Open Access
Add to BibSonomy
Abstract
A novel low complexity and energy-efficient scheme by controlling the toggle-rate of ONU with time-domain amplitude identification is proposed for a heavy load downlink in an intensity-modulation and direct-detection orthogonal frequency division multiplexing passive optical network (IM-DD OFDM-PON). In a conventional OFDM-PON downlink, all ONUs have to perform demodulation for all the OFDM frames in a broadcast way no matter whether the frames are targeted to or not, which causes a huge energy waste. However, in our scheme, the optical network unit (ONU) logical link identifications (LLIDs) are inserted into each downlink OFDM frame in time-domain at the optical line terminal (OLT) side. At the ONU side, the LLID is obtained with a low complexity and high precision amplitude identification method. The ONU sets the toggle-rate of demodulation module to zero when the frames are not targeted to, which avoids unnecessary digital signal processing (DSP) energy consumption. Compared with the sleep-mode methods consisting of clock recovery and synchronization, toggle-rate shows its advantage in fast changing, which is more suitable for the heavy load scenarios. Moreover, for the first time to our knowledge, the characteristics of the proposed scheme are investigated in a real-time IM-DD OFDM system, which performs well at the received optical power as low as −21dBm. The experimental results show that 25.1% energy consumption can be saved in the receiver compared to the conventional configurations.
© 2017 Optical Society of America
1. Introduction
With the rapid development of HD videos, cloud storage and other new emerging network service, the demand of network bandwidth in optical access network has grown exponentially [1]. The next-generation passive optical network (NG-PON) was proposed to meet the bandwidth requirement, which is divided into two parts: first stage NG-PON1 and second stage NG-PON2 [2]. NG-PON2 has been selected as the long-term next generation optical access network and provides no less than 40 Gb/s. OFDM-PON has demonstrated its unique advantages, such as high channel capacity, high spectrum efficiency, low construction cost, flexible resource allocation, scalable architecture [3]. Therefore, OFDM-PON essentially extends the trend of “software-defined” (DSP-based) optical communications to next-generation optical access [4], which can provide a high capacity, high-speed and multi-service network platform supporting residential, business, mobile backhaul, and special purpose applications [5]. However, OFDM-PON consumes more energy than other PONs due to its high speed advanced digital signal processing (DSPs) and analogue-to-digital converters /digital-to-analogue converters (ADCs/DACs) for modulation/demodulation with high sampling rate [6]. As a result, OFDM-PON system consumes ~70% more energy than time division multiplexing passive optical network (TDM-PON) and ~160% more energy than time and wavelength division multiplexed passive optical network (TWDM-PON) [6]. Moreover, optical network units (ONUs) in an OFDM-PON system account for ~80% energy consumption [6], which is much higher than TDM-PON and TWDM-PON [7]. Given the fact that the ONU energy consumption account for a large portion due to the demodulation DSP complexity, special attention is mainly focused on the ONU side in this paper. It is essential to develop an energy-efficient OFDM-PON from a long-term point of view.
