basic concept of PON system for multiservices-over-fiber [(a) basic concept scheme, and synchronous/asynchronous situations in both (b) time and (c) frequency domains].

Kang et al. VOL. 8, NO. 4/APRIL 2016/J. OPT. COMMUN. NETW. 231 Therefore, the sinc shape of different service bands is transformed, as they have different zero-crossing intervals compared with the other service bands, as shown in Fig. 1(c). Thus, the orthogonality of the inter-service band could be destroyed, although that of the intra-service band is maintained. Consequently, interference could be caused due to loss of orthogonality between the two service bands. Therefore, a more timing-offset-tolerant system for multiservices-over-fiber should be introduced. A proposed system, which mitigates inter-subband interference in the asynchronous situation, is introduced in the next subsection. B. Block Diagrams of UFMC and DMT The concept of UFMC compared with conventional DMT is depicted in Fig. 2. UFMC applies subband filtering using a finite-impulse-response (FIR) filter function [16,17]. In the upper part of Fig. 2, subband denotes a subset of subcarriers in the allocated total bandwidth. Thus, this scheme is based on dividing the signal band into several subbands in the frequency domain. Independent multiservices can be allocated to each subband, and the bandwidth of each subband can be adjusted flexibly by changing the subband composition. In UFMC, subband filtering is applied to protect service signals from a timing offset. This filtering can be done using convolution in the time domain or multiplication in the frequency domain between the FIR filter and the signals [17]. Through subband filtering, the sidelobe level of each subband can be suppressed. In this paper, this approach is more robust in the asynchronous situation, compared with conventional DMT for supporting multiservices-over-fiber. Detailed block diagrams are shown in Fig. 2, where the double-lined box is UFMC-only and the dotted-line box is DMT only. First, the allocated bandwidth composed of symbol-mapped signals are separated as i number of subbands. X i is the ith subband and the same bandwidth is assumed for all subbands to facilitate easy comparison between UFMC and DMT PON systems. In our work, UFMC subband filtering is done by multiplying the FIR filter function in the frequency domain. To achieve subband filtering in each subband, the center frequency of each filter is set at the middle of each subband. Next, both UFMC and DMT signals are loaded as orthogonal subcarriers in the IFFT block. Because our work is based on an IM/DD PON system, Hermitian symmetry was applied to the IFFT block to generate a real-valued signal. The parameters N, L, and CP are the total number of subcarriers, filter length, and CP length, respectively. The effective symbol length is 2 N L 1 in UFMC and 2 N CP in DMT according to Hermitian symmetry. Signals after the IFFT block are combined into the total allocated bandwidth. In order to protect signals from the asynchronous situation, conventional DMT adds CP. In contrast with DMT, UFMC does not use CP because of subband filtering [17]. By subband filtering, filter ramp-up and -down shape is added to each frame in the time domain. Filter ramp-up and -down shape means that each end of UFMC signal per frame seems to be sharp because of FIR filtering. Thus, relatively small energy is contained in these regions (start and end of each frame), and these shapes assist in providing protection from timing offsets. It will be shown experimentally in Section IV. After receiving both UFMC and DMT signals, simple receiver processes such as low-pass filtering (LPF), downsampling, and time synchronization are performed. In DMT, added CP is removed and one-tap equalization in the frequency domain is done. In UFMC, on the other hand, inverse subband filtering between FFT and one-tap equalization is done. Many alternate methods of inverse subband filtering are possible. Among them, a zero-forcing (ZF) equalizer was used in our work. This process is a basic transceiver of DMT and UFMC, and we apply these two types of multicarrier to IM/DD-based PON systems. C. Theoretical Analysis of UFMC Filter Properties Fig. 2. Block diagrams of both UFMC and conventional DMT systems (double-lined box is UFMC-only and dotted-line box is DMT only). UFMC uses subband filtering in each subband. To achieve this filtering, there are many alternative FIR filter functions in UFMC [16,17]. We present h io nlcos cos 1 πk β cos W k 1 k L cosh Lcosh 1 ; β 0 jkj L 1; where β cosh 1 L cosh 1 10 α : In this work, we used a Dolph Chebyshev filter function because this filter function maximized sidelobe (1)

