We have noticed that you are visiting from North American areas. Would you like to browse the US site (US.DFI.com) for a better experience?

Would you answer a brief survey?

Thank you for taking this survey. Your feedback will help us improve our official website and provide a better user experience.

Your feedback has been successfully submitted. Thank you very much for your time and participation.

Success Story

DFI Adds Horsepower to 5G Small Cellular and Multi Access Edge Computing

DFI Adds Horsepower to 5G Small Cellular and Multi Access Edge Computing

SLOT 14 OFDM symbols 1 ms 15 kHz SLOT 14 OFDM symbols 30 kHz 500 us SLOT 14 OFDM symbols 60 kHz 250 uS SLOT 14 OFDM symbols 120 kHz 125 us

Small 5G Cell Needs Big Changes

For the purpose of boosting up the speed of internet traffic and cloud-computing applications that require fast response, DFI gladly introduces you to the newest COM Express Type 7 module DV970. It provides not only enhanced computing performance and low-power consumption, but also versatile expansion capability. Therefore, a leading cellular carrier partnered with DFI to integrate DV970 in its 5G small cells.


Region: Taiwan

Application: 5G Small Cell POC (Proof-of-Concept)

Solution: DFI DV970


Scaling 5G for the Future

The popularity of 5G mobile networks is an irresistible trend. Compared to 4G, users can expect 100x faster download speeds, 10x lower latency, and support for 500x devices in the same geographic space. These enhancements are expected to help users in a wave of new applications from autonomous vehicles, delivery drones, augmented reality, to 4K/8K video streaming, which are all areas where 5G shows great potential. 5G is a unified connectivity fabric that will connect everything around us.

As an old saying goes, “great oaks grow from little acorns”. Before moving to the era of the Internet of things, an infrastructure that can support massive data processing must be established, and there must be enough margin to support the substantial increase in data usage in the next ten years, along with satisfying the low-latency requirements of real-time applications.

According to the Ericsson Mobile Report 2020 Q2, the total traffic driven by 5G networks will be multiplied by 4 in 2025 years, which means the small cell as the backbone of 4G networks must have more powerful computing performance and share the workload of cloud computing. Based on the consideration of reducing long-term operating costs arisen from dramatically increased small cells to cope with the 5G millimeter wave (mmWave) due to the shorter transmission distance, incorporating the spirit of software-defined networking (SDN) and network function virtualization (NFV), and adopting a COTS (Commercial-Off-The-Shelf) and general-purpose hardware platform is even more indispensable.

"Latency" matters more than ever

From near to far, the latency issues of 5G networks can be divided into two levels: "signal processing" and "quality of service."

According to the Time Division Duplex (TDD) architecture of 5G NR (New Radio), if the theoretical transmission bandwidth of 10Gb/s (carrier bandwidth 400MHz, sub-carrier spacing 120kHz) is to be achieved, the small cell needs to have a slot that processes 14 OFDM symbols (Symbol) within 125 seconds (µs), in other words, 8000 slots must be processed within one second.

Due to the trend of software-defined networking (SDN) and network function virtualization (NFV), general-purpose processors are becoming more and more suitable for computing cores of small cells. Intel Atom® processors are used as network processors (NPUs), combined with FPGA in charge of baseband processing to effectively backhaul the broadband data traffic of the local small cell to the core network, and bring more flexibility for application deployment.

To confirm the technical feasibility of 5G small cells, one leading cellular carrier began investigating the use of DFI's DV970 embedded with Intel Atom® C3958 Processor with FPGA and 5G communication modules to conduct a proof of concept (POC). However, each device needed to be synchronized at any time and periodically sent a signal pulse to ensure the coherence of data and actions. Therefore, it was a challenge to effectively reduce latency. Longer latency meant that the processor must consume higher utilization of computing power.

To overcome the latency problem, DFI took a two-pronged approach from both software and hardware. First, DFI optimized the hardware circuit to detect the signal pulse as soon as possible, and at the same time improved the PCIe connection between the processor and the 5G communication module to increase the overall system performance. Second, DFI strived for more processing time by using a low-latency Linux kernel and released more processor performance to support other network service requirements.

In order to realize both low latency and high bandwidth, the 5G base station divided the BBU (Baseband Unit) into CU and DU. CU (Centralized Unit) was responsible for processing non-real-time protocols and services, and DU (Distribute Unit) for physical layer protocols and real-time services.

By utilizing SDN/NFV technology, the BBU should be compatible with 4G/5G, and support could-radio access networks (C-RAN), distributed-RAN (D-RAN), and both 5G central and distributed units (CU/DU), which equips it with the robust ability for flexible deployment in the future. The new generation of modular baseband processing platforms based on Intel architecture has increased capacity, high-level integration, and multi-mode flexible networking features, to reduce the total cost of ownership (TCO).

FGPA 5G module PCle x2 5G 1slot length 125 us Trigger Atom@ C3958 NPU FGPA synchornize signal Pulse width 80ns 10 GbE Ethernet 125 us 土 180 ns
Core Network Backhaul Baseband Backhaul Core Network Core functions may be co-located with Cloud-RAN Units Cloud RAN Fronthaul Fronthaul RRH Centralized RAN (C-RAN) RRH C-RAN with NFV

Refined for edge computing server with industry-grade reliability

After achieving the 10Gb/s data transmission rate, there was another major requirement of 5G networks: network service quality of "less than one-millisecond latency", but this was not a problem that could be solved by simply deploying a 5G network. The servers that provide network services also needed to be closer to users and provided edge computing functions to utilize real-time applications. More importantly, the server should adopt the popular x86 architecture to achieve maximum flexibility of application deployment.