So far, various schemes have been proposed to reduce energy consumption of PONs. Recently, IEEE and ITU-T has standardized for TDM-PONs with sleep mode that the optical line terminal (OLT) should be in charge of invoking ONUs into sleep mode in the absence of frames to reduce energy consumption [8]. According to the most of researches in cyclic mode or doze mode of TDM-PON, various algorithms were proposed to obtain a balance between energy saving performance and quality of service (QoS) [9–11]. Meanwhile, some dynamic wavelength and bandwidth allocation (DWBA) algorithms and new frameworks for TWDM-PON have been proposed for energy-efficiency [12–14]. For OFDM-PON system, many schemes focus on ADC/DAC and DSPs for energy-efficiency, In general, the energy-efficient methods can be divided into techniques in physical layer and dynamic bandwidth allocation (DBA) scheduling policy with sleep modes in network layer. In physical layer, some schemes were proposed to achieve high energy-efficiency. Researchers of the University of London proposed that reducing output size of the last stage in FFT/IFFT module of OFDM system can efficiently reduce energy consummation without degradation of transmission quality [15]. The NTT Research Lab proposed an energy-efficient OFDM-PON based on dynamic signal to noise ratio (SNR) management and gave numerical analysis of its efficiency [16]. They further presented a new IM-DD OFDM-PON which combined dynamic SNR management and adaptive modulation technology. The NEC laboratory proposed that energy-efficient OFDM can be achieved by reducing the precision of digital signal processing and the sampling rate at the ONU side [17]. Xiaofeng Hu proposed an energy-efficient WDM-OFDM-PON which employs shared OFDM modulation modules at the OLT side [18]. An energy-efficient OFDM-PON based on time-domain interleaved OFDM technique was also proposed, which decreases the quality of service (QoS) and is complicated to be implemented as well [19]. Besides, a selective sampling receiving method was proposed [20], which caused delay and consumed extra energy due to the specially designed receiver for PN sequences calculation and the complexity of synchronization. In network layer, some DBA algorithms were proposed such as adaptive sleep-mode control and dynamic bandwidth allocation to save energy. Such DBA algorithms with sleep mode were complex and caused unacceptable QoS when the load is heavy [21,22]. However, these sleep modes are complicated and need the clock recovering and the network timing synchronizing and other delay problems, and they can’t deal with the heavy load OFDM-PON very well. Recently, we proposed and demonstrated an energy-efficient strategy for OFDM-PON based on ONU identification regarding the broadcast downlink characteristics in an offline approach [23]. Then we implemented the scheme in our real-time OFDM system [24]. To reduce the complexity of digital signal processing (DSP) complexity, we also proposed an improved stage-dependent minimum bit resolution map [25]. For the first time to our knowledge, there is still few energy-efficient experimental demonstration focused on investigating the characteristics of the downlink OFDM signal in a real-time manner
In this paper, an energy-efficient scheme for real-time OFDM-PONs by controlling the toggle-rate of demodulation module in ONUs with time-domain identification with amplitude decision [26] to reduce energy consumption of DSPs is proposed. In a conventional OFDM-PON downlink, all ONUs have to perform demodulation for all OFDM frames in a broadcast way no matter whether the frames are targeted to or not. In our scheme, LLIDs are inserted in downlink OFDM frames at the OLT side, at the ONU side, the ADC samples all frames first, then the symbol synchronization module demodulates LLID sequence and obtains LLIDs with amplitude decision. When the LLIDs are targeted to the ONUs, the ONUs perform normally, otherwise they will set the toggle-rate of demodulation modules to zero to decrease DSP energy consumption. Compared to sleep mode, our scheme can change the toggle-rate in a fast way in FPGA logic without complicated clock recovery and synchronization. Therefore it will not cause any delay in QoS even in a heavy load link. Experimental results show that 25.1% power in ONUs can be saved compared to the conventional configurations.
2. Operation principle of ONU LLID identification using amplitude decision in time-domain
In principle, the dynamic power consumption of DSP chips is proportional to signal toggle-rate as reported in [27–29]. Besides, it’s noted that dynamic power consumption takes 58% of the total power consumption at the 90nm technology node [30], which indicates that dynamic power consumption is a major component of the overall power consumption in DSP chips. So we focus on controlling the toggle-rate to reduce the dynamic power consumption.
In order to set the toggle-rate of demodulation module to zero when receiving unrelated downlink data stream, it is necessary for the ONU to identify whether the received downlink data stream contains information destined to itself to avoid the demodulation processing which is unavoidable in conventional OFDM receiver. In conventional OFDM frame structure, data sequence is packed behind the long training symbol. However, in our proposed frame structure, a LLID sequence is inserted between the header and data signal, which helps ONU to identify whether the received OFDM signal contains data destined to it. When the signal is not targeted to the receiver, the toggle-rate will be set to the zero. In this way, more energy will be saved in OFDM DSP demodulation process.