232 J. OPT. COMMUN. NETW./VOL. 8, NO. 4/APRIL 2016 Kang et al. Fig. 3. Theoretical properties of Dolph Chebyshev filter used for UFMC. Inset: case of L; 20α as (a) (20,20), (b) (20,100), and (c) (60,100). Fig. 3, the filter properties depend on two main parameters, L or α, by Eq. (1). As α increases with fixed L [Fig. 3(a) Fig. 3(b)], the 3 db bandwidth of the main lobe increases, as described in Eq. (1), and more signal can be loaded. In addition, as L increases with fixed α [Fig. 3(b) Fig. 3(c)], the 3 db bandwidth of the main lobe decreases, whereas the degree of sidelobe suppression increases. In this situation, moreover, computational complexity could be increased by increasing the length L (more FIR taps would be needed). In practice, filter-parameter optimization is required per our experiment, e.g., optimization of the subband bandwidth (or the number of subcarriers per subband). When applying subband filtering, it is important to stabilize the 3 db bandwidth of the filtered subband and suppress the sidelobe sufficiently to prevent interference. The method we chose to optimize the filter parameters in our experiment will be discussed in the next section. suppression [18]. Equation (1) is the Dolph Chebyshev filter function for subband filtering in the frequency domain when k is equal to the frequency index. W k consists of L and α, which determine the filter properties. L is the filter length in the time domain and α sets the Chebyshev norm of the sidelobes to 20α decibels (db). In other words, the degree of sidelobe suppression is equal to 20αdB. Through Eq. (1), larger L can increase the number of FIR filter taps, whereas α does not influence that number. α adjusts the sidelobe suppression ratio only in Eq. (1). Figure 3 shows theoretical Dolph Chebyshev filter properties according to L and α [18]. The normalized 3 db bandwidth of the main lobe is shown as the y axis. As shown in III. EXPERIMENTAL SETUP We have experimentally proposed a timing-offsettolerant UFMC-PON system as a proof-of-concept of multiservices-over-fiber, which is composed of several independent service signal bands. Figure 4 depicts the experimental setup, its parameters, and the method for adding timing offset. To exclude the influences of chromatic dispersion, received-power budget, and optical-amplifier noise, the experiment was verified using an optical backto-back link, which enabled us to focus only on the degradation of timing offset according to the sample delay. For performance comparison, DMT and UFMC were analyzed under both synchronous and asynchronous conditions. At Fig. 4. (a) Experimental setup, (b) parameters, and (c) process of adding timing offset to create an asynchronous environment.

Kang et al. VOL. 8, NO. 4/APRIL 2016/J. OPT. COMMUN. NETW. 233 the transmitter, the sampling frequency was fixed to six giga samples per second (G sps). We oversampled the transmitted signal by two times (Fs 2) in Fig. 4(b), and four subbands of the same bandwidth were used. Thus, a 6G 1 Fs 1 2 0.375 GHz bandwidth per subband was used, along with a total signal bandwidth of 1.5 GHz combining four subbands. As reported in Section II.C, we found the optimal point, where a sufficiently wide 3 db bandwidth is created, which is high enough to suppress interference from the sidelobe level. When x is the 3 db bandwidth in the inset of Fig. 3, x can be approximated so that x 1 is similar to 0.375G 6G. Therefore, x would be about 0.0625 π rad sample. We found the optimized point as (49,100), where the theoretical 3 db bandwidth was about 0.074 π rad sample, to stabilize signal bandwidth per subband. Thus, an L value that retains a sufficient 3 db bandwidth and suppresses sidelobe interference effects and an α value, which enables to stabilize a maximum 3 db bandwidth without loss of signal, were selected for presentation in Fig. 4(b). The number of total subcarriers used in the experiment was 512 (128 subcarriers per subband). In Fig. 4(a), two groups of signals were generated per frame. One is the odd-subband signal group (first and third subbands), and the other is the even-subband signal group (second and fourth subbands). Thus, the effective symbol length was 2 Fs N L 1 in UFMC and 2 Fs N CP in DMT. To maintain the same effective symbol duration for both signal formats, a CP length of 48 and an L value of 49 were used, as shown in Fig. 4(c). To realize the asynchronous environment between the odd- and even-subband signal groups, the odd-subband signal group at the transmitter is delayed in the signal sample. Although using an RF delay line could be practical, a digital sample delay