Open Access
Add to BibSonomyAbstract
We analyze the mobile fronthaul (MFH) bandwidth and the wireless transmission performance in the split-PHY processing (SPP) architecture, which redefines the functional split of centralized/cloud RAN (C-RAN) while preserving high wireless coordinated multi-point (CoMP) transmission/reception performance. The SPP architecture splits the base stations (BS) functions between wireless channel coding/decoding and wireless modulation/demodulation, and employs its own CoMP joint transmission and reception schemes. Simulation results show that the SPP architecture reduces the MFH bandwidth by up to 97% from conventional C-RAN while matching the wireless bit error rate (BER) performance of conventional C-RAN in uplink joint reception with only 2-dB signal to noise ratio (SNR) penalty.
© 2016 Optical Society of America
1. Introduction
Toward the future radio access beyond 2020, many small cell base stations (BS) will be densely deployed to enhance network capacity and accommodate the explosive increase in mobile traffic [1,2]. Flexible deployment of BSs is achieved by splitting BS into baseband unit (BBU) and remote radio head (RRH). The BBUs and RRHs are connected by optical fibers, and the links between them are called mobile fronthaul (MFH). In addition, the centralized/cloud radio access network (C-RAN) architecture [3,4] which centralizes BBU processing and connects distributed RRHs has been applied to mobile networks as shown in Fig. 1. C-RAN enhances the performance of coordinated multi-point (CoMP) transmission and reception such as joint transmission and reception [5].
The current MFH uses the common public radio interface (CPRI) [6] to forward the IQ data of the baseband signals. IQ data requires a very large optical bandwidth more than 10 times the original wireless data rate. For example, given the current long term evolution (LTE) system with wireless data rate of up to 150 Mbps, CPRI requires a 2.4-Gbps optical bandwidth. In future radio access, the wireless data rate will increase to several Gbps, and thus CPRI-based MFH bandwidth will increase to several tens of Gbps. This huge optical bandwidth will lead to an enormous optical transmission cost, since many small cells will be deployed with C-RAN and all will need to use expensive 40- or 100-Gbps optics. Therefore, to realize cost-effective MFH deployment, the MFH bandwidth must be reduced to less than 10 Gbps, which enable cost-effective optical transmission.
Typical approaches to reduce the MFH bandwidth include IQ data compression techniques [7–9] based on the signal processing of IQ data such as decimation and bit width reduction. The compression ratio of these techniques is about 0.5, and thus IQ data compression is insufficient to reduce the MFH bandwidth from several tens of Gbps to 10 Gbps or less. A more promising approach is changing the functional split point between BBU and RRH [10–13] such as the MAC-PHY split [10], which splits BS functions between MAC and PHY layers, and the functional split within PHY layer (PHY1 and PHY2 in [10]). This approach can significantly reduce the MFH bandwidth to almost the same as the wireless data rate by transferring some PHY layer functions to RRH, and existing packet networks can be used for the MFH since the MFH bandwidth varies with mobile traffic. However, distributing the PHY layer functions makes it difficult to realize CoMP with centralized processing functions such as joint transmission and reception [3]. Therefore, it is essential to create a functional split scheme that can realize both large MFH bandwidth reduction and high CoMP performance. Moreover, a detailed analysis of the relationship between MFH bandwidth and CoMP performance is important to select an appropriate functional split.
In this paper, we analyze the MFH bandwidth and wireless CoMP performance in the split-PHY processing (SPP) architecture which we have proposed [14,15] to reduce the MFH bandwidth while achieving wireless CoMP performance close to that of conventional C-RAN. The SPP architecture employs the functional split between the wireless channel coding/decoding and modulation/demodulation of LTE PHY layer functions. It also employs its own joint transmission and reception schemes in the downlink and the uplink to enhance wireless transmission performance under its functional split. We conducted numerical simulations to evaluate the MFH bandwidth in the SPP architecture. Simulation results show that the SPP architecture reduces the MFH bandwidth by up to 97% compared with conventional C-RAN. We also evaluate the wireless transmission bit error rate (BER) performance of uplink CoMP joint reception in the SPP architecture when the modulation scheme is quadrature phase shift keying (QPSK), 16 quadrature amplitude modulation (QAM) and 64QAM. Simulation results show that the SPP architecture matches the bit error rate (BER) performance of conventional C-RAN for all 3 modulation schemes and the signal to noise ratio (SNR) penalty is less than 2 dB.
