Open Access
Add to BibSonomy
Abstract
In this paper, we firstly demonstrate an advanced arraying scheme in the TDM-based analog mobile fronthaul system to enhance the signal fidelity, in which the segment of the antenna carrier signal (AxC) with an appropriate length is served as the granularity for TDM aggregation. Without introducing extra processing, the entire system can be realized by simple DSP. The theoretical analysis is presented to verify the feasibility of this scheme, and to evaluate its effectiveness, the experiment with ~7-GHz bandwidth and 20 8 × 8 MIMO group signals are conducted. Results show that the segment-wise TDM is completely compatible with the MIMO-interleaved arraying, which is employed in an existing TDM scheme to improve the bandwidth efficiency. Moreover, compared to the existing TDM schemes, our scheme can not only satisfy the latency requirement of 5G but also significantly reduce the multiplexed signal bandwidth, hence providing higher signal fidelity in the bandwidth-limited fronthaul system. The experimental result of EVM verifies that 256-QAM is supportable using the segment-wise TDM arraying with only 250-ns latency, while with the ordinary TDM arraying, only 64-QAM is bearable.
© 2018 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
Driven by the development of mobile Internet and emerging services, the technologies for the 5th generation mobile network (5G) are massively investigated. And on the road toward 5G, the access network is considered as one of the most critical segments where revolutionary network architecture and interface should be employed. For this reason, the concept of centralized radio access network (CRAN) [1] is proposed. In this centralized architecture, conventional baseband units (BBUs) are physically centralized. And the traditional base station (BTS) at remote side is simplified as remote radio head (RRH) that only keeps the radio functions. To carry the high volume traffic of radio signal in the link between BBU and RRH, namely the fronthaul, the optical fiber is supposed to be the desirable media [2]. Currently, the digital and the analog radio over fiber (RoF) technologies for the fronthaul interface are widely investigated. Whereas, due to the inherent bandwidth consuming characteristic of the digitalized fronthaul signal, the digital interfaces based on the common public radio interface (CPRI) [3–6] are still facing many technical challenges. Meanwhile, characterized by high bandwidth efficiency, the analog RoF captures people’s attention, and many schemes are demonstrated [7–17]. One primary concern for designing analog fronthaul interface is how to multiplex multiple baseband in-phase and quadrature (IQ) signals, namely the AxCs, in a simple and cost-efficient way without sacrificing signal fidelity. However, existing schemes usually require either the ideal components [7–11] like high-accurate oscillators and modulators, or complex digital signal processing (DSP) [12,13], otherwise the uniform high fidelity is not guaranteed [14].
To solve the preceding problems, the time division multiplexing (TDM)-based schemes [15–17] are proposed, in which each AxC flow is firstly divided into uniform subsets in time dimension, and the subset length is termed as the TDM granularity. The subsets from multiple flows are then multiplexed in TDM manner to constitute one serial signal. The merits of the TDM-based schemes are that no high-accuracy oscillator or modulator is required, and the DSP complicity of multiplexing can be ultra-low since only sampling data rearrangement is required. Furthermore, the TDM signal presents small peak to average ratio (PAPR) and slight inter-modulation-distortion (IMD), hence providing high signal fidelity [15]. Nevertheless, the reported TDM-based schemes are facing the conflict between transmission latency and spectrum efficiency. In this paper, for the sake of fair comparison, the aggregating latency caused by sample buffering at transmitter for unidirectional transport is concerned, and the latency discussed below refers to this type. As with [12,16], the aggregating latency is selected as the key parameter for the estimation of system latency performance. Take the symbol-wise TDM (Sy-TDM) scheme [15] for instance, as shown in Fig. 1(a), the TDM granularity is the length of one wireless symbol, which contains ~2192 samples for a typical 20-MHz long term evolution (LTE) signal. Due to the buffering process in cycle-1, the aggregating latency is ~66.7-us. Since the most stringent delay of the unidirectional transmission in fronthaul is ~100-us [18,19], only ~33.3-us is left for fiber transmission and other processing, which is impracticable. To lower the latency, the sample-wise TDM (Sa-TDM) scheme with one AxC sample as the new TDM granularity is proposed in [16]. And the corresponding schematic is presented in Fig. 1(b). However, in practice, due to the imperfect channel response (bandwidth limitation, passband ripple, etc.), the interference among adjacent samples can be easily introduced, leading to the interference among AxCs, which dramatically impairs the signal quality. For this reason, guide intervals (GIs) was inserted among adjacent samples, which, however, significantly decrease the time slot (TS) utilization. Correspondingly, compared to the GI-free one, the multiplexed signal occupies wider bandwidth and suffers worse deterioration caused by imperfect channel response. Therefore, the signal fidelity is decreased. To reduce the GI, the authors [16] introduced the multiple input and multiple output (MIMO)-interleaved arraying, where the samples from different AxCs but within a MIMO group are alternately transmitted. Yet the remaining GIs still occupy numerous TSs, leading to a low spectrum efficiency.
