Home > 5G Network > Fronthaul Design In Cloud Radio Access Networks: A Survey

Fronthaul Design In Cloud Radio Access Networks: A Survey

6  Fronthaul Compression

The fronthaul links carry quantized IQ samples from RRH to BBU pool in uplink; and from BBU pool to RRH in the downlink. The fronthaul links are supposed to meet the upcoming 5G standards in terms of low latency and high data rates. The enormous data rates produced by the quantized IQ samples exploit fronthaul capacity limitation, which is the main bottleneck in the commercial deployment of C-RAN. This fronthaul capacity limitation will degrade the Large Scale Collaborative Processing (LSCP) gain that could be achieved through C-RAN architecture. Ex:- With densely deployed RRHs and with a UE operating around a few MHz bandwidths will increase the data rate of fronthaul link that connects the particular RRH to BBU pool will scale up to several Gbps. Hence, uplink and downlink fronthaul compression are required. There are two possible ways to achieve fronthaul compression. One of them is partial C-RAN architecture. In this architecture, RRHs are given some responsibility to do some signal processing that could be done at the BBU-pool, but this will oppose the whole idea of easy network densification to achieve network capacity. Hence the design of robust fronthaul compression techniques is needed that can support the full-centralization of C-RAN architecture.

There are several fronthaul compression techniques available to reduce the fronthaul IQ data rate, which can be used depending on the requirement. The point-to-point compression technique can be used in low complexity scenarios. point-to-point compression techniques are generally the basic IQ sample quantization methods. But point-to-point compression alone cannot serve the purpose in many cases. point-to-point compression can be used along with network-aware compression techniques like distributed source coding and joint compression. These network-aware compression techniques bring complexity to the analysis. Spatial filtering and Compressive Sensing (CS) based compression are other types of compression techniques. Like a point-to-point compression technique, spatial filtering is also a low complexity technique. The original idea behind the CS-based compression is to exploit the signal sparsity among UEs. Hence CS-based compression and spatial filtering cannot be used for the downlink signal compression. As part of CS-based compression and spatial filtering, the received UE’s signal will be multiplied with the local RRH matrix. This will lead to a reduction in the dimension of the received signal. Similarly, distributed source coding can only be used for the uplink compression. The basic idea of distributed source coding is to exploit the correlation among the received signals at RRH. Another network-aware compression technique, joint compression, can be used at the BBU side for downlink compression.

In the following subsections, the uplink compression and downlink compression are discussed. Mostly, the compression techniques discussed under the uplink and downlink compression subsections are not simple point-to-point compression techniques. The quantization-based and similar point-to-point compression techniques are discussed separately under point-to-point compression subsection.

  • A. Uplink Compression:

    The uplink compression techniques are mainly two types, point-to-point compression, and multiterminal compression. The point-to-point compression at the uplink side is shown in Figure. 4, where the received signals at each RRH are compressed and transmitted to the BBU pool on a capacity limited fronthaul. At the BBU pool side, the compressed signals from each RRH are decompressed separately, but all these decompressed signals are decoded jointly. This is based on the assumption that the BBU does joint decoding based on all quantization values from all RRHs. But, in multiterminal compression, as shown in Figure. 5, the decompressor takes the correlation among the received RRH signals into account, and decompress accordingly.


    Since the signals received at different RRHs are correlated, the distributed source coding based fronthaul compression can be considered as an alternative to reduce the fronthaul data rate by removing the redundant information. However, to implement this technique, each RRH requires information about the joint statistics of the received signals across other RRHs. As part of this compression technique, at each step, an RRH compresses its received signal based on the statistical information of the compressed signal of active RRHs at the previous step. These RRHs are generally sensitive to uncertainties regarding such as correlation of the received signal’s information. The imperfect knowledge of the joint statistics of the signals received at RRHs can lead to performance degradation of distributed source coding. Distributed source coding is based on the idea of reducing the rate of the compressed stream by incorporating some uncertainty in the compressed stream. This uncertainty is resolved with the help of side information. The amount of rate reduction that is allowed without decompression errors is dictated by the quality of side information, which should be known at the cloud decoder. Hence, the lack of side information at the cloud decoder may result in decompression error of received signal at RRH.

    By taking these correlation errors into account, a robust distributed uplink compression scheme at multi-antenna RRHs is proposed in [51], where distributed compression is implemented with the help of sequential source coding and side information. The knowledge of joint statistics at each RRH is represented in a covariance matrix, which is considered to be imperfect. The uncertainty is modeled using a deterministic additive error model with bounds on the eigenvalues of the covariance matrix. Bounding these eigenvalues is equivalent to bounding any norm of the error. The problem is formulated as a worst-case deterministic approach problem. The stationary solution to this problem is achieved by solving KKT conditions. The simulation results suggest that, even with sizable errors, the proposed compression scheme gave the benefits of the distributed compression scheme. Whereas, the errors in the statistical model of the side information makes the distributed source coding useless.

