Comparing Lower Layer Splits for Open Fronthaul Deployments

Introduction

The transition to open RAN (Radio Access Network) based on interoperable lower layer splits is gaining significant momentum across the mobile industry. However, where best to split the open RAN is a complex compromise between radio unit (RU) simplification, support of advanced co-ordinated multipoint RF capabilities, consequential requirements on the fronthaul transport, including limitations on transport delay budgets as well as bandwidth expansion. To help in comparing alternative options, different splits have been assigned numbers with higher numbers representing splits “lower down” in the protocol stack, meaning less functionality being deployed “below” the split in the RU. Lower layer splits occur below the medium access control (MAC) layer in the protocol stack, with options including Split 6 – between the MAC and physical (PHY) layers, Split 7 – within the physical layer, and Split 8 – between the physical layer and the RF functionality.

Figure 1: Different Lower Layer Splits in the RAN Protocol Stack

This paper compares the two alternatives for realizing the lower layer split, the network functional application platform interface (nFAPI) Split 6 as defined by the Small Cell Forum (SCF) and the Split 7-2x as defined by the O-RAN Alliance.

Small Cell Splits

The Small Cell Forum took the initial lead in defining a multivendor lower layer split, taking its FAPI platform application programming interface (API) that had been used as an informative split of functionality between small cell silicon providers and the small cell RAN protocol stack providers, and enabling this to be “networked” over an IP transport. This “networked” FAPI, or nFAPI, enables the Physical Network Function (PNF) implementing the small cell RF and physical layer to be remotely located from the Virtual Network Function (VNF) implementing the small cell MAC layer and upper layer RAN protocols. First published by the SCF in 2016, the specification of the MAC/PHY split has since been labelled as “Split 6” by 3GPP TR38.801 that studied 5G’s New Radio access technology and architectures.

The initial SCF nFAPI program delivered important capabilities that enabled small cells to be virtualized, compared with the conventional macro-approach that at the time advocated using the Common Public Radio Interface (CPRI) defined split. CPRI had earlier specified an interface between a Radio Equipment Control (REC) element implementing the RAN baseband functions and a Radio Equipment (RE) element implementing the RF functions, to enable the RE to be located at the top of a cell tower and the REC to be located at the base of the cell tower. This interface was subsequently repurposed to support relocation of the REC to a centralized location that could serve multiple cell towers via a fronthaul transport network.

Importantly, when comparing the transport bandwidth requirements for the fronthaul interface, nFAPI/Split 6 does not significantly expand the bandwidth required compared to more conventional small cell backhaul deployments. Moreover, just like the backhaul traffic, the nFAPI transport bandwidth is able to vary according to served traffic, enabling statistical multiplexing to be used over the fronthaul IP network. This can be contrasted with the alternative CPRI split, also referred to as “Split 8” in TR38.801, that requires bandwidth expansion up to 30-fold and a constant bit rate connection, even if there is no traffic being served in a cell.

HARQ Latency Constraints

Whereas nFAPI/Split 6 offers significant benefits over CPRI/Split 8 in terms of bandwidth expansion, both splits are below the hybrid automatic repeat request (HARQ) functionality in the MAC layer that is responsible for constraining the transport delay budget for LTE fronthaul solutions. Both LTE-based Split 6 and Split 8 have a common delay constraint equivalent to 3 milliseconds between when up-link data is received at the radio to the time when the corresponding down-link ACK/NAK needs to be ready to be transmitted at the radio. These 3 milliseconds need to be allocated to HARQ processing and transport, with a common assumption being that 2.5 milliseconds are allocated to processing, leaving 0.5 milliseconds allocated to round trip transport. This results in the oft-quoted delay requirement of 0.25 milliseconds for one way transport delay budget between the radio and the element implementing the MAC layer’s up-link HARQ functionality.

The Small Cell Forum acknowledges such limitations when using its nFAPI/Split 6. Because the 0.25 milliseconds round trip transport budget may severely constrain nFAPI deployments, SCF defines the use of HARQ interleaving that uses standardized signaling to defer HARQ buffer emptying, enabling higher latency fronthaul links to be accommodated. Although HARQ interleaving buys additional transport delay budget, the operation has a severe impact on single UE throughput; as soon as the delay budget exceeds the constraint described above, the per UE maximum throughput is immediately decreased by 50%, with further decreases as delays in the transport network increase.

