Using Programmable Logic for Efficient and Effective DSP Design - Embedded.com

Using Programmable Logic for Efficient and Effective DSP Design

Voice over IP (VoIP) has become one of the fastest growing applicationsfor both wireline and wireless telecommunications equipmentmanufacturers. Not only does VoIP afford end-users affordable access tovoice calls, but it allows media gateway designers the ability toengineer equipment that serves many different purposes ” a situationthat typically would provide a cost benefit for both.

In the case of VoIP, however, theopposite is true. As VoIP gains greater and greater acceptance, profitmargins for equipment manufacturers are minimized based on thecompetitive nature of the application and the high cost of designing IPPBX and multiservice access nodes (MSANs)at many different channel densities. Engineering costs are rising andproduct prices are declining.

Although many companies have designed proprietary hardware/softwaremedia gateway solutions, they must make compromises in order to offercost-competitive products. As a result, most VoIP functionalities arebest offered today using DSP technology, but not without a new set ofchallenges.

As media gateway or IP PBX designers tackle thedaunting task of bringing a cost-effective product to market in atime-sensitive, ever-changing environment, the high-tech dilemmarecurs: performance vs. functionality. The main problem with DSPtechnology is finding the perfect balance of performance andinterfaces.

For example, designers often find that only the very high-end DSPson the market can support their most sophisticatedapplications/interfaces, including SGMII Ethernet or RapidIO.

So, when it comes time to interconnect these DSPs in larger systems,a field programmable gate array (FPGA) is often used to glue devicestogether and perform functions such as back-pressure on packettransmission or packet classification. This is where programmable logichas found its way into hardware design.

Application portability is one of the key values of DSPs usingprogrammable logic. The same underlying media processing technology canachieve performance results in application areas as diverse as:

– Large-scale media gateways
– IP PBX
– HMP media servers off-loading
– MSAN / Access equipment

Even though the basic feature set may be the same for all VoIPapplications, requiring the same list of codecs and voice qualityfeatures, the interfaces required for each may be quite different.

For example, a PCI card designed for host-based media processing(HMP) off-loading would require a PCI Express interface, whereas aMSAN may require a lightweight interface to an Ethernet PHY and to T1/E1framers, as well as a lot of glue logic to the variouscomponents on the board.

One of the beauties of DSPs for the designer is the inherent abilityto make the solution unique ” not an off-the-shelf, one-trick pony.This technology flexibility allows designers to explore morecost-effective ways of doing things, but they can often find that theyare limited by what their DSPs support.

This is where programmable logic ” FPGA ” provides the addedenhancement. For example, in an IP PBX, the USB, rather than Ethernet,is the ideal interconnect for low-cost systems.

Not only is it hot-pluggable by design, a USB hub device is lessexpensive and less complex than an Ethernet switch, which means thatthe base cost for a system is lower. In addition, as a system grows “particularly a blade-based system — the incremental cost is lower,because for each processing blade added to the system, very simple,low-cost RISC processors are available with USB interfaces.

Sadly, most DSPs provided to the VoIP gateway market do not offerUSB support, leaving designers with only the high cost product model.Happily FPGA vendors provide IP cores for such an interface, allowingfor newer, lower-cost innovation.

In a perfect world, a media gateway designer would be able to getexactly the interfaces needed, without the cost and side effects ofbuying a fully-featured DSP. But today, most high-end DSPs havemultiple high-speed interfaces ” many of which are not needed ” andhost a very large package with a very high pin count.

These large packages are dense ball grid arrays (BGAs) that areexpensive and require a printed circuit board (PCB) with a high numberof layers, which can all contribute to design delays.

Although some manufacturers have tried combining an FPGA with a DSPto achieve design nirvana — particularly in cases where the DSPinterfaces are judged inadequate or customization was required — thissolution can be overkill. It adds unnecessary cost and board realestate, only to correct short comings in the DSP.

The best approach to adding programmable logic – an FPGA – to afull-function DSP is to let each device do what it does best. In thismethod, the DSP serves only as a DSP.

