Why Should We Process LTE Digital Signal Processing Functions in the Cloud?

In the following post, I will try to explain the importance and interests of the DSP processing in the cloud, including my vision of how it can be done and why it should be enabled.

First of all, the processing of the DSP functions currently in the LTE is supported by the BBU (Baseband unit). BBUs are specialized hardware which process baseband signals coming from the RRUs (Radio Remote Units).

These elements are part of the C-RAN (Centralized/Cloudified Radio Access Network) architecture and the functions of the BBU depend on the actual functional split between RRU and BBU. In the 5G, these two blocks are called distributed units (DU) and centralized units (CU). In the scenario of a fully functional split or scenarios when the split is made inside PHY layer of the eNodeB/gNodeB, the DSP functions are part of the centralized units. This functional split has drawbacks of the high fronthaul data rates with low latency, while on the other side it is offering the highest flexibility (e.g. enabling network coding and CoMP)

Additionally, if we consider that the virtualization and cloudification technologies are one of the enablers of the 5G network we can benefit even more because the virtualized DSP functions can be scaled up and down.

DSP functions in LTE are (de)coding, (de)modulation and IFFT/FFT. The most intensive one is decoding.

When we speak about LTE DSP in the cloud, we are only focusing on the edge cloud, in the locations where are currently BBU pools placed.

In general, DSP in the edge cloud can be processed in two ways:

• By virtualization it can be placed on the GPP hardware (x86 platforms);
• Supported by the additional DSP specialized hardware, placed in the edge cloud.

In the first case, when full virtualization is considered, the DSP functions are actually processed by the multi-core processors of the x86 platforms. In this way, we are reaching the highest flexibility. The processes (virtual containers) can be scaled up and down in order to reach the desired performances. In this way, it is possible to increase utilization of the hardware and reduce the costs of the deployment, while the main drawback is on the complexity of the multi-thread processing. The more details are given in [1].

On the other side, when we are supporting the edge cloud with the specialized hardware [2], we can also benefit in different ways, because in this way the utilization of the hardware is still high, while the complexity of the virtualization is avoided, the functions of the rest of the LTE protocol stack can be virtualized and placed on x86 platforms.

[1] Gatherer, Alan, and Eric Biscondi, “Multicore DSP programming models [In the spotlight].” IEEE Signal Processing Magazine vol. 26, no.6, pp. 224-220, 2009

[2] NGMN Alliance, “Further study on critical C-RAN technologies”, Next Generation Mobile Networks, 2015.