Software radio is a buzzword that's been around for many years, with deep roots in the military. These were "be all and do all" receivers, the workhorses of military intelligence.1,2 As the cold war melted, software-radio enthusiasts found a new home for their technology: cellular-radio applications.3 This article reviews the concept, architecture, technology challenges, and economics of the software-defined radio.
Historically, the relatively low number of prevalent standards, as well as the state of the art and high cost of key components, have limited the benefits and use of software radios. The second generation of wireless systems has offered a variety of different modulation formats and multiple-access technologies to be covered by a single radio.
Dual-mode operation and compatibility requirements with analog systems make this task even more challenging. The main goal of current developments in dual-mode (and eventually triple-mode) transceivers, which cover drastically different data rates and modulation formats, is often reduction of cost, power, and size.4
Defining Software Radio
The most literal translation of software radio would be a radio where signals on the antenna, or perhaps at an intermediate frequency, are digitized with a high-performance analog-to-digital converter (ADC) and sent to a terminal (computer, mobile phone, etc.). Once digitized and inside the terminal, code would be used to select an RF channel and demodulate the signal (Fig. 1a). While this is a worthy goal, it's only now becoming practical for specific applications.
A more reasonable name for this desired technology would be a "digital reprogrammable radio." (Note that digital receivers can be designed to receive digitally modulated signals, as well as analog, or FM, signals.) As with a software radio, an ADC is used to digitize the signal at the antenna or at an intermediate frequency. Instead of processing the digitized data solely in software, however, a variety of flexible, reconfigurable ASICs and general-purpose digital signal processors (DSPs) are used to reduce system power dissipation, size, and cost (Fig. 1b). These ASICs are programmable and can be adjusted for different channel characteristics and modulation schemes. These implementations, which include ASICs or field-programmable gate arrays (FPGAs), are more economical than fully flexible DSP implementations.5
A practical definition of software radio includes radios with a set of predefined hardware modules (such as ASICs or FPGAs). These modules must be selectable through software as common hardware for several different systems.
These modules provide multirate signal-processing functions (like decimators and interpolators), digital down/up conversion capabilities, and filter programmability via RAM-coefficient, finite-impulse-response (FIR) filters. This approach enables the efficient realization of transceiver functions in terms of power consumption, minimal component count, and compactness. In effect, the filters and demodulation that would have run on the terminal have been generalized and committed to silicon with programmable characteristics.
Thus, a software radio that's practical for today is one where selected functions have been committed to silicon. But, enough flexibility must be retained in order to reconfigure it for a variety of different standards.
What Technology Is Needed?
In recent years, there's been significant improvement in critical technologies such as low-noise amplifiers (LNAs), mixers, data conversion, and DSPs. Only now do these enhancements make "software radios" possible.
Whether sampled at the antenna or at an intermediate frequency (IF), the signal must still be sampled with an analog-to-digital converter (ADC). In the case of the military archetype (Fig. 1a, again), the usual specification was for a 16-bit, 1-GHz sampler. Needless to say, if such a converter ever existed, it was quite expensive. Although advances have been made in RF-bandpass sigma-delta converters, a much more practical solution is to sample at an IF frequency.
A key breakthrough in the commercialization of software radios has been to limit the bandwidth of the receiver. The personal communications services (PCS) and cellular industries have done this through the licenses granted to operators (typically under 15 MHz per operator). Technically, this means that as long as the band of interest has bounds, images and other spurious signals can be managed and placed out of band. When applied to software radio, a defined bandwidth means that a system's sample rate and dynamic range can be reasonably limited.
This means that the IFs able to be selected also can be directly sampled with current ADC technology. Five years ago, data converters required that the RF signal be converted to baseband. Present technology allows the sampling of IF signals up to 250 MHz. An added benefit of IF sampling is that one or more downconvert stages can be eliminated. This results in very small receiver designs and, thereby, reduced cost.
Figure 2a shows a baseband-sampling receiver. This is a triple downconvert to near baseband with analog channel filtering. The final downconvert incorporates an IQ separation feeding separate baseband ADCs. The ADC datastream goes to the DSP, where demodulation is done in software. In addition to the RF band-select filter on the front end of this receiver, a channel-select filter is implemented in the analog domain.
Although the software could be altered to support a different air interface, the channel characteristics cannot be changed since the bandwidth of the analog filters is fixed. Certainly, different analog filters could be switched in and out. But often, these filters are quite expensive and add complexity.