2. Split-PHY processing architecture
2.1 Functional split of base station
The SPP architecture splits the BS functions between wireless coding/decoding functions and wireless modulation/demodulation functions as shown in Fig. 2, and so differs from conventional C-RAN, MAC-PHY split, PHY1 and PHY2 [10]. Note that LTE PHY layer includes baseband processing functions such as orthogonal frequency domain multiplexing (OFDM) processing and resource mapping, not shown in Fig. 2. Wireless channel coding is the first PHY layer function in the downlink and it adds parity bits to the LTE MAC frames for error correction. Wireless channel decoding is the last PHY layer function in the uplink and it extracts the received bit data from the log likelihood ratio (LLR) data output from wireless demodulation function. The LLRs are real numbers indicating the probability that each received data bit is zero or one. Thus, the SPP architecture packetizes and forwards the coded bit data in the downlink and the LLR data in the uplink over an optical interface (IF) such as 10 Gigabit Ethernet (GbE) which is different from CPRI.
The maximum MFH bandwidth is determined by the maximum downlink wireless data rate, because the wireless data rate is generally larger in the downlink than the uplink. The difference in the MFH bandwidth of the LTE MAC frame data forwarded in MAC-PHY split and the coded bit data forwarded in the SPP architecture depends on the coding rate. Since the coding rate used for the maximum wireless data rate is so high that the number of parity bits is small, the maximum MFH bandwidth is almost the same for the MAC-PHY split and the SPP architecture. On the other hand, forwarding coded bit data instead of IQ data in the SPP architecture has an advantage over PHY1 and PHY2 because forwarding IQ data requires a larger bandwidth than bit data. Therefore, the SPP architecture can greatly reduce the MFH bandwidth compared with conventional C-RAN, PHY1 and PHY2 which forward IQ data and so requires a very large bandwidth.
2.2 CoMP schemes
Among the CoMP schemes, joint transmission and reception can provide the highest performance. These schemes employ network multiple-input multiple-output (MIMO) [16] and demand centralized MIMO processing functions such as precoding and equalization. However, it is difficult for the SPP architecture to implement centralized precoding and equalization, because MIMO processing functions are distributed to each RRH. We employ joint transmission and reception schemes for the SPP architecture to resolve these issues.
Figure 3(a) shows the downlink joint transmission scheme in the SPP architecture while Fig. 3(b) shows that in conventional C-RAN. We assume 2 RRHs, each with 2 antennas in explaining these schemes. In the joint transmission scheme for the SPP architecture, BBU multicasts the coded bit data c1, c2, c3 and c4 output from wireless channel coding and forwards them over the MFH. Meanwhile, BBU in conventional C-RAN performs wireless channel coding, wireless modulation and centralized MIMO precoding and then forwards IQ data of precoded symbols s1, s2, s3 and s4 to each RRH. Each RRH in the SPP architecture performs wireless modulation and MIMO precoding using forwarded coded bit data in the same way as BBU in conventional C-RAN, and selects the signals to transmit from each RRH. Finally, precoded signals s1 and s2 are transmitted from RRH 1 as radio frequency (RF) signals s1RF and s2RF, and precoded signals s3 and s4 are transmitted from RRH 2 as RF signals s3RF and s4RF. Since these output RF signals are the same as the output RF signals obtained at each RRH in conventional C-RAN, the downlink joint transmission scheme in the SPP architecture can realize the same performance as conventional C-RAN.
