1. Introduction

The cloud-radio access network (C-RAN) architecture is shown in Fig. 1, which offers several benefits, such as improved network energy efficiency and improved network performance via coordinated multi-point [1]. The network segment from service gateway(S-GW/MME) of mobile core network to baseband units (BBUs) is defined as mobile backhaul (MBH). The segment from BBUs to remote radio heads (RRHs) is defined as mobile fronthaul (MFH). In C-RAN, signal processing and management are centralized in BBU pool to simplify the design of RRHs. The traditional MFH is primarily based on the common public radio interface (CPRI) [2, 3]. In CPRI, digitized baseband signals are transmitted via binary bit sequence. Such digital MFH technique offers excellent robustness against noise and nonlinearities by leveraging the existing digital intensity modulation direct detection (IM-DD) systems [4, 5]. However, it suffers from low spectral efficiency. For an individual typical LTE channel with 20-MHz bandwidth and 2 × 2 multiple-input and multiple-output (MIMO), CPRI requires a separate fronthaul data rate of 2.5-Gb/s [6].

 

Fig. 1 C-RAN architecture, including mobile backhaul (MBH) and mobile fronthaul (MFH).

Download Full Size | PDF

In order to increase the bandwidth efficiency of MFH, several methods have been proposed to aggregate multiple mobile channels in a single wavelength through analog signal processing. Analog MFH technology features simple implementations and high spectral efficiency, but are subjected to noise and nonlinear distortions due to the continuous waveforms and high peak to average power ratio (PAPR) [7–10]. To circumvent the PAPR issue, a digital MFH architecture based on delta-sigma modulation has been recently demonstrated, where the aggregated 32 4G-LTE signals are transformed into OOK signals and then transmitted via a single-λ 10-Gb/s IM-DD channel [11]. By employing the classic 1st-order delta-sigma modulation with 1-bit quantizer and 8 times oversampling ratio (OSR), an average EVM of ~9% could be achieved.

In this paper, a high order delta-sigma modulator with multi-bit quantizer is proposed to enable further enhancement of EVM performance. Compared to the conventional delta-sigma modulation in [11], the improvements lies in the following two facts: 1) The 2-bit quantizer is used to replace the 1-bit quantizer, which not only reduce the overall quantization noise, but also be compatible to the widely deployed 4-level pulse amplitude modulation (PAM-4) based transmission systems. Therefore, it is possible to deliver the delta-sigma modulated signal, using the mass-production 4x25-Gb/s PAM-4 optics/electronics modules. 2) High-order high-pass filter (HPF) is employed to replace the first-order HPF, which leads to a much better band-stop response. Thus, the EVM performances of the high frequency component carriers are further enhanced. By using the proposed scheme, a PAM-4 based mobile fronthaul transmission of 32 aggregated 4G-LTE signals with a CPRI equivalent data rate of 39.32-Gb/s is demonstrated in a single-λ 10-Gb/s IM-DD channel. Compared to the previous delta-sigma modulation based MFH, around 68% improvement is achieved in the average EVM performance.

2. Principal

Figure 2(a) shows the proposed PAM-4 based MFH employing high order delta-sigma modulator. In BBU, mobile signals are first aggregated using FFT/IFFT-based channel aggregation method [9]. The aggregated mobile signals are then transformed to PAM-4 signal by proposed high order delta-sigma modulator. The output of the delta-sigma modulator is then delivered to RRH via PAM-4 based IM-DD channel. A low-pass filter (LPF) is used in RRH to retrieve the original mobile signal from PAM-4 signals.

 

Fig. 2 (a) PAM-4 based MFH employing high order delta-sigma modulation (b) structure of the proposed high order delta-sigma modulator.

Download Full Size | PDF

As shown in Fig. 2(b), the proposed high order delta-sigma modulator first oversamples the input signal, and then exploits multi-bit quantization to transform the input signal into PAM-4 (for 2-bit quantization) signal. Then, the transformed PAM-4 signal are sent into a noise shaping technique block to push the quantization noise out-of-band. The noise shaping technique block consists of a number of canscaded feedback loops, where the Z-transform function could be formulated as 1/(1-Z⁻¹). Each feedback loop could be seen as a high pass filter (HPF), and the number of feedback loops indicates the order of the HPF.