Moreover, considering the stable physical channel characteristic in IM-DD OFDM system, a new OFDM frame structure is proposed. One preamble is inserted before conventional OFDM frame, so that OFDM symbol synchronization and ONU LLID can be identified in time-domain, as depicted in Fig. 1(a). Figure 1(b) shows the preamble amplitude quantification, which contains zero sequence, sync-header and biphasic code LLID. The continuous zeros are used for coarse synchronization. The sync-header contains two sharp peaks of timing metric, which can accurately define the location of timing synchronization by using amplitude decision method in our previous work [23]. The LLID length is 10 bits encoded by 8B/10B encoding from 8 bits, the 8 bits can represent 256 ONUs at most. Every ONU has its own unique LLID, only one LLID of “1010101010” is shown in this paper in Fig. 5. The biphasic code LLID is encoded by 8B/10B, which performs well for “DC-balanced” in high speed channels. The way how the ONU identify the LLIDs and control the toggle-rate when frames come is just as Fig. 1(c) shows.
Fig. 1 (a) A new OFDM frame structure. (b) The preamble amplitude quantification. (c) Toggle-rate based energy-efficient scheme.
3. Experimental verification for LLID identification in real-time IM-DD OFDM-PON system
3.1 Experimental setup
To experimentally demonstrate the toggle-rate based energy-efficient scheme and verify the time-domain LLID recognition accuracy, we set up a real-time IM-DD OFDM system, and its corresponding DSP blocks in OFDM transceiver are illustrated in Fig. 2. The OFDM frames in time-domain and the corresponding spectrum are also illustrated in Fig. 2. The generated OFDM signal has a periodic frame structure, as illustrated in Fig. 3. The 80 zero-valued samples header performs symbol synchronization, and the two 1200-valued sync-header is inserted to find the accurate synchronization position. 30 samples’ LLID is for ONU identification, followed are two FFT sized-sample training sequences (TSs) and 500 data-carrying OFDM symbols. PAD means zero-valued padding, The 16 zero-valued padding and a cyclic prefix of 32 samples points for training sequences are adopted to ease the hardware design complexity of 32-parallel real time OFDM receiver and to reduce the interference of the time domain LLID sequence. Channel estimation is achieved by the two FFT sized-sample training sequences (TSs). The 16 samples CP is for mitigating signal degradation caused by ICI. The real-time experimental demonstration of prototype OFDM system is shown in Fig. 4. The key transceiver and system parameters are summarized in Table 1. In this paper, two Xilinx FPGA boards namely ML605 equipped with a Virtex-6 XC6VLX240T separately are used to implement the OFDM system.
Table 1. OFDM-PON System Parameters
At the transmitter, an incoming PRBS sequence with a word length of 215-1 is adaptively encoded using signal modulation formats varying within 2/4/8/16/32/64/128QAM. The 1st subcarrier is deactivated because of the impairments caused by low-pass filters and AC-coupling, whilst all other information-bearing subcarriers are arranged to satisfy the Hermitian symmetry with respect to their conjugate counterparts to generate real-valued OFDM symbols after the floating-point IFFT of 64points. After the IFFT function, a cyclic prefix of 16 sample points is inserted to mitigate the inter-symbol interference (ISI) caused by chromatic dispersion. After having applied 12-bit quantization and 12 dB digital clipping, 80 zero-valued samples header, two 1200-valued sync-header and 30 LLID time-domain sequences with the absolute amplitude value set to 1200 are inserted in the head, followed are two FFT sized-sample training sequences (TSs) and 500 data-carrying OFDM symbols. To compensate the non-ideal accuracy of the symbol synchronization due to the optical/electrical noise, and to ensure the accuracy of demodulation of received identification symbols, each identification symbol must comprise a number of sample points, and it is obvious that the more points there are, the more accurate the result would be. However, more sample points mean lower signal efficiency, so that we have to make a tradeoff between these two factors. Here we use three samples