Figure 4(a) shows the uplink joint reception scheme in the SPP architecture, while Fig. 4(b) shows that in conventional C-RAN. We assume 1 user equipment (UE) with 4 antennas, and 2 RRHs, each with 2 antennas. Each RRH in the SPP architecture performs 2 x 4 MIMO equalization using only 2 signals received at each RRH out of the 4 received signals r1, r2, r3 and r4, while each RRH in conventional C-RAN forwards IQ data of received signals to BBU and BBU in conventional C-RAN performs 4 x 4 MIMO equalization using all 4 received signals. The matrix sizes in MIMO equalization means that conventional C-RAN uses a larger number of receive antennas than the SPP architecture, yielding higher wireless transmission performance. The joint reception scheme in the SPP architecture complements the smaller number of received antennas in MIMO equalization with LLR combining in BBU. Each RRH in the SPP architecture obtains the LLRs after 2 x 4 MIMO equalization and wireless demodulation. l11, l12, l13 and l14 denote the LLRs corresponding to the transmitted signals s1RF, s2RF, s3RF and s4RF obtained in RRH 1, while l21, l22, l23 and l24 denote LLRs obtained in RRH 2. Then, each RRH forwards them over the MFH, and BBU in the SPP architecture combines these LLRs. For example, combined LLR lc1 is obtained by simply adding l11 and l21. Finally, BBU performs wireless channel decoding using the combined LLRs and so can enhance the wireless transmission performance.
3. Performance analysis
3.1 Mobile fronthaul bandwidth
We evaluated the radio related MFH bandwidth in the SPP architecture and compared it with that in conventional C-RAN by numerical simulation, and so we excluded the factors for control overhead such as 16/15 [6] and line coding such as 8B10B or 64B66B in CPRI or other optical IF. We assumed that multiple UEs were randomly distributed in a 3-sector macro cell area and calculated the SNR of each UE from received RF power and noise power. The MFH bandwidth in the SPP architecture is derived from the modulation scheme of each UE according to its SNR [17]. When UE k selects its modulation scheme, the MFH bandwidth in the downlink BSPP-DL and that in the uplink BSPP-UL are derived as
where TTTI, Nmod-DL,k, Nmod-UL,k, Nsym, NSC, NRB, Nst-DL, Nst-UL, Nq-LLR are the transmission time interval (TTI), the modulation order of UE k in the downlink, the modulation order of UE k in the uplink, the number of symbols within TTI, the number of subcarriers per resource block (RB), the number of RBs per UE, the number of MIMO streams in the downlink, the number of MIMO streams in the uplink and the number of quantization bits for LLR, respectively. On the other hand, the MFH bandwidth of CPRI in conventional C-RAN BC is derived aswhere fs, Nq-IQ, Nant and NCA are the sampling frequency, the number of quantization bits for IQ data, the number of antennas in the RRH and the number of carrier aggregation (CA) [18], respectively. Sampling frequency fs is 30.72 MHz corresponding to wireless signal of 20-MHz bandwidth, and system bandwidth is 100 MHz which is a carrier aggregation of 5 component carriers of 20 MHz. Table 1 shows the simulation parameters which are derived from LTE and CPRI specifications [6,18]. The number of quantization bits for LLR was 2 in the SPP architecture, because we assume that non-linear quantization is employed for LLR quantization, where the quantization boundaries and the quantization levels are determined according to the probability distributions of the LLR values [19].
Table 1. Simulation Parameters for Bandwidth Evaluation
Figure 5 shows the MFH bandwidth per link versus UE number for conventional C-RAN and the SPP architecture. The maximum values when all UEs used the highest-order modulation scheme are also plotted in addition to simulated average values for the SPP architecture. As seen from Fig. 5, the conventional C-RAN requires the fixed bandwidth of 36.9 Gbps which corresponds to CPRI bandwidth of 49.2 Gbps, and it is independent of the UE number. Meanwhile, the bandwidth for the SPP architecture varies with UE number. When 100 UEs use all radio frequency resources (500 RBs), the average values are 1 Gbps in both directions, while the maximum values are 2.3 Gbps in the downlink and 1.7 Gbps in the uplink. These results confirm that the MFH bandwidth in the SPP architecture is reduced by 94% for maximum values and 97% for average values compared to conventional C-RAN. The SPP architecture also achieves the reduced bandwidth target of 10 Gbps as mentioned above. Moreover, the SPP architecture can obtain statistical multiplexing gain since the MFH bandwidth varies according to mobile traffic.