Importantly, 5G New Radio does not implement the same synchronous up-link HARQ procedures and therefore does not suffer the same transport delay constraints. Instead, the limiting factor constraining the transport budget in 5G fronthaul systems is the operation of the windowing during the random access procedure. Depending on the operation of other vendor specific control loops, e.g., associated with channel estimation, this may enable increased fronthaul delay budgets to be used in 5G deployments.

O-RAN Alliance

The O-RAN Alliance published its “7-2x” Split 7 specification in February 2019. All Split 7 alternatives offer significant benefits over the legacy CPRI/Split 8, avoiding Split 8 requirements to scale fronthaul bandwidth on a per antenna basis, resulting in significant lower fronthaul transport bandwidth requirements, as well introducing transport bandwidth requirements that vary with served traffic in the cell. Moreover, when compared to Split 6, the O-RAN lower layer Split 7-2x supports all advanced RF combining techniques, including the higher order multiple-input, multiple-output (MIMO) capability that is viewed as a key enabling technology for 5G deployments, as shown in Table 1, that can be used to contrast Split 6 “MAC/PHY” with Split 7 “Split PHY” based architectures.

Table 1: Comparing Advanced RF Combining Capabilities of Lower Layer Splits

However, instead of supporting individual transport channels over the nFAPI interface,  Split 7-2x defines the transport of frequency domain IQ defined spatial streams or MIMO layers across the lower layer fronthaul interface. The use of frequency domain IQ symbols can lead to a significant increase in fronthaul bandwidth when compared to the original transport channels. Figure 2 illustrates the bandwidth expansion due to Split 7-2 occurring “below” the modulation function, where the original 4 bits to be transmitted are expanded to over 18 bits after 16-QAM modulation, even when using a block floating point compression scheme.

Figure 2: Bandwidth Expansion with Block Floating Point Compressed Split 7-2x

The bandwidth expansion is a function of the modulation scheme, with higher expansion required for lower order modulation, as shown in Table 2.

Table 2: Bandwidth Expansion for Split 7-2x with Block Floating Point Compression compared to Split 7-3

Such a bandwidth expansion was one of the reasons that proponents of the so called Split 7-3 advocated a split that occurred “above” the modulation/demodulation function. In order to address such issues, and the possible fragmentation of different Split 7 solutions, the O-RAN Alliance lower layer split includes the definition of a technique termed modulation compression. The operation of modulation compression of a 16-QAM modulated waveform is illustrated in Figure 3. The conventional Split 7-2 modulated constellation diagram is shifted to enable the modulation points to lie on a grid that then allows the I and Q components to be represented in binary instead of floating point numbers. Additional scaling information is required to be signalled across the fronthaul interface to be able to recover the original modulated constellation points in the RU, but this only needs to be sent once per data section.

Figure 3: User Plane Bandwidth Reduction Using Modulation Compression with Split 7-2x

Because modulation compression requires the in-phase and quadrature points to be perfectly aligned with the constellation grid it can only be used in the downlink.  However, when used, it decreases the bandwidth expansion ratio of Split 7-2x, where the expansion compared to Split 7-3 is now only due to the additional scaling and constellation shift information. This information is encoded as 4 octets and sent every data section, meaning the bandwidth expansion ratio will vary according to how many Physical Resource Blocks (PRBs) are included in each data section. This value can range from a single PRB up to 255 PRBs, with Table 3 showing the corresponding Split 7-2x bandwidth expansion ratio over Split 7-3 is effectively unity when operating using large data sections.

Table 3:  Bandwidth Expansion for Split 7-2x with Modulation Compression compared to Split 7-3

Note, even though modulation compression is only applicable to the downlink (DL), the shift of new frequency allocations to Time Division Duplex (TDD) enables a balancing of effective fronthaul throughput between uplink (UL) and downlink. For example, in LTE, 4 of the 7 possible TDD configurations have more slots allocated to downlink traffic, compared to 2 possible configuration that have more slots allocated in the uplink. Using a typical 12-to-6 DL/UL configuration, with 256-QAM and 10 PRBs per data section, the overall balance of bitrates for modulation compression in the downlink and block floating point compression in the uplink will be (1.03 x 12) to (2.33 x 6), or 12.40:13.98, i.e., resulting in a relatively balanced link as it relates to overall bandwidth.