to represent one bit, which shows a good result. Compared to the scheme in [23], we only use 30 sample points to represent 256 ONUs, which occupies only 0.07% in OFDM frames to get higher frame utilization. The most important is that the proposed method has very low complexity and is easy to be implemented in real-time system. The LLID sequence for 256 ONUs consists of 8 bits is coded into 10 bits by 8B/10B encoder at the OLT side. As shown in Fig. 3. The 16 zero-valued padding and a cyclic prefix of 32 sample points for training sequences are adopted to ease the hardware design complexity of 32-parallel real time OFDM receiver and to reduce the interference of the time-domain LLID sequence. As a result, The OFDM frame duration is 10.072μs ((80 + 2 + 30 + 16 + 32 + 64 × 2 + 80 × 500) × 0.25 ns = 10.072μs). After that, the generated OFDM signal is transferred into the internal RAM of a Xilinx ML605 FPGA board with a Virtex-6 XC6VLX240T FPGA via the UDP protocol. The ML605 FPGA board operating at 125 MHz feeds the digital OFDM signal into a 4GS/s@12-bit DAC. The baseband OFDM signals generated by the DAC are then adjusted to 2Vpp via a variable attenuator and a 13dB amplifier. A narrow line-width distributed feedback laser (DFB-LD) is finally used to convert the electrical OFDM signal into the optical domain before injecting into a 25km SSMF.
At the receiver, a variable optical attenuator (VOA) is employed to adjust the received optical power. After converting the optical signal to the electronic domain by a 2.7 GHz PIN detector, the electrical OFDM signal is amplified by a variable electrical amplifier (VEA) to ensure that the signal occupies the entire dynamic range of a 4GS/s@10-bit ADC to convert the received analogue electrical OFDM signal into digital domain. The high-speed sampled digital signals are first de-multiplexed to 32 parallel channels, and then the receiver performs symbol synchronization, which is achieved by the leading-zeros as show in Fig. 3. Each receiver performs symbol synchronization to obtain the LLID sequence first, and then it uses amplitude decision in time-domain to achieve the address where these frames are targeted. The LLID is decided by the second point of three points, which is decoded by 10B/8B decoder after that. Once the LLID is recognized, the receiver will compare it with its own LLID, when the LLIDs are not theirs, they will drop the frames and set the toggle-rate of the real-time demodulator DSP modules to zero. When the symbol synchronization and LLID identification are done, two training sequences and a cyclic prefix of 16 samples at the beginning of OFDM symbol are removed. The next procedure is the 64-point FFT to realize OFDM demodulation. After that is the channel estimation achieved by TS. Then the recovered 50000 symbols are transmitted to the bit error ratio (BER) calculator, in which the received bits are compared with the transmitted PRBS pattern, and error bits are counted in every OFDM symbol.
3.2 Experimental verification of the accuracy of the LLID recognition in time-domain
The LLID of ‘1010101010’ being part of OFDM frames captured by Chipscope is shown in Fig. 5, which corresponds to the Fig. 3 as descripted in experimental setup section. Note that the waveform is in a reversed order in Y-line due to our two amplifiers. To evaluate the accuracy of our amplitude decision method, over 25,000,000 OFDM frames with different received optical power using 4QAM, 8QAM, 16QAM are tested. The received error frames are counted by identifying the second point in three points of LLIDs using amplitude decision. The results achieved by Xilinx integrated logic analyzer ChipScope and then transported via Joint Test Action Group (JTAG) cable to host PC with ChipScope analyzer for observation or further analysis. Obviously, when the threshold is too low, the noise causes a fail identification.
The different amplitudes to get the error frame numbers is shown in Table 2. As we can see, the threshold of 80 performs very well for our system. When the received optical power get lower, the OFDM frames are affected by the noise seriously in channel, which causes synchronization error, so the recognition rate will decrease, actually, this depends on the threshold.