Fig. 1 Arraying of (a) the symbol-wise TDM and (b) the basic sample-wise TDM for 3 AxCs. TS: time slot.
In this paper, for the first time, by jointly considering the bandwidth efficiency, signal fidelity and the latency performance, we propose to use the segment of AxC signal as the new TDM granularity in analog TDM-based fronthaul and systematically analyze how the granularity size impacts the performance. Furthermore, the scheme of combining segment-wise granularity with MIMO-interleaved arraying is firstly proposed and verified in this paper, which significantly improves the signal fidelity while keeping the latency satisfying 5G requirement [18,19]. Besides, no additional DSP complicity is introduced in this segment-wise TDM (Se-TDM) arraying. To evaluate the Se-TDM fronthaul, we conduct an experiment, where a 7-GHz bandwidth system is used to carry 20 8 × 8 MIMO signals, namely 160 20-MHz-AxC signals or equivalent ~162-Gb/s CPRI signals. Such a large volume of signals are aggregated and transmitted without the assistance of any equalizer. Compared to the Sa-TDM signal, the bandwidth of the Se-TDM signal is decreased from 6-GHz to 4.96-GHz, which highly enhances the signal fidelity. The EVM results show that 256-QAM signal is bearable for all MIMO groups with the Se-TDM arraying, while with the existing Sa-TDM arraying, the EVM-satisfied modulation order declines to 64-QAM. The trade-off between fidelity performance and latency is also studied in the Se-TDM system. The result shows that the system can produce almost the optimal fidelity with only 250-ns aggregating latency.
2. Principle of the segment-wise TDM
2.1 Proposed segment-wise TDM with MIMO-interleaved arraying
To further reduce the number of GI and hence improve the spectrum efficiency while keeping low aggregating latency, we propose the new TDM granularity, which is the consecutive signal segment with well-designed length from the original AxC stream. With this new granularity, the TDM process works as follows. Each AxC stream is firstly divided into uniform segments. Within a TDM cycle, each AxC signal contributes one segment, and the segments belonging to the same MIMO group are firstly interleaved to enable the MIMO interleaved technology. Then, multiple MIMO groups are multiplexed in TDM manner to fulfill a cycle. For better understanding, the Se-TDM arraying with the 5-sample-length segment and three 4 × 4 MIMO groups is illustrated in Fig. 2. Here, Si-j stands for the sample from the AxC-i at sampling instant j, where sampling instant represents the time when the baseband sample is captured. Note that in the Se-TDM scheme, GI is only inserted between samples belonging to different MIMO groups. In this way, with an appropriate segment length, the number of GI can be significantly reduced.
Fig. 2 Arraying of MIMO-interleaved segment-wise TDM with three 4 × 4 MIMO groups and 5-sample-length segment. Si-j: the sample from AxC-i at sampling instant j; GI: guard interval.
2.2 Principle of interference elimination by MIMO-interleaved arraying
The MIMO-interleaved arraying introduced in the proposed Se-TDM is an effective method to eliminate the inference among adjacent samples. In this section, the mathematical expression is derived to disclose the principle and show how it affects the signal performance in the Se-TDM and the Sa-TDM scheme.