3.2 Wireless transmission performance
We evaluated the wireless transmission BER performance of the uplink joint reception scheme in the SPP architecture, since we need to evaluate the performance improvement achieved by LLR combining. We assumed that a UE with 2 antennas was located at the midpoint of 2 coordinated RRHs, each with 8 antennas, and thus the received SNR values at RRH 1 and RRH 2 were the same. We compared the BER performance in 3 cases: joint reception in conventional C-RAN, the joint reception in the SPP architecture and the BER performance without joint reception. Table 2 shows the simulation parameters; 3 modulation and coding schemes (MCSs) of QPSK, 16QAM and 64QAM were used. We also employed non-linear quantization utilizing probability distribution of LLR in the SPP architecture. Since we focus on the wireless transmission performance, we assumed the optical transmission which meets the optical BER requirement such as 10−12 stated in CPRI [6].
Table 2. Simulation Parameters for Wireless Transmission Performance Evaluation
Figure 6 shows the wireless transmission BER performance of uplink CoMP joint reception for the 3 cases with the 3 MCSs. The required BER values which achieve the 10% block error rate target [20] are 1.8 x 10−4 for QPSK, 4.1 x 10−5 for 16QAM, 2.4 x 10−5 for 64QAM. The degradation in required SNR for the SPP architecture from conventional C-RAN is less than 2 dB for all 3 MCSs examined, and the improvement in required SNR for the SPP architecture from the performance without joint reception is more than 3 dB for all 3 MCSs. These results confirm that the uplink joint reception scheme in the SPP architecture achieves wireless transmission performance close to that of conventional C-RAN.
4. Conclusion
Toward the future radio access beyond 2020, we analyzed the MFH bandwidth and the wireless CoMP performance in the SPP architecture which we have proposed to reduce the MFH bandwidth to less than 10 Gbps and realize high CoMP performance close to that of conventional C-RAN. Simulation results confirmed that the SPP architecture reduces the MFH bandwidth by 94% for maximum values and 97% for average values compared to conventional C-RAN, while matching the wireless BER performance of conventional C-RAN in uplink joint reception with only 2-dB SNR penalty for QPSK, 16QAM and 64QAM. These results confirmed validity of the SPP architecture to realize cost-effective MFH deployment.
References and links
1. NTT DOCOMO white paper, “DOCOMO 5G white paper, 5G radio access: requirements, concept and technologies,” https://www.nttdocomo.co.jp/english/binary/pdf/corporate/technology/whitepaper_5g/DOCOMO_5G_White_Paper.pdf.
2. NGMN whitepaper, “NGMN 5G white paper,” https://www.ngmn.org/uploads/media/NGMN_5G_White_Paper_V1_0.pdf.
3. China Mobile Research Institute white paper, “C-RAN the road towards green RAN white paper version 3.0,” http://labs.chinamobile.com/cran/wp-content/uploads/2014/06/20140613-C-RAN-WP-3.0.pdf.
4. NGMN Technical Report, “Suggestions on potential solutions to C-RAN,” http://www.ngmn-ic.info/uploads/media/NGMN_CRAN_Suggestions_on_Potential_Solutions_to_CRAN.pdf.
5. 3GPP, TS 36.819, v11.2.0, “Coordinated multi-point operation for LTE physical layer aspects,” http://www.3gpp.org/DynaReport/36819.htm.