between subbands was added to realize accurate timing offsets. The delay used in the experiment was 0, 48, 200, 400,, 1048,, 1200, 1400,, 2096 signal samples, where 0 and 2096 are the start and end points of the symbol duration, respectively. The 1048 signal sample is half of one signal sample, which is the same as a middle point between the 0th and the 2096th offsets. Both signal-generation and timing-offset additions were processed by using offline processing and an arbitrary waveform generator (AWG) (Tektronix 70002A). At an input port of the AWG, two signal groups having different sample delays were electrically combined. The combined signal, which had a timing offset among each subband by a digital sample delay, was applied to an output port of the AWG. The asynchronous situation was realized by this process. For the experimental setup depicted in Fig. 4, a tunable light source (TLS) with a center wavelength at 1549.99 nm was used as an optical carrier. Both DMT and UFMC signals were modulated by a 10 GHz Mach Zehnder modulator (MZM) at the quadrature point for linear intensity modulation. A variable optical attenuator-power monitor (VOA-PM) was used to maintain the received optical power at the photodetector (PD) at 4.2 dbm. After receiving the optical signal at PD, O-E conversion is done, and the signal is amplified in the electrical domain. Finally, the signal is received by a digital phosphor oscilloscope (DPO, Tektronix 71604C) with a sample rate of 25 G sps. By analyzing the signal received from the DPO, the performance of DMT and UFMC under asynchronous conditions is compared and evaluated. IV. RESULTS AND DISCUSSION A. Input and Output Signal Comparison Figure 5 shows the input signal comparison between DMT and UFMC in the time and frequency domains. In Fig. 5(a), input signals in the time domain were captured by an AWG. In Fig. 5(b), RF spectra were measured around each odd-subband signal group (first and third subbands) and even-subband signal group (second and fourth subbands). Notice that one step of y axis in Fig. 5(b) signified 10 db in our results of RF spectra. In Fig. 5(a), as mentioned in Section II.B, DMT used CP, which was a copy of the end part of the symbol duration. This CP protected the signal from interference due to timing offset. On the other hand, because of FIR subband filtering in UFMC, Fig. 5. Input signal comparison in (a) time and (b) frequency domains.

234 J. OPT. COMMUN. NETW./VOL. 8, NO. 4/APRIL 2016 Kang et al. Fig. 6. Received signal RF spectrum in each service. each end of UFMC signal per frame seemed to be sharp in the labeled region of Fig. 5(a). Note that interference in the time domain could be softly suppressed without CP due to filter ramp-up and -down shape because it has relatively small loaded energy [17]. In Fig. 5(b), DMT shows higher sidelobe levels due to sinc shapes in the frequency domain. However, these levels could be suppressed by more than 10 db using UFMC. Using the filter-optimization process detailed in Sections II and III, we set L 49 and 20α 100 db. Figure 6 shows RF spectra of the received odd- and even-subband signal groups captured by a RF spectrum analyzer. Based on the fact that the received optical power was fixed at 4.2 dbm, the sidelobe level of UFMC was lowered by 5 db compared to DMT. B. Initial Channel State Measurement in DMT and UFMC Figure 7 shows the results of channel state measurement by variation of timing offset. A probe signal that was mapped with quadrature amplitude modulation (4-QAM) to each subcarrier was transmitted initially. The channel error-vector magnitude (EVM) was measured by changing the timing offset in the odd-subband signal group depicted in Fig. 4(c). The effective symbol duration of DMT and UFMC was set as 2096 for a fair comparison. In the case of DMT [Fig. 7(a)], if the timing offset was lower than the CP length (0 and 48 sample delay), interference due to the timing offset could be prevented by CP. In the 0 and 48 sample delays, perfect synchronization was possible, and the orthogonality in both intra- and intersubbands was maintained. Thus, low EVM was achieved, but as the timing offset increased over CP length, the EVM continuously increased up to 1048, which is half of the effective symbol duration. Owing to the sinc-shaped sidelobe of the subband spectra, timing offset led to inter-subband interference among different signal groups. Loss of orthogonality in the inter-subband occurred due to its sidelobe problem, whereas orthogonality in the intra-subband was maintained in the asynchronous environment. Consequently, these losses of orthogonality in Fig. 7. Channel EVM of (a) DMT