Abstract
In cell-free massive MIMO networks, a large number of distributed access points (APs) provide service to a much smaller number of mobile stations (MSs) over the same time/frequency resources. The key idea is to use a central processing unit (CPU) to manage such a densely populated network of APs. This centralization helps reducing operational costs and eases implementation of joint power control and coherent signal processing through a proper orchestration of the functional split between the CPU and the APs. Cell-free massive MIMO networks, however, are often subject to stringent capacity requirements on the fronthaul links connecting the APs to the CPU and thus, low-resolution ADCs must be used to quantize the signals shared among CPU and APs. In this paper, analytical closed-form expressions for the achievable user rates on both the uplink (UL) and downlink (DL) of a fronthaul-capacity constrained cell-free massive MIMO network using low-resolution ADCs are obtained. These expressions, jointly with the use of theoretical models characterizing the fronthaul capacity consumption of different CPU-AP functional splits, allow posing max-min fairness power control optimization problems that can be solved using standard convex optimization algorithms. Numerical results show that, under fronthaul capacity constraints, CPU-AP functional splits where the precoding/decoding schemes are implemented at the APs are clearly outperformed by those functional splits in which, thanks to sharing CSI among APs and CPU, the precoding/decoding functions are implemented at the CPU. In contrast, if the limiting factor is the resolution of the ADCs used to quantize the samples to be transmitted on the fronthaul links, the preferred CPU-AP functional splits are those in which the baseband processing is performed at the APs. Moreover, they also reveal that in such functional splits there is always an optimal range of values of the UL fronthaul capacity fraction allocated to share the CSI.
Highlights
Massive multiple-input multiple-output (MIMO) has recently emerged as one of the fundamental physical layer pillars of the so-called 5G and beyond-5G wireless networks [1], [2]
The impact of using low-resolution analog-to-digital converters (ADCs) to transmit over capacity-constrained fronthaul links has been addressed by resorting to the use of a linear additive quantization noise model that is based on the Bussgang decomposition and is sometimes referred to as the AQNM
If the limiting factor is the resolution of the ADCs used to quantize the samples to be transmitted on the fronthaul links, the preferred central processing unit (CPU)-access points (APs) functional splits are those in which the baseband processing is performed at the APs
Summary
Massive multiple-input multiple-output (MIMO) has recently emerged as one of the fundamental physical layer pillars of the so-called 5G and beyond-5G wireless networks [1], [2]. Our main aim in this paper is to fill in the gap left by previous research work on this topic by presenting a realistic general framework allowing a fair comparison between different CPU-AP functional splits in both the UL and DL of a fronthaul-constrained cell-free massive MIMO network using low-resolution ADCs. Specific contributions of this paper can be summarized as:. The quantizer noise model described by Mezghani and Nossek in [30], which is based on the Bussgang decomposition [31] and is sometimes referred to as the additive quantization noise model (AQNM) [32], is used to characterize the fronthaul bandwidth consumption of different CPU-AP functional splits, allowing us to provide a thorough comparison among them and discuss the impact they may have on the global performance of the network As it will be shown over the sections, CPU-AP functional splits performing the baseband signal processing operations at the VOLUME 8, 2020. CN (m, R) denotes a complex Gaussian vector distribution with mean m and covariance R, whose zero-mean part constructed by subtracting its mean is circularly symmetric, and N (0, σ 2) denotes a real valued zero-mean Gaussian random variable with standard deviation σ
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