Compared to the conventional delta-sigma modulation in [11], two major improvements are illustrated as shown in Fig. 3. Firstly, considering the fact that PAM-4 has been recently considered as the main standard for data center communications, a 2-bit quantization for delta-sigma modulator should also be a promissing technology due to the mass production of PAM-4 based IM-DD optics/electronic modules [12, 13]. Thus, 1-bit (2-level) quantization as shown in Fig. 3(a) is replaced by 2-bit (4-level) quantization as shown in Fig. 3(b) using the same oversampling ratio (OSR), where f_b is the bandwidth of signal, and fs is the original sampling rate. It is noted that the quantization noise is evenly distributed within the oversampled frequency range. However, the noise intensity after 2-bit quantization is smaller than that after 1-bit quantization [14].

 

Fig. 3 Illustration of comparison between conventional delta-sigma modulation and the high order delta-sigma modulation using the same OSR.

Download Full Size | PDF

Moreover, conventional delta-sigma modulation employs the 1st-order HPF to shape the noise as shown in Fig. 3(a). However, the 1st order modulator shows unevenly distributed band-stop response for the quantization noise in frequency range of [0, f_b]. Pre-emphasize is thus necessary for the aggregated MFH signals to accommodate more component carriers [11]. The proposed high order delta-sigma modulation employs high order HPF to reduce the noise intensity after shaping as shown in Fig. 3(b). Figure 3(c) shows the characteristics of HPF response in delta-sigma modulation with 1st order, 2nd order and 4th order modulator, respectively. The responses of HPF are measured with 8-times OSR, and the horizontal axis represents the normalized frequency. It can be observed that the band-stop response is much more evenly distributed for the 4th order modulator [15]. Therefore, the shaped noise located in the frequency range of [0, f_b] is much smaller than that of the conventional delta-sigma modulation.

3. Experimental setup

Figure 4 shows the experimental setup for the MFH transmission of aggregated 32 4G-LTE signals based on delta-sigma modulation. As a typical example, the equivalent CPRI data rate needed for aggregating 32 20-MHz LTE signals (with 30.72-MHz sampling rate) in the case of four 8 × 8 MIMO antennas is as high as 39.32 Gb/s ( = 4 × 8 × 2 × 16 × 30.72 × 10/8Mbit/s) [6]. In the transmitter, 32 component carriers (CC) of mobile signals are aggregated by DFTS-OFDM [7] with sampling rate of 1.25 GSa/s. Each LTE CC occupies 1200 loaded data-carrying subcarriers in DFTs-OFDM symbol with 15.26-kHz subcarrier spacing. 32 CCs are closely allocated with 76.3 kHz guard band, whose aggregated spectrum is shown in Fig. 4(a). Hermitian symmetry is used to obtain a real output sequence. Then, the real sequence is fed into a high order delta-sigma modulator with OSR = 8 and a bit resolution of 2. The output sampling rate from delta-sigma modulator is 10-GSa/s, where the effective bandwidth for mobile signals is 1.25 GHz. After a single-channel IM-DD system, including a 10-Gb/s direct modulate laser (DML), a variable optical attenuator (VOA), 20-km standard single-mode fiber (SSMF), and commercial 10-GHz photo detector (PD), a LPF is used to retrieve the original mobile signal. Figure 4(b) shows the received spectrum of a delta-sigma modulated signal with 8-times oversampling ratio. The carrier de-aggregation is the inverse process of carrier aggregation, which is performed offline after the LPF. Finally, the EVM is calculated for each CC to evaluate the performance of proposed method.

 

Fig. 4 Experimental setup for the transmission of 32 4G-LTE signals by carrier aggregation and delta-sigma modulation, (a) aggregated spectrum of 32 4G-LTE signals with 1.25-GSa/s sampling rate, and (b) received spectrum of an 8-times oversampling delta-sigma modulated signal with 10-GSa/s sampling rate.

