'Soft' basestation runs on commodity server

A traditional software-defined radio (SDR) uses digital signal processors or field-programmable gate arrays for signal processing, but this is not the only way to build SDRs. Our company has focused on all-software designs where the signal-processing code runs as a normal application process on top of commodity general-purpose processors and operating systems.

Building a software radio this way offers technical, economic and time-to-market advantages. These advantages derive from minimizing or eliminating hardware design effort, maximizing software portability for reuse across markets and hardware generations, and exploiting the sophisticated software available off-the-shelf for these platforms.

A recent example of this design approach is a Vanu GSM cellular telephone basestation, which completed successful field trials in DeLeon and Gorman, Texas, in October. The trial installation consisted of two basestation transceivers (BTS) and a basestation controller (BSC), each running on an industry-standard Hewlett-Packard ProLiant DL380 server with dual Intel Xeon 2.8-GHz processors. All of the signal processing, protocol processing and BSC functionality was implemented as application-level software, written in portable C++, running on top of an un-modified Linux operating system.

The complete installation also included the ADC Digivance Long Range Coverage Solution as an RF interface, a Telos SoftSwitch for PSTN connectivity, and HP and Cisco switches and routers for the all-Internet Protocol radio-access network. Each BTS supported two 200-kHz GSM carriers, permitting 14 simultaneous voice calls per server. Benchmarks suggest that the HP DL380 is fast enough to support four GSM carriers and 32 simultaneous calls. The field trial demonstrated full BTS and BSC functionality, including authentication, registration, handover, mobile-to-mobile, mobile-to-land line and land line-to-mobile calls.

Implementing a high-performance radio using an all-software commodity-processor approach is challenging. The processors, buses and peripherals on commodity boards are designed for workloads with high spatial and temporal data-access locality-exactly the opposite of the streaming pattern characteristic of signal processing. The signal-processing algorithms widely used in the industry were designed with a different performance and implementation model in mind. (Consider that a floating-point multiply-accumulate operation, the standard metric for signal processing, costs only one cycle on a high-end commodity processor, while a second-level cache miss can consume more than 100 cycles. An algorithm with an order-of-magnitude worse performance measured in MACs may be substantially faster because of better memory locality.) Also, familiar software design methods, which rely on sample-by-sample predictable processing costs, must be reconsidered because of the unpredictable nature of I/O and OS interrupts.

Overcoming these challenges required a ground-up approach to implementing the functions of a communications device, with innovations in all of these areas. Vanu systems use custom-developed signal-processing middleware that separates data and control flow, and dynamically sizes data payloads to match the cache. This minimizes the number of memory accesses associated with each trip down the processing pipeline. Novel algorithms include a replacement for the Viterbi decoder that executes much more quickly on a serial processor, and automatic generation of finite-impulse response filter and fast-Fourier transform code tuned to the characteristics of a particular device. Signal-processing computation is decoupled from I/O flows, so processing jitter due to interrupts does not affect the smooth flow of data through the system. Papers covering all these areas are available at www.vanu.com.

Many benefits
Using efficient, portable software on commodity platforms to implement communications systems provides benefits in many areas, starting from the lowest technical level: These platforms are a great software development target. Because of their high volume and wide applicability, substantially greater investment has gone into tools for tasks such as compilation, code generation, testing, debugging, benchmarking and monitoring than for any other processing platform. The availability of great tools substantially reduces software development cost and time.

The commodity platforms also offer a variety of useful higher-level features. For example, the Linux OS used for the Vanu GSM basestation supports encrypted remote login and file transfer via Secure Shell authentication, and the underlying HP server offers Ethernet connectivity. As a result, Vanu engineers were able to remotely monitor performance and perform software upgrades over the public Internet without any security or privacy concerns. Remote high-speed access substantially reduced cost and duration of the field trial.

The highest benefits come from the use of portable software. Increased capacity can be offered to customers over time by simply moving the software to faster platforms as they are developed. Customer requirements in particular market niches, for features such as NEBS compliance, environmental ruggedness or integration with other equipment, can be met with little target-specific engineering effort. For example, one customer requested a GSM basestation in a luggable form factor. Vanu acquired an appropriate embedded server and was able to bring the software up on that platform in two days.

For markets where this design approach meets unit size and power consumption requirements, commodity server-based SDR provides substantial time-to-market and economic advantages. The successful field trial of the GSM basestation represents an important milestone in commercial acceptance of this concept.

John Chapin (jchapin@vanu.com) is chief technology officer at Vanu Inc. (Cambridge, Mass.).

See related chart