and (b) UFMC by timing offset. each subband caused inter-subband interference at the boundaries of each subband. Thus, each subband boundary was severely degraded, and EVM increased in these regions. Furthermore, the middle regions of each subband were damaged by the loss of orthogonality in the adjacent subband. In the most asynchronous case (1048 timing offset), the EVM was maximized. Because the in-flow rate of the adjacent frame in symbol duration was maximized at this point, this timing offset could realize the most asynchronous situation. From 1200 to 2096, the EVM decreased gradually. At the point of 2096, whose timing offset is an integer times the symbol duration, there was no interference among different signals. Thus, the EVM was equal to that of the zero-offset case. On the other hand, UFMC was more robust in the asynchronous environment compared with DMT. In Fig. 7(b), the EVM also increased up to the 1048 case and decreased up to the 2096 case in DMT. However, UFMC exhibits better performance upon comparison to Fig. 7(a). Owing to subband filtering, the sidelobe level of each subband was lowered. Therefore, channel state was improved over all subbands in general. In addition, the middle regions of each subband were well protected from interference, unlike in DMT, due to lower sidelobe level. Therefore, interference between service groups was lowered in UFMC compared to DMT. C. System Performances after Adaptive Modulation Figure 8 shows bit-loading profiles after adaptive modulation [perfectly synchronized case (open squares) and the most asynchronous case (solid circles)].

Kang et al. VOL. 8, NO. 4/APRIL 2016/J. OPT. COMMUN. NETW. 235 Fig. 8. Bit-loading profiles of (a) DMT and (b) UFMC for adaptive modulation. To maximize channel capacity, we adaptively modulated and retransmitted the signal. In our experiment, adaptive modulation is based on feedback using an estimated channel state per subcarrier. Adaptive modulation was applied in the perfectly synchronous case and the most asynchronous case. Figure 8(a) shows bit-loading profiles of DMT for adaptive modulation. In the most asynchronous case, DMT only loaded a small number of bits in the boundaries of subbands owing to sidelobe influences. This boundary interference generated approximately 10% loss per subband. Therefore, these results could be regarded as a severe degradation of channel capacity. However, UFMC could load more bits than DMT on boundaries. In the most asynchronous case in Fig. 8(b), higher bits could not be allocated in each boundary, and boundary interference generated approximately 3% loss per subband. This loss was effectively reduced compared with the same conditions in DMT [Fig. 8(a)]. Signal protection from inter-subband interference was effectively achieved using sidelobe suppression among subbands. Thus, more bits were loaded in the entire band and in the boundaries of subbands, compared with the DMT case. Therefore, the robustness of UFMC in the asynchronous case was verified from channel state measurement and bit-loading profiles for adaptive modulation. Figure 9 shows the performance evaluation after adaptive modulation within the forward-error-correction (FEC) limit (BER < 2 10 3 ), and Table I summarizes the data plotted in Fig. 9, which calculated total bit rate and effective bit rate. As the timing offset was getting closer to that in the most asynchronous case, more subband boundaries severely deteriorated. Although this tendency was similar in both cases, the throughput performance of UFMC was superior to that of DMT. In terms of the total bit rate and effective bit rate, UFMC shows better performance of 32% higher than DMT owing to suppression of sidelobe interference. When target BER satisfies the FEC limit in both timingoffset cases mentioned, Fig. 9 shows the SE in each case. Even if DMT contains CP to prevent interference, usage of CP per frame was too short to endure various asynchronous situations. Therefore, DMT shows remarkable performance degradation by inter-subband interferences. On the other hand, UFMC can achieve total SE up to 4.42 bit/s/hz with sidelobe suppression using subband filtering. This SE was improved by up to 32.5% compared with DMT, and an effective SE was improved up to 32.7% compared with conventional systems. Therefore, adoption of UFMC into the PON system facilitated mitigation of inter-subband interference using subband filtering. Consequently, system performance improvements in terms of channel state, throughput, and SE were achieved Fig. 9. System performance comparison of throughput and spectral efficiency at BER under forward-error- correction (FEC) limit.