In MIMO-interleaved Sa-TDM scheme, the baseband IQ signal should be firstly up-converted onto a low-frequency intermediate frequency (IF) to form one real-valued passband signal, which is still denoted as AxC for simplicity. Then, the AxC samples are used for TDM arraying. Figure 3(a) is the schematic diagram of the MIMO-interleaved arraying for a Sa-TDM fronthaul containing three 4 × 4 MIMO groups. Here, S11~S14 are 4 samples from AxC-1~AxC-4 belonging to group-1 respectively. Samples of group-2 and 3 are denoted in the same way. With the MIMO-interleaved arraying, the interference among AxCs caused by channel transmission can be divided into two types, the inter-MIMO interference (inter-MI) and the intra-MIMO interference (intra-MI), which are shown in Fig. 3(b). The inter-MI occurs among the samples from different AxCs belonging to different MIMO-decoding groups. This interference can be considered as random noise and cannot be easily removed, hence the introduction of GI is necessary. Correspondingly, the intra-MI occurs among the samples from different AxCs belonging to the same MIMO-decoding group. This intra-MI can be regarded as introducing cross-correlation among the subchannels (the channel between one Tx antenna and one Rx antenna) within a MIMO channel, which is removable through existing MIMO decoding process [20]. Thus, it is relatively not necessary to insert GI for intra-MI elimination.
Fig. 3 (a) MIMO-interleaved arraying in sample-wise TDM, (b) schematic of the received signal. Sij: the sample from jth antenna of MIMO group-i.
For the Se-TDM scheme, compared to the Sa-TDM one, due to the enlarged TDM granularity, the inter-MI is reduced. Besides, although the intra-MI is increased, it can still be eliminated by the MIMO-interleaved arraying. Here, the intra-MI is further divided into two types, the intra-MI at the same sampling instant (intra-MI-intra-SI) and the intra-MI between adjacent sampling instants (intra-MI-inter-SI), which are depicted in Fig. 4(a). The latter only exists in the Se-TDM scheme. In this figure, the index Si-j stands for the sample from ith antenna at sampling instant j. Here, the intra-MI-intra-SI is firstly analyzed in the discrete-time domain, and the sampling period of the discrete-time domain is one TS. As depicted in Fig. 4(b), a channel with low-pass and passband-ripple property can be represented by the impulse response sequence h(n). Without loss of generality, the peak value of h(n) is denoted as h(0). The sample from ith antenna of a MIMO group at kTs is denoted as si(k). Here, Ts represents the sampling period of the up-converted AxC and hence kTs represents the sampling instant of AxC signal. si(k) is loaded into ((k-1)*N + i)th TS, then, after the imperfect channel transmission, the received sample value ri(k) at this TS is
It can be drawn from Eq. (1) that for a fixed i, the relationship between s(k) and r(k) does not change with k, and Eq. (1) can be further written in the continuous-time domain:where the sj(t) denotes the AxC signal for jth antenna before multiplexing, and ri(t) denotes the AxC-i signal after demultiplexing. The corresponding matrix presentation iswhere ,and Hintra denote the N × 1 MIMO input signal vector, the N × 1 signal vector deteriorated by intra-MI-intra-SI and the N × N intra-MI-intra-SI transition matrix respectively. Then, take the downlink for instance, after the transmission of MIMO channels, the final received signal is given in frequency domain:Here, S(ω), R(ω) are the frequency-domain counterparts of and respectively, HMIMO is an N × N matrix representing the wireless MIMO channel, and NW is an N × 1 vector representing the additive white Gaussian noise. Based on Eq. (4), and with the help of the existing MIMO processing which estimates and compensates the entire channel matrix HMIMO(ω) Hintra(ω), the intra-MI-intra-SI can be removed. Note that although intra-MI-intra-SI can be eliminated, the MIMO channel diversity is decreased, thus decreasing the MIMO gain to some degree. However, this impairment is relatively mild compared to the un-removable inter-MI case.Fig. 4 (a) Intra-MI-intra-SI and Intra-MI-inter-SI in segment-wise TDM with 4 × 4 MIMO-interleaved arraying, Si-j: sample from ith antenna at sampling instant j. (b) Impulse response sequence of a low-pass system.