The European Telecommunications Standards Institute (ETSI) proposed Network Function Virtualization (NFV) in 2012, attempting to integrate different types of network equipment with virtualization technology into a common hardware platform based on x86 servers to simplify network equipment and overall architecture. It gave birth to Cloud RAN (Radio Access Network), which centralizes the baseband signal processing in the cloud data center, to reduce overall operating costs.

Based on NFV, ETSI further proposed multi-access edge computing (MEC, Multi-access Edge Computing) in 2015, so that it could be deployed close to the small cell to provide ultra-low-latency and real-time services, and even combining both as one unity with powerful computing power, complementary to Massive MIMO and Beamforming, which allowed small cells and multiple users to transmit data simultaneously within same frequency resources. When the architecture of access networks adopted separate CU (Centralized Unit) and DU (Distributed Unit), edge computing of CU could be deployed on MEC platforms to provide ultra-low latency services. Furthermore, according to the viewpoint from ETSI, Cloud RAN and MEC were a perfect pairing, because of the benefits of co-deployment.

The Intel Atom® C3000 series adopted by DFI DV970 has a wide range of SKUs (2-16 cores), QuickAssist Technology (QAT) that can accelerate encryption, decryption, and decompression, up to 16 x SATA, 16 x PCIe 3.0, 4 x USB 3, and 4 x 10 GbE Ethernet. It addresses extreme low power requirements (less than 10W TDP) and high-density form factors such as MEC. For boosting up the speed of internet traffic and cloud-computing applications that require fast response, DFI DV970 based on COM Express Type 7 provides not only enhanced computing performance and low-power consumption, but also versatile expansion capability. The DV970 supports 4 ports of 10GbE-KR Ethernet to strengthen the connection between server and equipment and is capable of a 24/7 stable operation in an extended-temp environment, therefore being an ideal solution for 5G Small Cell and MEC.

MEC servers and numerous 5G small cells are often deployed in a relatively harsh environment within limited space. In addition to the wide temperature support, the overall design must adopt industrial computer specifications, which is by no means competent with a general server. DFI has been working in the field of industrial-grade computers for a long time across industrial motherboards, system-on-modules, industrial computers, and industrial panel PC & displays. It also provides fast customized services for various types of applications to meet customer’s requirements.

RemoGuard Ensure the Availability of Numerous Small Cells and Edge Computing Servers

Many 5G small cells and edge computing servers are deployed in a very wide range. If the operating system crashes and the general in-band remote management cannot be used, it is necessary to allocate manpower for on-site maintenance. The more the number of devices, the higher the costs, and it will lead to the rapid growth of the overall operation expenses (OPEX), and lower quality of service.

RemoGuard, a cloud management platform co-developed by both DFI and Innodisk, introduces out-of-band management and moves the entire management system to the cloud. The platform can automatically backup the data and recover the operating system remotely when the edge devices crash, which exactly meets the managing attributes of 5G small cells – to maintain numerous stations allocated at multiple sites.

Besides remote OS recovery, RemoGuard also has device monitoring services. The platform updates the real-time updated data of logging temperature, input/output voltage, and power consumption for in-time action, and proactively estimates the lifespan of SSDs, turning passive notification to active prediction to pin down a precise benchmark of replacement timing as well as yielding benefits for inventory control, maintenance efficiency, and service continuity of the 5G small stations.

As the management is conducted through both in-band and out-of-band, both the connection to the cloud and the data itself should be protected. The advanced AES encryption is adopted to prevent data from tampering, and the Transport Layer Security Protocol (TLS) adopted by Azure Sphere further ensures communication confidentiality. Moreover, in the case of potential data breaches or improper violation of the physical equipment, secure erasure and self-destruction can also be triggered as specially required in mission-critical settings, ensuring comprehensive data security under the transmission.

DFI assists customers to scale 5G into the future

In the advent of the 5G era, whether from radio access network (RAN), multi-access edge computing (MEC), or evolved data packet core (EPC), it is necessary to create a good 5G application environment and take into account the capital expenditures (CAPEX) and operating expenses (OPEX). The x86 server platform with high application flexibility, high reliability, and high availability is the most ideal choice, fully implementing the spirit of software-defined networking (SDN) and network function virtualization (NFV) core.

DFI, which focuses on industrial computers, not only provides complete hardware solutions and cloud management systems for 5G small cell and edge computing servers but also combines 16-core Atom C3958 processor and 5G small station by its DV970 COM Express module to provide sufficient performance for today’s workloads. DFI has enough expansion space to adapt to the ever-increasing data flow in the future and makes the vision of the internet of things no longer unattainable and unreachable.

MEC APP VM MEC Application Platform MEC Virtualiation Infrastructure Manager MEC Application Platform MEC Virtualiation Layer (e.g. OpenStack, Kubernete) MEC Operting System MEC Hardware Resources(e.g. Intel Hard ware) MEC Hosting Infrastructure MEC APP VM MEC APP Management