236 J. OPT. COMMUN. NETW./VOL. 8, NO. 4/APRIL 2016 Kang et al. TABLE I THROUGHPUT AND SPECTRAL EFFICIENCY PERFORMANCE COMPARISON Waveform Type DMT UFMC Offset case w/o w/ w/o w/ Total bit rate [Gbps] 7.42 5.01 7.88 6.63 Effective bit rate [Gbps] 6.52 4.40 6.93 5.83 Total SE [bit/s/hz] 4.94 3.34 5.25 4.42 Effective SE [bit/s/hz] 4.35 2.93 4.62 3.89 by adopting UFMC into the PON system for multiservicesover-fiber. However, the trade-off in UFMC should be considered based on specific circumstances. Sidelobe suppression of a conventional DMT system was achieved by simple FIR subband filtering, whereby relaxation of inter-subband interference could be realized in the asynchronous case. On the other hand, insertion of FIR and FIR inverse filters in each subband could increase computational complexity. We adopted FIR-filter multiplication for each subband in the frequency domain to make a convolution of the FIR filter and signal in the time domain. However, computational complexity of IFFT and FFT was dominant in both multicarrier systems. In the UFMC system, it amounted to no more than adding a constant multiplier composed of a FIR filter function in the frequency domain. Therefore, DSP technology, which is rapidly progressing, would sufficiently overcome these computational complexities. V. CONCLUSIONS A timing-offset-tolerant UFMC PON was experimentally verified in an asynchronous environment having two independent signal service groups. The proposed system used simple FIR subband filtering to suppress the sidelobe level of a conventional DMT system. An adaptively modulated UFMC-PON for supporting multiservices-overfiber was transmitted in an asynchronous environment. The asynchronous situation was realized using a signal sample delay between independent service signals. Although this situation caused severe inter-subband interference at subband boundaries in a conventional DMT system, this interference could be mitigated in a UFMC system. Performance improvements such as channel EVM, throughput, and SE have been verified experimentally with effective sidelobe suppression. Experimental results demonstrated that UFMC was robust in an asynchronous environment with throughput and SE improvements at the boundaries of subbands. The effective bit rate and SE of UFMC were improved by more than 30% compared with those of conventional DMT in the most asynchronous case. Therefore, UFMC could be a promising solution for multiservices transmission, e.g., wired, wireless, and IoT signals in asynchronously multiplexed PONs through simple subband filtering. ACKNOWLEDGMENT This work was supported by the ICT R&D programs of MSIP/IITP, Republic of Korea [R0101-15-0086, Research on 25 Gbps optical access network based on discrete multitoned enabling dynamic network resource management]. REFERENCES [1] M. A. Ali, G. Ellinas, H. Erkan, A. Hadjiantonis, and R. Dorsinville, On the vision of complete fixed-mobile convergence, J. Lightwave Technol., vol. 28, no. 11, pp. 2343 2357, 2010. [2] D. Nesset, NG-PON2 technology and standards, J. Lightwave Technol., vol. 33, no. 5, pp. 1136 1143, 2015. [3] T. Pfeiffer, Next generation mobile fronthaul architectures, in Optical Fiber Communication Conf. (OFC), 2015, paper M2J.7. [4] H. Feng, C. J. Chae, and A. Nirmalathas, Flexible, lowlatency peer-to-peer networking over long-reach WDM/TDM PON systems, in 16th Opto-Electronics and Communications Conf. (OECC), 2011, pp. 186 187. [5] T. A. Truong, M. Arzel, H. Lin, B. Jahan, and M. Jezequel, New low-complexity and robust time synchronization technique for optical IMDD OFDM transmissions, Opt. Express, vol. 22, no. 12, pp. 14322 14340, 2014. [6] X. Hu, P. Cao, L. Zhang, L. Jiang, and Y. Su, Energy-efficient optical network units for OFDM PON based on time-domain interleaved OFDM technique, Opt. Express, vol. 22, no. 11, pp. 13043 13049, 2014. [7] L. A. Neto, D. Erasme, N. Genay, J. Guillory, B. Charbonnier, P. Chanclou, T. A. Truong, and C. Aupetit-Berthelemot, Experimental investigation of dispersion-induced distortions in IMDD OFDM PON transmissions, in European Conf. on Optical