Then, the effect of the intra-MI-inter-SI is analyzed. Since the effective part of the impulse response only last for several sampling periods, it is reasonable to assume that intra-MI-inter-SI only happens between adjacent sampling instants. Based on this assumption, considering the intra-MI-intra-SI and the intra-MI-inter-SI, the received sample value r'i(k) is
Correspondingly, the final received signal R'(ω) in frequency domain isHere, Hinter denotes the N × N intra-MI-inter-SI transition matrix. Based on Eq. (6), the entire channel matrix HMIMO(ω)(Hintra(ω) + Hinter(ω)) can still be estimated and compensated by MIMO processing. It is worth noting that Hinter(ω) contains e-jωTs and e+jωTs components, which result in the multipath effect on AxC signal. Fortunately, the maximum time difference between two paths is only 2Ts, namely less than 50-ns, which is negligible considering ~4.7-us cyclic prefix (CP) is provided in LTE signal to overcome the multipath effect. Moreover, without introducing additional devices or interference among AxC signals, the Se-TDM scheme can well inherit the applicability of the ordinary TDM scheme and serve wide 5G scenarios. Besides, the operating scenario can also be a non-MIMO case. In this case, the segment from non-MIMO signal can be considered as a segment from MIMO group with MIMO order reduced to one, and it can still be multiplexed with other MIMO segments in TDM manner, making the Se-TDM a widely applicable scheme.2.3 Spectrum efficiency improvement and the sacrifice of latency performance brought by the Se-TDM
To quantify the TS utilization and the signal bandwidth in different arraying schemes, we consider one link that carries M N × N MIMO groups. For the Sa-TDM scheme, the optical GI length is denoted as NGI time slots, and the TS utilization μsample can be expressed as. Let Fs denote the minimal sampling rate required for the analog-to-digital/digital-to-analog converter (ADC/DAC) in the system, and the bandwidth of the multiplexed signal is further defined as the half of Fs. Thereby, the bandwidth of the multiplexed signal BWsample is calculated as
Here, fsAxC denotes the sampling rate of the up-converted AxC signal. For the Se-TDM arraying, let L denote the segment length and N'G denote the length of the optimal GI, then, the TS utilization μsegment can be expressed as. Correspondingly, the bandwidth of the Se-TDM signal isAccording to Eq. (8), BWsegment is reduced with the growth of L.As for the latency performance, in the Se-TDM scheme, the aggregating latency is extended to L × Ts. Although the latency is enlarged to exchange a better TS utilization, this sacrifice is still worthwhile. The following section will prove that the Se-TDM arraying with 250-ns latency can significantly outperform the Sa-TDM one in terms of signal fidelity. And considering the 100-us-level latency requirement, the 250-ns here is still negligible.
To sum up, the Se-TDM can transform the inter-MI to the intra-MI, thus reducing the number of GIs, increasing the TS utilization and improving the spectrum efficiency. Consequently, in a system with imperfect channel response, to carry the same number of AxCs, the Se-TDM signal can offer better signal fidelity.
3. Experimental setup and result discussion
To verify the improvement of the signal fidelity brought by Se-TDM arraying, we conduct an uplink experiment setup including wireless MIMO channel and bandwidth-limited optical channel, as depicted in Fig. 5. In this experiment, 20 8 × 8 MIMO signals, namely 160 AxCs with each carrying 20-MHz LTE-A signal are aggregated. Note that the wireless signal with broader bandwidth can also be supported, and although only orthogonal frequency division multiplexing (OFDM) is tested, the Se-TDM is transparent to the 5G modulation formats. For each MIMO group, the sub-channels within the MIMO channel are independent Rayleigh fading channels. After the MIMO channel transmission, the signal is down-converted to the 14-MHz IF, and the signal @IF is digitalized by a 60-MHz ADC. After that, all AxCs are multiplexed with the Se-TDM or the Sa-TDM arraying. The aforementioned flow is achieved by offline processing, and the signal degradation from radio hardware circuit and noise caused by hardware and wireless channel is neglected. Next, the multiplexed signal is launched by an arbitrary waveform generator (AWG, Tektronix 7122C) with configurable sampling rate. Then, it is transmitted over 20-km fiber @1550nm in an intensity modulation direct detection (IM/DD) system. The Mach-Zehnder modulator (MZM) and the photodetector (PD) with 3-dB bandwidth being ~10-GHz and ~8-GHz respectively are used in this system, providing ~7-GHz system bandwidth, and the frequency-domain and time-domain response of the entire system are shown in Fig. 5 (i,ii). At BBU side, the signal detected by the PD is further sampled by an oscilloscope (LeCroy SDA 845Zi-A) at 20-GSa/s. Afterward, the demultiplexing and demodulation are achieved by offline processing, and the maximal ratio combining (MRC) [21] is employed in MIMO demodulation.