    Moreover, the network energy efficiency problem is addressed by selecting active RRHs. A joint optimization problem of compression and RRH selection is formulated by introducing sparsity inducing terms in the objective function. To solve this problem, an iterative block-coordinate ascent algorithm is proposed, which is converged to a locally optimal point. The quantization noise introduced while doing the compression operation, which is the critical parameter of fronthaul compression. The quantization noise level is optimized based on a per-RRH base.

    The performance evaluation of standard point-to-point and multiterminal fronthaul compression techniques for both uplink and downlink of C-RAN are studied in [52]. The simulation results are carried over standard cellular networks by focusing on performance metrics such as proportional-fairness utility, sum-rate, and cell-edge throughput. The multiterminal fronthaul compression showed 60% performance gain more compared to that of standard point-to-point fronthaul compression. Hence, it is claimed that the point-to-point fronthaul compression techniques fail to achieve optimal performance in even the simplest multiterminal settings of C-RAN.

    The work proposed in [53] deals with the optimization of fronthaul compression. Here uplink C-RAN model is considered, where the fronthaul link is noiseless and has finite sum capacity constraint. The independent messages sent from each UE within a cooperating cluster, interfere with other UE’s messages at their respective RRH. The RRHs compress and forward these received UE messages, which are decoded successively at the centralized processor. The quantization codewords are decoded initially. Later, based on the quantized signals from all the RRHs, the UE’s messages are decoded. This compression, decompression, and decoding operations are done within the constant gap to the sum capacity of the network. In this way, the uplink C-RAN model can be thought of as a Virtual Multiple Access Channel (VMAC) between UEs and BBU pools while RRHs acting as relays. Distributed Wyner-Ziv (WZ) coding, and Single-User (SU) compression are considered as the candidates for compression at RRH. These compression techniques are named by the authors as VMC-WZ and VMAC-SU, respectively. The main objective of this work is to set the optimization levels for quantization noise. The quantization noise level of all RRHs is optimized jointly. The optimization of quantization noise levels for VMAC-ZV is formulated as a weighted sum-rate maximization under fronthaul capacity constraints, which is solved through alternating convex optimization algorithm. For VMAC-SU, the problem is reformulated in terms of optimizing fronthaul capacities. It was observed that, during the high Signal-to-Quantization-Noise-Ratio (SQNR) times, setting the quantization noise levels proportional to background noise levels, irrespective of channel conditions and transmission power is near-optimal to attain maximum sum rate. With that choice of quantization noise level, VMAC-ZV schemes achieve the sum capacity of the uplink C-RAN model within the constant gap. VMAC-SU produces a similar constant gap with diagonally dominant channel conditions. Since VMAC schemes have low decoding complexity and low decoding delay as compared to joint decoding, these consistent gap results motivate to extend this VMAC scheme in practical C-RAN.

    The benefits of exploiting the uplink signal sparsity are not utilized in [52], [53], and hence the uplink signal sparsity structure is ignored in those works. Because of the uplink delay-sensitive services or the random access of the UEs in the uplink transmission, the transmission is bursty, which leads to the sparsity of the uplink packets. This uplink sparsity can be used to reduce the fronthaul loading. Classical distributed CS technique can’t be used directly for C-RAN. In traditional Classical distributed CS, signals are compressed distributively at different sensors, and later the joint recovery of these signals is made. Like conventional distributed CS technique, the uplink signals are compressed locally at RRHs. Still, these signals are an aggregation of multi-users transmitted from different UEs over multiaccess fading channels. Hence in C-RAN, the target of joint recovery at the BBU pool is the transmitted signals from UEs rather than the locally received signals from RRHs. So, if a distributed CS algorithm is used for joint signal recovery in C-RAN, it should incorporate multiaccess fading effects of C-RAN in it. For Robust CS recovery, restricted isometry property (RIP) is commonly adopted to derive sufficient conditions; but establishing adequate conditions for the RIP of the associated measurement matrix is non-trivial. Conventional results about the RIP condition can’t be applied for the aggregate CS measurement matrix, because of the complicated multiaccess fading channels between UE and RRHs in C-RAN system. Hence, by embracing multiaccess fading, a new characterization of the sufficient conditions for robust CS recovery should be done. While applying CS signal recovery to achieve fronthaul compression, it is also important to quantify a closed-form trade-off between uplink C-RAN capacity and the fronthaul loading at each RRH. All these challenges are addressed in [54] by proposing an uplink C-RAN model and distributed fronthaul compression at each RRH. A joint signal recovery is made at the BBU pool by using distributed CS, which exploits multiaccess fading between the UEs and RRHs and the signal sparsity of UEs. The aggregated CS measurement matrix contains both the multiaccess fading and distributed fronthaul compression. The performance of the end-to-end signal recovery algorithm is analyzed, and it was showed that the aggregated CS measurement matrix satisfies the RIP conditions with high probability under some mild conditions. Based on these results, the correct probability of active user detection is analyzed. Furthermore, the achievable uplink capacity is characterized in terms of the fronthaul compression rate, which will help conclude uplink performance and fronthaul loading.