Table 2. Received error frames in different amplitude of threshold (ROP = −18dBm, 16QAM)
We can get the frame errors at the −9dBm and the −21dBm received power in Fig. 6. Meanwhile, the successful recognition of OFDM frames is shown in Fig. 7. As we can see, when the received optical power is between −19dBm and −11dBm, the recognition rate is about 100%. Even at the −21dBm received power, the recognition rate is over 99.7%, which shows a good identification accuracy using 4QAM, 8QAM, 16QAM in our OFDM system.
Fig. 6 (a) Received frames counters and error frame counters in Chipscope (ROP = −9dBm, 4QAM) (b) Received frames counters and error frame counters in Chipscope (ROP = −21dBm, 4QAM)
To compare the proposed scheme with the conventional OFDM-PON system, the real-time measured BER curves for optical 25-km SSMF transmission case are depicted in Fig. 8(a). Corresponding constellations of the 4QAM and 16QAM signals in the energy-efficient OFDM-PON are provided at a received power as low as −21dBm. As we can see, the energy-efficient OFDM-PON performs a little worse than the conventional one. To enhance the performance of the real-time IM-DD OFDM system, we use adaptive modulation, where the modulation formats are 2-QAM, 4-QAM, 8-QAM, 16-QAM, 32-QAM and 64-QAM, 128-QAM. Figure 8(b) shows the corresponding raw signal bit rate (RBR) and net bit rate (NBR) in both circumstances respectively. As depicted in the figure, even if the received optical power ranges from −21dBm to −11dBm, we can see that, the proposed energy-efficient scheme does not degrade the system performance of the receivers.
Fig. 8 (a) BER versus received optical power with modulation format of 4QAM and 16QAM for conventional and proposed OFDM-PON after 25km SSMF transmission. (b) Signal bit rate in the conventional and proposed energy-efficient OFDM-PON downlink system using adaptive modulation respectively.
4. Energy-efficiency with controlling toggle-rate based on LLID identification
4.1 Energy-efficiency experimental verification with controlling toggle-rate
Then we investigate the relationship between dynamic power consumption and signal toggle-rate, we first measure the energy consumption of receiver modules when the toggle-rate is zero. The DSP modules of real-time OFDM receiver include Synchronization, FFT, Channel Estimation & Equalization, Demodulation Mapper and BER Analysis as shown in Fig. 2. We set the toggle-rate of these modules to zero one by one in a reversed order, and the experimental power consumption is shown in Fig. 9. When the receiver performs normally, we can measure and calculate the total power consumption. When the toggle-rate of the baseband in receiver is set to zero, we can get the static power. The dynamic power can be obtained by subtracting the static power from the total power. In fact, as we can see, modules of FFT, channel estimate account for a very great proportion of the energy consumption due to their DSP consumption. So, in a conclusion, the dynamic power can be saved by setting the toggle-rate to zero.
4.2 Numerical analysis for energy-efficiency combining controlling toggle-rate with LLID amplitude identification in time-domain
In Fig. 10(a), it’s a point-to-point circumstance, to prove our scheme that can be applied in heavy load, at the transmitter, the period of frame shown in Fig. 3 is 10.072μs, we select 8 periods’ frames (80.576μs) as one cycle and set the 1/8, 2/8 … of these 8 periods to zero, this method is equivalent to control the offered load, the result is shown as the red line, actually we can achieve a satisfactory energy-efficiency, in which the offered load is at the level of microsecond from a heavy load. According to Fig. 10(a), the black line shows the power savings of baseband in theory, as we can see, these two results perform almost the same. And when the toggle-rate is zero, the static power consumption can be achieved, meanwhile, we can get the dynamic power consumption when the toggle-rate increases. In Fig. 10(b), 8 ONUs receive the unicast frames from the OLT, in conventional OFDM-PON, all ONUs will process the complete downlink OFDM frames no matter what the offered load is. However, in our proposed scheme, the ONUs will set the toggle-rate of demodulation module to zero when the frames are not target to by identifying the LLIDs, which can save much power. The red line in Fig. 10(b) shows the 8 ONUs at a normal energy consumption state. However in our proposed scheme, the energy consumption of an ONU changes with different offered load, the other seven ONUs are in a low energy consumption state, so we can calculate the total energy consumption of 8 ONUs as the blue line in Fig. 10(b) shows. According to our real-time experimental result, when the offered load is 100%, up to 25.1% energy can be saved compared to the conventional scheme.