Fig. 5 System processing flow and setup. (i) The frequency response and (ii) the time-domain impulse response of the entire system. (iii) The spectrum of the sample-wise TDM signal and the segment-wise TDM signal. AxC: antenna carrier signal; ADC: analog to digital converter; TDM MUX/DEMUX: time division multiplexing/demultiplexing; AWG: arbitrary waveform generator; LD: laser diode; PD: photodetector; OSC: oscilloscope. RF: radio frequency; IF: intermediate frequency; BB: baseband.
Before comparing the performance of the Se-TDM and the Sa-TDM, the optimal GI length for each scheme should be determined. Figure 6(a, b) shows the error vector magnitude (EVM) & TS utilization versus the length of GI for the Sa-TDM and the Se-TDM with the segment length of 60 samples respectively. Note that the GI-axis is consistent with the signal-bandwidth-axis. It can be seen from the results that GI with either too large or too short length cannot provide a satisfying EVM result. For the case of short GI length, the elimination of inter-MI is not complete, and with large GI length, the duration of each TS is reduced, leading to a more severe intra-MI which would limit the performance of MIMO as mentioned in section 2.2. From these results, the optimal GI length is 2 time slots and ~4 time slots for the Sa-TDM and the Se-TDM respectively. This result is given under −1-dBm received optical power (ROP), where the signal to noise ratio (SNR) is ~25 dB, which is a typical value for fronthaul scenario [22]. Figure 6(a, b) also presents the correspondence between the length of GI and the signal bandwidth. With the optimal GI length, the bandwidth of the multiplexed signal is 6-GHz in the Sa-TDM scheme, while only 4.84-GHz in the Se-TDM one, hence the spectrum efficiency is improved by 24%. And to generate the multiplexed signal, at least 12-GSa/s DAC is required in the Sa-TDM scheme, while only 9.68-GSa/s one in Se-TDM scheme. This result verifies that the system sampling rate can also be significantly reduced in the Se-TDM system.
Fig. 6 EVM & TS utilization versus GI length in (a) Sa-TDM scheme and (b) Se-TDM scheme with the aggregation of 20 8 × 8 MIMO groups. (c) EVM of Se-TDM scheme versus segment length. TS: time slot; Sa-TDM: sample-wise TDM; Se-TDM: segment-wise TDM.
We then test the EVM performance versus the segment length. Based on the optimal GI length and the frequency response shown in Fig. 5(i), we simulate the EVM performance versus the segment length under 25-dB SNR, and the result is shown in Fig. 6(c). It can be seen from the result that if the segment length is shorter than ~15 samples, the EVM sharply decreases with the growth of segment length. And afterward, the curve becomes flat, and the EVM is converged to ~2%. Based on this, 4 representative segment lengths, 3, 15, 60 and 300 samples are tested in the following experiment. With the optimal GI lengths and the chosen segment lengths, the EVM versus ROP with the Se-TDM arraying and the Sa-TDM arraying are measured, the results are shown in Fig. 7(a). We believe that a satisfying balance between EVM performance and latency is achieved in the Se-TDM scheme with 15-sample segment length since the performance is almost optimal and the latency is only 250-ns. With this configuration, the TS utilization is 96.7%, while in Sa-TDM, the TS utilization is 80%. Correspondingly, the signal bandwidth is decreased from 6-GHz to 4.96-GHz as shown in Fig. 5(iii). This improvement is reflected in the signal fidelity. Considering that EVM thresholds for 64-QAM and 256-QAM are 8% and 3.5% respectively, the highest bearable modulation order is 256-QAM in the Se-TDM scheme with the 15-sample segment, while only 64-QAM in Sa-TDM. This means much better signal fidelity can be achieved in the Se-TDM scheme, and the system capacity is increased by 33% (from 6-bit/subcarrier to 8-bit/subcarrier). As for the receiver sensitivity, consider the threshold of 64-QAM, the receiver sensitivity in Se-TDM is increased by ~3-dB compared to the one in Sa-TDM. The signal constellation and the detailed EVM of individual MIMO group are given in Fig. 7 index (i-iii). The BtB results for the Sa-TDM and the Se-TDM with 15-sample segment length is presented in Fig. 7(b), where the Se-TDM scheme continuously outperforms the Sa-TDM one.
Fig. 7 (a) EVM versus ROP for different segment lengths after 20-km fiber. (b) EVM of the Sa-TDM and the Se-TDM with 15-sample segment length via 20km and BtB transmission. (c) EVM versus the number of aggregated MIMO groups, the segment length for Se-TDM is 15 samples. Sa-TDM: sample-wise TDM; Se-TDM: segment-wise TDM.