    CS is an effective method for signal acquisition and processing. The key idea of this technique is that the collected signals are not purely random, and hence can be compressed because of the redundancy. So, this nature can be taken as prior knowledge for compression and recompression of the collected signals. Therefore, CS provides a new sampling paradigm for sparse signals and enables the decoder to retrieve the original signal accurately from the reduced number of samples. The CS-based distributed fronthaul compression in [54] considers the uplink random access channel scenario between UEs and RRHs. Still, in general, only the active UEs and the uplink signals are sparse. Thus, by considering the correlation of the received uplink baseband signals at different RRHs within one cluster, a compressive sensing based uplink compression is proposed in [55]. To save the transport expenses in a multi-clustered C-RAN, one cluster Head (CH) collects the information from all the RRHs and sends it to the BBU pool. Since the number of RRHs are huge in a C-RAN cluster, the amount of information collected at CH would be too high to transmit over a capacity limited fronthaul. Whereas, the collected information at different RRHs within one cluster is correlated in nature because of the multiaccess fading effect. So, the CH performs CS-based fronthaul compression before transmission, and the BBU implements the joint recovery to reconstruct the gathered uplink signals by RRHs. Later, the original baseband signals of all the UEs are decoded at the end using a Successive Interference Cancellation (SIC) mechanism. The uplink rate increases with the UE SNR and the SIC implemented at the BBU side. It was also observed that the SNR increases with the fronthaul capacity, and the maximum SNR can be reached when the measurement channel capacity equals the fronthaul capacity.

    All the works till now studied the fronthaul compression schemes, in which the RRH communicates to the BBU pool by itself. The fronthaul compression scheme for multihop C-RAN is considered in [56], where each RRH manages to communicate with BBU through a set of intermediate RRHs. In this work, the generalized C-RAN model is followed, where each RRH either compresses and forwards or forwards the received baseband signal without any demodulation. Initially, a baseline Multiplex-and-Forward (MF) scheme is studied, in which each RRH receives the bitstream from a connected RRH and forwards it without any processing. It was observed that the MF scheme leads to performance degradation in the case when each RRH has many incoming links from other RRHs. To overcome this problem, the Decompress-Process-and-Recompress (DPR) scheme is proposed. In the DPR scheme, RRH decompresses the received bitstream and performs the linear in-network processing of the decompressed signals. Both MF and DPR fronthaul schemes are designed to maximize the sum-rate under the limited fronthaul capacity constraints. The advantage of in-network processing is highlighted by comparing the performance of MF and DPR strategies. The optimal solutions in the DPR scheme demand full CSI, but acquiring full CSI at every RRH is difficult in the case of dense C-RAN. Hence a decentralized DPR scheme for the optimal solution is proposed, where the RRHs require only local CSI to compute linear processing. The future extension of this work is the consideration of non-linear in-network processing at the RRHs, investigating the ways to utilize the side information available at intermediate RRHs, and the consideration of imperfect CSI of other RRHs at every RRH.

    Multiple antenna RRHs are considered while study-ing the uplink C-RAN compression technique in [57]. In order to exploit this multi-antenna spatial sparsity, a Spatial-Compression-and-Forward (SCF) technique is proposed. Rather than using a complex quantization scheme like the authors did in [51], [53], a less complex linear filter is considered for compressing the received correlated signals at all the antennas of each RRH. An optimization problem is framed to maximize the minimum SNR of all the users. The optimization problem considers the joint optimization of the user’s power allocation, RRH’s spatial filter design and quantization bits allocation, and BBU’s receive beamforming.

    A localization problem of positioning a single antenna radio transmitter in the C-RAN system is studied in [58]. The problem is formulated as the quantization strategy optimization while minimizing the worst case localization error under fronthaul capacity constraints. The Charnes-Cooper transformation and difference-of-convex programming are utilized to design corresponding optimization algorithms. From the numerical simulations, it was observed that the localization becomes better with the large fronthaul capacity. It was also shown that the fronthaul rate requirements are less for localization when compared to data communication.

Pages ( 9 of 12 ): « Previous1 ... 78 9 101112Next »

Leave a Comment:

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.