Communication (ECOC), 2012, paper P6.08. [8] N. Cvijetic, OFDM for next generation optical access networks, J. Lightwave Technol., vol. 30, no. 4, pp. 384 398, 2012. [9] J. Armstrong, OFDM for optical communications, J. Lightwave Technol., vol. 27, no. 3, pp. 189 204, 2009. [10] J. Zhao, S. K. Ibrahim, D. Rafique, P. Gunning, and A. D. Ellis, Symbol synchronization exploiting the symmetric property in optical fast OFDM, IEEE Photonics Technol. Lett., vol. 23, no. 9, pp. 594 596, 2011. [11] X. Lin, L. Jiang, and J. G. Andrews, Performance analysis for asynchronous multicarrier wireless networks, IEEE Trans. Commun., vol. 63, no. 9, pp. 3377 3390, 2015. [12] J. P. Mack, J. M. Garcia, H. N. Poulsen, E. F. Burmeister, B. Stamenic, G. Kurczveil, J. E. Bowers, and D. J. Blumenthal, End-to-end asynchronous optical packet transmission, scheduling, and buffering, in Opt. Fiber Communications Conf. (OFC), 2009, paper OWA2. [13] G. Gao, X. Chen, and W. Shieh, Analytical expressions for nonlinear transmission performance of coherent optical OFDM systems with frequency guard band, J. Lightwave Technol., vol. 30, no. 15, pp. 2447 2454, 2012. [14] E. D. V. Pereira, R. D. O. Helder, R. B. Nunes, M. E. Segatto, and J. A. Silva, Impact of optical power in guard-band reduction of an optimized DDO-OFDM system, J. Lightwave Technol., vol. 33, no. 23, pp. 4717 4725, 2015. [15] A. J. Lowery, Reducing cyclic prefix overhead in optical OFDM systems, in European Conf. on Optical Communication (ECOC), 2009, paper 1.3.4. [16] G. Wunder, P. Jung, M. Kasparick, F. Schaich, Y. Chen, S. T. Brink, A. Lucent, N. Michailow, A. Festag, L. Mendes, D. Kténas, S. Pietrzyk, B. Eged, P. Vago, and F. Wiedmann,

Kang et al. VOL. 8, NO. 4/APRIL 2016/J. OPT. COMMUN. NETW. 237 5GNOW: Non-orthogonal, asynchronous waveforms for future mobile applications, IEEE Commun. Mag., vol. 52, no. 2, pp. 97 105, 2014. [17] Y. Chen, F. Schaich, and T. Wild, Multiple access and waveforms for 5G: IDMA and universal filtered multi-carrier, in Vehicular Technology Conf. (VTC), 2014, pp. 1 5. [18] F. J. Harris, On the use of windows for harmonic analysis with the discrete Fourier transform, Proc. IEEE, vol. 66, no. 66, pp. 51 83, 1978. research interests include optical access networks and coherent optical communications. Sang-Min Jung received a B.S. in electrical and electronic engineering from Yonsei University, Seoul, South Korea in 2011, and is currently working toward a Ph.D. degree in electrical and electronic engineering at Yonsei University, Seoul, South Korea. His research interests include optical OFDMA-PON and coherent optical access networks. Soo-Min Kang was born in Jinju, South Korea in 1990. She received a B.S. in electronic engineering from Sogang University, Seoul, South Korea in 2014, and is currently working toward a Ph.D. in electrical and electronic engineering at Yonsei University, Seoul, South Korea. Her research interests include optical access networks, the Internet of Things for optical networks, and software-defined optical networks. Chang-Hun Kim received a B.S. in electronic engineering from Kwangwoon University, Seoul, South Korea in 2012, and is currently working toward a Ph.D. in electrical and electronic engineering from Yonsei University, Seoul, South Korea. His Sang-Kook Han (M 95 SM 2015) became a member of IEEE in 1995, and a senior member in 2015. He received a B.S. degree in electronic engineering from Yonsei University, Seoul, South Korea in 1986, and M.S. and Ph.D. degrees in electrical engineering from the University of Florida, Gainesville, FL, USA, in 1994. From 1994 to 1996, he was involved in the development of optical devices for telecommunications at the System IC R&D Laboratory, Hyundai Electronics Ltd., South Korea. He is currently a professor in the Department of Electrical and Electronic Engineering, Yonsei University, Seoul, South Korea. His current research interests include optical devices/systems for communications, optical OFDM transmission, passive optical networks, software-defined optical networks, and LED-based visible-light communications.