Download Full Size | PDF

4. Results and discussion

The quantization noise can be effectively reduced by increasing the levels of quantizer. As shown in Fig. 5, the original signal waveform before delta-sigma modulation is represented as the red line, the delta-sigma modulated signal waveform is represented as the black line, and the retrieved signal waveform with low pass filter (LPF) at the receiver is represented as the blue line. The mean squared error (MSE) between the original transmitted signal and the retrieved original signal with 1-bit (2-level), 1.5-bit (3-level), 2-bit (4-level) and 2.5-bit (5-level) quantizer is measured as ${\sigma }_{err}$ = 0.0161,0.0139,0.0061, and 0.0050, respectively. Here, ${\sigma }_{err}=E\left(\sqrt{\frac{{(y-x)}^2}{y^2}}\right)$, $x$ denotes retrieved signal, $y$ denotes the original signal, and $E(\cdot )$ denotes the arithmetic mean value.

 

Fig. 5 Signal waveforms before delta-sigma modulation (blue), after delta-sigma modulation (black) and after received LPF (red) with 1-bit, 1.5-bit, 2-bit and 2.5-bit quantization, respectively.

Download Full Size | PDF

Furthermore, we measure the EVM of each component carrier using the delta-sigma modulator with the same 1st order HPF, OSR = 8, and different quantization resolutions from 1-bit to 2.5-bit at the optical back to back condition with received optical power equaling to −6 dBm. As shown in Fig. 6, the average EVM of 1-bit and 1.5-bit quantization is 9.1% and 7.3%, respectively. By increasing the quantization resolution to 2-bit, the average EVM is reduced to 5.6%. Further increasing the quantization resolution, i.e. 2.5-bit, the average EVM is 5.5%, which brings little improvement. It is noted that the 2-bit quantization results in a 4-level output sequence, which is compatible with the widely deployed PAM-4 based IM-DD system. Thus, the 2-bit quantization is chosen for the rest of the experiments.

 

Fig. 6 Measured EVM of each CC using 1st order HPF, OSR = 8, and different quantization resolutions from 1-bit to 2.5-bit, respectively.

Download Full Size | PDF

Then, we employ the delta-sigma modulator with the same 2-bit quantization, OSR = 8, and 1st, 2nd and 4th order HPF, respectively. Figure 7 shows the measured EVM of each CC at optical back to back experiment. Due to the better band-stop response of the high order HPF, the EVMs of the high frequency CCs (large carrier indexes) are greatly reduced. Considering the EVM requirement of 3GPP specifications (256-QAM 3.5%, 64-QAM 8%, 16-QAM 12.5%, QPSK 17.5%), only the first 2 CCs can be loaded by 256-QAM, the middle 3~30 CCs can be loaded by 64-QAM, and the last two CCs can be loaded by 16-QAM, with 1st order HPF. However, with 4th order HPF, 256-QAM can be loaded for almost all of the CCs (except for the last two). The average EVM with 1st, 2nd and 4th order HPF are 5.6%, 3.4% and 2.9%, respectively. Compared to the scheme that employing the 1st order HPF, around 48% ( = (5.6%-2.9%)/5.6%) EVM performance improvement is obtained with the 4th order HPF.

 

Fig. 7 Measured EVM of each CC at optical back to back, using 2-bit quantization, OSR = 8, and various 1st, 2nd and 4th order HPF, respectively.