Fig. 10 (a) Normalized energy consumption of one receiver in the baseband where toggle-rate to zero and time percentage where toggle-rate to zero in transmitter respectively. (b) Power consumption in the conventional and proposed energy-efficient OFDM-PON system for 8 ONUs.
The power consumption of chip consists of static power and dynamic power in one receiver. The static power can be achieved when the toggle-rate is zero. When the toggle-rate of receiver performs normally, we can get the total power. The total energy. The energy-efficiencycan be achieved by Eq. (1). The ONU numbers are set to, and the is the total power consumption of the receivers. So the total PON’s energy consumption is shown in Eq. (2), is offered load. One receiver and the PON’s energy-efficiency are expressed by Eq. (3) and Eq. (4) separately. Also, we define energy-efficiency compared to conventional scheme as expressed in Eq. (5). The ratio of dynamic power consumption to static power consumption is set as.
The ratio of energy consumption of energy-efficient scheme compared to conventional scheme with different ONU numbers in different occupation of dynamic power is shown in Fig. 11. When the ONU number is less than 8, the energy consumption reduction is obvious. When the ONU number gets higher, even to 80 or more, due to the limit of dynamic power in a chip decided by the chip technology, the energy-efficiency is closer to the occupation of dynamic power consumption to total chip power.
So, under a heavy load OFDM-PON circumstance, we can set the toggle-rate of ONU which receives unrelated frames to zero to save energy without introducing time delays under heavy traffic. Moreover, when the OFDM-PON system has over 8ONUs, our scheme can get a reasonable energy-efficiency.
5. Conclusions
In this paper, a novel toggle-rate based energy-efficient scheme for heavy load IM-DD OFDM-PON by using time-domain ONU identification of 8B/10B code is proposed and experimentally demonstrated in a real-time manner. The ONUs do not have to process all received frames but only the frames contain their own LLIDs in downlink. The ONU sets the toggle-rate of demodulation module to zero when the frames are not targeted to. Experimental verifications on the real-time IM-DD OFDM-PON system shows that up to 25.1% power consumption of FPGA chip in ONUs can be saved in baseband modules compared to the conventional configurations. Besides, we can save power without decreasing the QoS under heavy traffic load. Of course, under the light load scenario, doze or sleep modes can save more power. In the future work, OFDM systems with sleep-mode approach in low peak time can be combined with our scheme in heavy load time to further improve the energy-efficiency.
Funding
Natural Science Foundation of China (Project No. 61420106011, 61601277, 61601279); China Postdoctoral Science Foundation funded project (2016M601564); Shanghai Science and Technology Development Funds (Project No. 14dz1104800, 15511105400, 15530500600, 16511104100, 16YF1403900).