6. CPRI, “CPRI specification v6.1,” (2014), http://www.cpri.info/jp/spec.html.
7. B. Guo, W. Cao, A. Tao, and D. Samardzija, “LTE/LTE-A signal compression on the CPRI interface,” Bell Labs Tech. J. 18(2), 117–133 (2013). [CrossRef]
8. K. Tanaka and A. Agata, “Next-generation optical access networks for C-RAN,” in Optical Fiber Communication Conference, OSA Technical Digest (online) (Optical Society of America, 2015), paper Tu2E.1.
9. ETSI specification, “Open radio equipment interface (ORI); ORI interface specification (release-4),” (2014), http://www.etsi.org/technologies-clusters/technologies/ori.
10. U. Dötsch, M. Doll, H.-P. Mayer, F. Schaich, J. Segel, and P. Sehier, “Quantative analysis of split base station processing and determination of advantageous architectures for LTE,” Bell Labs Tech. J. 18(1), 105–128 (2013). [CrossRef]
11. A. Maeder, M. Lalam, A. De Domenico, E. Pateromichelakis, D. Wübben, J. Bartelt, R. Fritzsche, and P. Rost, “Towards a flexible functional split for Cloud-RAN networks,” in Proceedings of 2014 European Conference on Networks and Communications (EuCNC, 2014), pp. 1–5. [CrossRef]
12. NGMN, “Future study on critical C-RAN technologies,” (2015), https://www.ngmn.org/uploads/media/NGMN_RANEV_D2_Further_Study_on_Critical_C-RAN_Technologes_v1.0.pdf.
13. China Mobile Research Institute, “White Paper of Next Generation Fronthaul Interface,” (2015), http://labs.chinamobile.com/cran/wp-content/uploads/2015/09/NGFI-Whitepaper_EN_v1.0_201509291.pdf.
14. K. Miyamoto, S. Kuwano, J. Terada, and A. Otaka, “Split-PHY processing architecture to realize base station coordination and transmission bandwidth reduction in mobile fronthaul,” in Optical Fiber Communication Conference, OSA Technical Digest (online) (Optical Society of America, 2015), paper M2J.4. [CrossRef]
15. K. Miyamoto, S. Kuwano, J. Terada, and A. Otaka, “Uplink joint reception with LLR forwarding for optical transmission bandwidth reduction in mobile fronthaul,” in Proceedings of Vehicular Technology Conference2015-Spring (VTC2015-Spring), pp. 1–5.
16. M. Ding, J. Zou, Z. Yang, H. Luo, and W. Chen, “Sequential and incremental precoder design for joint transmission network MIMO systems with imperfect backhaul,” IEEE Trans. Vehicular Technol. 61(6), 2490–2503 (2012). [CrossRef]
17. 3GPP, R1–081483, Qualcomm Europe, “Conveying MCS and TB size via PDCCH,” (3GPP, 2008), http://www.3gpp.org/DynaReport/TDocExMtg–R1-52b–26911.htm.
18. 3GPP, TS 36.300, v13.1.0, “Evolved universal terrestrial radio access (E-UTRA) and evolved universal terrestrial radio access network (EUTRAN); Overall description; Stage 2,” (3GPP, 2015), http://www.3gpp.org/DynaReport/36300.htm.
19. S. Schwandter, P. Fertl, C. Novak, and G. Matz, “Log-likelihood ratio clipping in MIMO-BICM systems: Information Geometric Analysis and Impact on System Capacity,” in Proceedings of International Conference on Acoustics, Speech and Signal Processing (ICASSP, 2009), pp. 2433–2436. [CrossRef]
20. 3GPP, TS 36.912, v12.0.0, ”Feasibility study for future advancements for E-UTRA (LTE-Advanced),” (3GPP, 2014), http://www.3gpp.org/DynaReport/36912.htm.
21. 3GPP, TS 36.104, v13.1.0, “Evolved universal terrestrial radio access (E-UTRA); Base station (BS) radio transmission and reception,” (3GPP, 2015), http://www.3gpp.org/DynaReport/36104.htm.