Download Full Size | PDF

By applying the delta-sigma modulator with 2-bit quantization and 4th order HPF, the OSR can be reduced to provide similar throughput compared to the delta-sigma modulator with 1-bit quantization and 1st order HPF. Figure 8 shows the measured EVM of each CC using: 1) OSR = 8, 1-bit quantization, 1st order HPF 2) OSR = 8, 2-bit quantization, 4th order HPF, 3) OSR = 6, 2-bit quantization, 4th order HPF, 4) OSR = 4, 2-bit quantization, 4th order HPF. The first case is measured as the baseline [11], where the highest EVM is around 15% for CC32, the lowest EVM is around 4.8% for CC1 and the average EVM is around 9.1%,. When the OSR remains 8, the EVMs of all CCs are evenly below 4.0% by using the proposed delta-sigma modulator with 2-bit quantization and 4th-order HPF. Most CCs (CC1~CC30) can be loaded by 256-QAM for the EVMs being below 3.5%. Only the EVMs of CC31 and CC32 are a little beyond 3.5%, thus loaded by 64QAM. The average EVM is only 2.9%, which means that 68% EVM performance improvement is obtained. Reducing the OSR from 8 to 6 with the proposed high order delta-sigma modulator, we can observe that EVMs for all CCs are also decreased, especially for low-frequency CCs with indexes less than 25, and the average EVM is decreased to 6.2%. Further reducing the OSR to 4, the EVMs of CC1~CC17 are still lower than the baseline method. However, the EVMs of the high-frequency CCs (CC18~CC32) are higher than the baseline method, and the average EVM is increased to 12.1%. From the comparison results as shown in Fig. 8, we can observe that for the purpose of saving bandwidth, OSR = 6 could be selected to provide even better EVM performance with less bandwidth cost. If we need to deliver mobile data most with 256QAM format, the OSR of 8 could be selected, which needs the same bandwidth as that of the baseline method.

 

Fig. 8 Measured EVM of each component carriers using: 1) OSR = 8, 1-bit, 1st order 2) OSR = 8, 2-bit, 4th order, 3) OSR = 6, 2-bit, 4th-order, and 4) OSR = 4, 2-bit, 4th order.

Download Full Size | PDF

Besides this, the EVM performances of 32 CCs are measured after the transmission over 20-km SSMF by employing the proposed high order delta-sigma modulator with OSR = 8, at received optical power of −6 dBm. The average EVM is 3.6%, and slight performance degradation is found compared to back-to-back (B2B), as shown in Fig. 9. The EVMs for the first 22CCs are below 3.5%, thus can be loaded with 256 QAM according to the EVM requirement of 3GPP specifications. And the EVMs of last 10 CCs are slightly increased from 4% to 5.3%, which is much lower than the required 8% threshold of 64 QAM. Therefore, better EVM performance can be obtained by loading with 64 QAM format. The restored constellations are also shown in the insets for the corresponding CCs.

 

Fig. 9 Measured EVM of each component carrier after 20-km transmission using 2-bit quantization, 8 times oversampling and 4th order modulator.

Download Full Size | PDF

Lastly,the mean EVM of all CCs as a function of OSNR and as a function of the received optical power by employing the proposed high order delta-sigma modulator with OSR = 8 are shown in Figs. 10(a) and 10(b), respectively. At the required EVM threshold of 8% for 64-QAM, the required OSNR for back to back is 29.8dB, and the corresponding received optical power is −22.5dBm. We further conduct the proposed scheme after 5-km and 20-km SSMF transmission, respectively. After 5-km transmission, slight performance degradation is found compared to the back to back case. After 20-km transmission, around 2.7dB and 2.9dBm degradation of OSNR and received optical power requirement could be observed. To obtain an EVM of below 8%, the required OSNR and received optical power is higher than 32.6dB, and −19.4dBm, respectively.

 

Fig. 10 Mean EVM of all CCs as a function of (a) OSNR, and (b) received optical power, respectively.

Download Full Size | PDF

5. Conclusion

An improved mobile fronthaul architecture employing high order delta-sigma modulator and PAM-4 format has been proposed in C-RAN. Compared to the previous OOK based MFH with delta-sigma modulator, reduced overall quantization noise was obtained by using 2-bit quantization and shaped noise was reduced by using 4-th order HPF, which enables lower and more evenly distributed EVMs for component carriers in the aggregated signal. The proposed method has been verified by carrier aggregation of 32 4G-LTE signals with a CPRI equivalent data rate of 39.32-Gb/s in a single-λ 10-Gb/s IM-DD channel. Around 68% improvement is achieved for the average EVM performance compared to the previous MFH based on delta-sigma modulation.