References and links
1. D. Nesset, “NG-PON2 Technology and Standards,” J. Lightwave Technol. 33(5), 1136–1143 (2015). [CrossRef]
2. Y. Luo, X. Zhou, F. Effenberger, X. Yan, G. Peng, Y. Qian, and Y. Ma, “Time- and Wavelength-Division Multiplexed Passive Optical Network (TWDM-PON) for Next-Generation PON Stage 2 (NG-PON2),” J. Lightwave Technol. 31(4), 587–593 (2013). [CrossRef]
3. Q. W. Zhang, E. Hugues-Salas, Y. Ling, H. B. Zhang, R. P. Giddings, J. J. Zhang, M. Wang, and J. M. Tang, “Record-high and robust 17.125 Gb/s gross-rate over 25 km SSMF transmissions of real-time dual-band optical OFDM signals directly modulated by 1 GHz RSOAs,” Opt. Express 22(6), 6339–6348 (2014). [CrossRef] [PubMed]
4. N. Cvijetic, “OFDM for next-generation optical access networks,” J. Lightwave Technol. 30(4), 384–398 (2012). [CrossRef]
5. N. Cvijetic, M. Cvijetic, M. F. Huang, E. Ip, Y.-K. Huang, and T. Wang, “Terabit Optical Access Networks Based on WDM-OFDMA-PON,” J. Lightwave Technol. 30(4), 493–503 (2012). [CrossRef]
6. B. Skubic, E. D. Betou, T. Ayhan, and S. Dahlfort, “Energy-efficient next-generation optical access networks,” IEEE Commun. Mag. 50(1), 122–127 (2012). [CrossRef]
7. P. Vetter, D. Suvakovic, H. Chow, P. Anthapadmanabhan, K. Kanonkis, K. Lee, F. Saliou, X. Yin, and B. Lannoo, “Energy-efficiency improvements for optical access,” IEEE Commun. Mag. 52(4), 136–144 (2014). [CrossRef]
8. A. F. Y. Mohammed, S. H. S. Newaz, M. R. Uddin, G. M. Lee, and J. K. Choi, “Early wake-up decision algorithm for ONUs in TDM-PONs with sleep mode,” J. Opt. Commun. Netw. 8(5), 308–319 (2016). [CrossRef]
9. S. H. S. Newaz, M. S. Jang, A. F. Y. Mohammed, G. M. Lee, and J. K. Choi, “Building an Energy-efficient Uplink and Downlink Delay Aware TDM-PON System,” Opt. Fiber Technol. 29, 34–52 (2016). [CrossRef]
10. F. Y. M. Alaelddin, S. H. S. Newaz, J. Lee, M. R. Uddin, G. M. Lee, and J. K. Choi, “Performance analysis of TCP traffic and its influence on ONU’s energy saving in energy efficient TDM-PON,” Opt. Fiber Technol. 26, 190–200 (2015). [CrossRef]
11. S. W. Wong, S. H. Yen, P. Afshar, S. Yamashita, and L. G. Kazovsky, “Demonstration of Energy Conserving TDM-PON with Sleep Mode ONU using Fast Clock Recovery Circuit”, inProceedings of Optical Fiber Communication Conference and Exposition and the National Fiber Optic Engineers Conference (California, 2010), paper OThW7. [CrossRef]
12. M. P. I. Dias, E. Wong, D. P. Van, and L. Valcarenghi, “Offline energy-efficient dynamic wavelength and bandwidth allocation algorithm for TWDM-PONs,” in Proceedings of IEEE International Conference on Communications, (IEEE 2015), pp. 5018–5023. [CrossRef]
13. A. Dixit, B. Lannoo, D. Colle, and M. Pickavet, “Novel DBA algorithm for energy efficiency in TWDM-PONs” in Proceedings of European Conference and Exhibition on Optical Communication, (ECEOC 2013), pp. 1–3. [CrossRef]
14. I. D. M. Pubuduni, D. P. Van, L. Valcarenghi, and E. Wong, “Energy-efficient framework for time and wavelength division multiplexed passive optical networks,” J. Opt. Commun. Netw. 7(6), 496–504 (2015). [CrossRef]
15. R. Bounziane, R. Koutsoyannis, P. Miilder, and Y. Benlachtar, “Optimizing FFT precision in optical OFDM transceivers,” IEEE Photonics Technol. Lett. 23(20), 1550–1552 (2011). [CrossRef]
16. H. Kimura, H. Nakamura, S. Kimura, and N. Yoshimoto, “Numerical analysis of dynamic SNR management by controlling DSP calculation precision for energy-efficient OFDM-PON,” IEEE Photonics Technol. Lett. 24(23), 2132–2135 (2012). [CrossRef]
17. J. Zhang, J. Hu, D. Qian, and T. Wang, “Energy efficient OFDM transceiver design based on traffic tracking and adaptive bandwidth adjustment,” Opt. Express 19(26), B983–B988 (2011). [CrossRef] [PubMed]
18. X. Hu, L. Zhang, P. Cao, K. Wang, and Y. Su, “Energy-efficient WDM-OFDM-PON Employing Shared OFDM Modulation Modules in Optical Line Terminal,” Opt. Express 20(7), 8071–8077 (2012). [CrossRef] [PubMed]
19. 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 22(11), 13043–13049 (2014). [CrossRef] [PubMed]
20. H. He and W. Hu, “Physical-layer energy-efficient receiving method based on selective sampling in orthogonal frequency division multiplexing access passive optical network,” Jpn. J. Appl. Phys. 42(4B), 2176–2181 (2003).