A more comprehensive analysis by the O-RAN Alliance has examined control and user-plane scaling requirements for Split 7-2x with modulation compression and compared the figures with those for Split 7-3. When taking into account other overheads, this analysis indicated that the difference in downlink bandwidth between Split 7-3 and Split 7-2x with Modulation Compression was estimated to be around 7%. Using such analysis, it is evident why the O-RAN Alliance chose not to define a Split 7-3, instead advocating a converged approach based on Split 7-2x that can be used to address a variety of lower layer split deployment scenarios.

Comparing Split 7-2x and nFAPI

Material from the SCF clearly demonstrates that, in contrast to Split 7, their nFAPI/Split 6 approach is challenged in supporting massive MIMO functionality that is viewed as a key enabling technology for 5G deployments. However, massive MIMO is more applicable to outdoor macro-cellular coverage, where it can be used to handle high mobility and suppress cell-edge interference use cases. Hence, there may be a subset of 5G deployments where massive MIMO support is not required, so let’s compare the other attributes.

With both O-RAN’s Split 7-2x and SCF’s nFAPI lower layer split occurring below the HARQ processing in the MAC layer, both are constrained by exactly the same delay requirements as it relates to LTE HARQ processing and fronthaul transport budgets. Both O-RAN’s Split 7-2x and SCF’s nFAPI lower layer split permit the fronthaul traffic load to match the served cell traffic, enabling statistical multiplexing of traffic to be used within the fronthaul network. Both O-RAN’s Split 7-2x and SCF’s nFAPI/Split 6 support transport using a packet transport network between the Radio Unit and the Distributed Unit.

The managed object for the SCF’s Physical Network Function includes the ability for a single Physical Network Function to support multiple PNF Services. A PNF service can correspond to a cell, meaning that a PNF can be shared between multiple operators, whereby the PNF operator is responsible for provisioning the individual cells. This provides a foundation for implementing Neutral Host. More recently, the O-RAN Alliance’s Fronthaul Working Group has approved a work item to enhance the O-RAN lower layer split to support a “shared O-RAN Radio Unit” that can be parented to DUs from different operators, thus facilitating multi-operator deployment.

Both SCF and O-RAN Split 7-2x solutions have been influenced by the Distributed Antenna System (DAS) architectures that are the primary solution for bringing the RAN to indoor locations. The SCF leveraged the approach to DAS management when defining its approach to shared PNF operation. In contrast, O-RAN’s Split 7-2x has standardized enhanced “shared cell” functionality where multiple RUs are used in creating a single cell. This effectively uses the eCPRI based fronthaul to replicate functionality normally associated with digital DAS deployments.

Comparing fronthaul bandwidth requirements, it’s evident that  the 30-fold bandwidth expansion of CPRI was one of the main reasons for SCF to embark on its nFAPI specification program. However, the above analysis highlights how O-RAN has delivered important capabilities in its Split 7-2x to limit the necessary bandwidth expansion and avoid fragmentation of the lower layer split market between alternative split PHY approaches. Hence, the final aspect when comparing these alternatives is how much the bandwidth is expanded when going from Split 6 to Split 7-2x. Figure 1 illustrates that the bandwidth expansion between Split 6 and Split 7-3 is due to the operation of channel coding. With O-RAN having already estimated that Split 7-3 offers a 7% bandwidth savings compared to Split 7-2x with Modulation Compression, we can use the channel coding rate to estimate the bandwidth expansion between Split 6 and Split 7-2x. Table 4 uses typical LTE coding rates for 64QAM modulation to calculate the bandwidth expansion due to channel coding. This is combined with the additional 7% expansion due to Modulation Compression to estimate the differences in required bandwidth. This table shows that the difference in bandwidth between nFAPI/Split 6 and Split 7-2x is a function of channel coding rate and can be as high as 93% for 64QAM with 1/2 rate code, and as low as 16% for 64 QAM with 11/12 rate code.

Table 4: Example LTE 64QAM Channel Coding Bandwidth Expansion

Whereas the above analysis indicates that the cost of implementing the Channel Coding above the RU in Split 7-2x is a nominal increase in bandwidth, the benefit to such an approach is the significant simplification of the RU by removing the need to perform channel decoding. Critically, the channel decoder requires highly complex arithmetic and can become the bottleneck in physical layer processing. Often, this results in the use of dedicated hardware accelerators that can add significant complexity and cost to the nFAPI/Split 6 Radio Unit. In contrast, O-RAN’s split 7-2x allows the decoding functionality to be centralized, where it is expected that it can benefit from increased utilization and associated efficiencies, while simplifying the design of the O-RAN Radio Unit.

Source: cisco.com