At last, since the requirement on transmission capacity and signal fidelity can be different in various scenarios, we also present the simulation result of EVM performance versus the number of multiplexed MIMO groups. In the simulation process, the extracted channel parameters, as shown in Fig. 5(i), are firstly used to generate the impulse response sequence. Then, the TDM signal is convolved with this sequence to suffer the imperfect channel response. Following the convolution, Gaussian white noise is added to the signal and the SNR fixed at 25-dB. The result is shown in Fig. 7(c), and for any MIMO group number configuration, the result is given with the corresponding optimal GI length. And the segment length of the Se-TDM is fixed at 15 samples. It can be seen from the result that with the growth of aggregated MIMO group number, the EVM performance is continuously deteriorated in both schemes, which is due to the increase of the signal bandwidth after TDM aggregation. Whereas the Se-TDM scheme always outperforms the Sa-TDM one with any number of aggregated MIMO groups. This result is easily understood since the Se-TDM signal is always more spectrum-effective than the Sa-TDM one. Thus, in real application scenarios, although the requirement on aggregated MIMO group number and the signal fidelity may vary, the Se-TDM can always bring noticeable performance improvement, making it a universal and feasible solution.
4. Conclusion
A segment-wise TDM with MIMO-interleaved arraying scheme is proposed for bandwidth-efficient mobile fronthaul transmission. The theoretical analysis illustrates that the TS utilization and the spectrum efficiency are significantly improved by Se-TDM arraying, which is compatible with existing MIMO-interleaved arraying. In our experiment with 20 8 × 8 MIMO groups configuration, compared to the Sa-TDM scheme, the signal bandwidth in the Se-TDM scheme is reduced from 6-GHz to 4.96-GHz. As for signal fidelity, the maximum bearable QAM order is raised from 64-QAM to 256-QAM. Meanwhile, the aggregating latency is only 250-ns. Regardless of the fronthaul capacity, the Se-TDM can always notably improve the signal fidelity. Furthermore, the Se-TDM arraying requires no linear operation on the signal and no equalization for channel compensation, making the DSP complexity extremely low. All these results verify that it is a feasible solution for the bandwidth-efficient and delay-sensitive mobile fronthaul transmission.
Funding
National Natural Science Fund of China (NSFC) (No. 61431009, No. 61501157 and 61521062); National Science and Technology Major Project of the Ministry of Science and Technology of China (2015ZX03001021).
References and links
1. China Mobile, “C-RAN: the road towards green RAN,” White Paper, version 2.5 (2013).
2. P. Chanclou, A. Pizzinat, F. Le Clech, T.-L. Reedeker, Y. Lagadec, F. Saliou, B. Le Guyader, L. Guillo, Q. Deniel, S. Gosselin, S. D. Le, T. Diallo, R. Brenot, F. Lelarge, L. Marazzi, P. Parolari, M. Martinelli, S. O’Dull, S. A. Gebrewold, D. Hillerkuss, J. Leuthold, G. Gavioli, and P. Galli, “Optical fiber solution for mobile fronthaul to achieve cloud radio access network,” Future Network and Mobile Summit IEEE, 2013.
3. CPRI Specification V7.0, “Common Public Radio Interface (CPRI); Interface Specification,” October 2015.
4. D. Samardzija, J. Pastalan, M. MacDonald, S. Walker, and R. Valenzuela, “Compressed transport of baseband signals in radio access networks,” IEEE Trans. Wirel. Commun. 11(9), 3216–3225 (2012). [CrossRef]
5. S. H. Kim, H. S. Chung, and S. M. Kim, “Experimental demonstration of CPRI data compression based on partial bit sampling for mobile front-haul link in C-RAN,” Optical Fiber Communication Conference, W1H.5. 2016. [CrossRef]
6. M. Xu, X. Liu, N. Chand, F. Effenberger, and G. K. Chang, “Fast Statistical Estimation in Highly Compressed Digital RoF Systems for Efficient 5G Wireless Signal Delivery,” Optical Fiber Communication Conference, M3E.7. 2017. [CrossRef]
7. D. Wake, A. Nkansah, and N. J. Gomes, “Radio Over Fiber Link Design for Next Generation Wireless Systems,” J. Lightwave Technol. 28(16), 2456–2464 (2010). [CrossRef]
8. M. Zhu, X. Liu, N. Chand, F. Effenberger, and G. Chang, “High-capacity mobile fronthaul supporting LTE-advanced carrier aggregation and 8× 8 MIMO,” Optical Fiber Communication Conference, M2J.3. 2015. [CrossRef]
9. X. Liu, H. Zheng, and F. Effenberger, “Bandwidth-Efficient Synchronous Transmission of I/Q Waveforms and Control Words via Frequency Division Multiplexing for Mobile Fronthaul,” Global Communications Conference. IEEE, 2015.
