Multi USER DETECTION
Consider again the functional split option, Rx Data Forwarding
(C) in Figure 3. In this case, I/Q samples are forwarded over highcapacity
BH links to the central processor that performs joint
multiuser detection (MUD) using the Rx signals of several RAPs.
The joint processing of many RAPs implements a virtual MIMO
architecture and the huge computational power offered by the
cloud-platform allows for aggressive RRM across the RAPs. However,
due to the heterogeneous nature of BH networks, it is also
beneficial to use a mix of local processing at RAPs, cooperative
processing among RAPs, and central processing in the cloud-platform.
Promising techniques that are adaptable to changing
BH
and radio access parameters are, among others, multipoint
turbo detection (MPTD) and in-network processing (INP).
The underlying idea of MPTD [33] is to schedule (edge)
users attached to different RAPs on the same resource. Then, a
joint detection of these users through a turbo processing
approach is performed [21], [34]. Such processing could be
done either centrally on the cloud-platform or locally in each
RAP. If it is fully centralized, MPTD benefits from high degree
of spatial diversity due to the different locations of the
involved RAPs. Due to this spatial diversity increase, the centralization
gain can be quite significant compared to a classical
distributed detection.
This split of functionality may offer significant centralization
gains compared to distributed detection methods. This is
illustrated in Figure 6 for an UL scenario with NUE = 2 users
each equipped with NT = 1 transmit antenna. Both users
interfere with each other at NRAP = 2 RAPs each equipped
with NR = 2 receive antennas. We assume the worst case of
identical path-losses. In addition, these results consider Rayleigh
channel fading and LTE-compliant MCSs [20]. Figure 6
shows that at a frame error rate (FER) of 0.01 a centralization