21. C. Zhang, N. Xiao, C. Chen, W. Yuan, and K. Qiu, “Energy-efficient orthogonal frequency division multiplexing-based passive optical network based on adaptive sleep-mode control and dynamic bandwidth allocation,” Opt. Eng. 55(2), 026108 (2016). [CrossRef]
22. J. Zhang and N. Ansari, “Toward energy-efficient 1G-EPON and 10G-EPON with sleep-aware MAC control and scheduling,” IEEE Commun. Mag. 49(2), s33–s38 (2011). [CrossRef]
23. Z. W. Xue, J. J. Zhang, C. Qian, B. Cao, and Q. W. Zhang, “An Energy Saving Strategy for OFDM-PON Based on ONU Data Identification,” in Asia Communications and Photonics Conference, OSA Technical Digest (Optical Society of America, 2015), paper AM1F.6.
24. Y. X. Qin, J. J. Zhang, J. J. Ma, W. L. Wang, T. F. Sun, X. Z. Peng, J. Chen, and Q. W. Zhang, “A Novel Toggle-rate based Energy-Efficient Scheme for Real-time IMDD OFDM-PON,” in Asia Communications and Photonics Conference, OSA Technical Digest (Optical Society of America, 2016), paper AS3C.1.
25. J. J. Zhang, Z. H. Tang, R. Giddings, Q. Wu, W. Wang, B. Cao, Q. Zhang, and J. M. Tang, “Stage-Dependent DSP Operation Range Clipping-Induced Bit Resolution Reductions of Full Parallel 64-Point FFTs Incorporated in FPGA Based Optical OFDM Receivers,” J. Lightwave Technol. 34(16), 3752–3760 (2016). [CrossRef]
26. W. L. Wu, J. J. Zhang, Z. Zhang, H. Dun, and Q. W. Zhang, “A Novel Amplitude Decision based Symbol Synchronization Method for Real-time IMDD-OOFDM Systems,” in Proceedings of the Opto Electronics and Communications Conference (IEEE 2016), paper ThB1–5.
27. N. S. Kim, T. Austin, D. Blaauw, T. Mudge, K. Flautner, and J. S. Hu, “Leakage Current: Moore’s Law Meets Static Power,” Computer 36(12), 68–75 (2003). [CrossRef]
28. D. Nguyen, A. Davare, M. Orshansky, D. Chinnery, B. Thompson, and K. Keutzer, “Minimization of dynamic and static power through joint assignment of threshold voltages and sizing optimization,” inProceedings of the2003International Symposium of Low Power Electronics and Design (IEEE 2003), pp. 158–163.
29. J. H. Anderson and F. N. Najm, “Switching activity analysis and pre-layout activity prediction for FPGAs,” in Proceedings of the International Workshop on System-Level Interconnect Prediction (ACM 2003), pp. 15–21. [CrossRef]
30. J. Kao, S. Narendra, and A. Chandrakasan, “Subthreshold Leakage Modeling and Reduction Techniques,” in Proceedings of International Conference on Computer-Aided Design, (IEEE 2002), pp. 141–149.