The physical package contains a high-performance, multi-core deviceand uses a single memory interface to communicate with the FPGA (Figure 1 below ). In this case, thereis no carrying cost for the interfaces inside the DSP, keeping area andpower to an absolute minimum.

The DSP vendor provides turn-key FPGA designs for each majorapplication, which can then be customized by the OEM if desired. TheFPGAs used can be high-end devices or very small, low-cost devices,depending on the target application.

Figure1: A single memory interface links the DSP and FPGA

In a cost-sensitive, time-critical market, the biggest advantage tothe FPGA/DSP approach is its flexibility. FPGAs can be re-programmed inthe field if necessary. Furthermore, FPGAs are always among the firstdevices to be released at each new process node.

Therefore, they always support the latest I/O technologies withoptimal power performance. As interconnect standards continue toevolve, it is possible to retarget a design very quickly to the latestfabric, whether it's PCI express, RapidIO or a proprietary interface.

As designers evaluate a new product, it is important that they lookat the total cost of the solution, as opposed to summing the costs ofeach device in a vacuum. The combined DSP and FPGA approach offers manycost savings. For example, there is a plurality of Ethernet standardsavailable: MII, RMII, GMII, RGMII, SGMII, and so on.

When deciding how to interconnect many VoIP DSPs together, hardwaredesigners can choose from a large number of Ethernet switch ICs, orconsider a FPGA solution. If a certain Ethernet switch offers the idealprice point, but does not support the interfaces provided by the DSP,then a more expensive switch must be used.

The cost savings can be further extended beyond the electricalinterface. For example, many large OEMs use proprietary headers orflags to help classify packet traffic.

With an FPGA front-end, this becomes trivial and can greatlyincrease performance by off-loading certain tasks from the DSP. Thistype of header manipulation is best done in hardware by an FPGA. Thisoffers deterministic performance, and keeps control of the design inthe hands of the OEMs.

A media gateway DSP often incorporates a TDM interface, that isconnected to Timeslot interchangers (TSIs) or T1/E1 framers. Veryoften, an FPGA is already present between the DSP and these TDM-basedevices, or the TDM backplane.

FPGAs are currently used because custom logic is required, orbecause the FPGAs offer more reliable I/Os that can sustain the highvoltage spikes often seen on backplanes. These FPGA tasks can now beincorporated into the front-end FPGA saving more money and board area.

Media gateway applications are emerging at an unprecedented pace.Products currently under development for VoIP may need to accommodatevideo applications within months. Other applications ” such as wireless” are on the horizon.

As such, another major advantage of using an FPGA front-end is tomake life easier for designers who are faced with the near daily taskof imagining the future.

Today's DSP manufacturers release products with feature sets andinterfaces for specific applications that target the largest marketsegments. Therefore, it is easy to find a DSP that can serve as ageneric TDM to VoIP gateway.

However, if the designer is presented with the challenge of buildinga high-performance telepresence video over IP conferencing unit, thenthe choice of DSPs is quite slim.

The amount of bandwidth required for passing compressed orun-compressed video streams between DSPs is immense. Finding a DSP withthe right interface for this or any other emerging market is quitedifficult and costly.

With an FPGA front-end that allows access to the latest I/Otechnology, as well as IP blocks for the latest standards, designershave flexibility to choose the right interfaces, without making changesto the DSP software. In organizations with large FPGA teams, thedesigner may choose to develop a proprietary interface that meetscustomer needs today and provides a competitive edge.

Challenges such as increased price pressure and shortened designcycles continue to plague media gateway designers. However, designersusing a DSP that is tightly coupled to an FPGA are able to get exactlythe interfaces they need without the added cost or debugging of uselessI/O banks.

This allows OEMs to add their own unique features and differentiatetheir systems while greatly simplifying the design. This kind ofinnovation is what lets OEMs address current and future customerrequirements while maintaining a common DSP software platform.

Doug Morrissey is Vice Presidentand Chief Technology Officer atOctasic. Prior to joining Octasic ,He holds a BSc from RochesterInstitute of Technology.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.