10. A. Gatto, P. Boffi, L. Combi, P. Parolari, U. Spagnolini, R. Brenot, and M. Martinelli, “LTE-A Mobile Fronthaul Exploiting Pulse-Width Modulation in a RSOA-based WDM PON,” Optical Fiber Communication Conference, W3C.6. 2016. [CrossRef]
11. D. Che, F. Yuan, and W. Shieh, “Ultrahigh-Fidelity Mobile Fronthaul Using Analog Angle modulation,” European Conference on Optical Communication, 2016.
12. X. Liu, H. Zeng, N. Chand, and F. Effenberger, “Efficient Mobile Fronthaul via DSP-Based Channel Aggregation,” J. Lightwave Technol. 34(6), 1556–1564 (2016). [CrossRef]
13. L. Haibo, Q. Yang, M. Luo, R. Hu, P. Jiang, Y. Liu, X. Li, and S. Yu, “Demonstration of Bandwidth Efficient and Low-Complexity Mobile Fronthaul Architecture via CDM-Based Digital Channel Aggregation,” Optical Fiber Communication Conference, TH3A.5, 2017. [CrossRef]
14. J. Wang, Z. Yu, K. Ying, J. Zhang, F. Lu, M. Xu, and G.-K. Chang, “Delta-Sigma Modulation for Digital Mobile Fronthaul Enabling Carrier Aggregation of 32 4G-LTE / 30 5G-FBMC Signals in a Single-λ 10-Gb/s IM-DD Channel,” Optical Fiber Communication Conference, W1H.2, 2016. [CrossRef]
15. C. Ye, K. Zhang, Q. Chang, Z. Gao, X. Hu, X. Huang, and X. Sun, “A DSP-Assisted Symbol-Cascade Mobile Fronthaul Solution with Large Capacity and Neat RRHs,” European Conference on Optical Communication, 2015. [CrossRef]
16. F. Lu, M. Xu, L. Cheng, J. Wang, S. Shen, C. Su, and G.-K. Chang, “Efficient mobile fronthaul serving massive MIMO new radio services using single-IF with Sample-wise TDM for reduced RRH complexity and ultra-low latency,” Optical Fiber Communication Conference, TH3A.4. 2017. [CrossRef]
17. X. Liu, H. Zeng, N. Chand, and F. Effenberger, “CPRI-Compatible Efficient Mobile Fronthaul Transmission via Equalized TDMA Achieving 256 Gb/s CPRI-Equivalent Data Rate in a Single 10-GHz-Bandwidth IM-DD Channel,” Optical Fiber Communication Conference, W1H.3. 2016.
18. China Mobile Research Institute, “White Paper of Next Generation Fronthaul Interface,” White paper v1.0, 2015.
19. A. Pizzinat, P. Chanclou, F. Saliou, and T. Diallo, “Things you should know about fronthaul,” J. Lightwave Technol. 33(5), 1077–1083 (2015). [CrossRef]
20. L. Cheng, X. Liu, N. Chand, F. Effenberger, and G.-K. Chang, “Experimental demonstration of sub-Nyquist sampling for bandwidth-and hardware-efficient mobile fronthaul supporting 128× 128 MIMO with 100-MHz OFDM signals,” Optical Fiber Communication Conference, W3C.3. 2016. [CrossRef]
21. Y. S. Cho, J. Kim, W. Y. Yang, and C. G. Kang, MIMO-OFDM wireless communications with MATLAB (John Wiley & Sons, 2010).
22. X. Liu, H. Zeng, N. Chand, and F. Effenberger, “Efficient Mobile Fronthaul via DSP-Based Channel Aggregation,” J. Lightwave Technol. 34(6), 1556–1564 (2016). [CrossRef]
