Reconfigurable signal processing key in base station design

Reconfigurable signal processing key in base station design
By Alan Marshall, Chief Technology Officer, Elixent Limited,Bristol, England , EE Times
January 3, 2002 (5:33 p.m. EST)
URL: http://www.eetimes.com/story/OEG20011205S0055

The verification of any sufficiently complex system quickly becomes a huge bottleneck for basestation designers as well as system designers. Basestations use algorithms that are difficult to implement and verify. Keeping up with evolving standards adds more complications on the design end. In the telecommunications industry, public demands include cries of "faster," "cheaper" and "more reliable." Consequently, new standards must be implemented into a design quickly to ensure market success.

But basestations are complex beasts, and design takes time. Herein lies the problem: A design that is state of the art when conceived may be anything but when actually completed. So how do you design a basestation last in the fast-paced, rapidly changing world of consumer telecommunications?

Current designs tend to call for extensive redundancy as a matter of course, in order to provide easily implementable backup should problems occur. This approach helps, but it isn't ideal, and that makes the redundancy issue a tough one to solve. While you need it to make the design reliable, redundancy also costs more and leads to rack upon rack of high-performance cards.

With very aggressive cost targets, an obvious way to help is to reduce the number of cards you need and to design in less redundancy. In short, you need to design a better, more reliable system that's capable of evolving in parallel with the breakthroughs in algorithms and standards that are occurring all the time.

As the system performs a standardized function, cost of ownership is a critical differentiating factor in the marketplace. Reliability and quality of service are crucial issues for designers, and redundancy is employed extensively — and at considerable cost. Systems are also designed for easy maintainability via card swapping. Rapid development and easy upgrade capabilities are very important for companies wishing to gain a commercial advantage, and one of the biggest issues for basestation companies is handling the design issues posed by evolving standards quickly and cost-effectively.

Typically, current wireless communications basestations are implemented by racks of high-performance cards. The system contains the front-end IF subsystem, followed by a very high-performance digital subsystem. If we took just the receive and digital down-conversion functions as a discussion example, the system would consist of channel filtering, channel equalization and compensation, demodulation and despreading, and demux and channel decoding stages — plus the back-haul and system controller modules.

Traditionally, the digital down converters are implemented via an ASIC or dedicated device because of the sampling rates involved. The baseband-processing functions are typically based on state-of-the-art DSPs and FPGAs, and replicated many times depending on the size of the basestation. Basestation builders divide the system into various logical stages and implement them using bus boards like CompactPCI for easy system building and configuration for different world markets.

Three major functions of the digital down-conversion chain are typically implemented by a signal-processing solution: channel equalization and compensation; demodulation and despreading; and demux and channel decoding. Essentially, these baseband functions can be implemented via the more traditional design approaches of DSP, ASIC or FPGA, or using the new design approach of reconfigurable signal processing (RSP).

The traditional ASIC approach generates the most cost-effective and high-performance solution to the algorithmic and system challenges of baseband design. This solution, however, is not really practical in a product environment where the specifications are not fixed and the system complexities are high. There is a very real probability that the system will not work, and will need to be recycled for every change of specification. This represents a product risk that's much too high when the pressures of ti me-to-market and short product life cycles are considered.

The typical approach is one based on DSPs, which offer the advantages of programmability and give sufficient performance to meet the system needs at a reasonable cost. In meeting the processing challenges of high-speed data services, however, DSPs are becoming increasingly complex. This means that the generation of efficient algorithmic code is difficult, which leads to poor code and computational density in the final system.

FPGAs provide an alternative approach. They offer the ability to adapt to system changes, but are generally not tuned to computational tasks, which ultimately leads to increased overall system costs.

RSP merges the programmability of the FPGA approach with an architecture that is tuned to signal-processing tasks. This means that the RSP solution is high-performance, but is also silicon- and computationally efficient. In addition, RSP offers the ability to handle standards diversity and "specification creep" in softwa re.

Basestation OEMs typically handle the global diversity in mobile communications standards via modularity, keeping as many system functions as possible in common, and implementing standards-dependent functions on separate modules. Although third-generation (3G) systems will reduce this problem, they will certainly not eliminate it.

One example of diversity is the variety of data modulation schemes employed by different standards. It is easy to see that different modulation forms can be handled by RSP. Although this can be implemented at the manufacturing stage using one standard DSP card with different firmware, a compelling advantage of the RSP approach is that it becomes easy to configure cards at the installation or maintenance stage, or actually in service via firmware download (even dynamically if required).

Developing this argument one step further, basestation OEMs are currently producing prototype products for 3G, before the standards are fixed. It is easy to predict that some aspec t of the system might change as field trial results come in. However, the global markets for such systems will be huge, and first-to-market status will undoubtedly positively influence market share. As a result, OEMs are currently committing huge design resources during these early stages.

However, this is extremely costly. For example, both the U.K. and Japanese trials have run into problems, necessitating expensive upgrades and possibly redesigns. If RSP is employed it's possible to upgrade to meet specification changes — even major ones — simply by utilizing a software update.

In short, in current systems, while a component is out of service the basestation is off-air, and in many parts of the world that means no phone coverage. RSP is intended to be reconfigured semidynamically. Therefore, instead of taking the basestation off-line, the RSP can be reconfigured between operations, and thus have no effect on normal service.

But there are challenges that occur when basestations are de signed specifically with reconfigurability in mind. Predicting which parts of the processing are most vulnerable to specification changes and would therefore benefit from being reconfigurable is an extremely important consideration that can be both difficult and time-consuming. This is an ongoing concern due to sheer design complexity, but using RSP to deal with specific functions minimizes the overall design risk.

Where reconfiguration is done on the fly in order to share hardware across multiple functions, the scheduling of functionality switching needs to be carefully considered. Currently, the air and network interfaces of a basestation are very tightly controlled and there is a large amount of paranoia involved in changing them. You simply cannot allow anything undefined to happen on these interfaces. Thus, reconfiguration has to be performed in a very controlled environment, with every reconfiguration type-tested and approved. Once this approval is achieved, the reconfiguration can be done automati cally , which means that the basestations will virtually never have to be taken off-line, increasing productivity, providing consistent coverage and saving costs.

With shrinking market windows and short product life cycles, RSP has a crucial role to play. Every system-on-chip will contain increasing degrees of reconfigurability. Technology such as RSP will be a key element in helping future-proof basestations and a multitude of other designs in a range of application areas.