• RapidIO Members News
• RapidIO Articles
• RapidIO Trade Association Press Releases
• News Release Search and Archive
• Newsletter
  • 2007
  • 2006
  • 2005
• Subscribe to Newsletter
• Photos and Diagrams

RapidIO Connections Newsletter - Fall 2005

In this issue...

Design It:

RapidIO® Technology for Wireless Infrastructure

The wireless infrastructure is an excellent target for the use of RapidIO technology.

The cellular wireless infrastructure network can be divided into three main parts — the MSC/GSN, the radio network controller and basestation. Data and calls entering the system from the network arrive via a mobile switching center (MSC) from the public switched telephony network (PSTN) or via a general packet radio system (GPRS) support node (GSN) from the IP network. There are many functions performed on the data at the network interface, including protocol termination and translation, routing switching, and transcoding and transrating of telephony and video signals. Normally, the number of MSC and GSN nodes in a network is kept to a minimum, so that each node processes a large number of calls. Therefore, the data rates are large, on the order of 10-100 Gigabit/s (Gbps), and high-bandwidth switching and routing occurs. Transcoding and transrating large quantities of voice and video signals requires banks of DSP farms that are connected to the routers by high-speed interconnect.

Once the data has been formatted for the wireless infrastructure, it passes to the radio network controller (RNC) in the radio access layer of the network. The RNC controls access to the basestations and is responsible for setting up and tearing down connections between users and the network. The RNC is responsible for managing the mobility of a user who is moving around between cells and networks. A certain amount of signal processing occurs in the RNC to allow processing of data from multiple basestations.

The basestation is responsible for maintenance of the physical link to the user. Each basestation keeps in contact with the users who are physically located within its cell. A great deal of signal processing occurs in the basestation. Like the MSC and GSN nodes, a basestation will typically house racks of DSP - and ASIC-laden digital baseband boards. RF processing and power amplification also occurs in the basestation. All of the RF and digital baseband cards are connected to each other and to miscellaneous control and interface cards via high-speed backplane links. The chip-to-chip data rate on these boards is also quite high and rising as higher data rate technologies such as 3G are deployed.

There are several standards in use today for wireless infrastructure. These include WCDMA, GSM, EDGE, IS95 and CDMA2000. The multitude of standards is due to competition between companies and national organizations as well as the evolution of standards to provide new features and higher data rates. The industry trend is toward higher data rates and more capacity in a single cell, leading directly to increased data rate requirements in the wireless infrastructure equipment itself.

Mobile Switching Center/GPRS Support Node
As was mentioned, the wireless infrastructure equipment can be classified into three main parts, the MSC/GSN, the radio network controller and basestation. The challenges for data flow in each of these parts are unique. This section describes how data flow in MSC/GSN equipment has been handled in the past and highlights some of the problems that need to be solved in future systems.

Core Network Interface
The core network interface can be divided into protocol processing and transcoding. Telephony traffic coming from the telephone service network is coded by pulse coded modulation (PCM) and has a signaling layer, typically SS7, applied so that the calls can be properly routed. Thousand of calls are processed by one MSC. The MSC is made up of racks of DSP cards, with each card containing tens of DSPs and processing hundreds of calls. A local controller deals with the signaling and routing of the calls and the DSP performs the translation of the signal from PCM into the appropriate speech coding standard used by the targeted cellular standard.

Calls are constantly being set up and torn down; therefore the data flow in the system is very dynamic. Microprocessors are typically employed as control processors responsible for managing the flow of data. They will perform load balancing on the boards to ensure that the maximum number of calls can be handled, even with a rapidly changing traffic pattern. Therefore, there is a mixture of data and control information flowing to and from the boards.

Voice calls are very latency-sensitive. Too much delay in the transmission of the voice signal and the quality of the phone call will suffer. Echo cancellation processing must also be performed to remove echoes from the conversation. This leads to very strict latency constraints in the MSC processing.

Transcoding and Transrating
Transcoding refers to the conversion of the PCM voice data from the traditional telephone network to the compressed wireless standard format used by the wireless infrastructure and back again. Transrating is the conversion of a signal stream (usually video in this case) from one data rate to another. An example of transrating is when a multicast video stream is sent to several user devices, with each user device requiring a different display size and resolution. In order to make optimal usage of bandwidth of the wireless channel, the video stream should be converted to the right format for display on each of the target devices before transmission. The data rate requirements for video and picture phones are higher than for voice only phones. As the use of video and pictures increases, the individual data rate to each DSP will increase as well, with an expectation of several hundreds of megabits per second (Mbps) in throughput to each DSP and aggregate data rates reaching into the gigabits per second range.

The DSP farm that is used to process this data is a classic embedded processing application. In a typical board level architecture, the data enters and leaves the system in packets that are transmitted across a backplane. Each packet is sent to the controller (usually a microprocessor of some kind), which extracts the channel information so that the destination of the packet can be ascertained. The controller then directs the protocol and routing device to route the packet to the DSP that is responsible for processing that channel. The protocol termination and routing chip is usually an FPGA or ASIC designed specifically for the board. Network processors are now seeing some use in this application. The controller must also be able to receive and transmit control information to and from each DSP so that the state of the DSP is known and changes in the channel allocation of a DSP can be communicated. Clearly, there is a lot of routing of both control and data information through a system like this. In addition, because current systems lack a bus that is scalable to the backplane, the protocol on the backplane is often different from that on the board, forcing the board controller and protocol termination devices to bridge between different bus protocols between the board and the backplane.

Radio Network Control
Radio network controllers are responsible for handling mobility management, which keeps track of users, and resource management, which handles radio and network resources. Mobility management localizes where the user is in the network (logically), and carries out handoffs between basestations. This occurs as a cellular user travels within a region. For example while driving at 60 miles per hour handoffs between basestations will occur approximately every ten minutes. The RNC is responsible for coordinating the signals received and transmitted by the basestations and making the decision about when any given basestation should pick up and/or release responsibility for a call.

The resource management function of the RNC refers to the RNC responsibility for keeping track of the available radio spectrum network capacity, and determining which of the available spectra should be used for any given call.

There is a high basestation to RNC ratio, ranging from 100 to 1000 to 1. Consequently, data rates at the RNC are quite high; they are about the same order as at MSC level. The need for high-speed interconnects at the chip, board and subsystem level have grown quite acute.

Basestation
In a basestation, there are several cards that need to be connected. In most systems these cards will all reside in the same rack. In larger basestations there can be multiple racks. The network interface card is connected to a transport network based on SONET/SDH, ATM or possibly IP. This is the communications link to the RNC. The transport protocol is formally terminated in the network card and converted to the internal system protocol. In some cases, the transport network protocol or a derivative is used as the internal protocol. The ATM-like protocol UTOPIA is sometimes used, and recently the OBSAI specification has suggested the use of Ethernet as the internal protocol. RapidIO has been considered by some vendors as a suitable internal system protocol as well.

The controller card decodes the basestation-specific control messages and sets up the baseband modem cards to correctly transmit and receive user data. There may be several baseband modem cards and several analog/RF cards in a single rack. The baseband modem card is sometimes split into transmit and receive cards, which have communication requirements owing to closed power control loops. The data rate requirement for the network-to-baseband modem cards is in the tens of Mbps. The data rate requirement for the baseband -to-analog interface is several Gbps. Both interfaces require a mixture of control and data.

Basestations come in a variety of sizes, from picostations that serve a single building to large macro cells that can cover hundreds of square kilometers in rural settings. The number of voice users supported by a single basestation varies from tens to thousands. Different basestation rack configurations have been developed to cope with this variety, but since they share many components a flexible, scalable interconnect is required.

The baseband modem card itself presents additional interconnect challenges. A typical third-generation WCDMA modem card architecture with the on-chip interconnect, poses several challenges:

• Data rate: the data rate from the analog cards can be on the order of several Gbps. The data needs to be routed to all of the front-end ASICs. Since data flow from the ASICs to the DSPs can also amount to Gbps, the ASIC and DSP pin requirements would become quite severe for a traditional parallel bus running close to 100 MHz.
• Interconnect complexity: there are several 'chip rate' DSPs that connect to the ASICs on the board, all of which connect to the 'symbol rate' DSP that does high-layer processing of the data. Control and status information flowing from the analog boards is usually connected on a separate link to the symbol rate DSP, which needs to process the data and gather statistics as well as monitor for faults.
• Lack of suitable interfaces and a variety of interfaces: typically a memory interface is used to connect from the chip rate DSP to the ASICs. This is not an ideal point-to-point interface and may require up to 100% overhead. Different interfaces are used for the DSP-to-DSP connection. Sometimes these are serial ports and sometimes host and memory interfaces. Generally, the lack of high-speed links on the board leads to a large amount of routing and a difficult, expensive board design. Legacy interfaces may not suit new architectures, and new components that have desirable functionality may not have the required legacy interfaces. Software can also be dependent on the interface used, and legacy interfaces often are used just to minimize the software rework that might otherwise be required.

Simplification with RapidIO Interconnect Technology
In the MSC/GSN side there are DSP farms whose cards require low-latency communication from the backplane to a farm of DSPs. There is a need for control for data routing and load sharing, as well as packet data transmission. A switch fabric capable of handling both data and control information provides a cost-effective, scalable solution. As the mobile network moves from mainly voice to higher-bandwidth multimedia applications, DSP farms will be required to process higher data rates in media gateways. In a RapidIO-based transcoding board architecture a RapidIO switch routes both the control and data and also does the address translation from the channel address t the DSP. The board controller is used to calculate the load balancing and update the necessary route tables in the switch. RapidIO is also suitable to be extended to the backplane. This significantly simplifies the task of the RapidIO switch, which replaces the protocol termination and touring ASIC or FPGA that a vendor would have had to develop in prior systems.

In the basestation, a variety of protocols are currently used to connect several boards together. The baseband modem board makes use of a variety of mostly proprietary interconnects, some of which require significant bandwidth and low latency. Both control and data information are transmitted through the system. A single protocol that can be used for both the control and data information and for both backplane and chip-to-chip communication minimizes chip count, simplifies the development of different-sized basestations, reduce development cost and can also reduce time to market.

The data rate requirements for basestations continue to increase as the move to third generation (3G) wireless standards brings higher user data rates. The number of antennas supported in a single basestation is also increasing, owing to increased use of techniques such as sectorization and adaptive antenna array processing. A protocol that can support a variety of data rates can be instrumental in reducing the cost and power of the baseband card.

Though the analog card to baseband card interface could be supported by RapidIO, recent moves to standardize this interface in OBSAI and CPR have focused on frame-based, rather than packet-based protocols. This is because the data transfer is regular and very low jitter is desired. The data is being generated by analog-to-digital converters (ADCs) or is being sent to digital-to-analog converters (DACs). Lower jitter leads to less buffering on the analog card. But there is no reason that streaming write transactions with relatively small packets could not be used in RapidIO to achieve a similar effect. Independent studies at Texas Instruments and Ericsson have shown that the physical area of a RapidIO port that is constrained to a streaming write is about the same as that of an OBSAI or CPRI port. With RapidIO, this interface would allow the use of standard RapidIO switches and routers, leading to less expensive, more flexible solutions. However, as at this point the industry push is to converge on a frame-based protocol, RapidIO switches with OBSAI and CPRI bridges may well be required to serve this market.