Figure 2b depicts a similar IF sampling receiver. In this case, a single analog mixer is used to downconvert to a convenient IF where the signal is digitized. I and Q separation is done digitally in the receive-signal-processor (RSP) chip, along with channel-filter and data-rate selection. In this instance, the DSP is only used for demodulation and the receiver is fully programmable. Both the channel characteristics and demodulation methodology can be changed. In a narrow sense, software radios as presented here are "future proof" in that they permit incremental channel or standards changes with little or no impact on the hardware. Since this architecture lends itself well to IF sampling, such receivers are both smaller and cheaper.6
Bridging The 2G/3G Gap
Development of multicarrier, second-generation picocell and microcell base stations has become practical, thanks to progress in fully digital modem techniques. These techniques including synchronization, equalization, and multirate signal processing. In a multicarrier base-transceiver-station (BTS) receiver, the constituent RF and bearer channels are only treated as individual signals once they've entered the digital domain. This permits the radio to be independent of modulation, access methodology, and channel spacing.
An efficient hardware design takes advantage of digital algorithmic approaches. And, with proper architectural partitioning, it also makes software-radio communications products practical. The flexibility of software-radio-based solutions enables the migration to third-generation system designs. Software-radio techniques also can provide a seamless evolutionary path from the second- to third-generation systems, thereby reducing network operators' future capital costs.
ADC And DAC Limitations
In typical base-station implementations, a wideband ADC may convert an entire system band at the IF (extended GSM = 35 MHz, IS-136 = 25 MHz, IS-95 = 25 MHz). This allows digital channelization and demodulation.
From a software-radio point of view, there are engineering limits related to data-converter technology and ASIC reprogrammable functions. These include the bandwidth and dynamic range of the ADCs and DACs, as well as the processing capacity of the digital-processing hardware, including reprogrammable ASICs, FPGAs, DSP chips, and general-purpose processors.
ADCs, in particular, have always been seen as key components of signal-processing systems. They often dictate system architectures due to their limitations on sampling rate, resolution, and dynamic range. Over the last decade, most ADC research has been aimed at monolithic, power-efficient ADCs, rather than high-performance, high-power converters.
Over the last few years, however, the maturation of monolithic converter technology has returned the focus to high performance. High-speed DACs have been considered easier to implement when compared to high-speed ADCs. DAC specifications have been reviewed in the light of wideband, multicarrier transmission requirements where high-speed DACs should operate at medium-to-high IFs. Wider, dynamic-range DACs are needed for multitone applications, which transmit many channels of information over several MHz of bandwidth and have a high peak-to-average output-signal ratio.
The illustrations that follow provide insight into the capabilities of current ADC technology. While some air interfaces can't yet take advantage of wideband sampling, converter technology has matured to the point where many popular standards can now be handled. Notable standards potentially implementing this technology are PHS, PDC, IS-136, AMPS, and GSM picocells. The remainder of current standards will be implemented shortly, when next-generation converters become available.
The graphs shown in figure 3 illustrate the spurious response from a prototyped receiver which uses the architecture shown in Figure 2b. This receiver was programmed to filter an IS-136 channel (Fig. 3a), then reprogrammed to filter a Groupe Speciale Mobile (GSM) channel (Fig. 3b). In each case, the RSP was tuned across the band to illustrate receiver performance. These figures show blocker rejection, along with the performance required by the standard.
One key reason that the setup fails to meet minimum, standard-defined performance levels is the limitation in the spurious performance of the ADC. Current ADC technology provides about 80 dB of dynamic range without dither. The converter used in this case is an IF-sampling ADC (AD6640). Standard vendor-supplied evaluation boards are used (Fig. 4). If dither techniques are utilized, spurious performance will be improved from 15 to 20db to 100 dB.7
DSP Limitation Factors
Software-radio solutions set the demand for high-speed components for IF and baseband processing. Third-generation architectures will require 1000 MIPS of digital-IF processing power, and up to 2000 MIPS of baseband DSP power. Some functions of a wideband code-division multiple access (W-CDMA), third-generation technology receiver system, such as matched filtering and despreading, are computationally intense. They'll probably be implemented as dedicated ASICs, or at least hardware accelerators within a DSP core.
The peak computational demand of a software radio for a 4.096-MHz W-CDMA mode is about 400 MIPS per finger, or 1600 MIPS. An FPGA or DSP-core based ASIC could more easily deliver the required computational capacity today. The software-radio DSP could be reprogrammed for GSM, IS-136, or IS-95. Advanced decision-/data-directed or nondecision-directed techniques can be implemented within a DSP for time-division multiple-access (TDMA) receivers.8
The DSP industry is currently undergoing a generational change driven by three factors: architectural innovation, process technology, and system integration. Architecturally, DSPs have to improve their performance by means of higher levels of parallelism and incorporating multiple data paths and execution units. Process technology remains the primary driver of performance by increasing clock rates and transistor counts. At the same time, DSP power consumption should decrease. DSPs with drawn geometries of 0.35-µm in the 1995 to 1996 timeframe are now being delivered in 0.25-µm, with 0.18-µm expected in late 1999.
A software radio could offer more flexibility by realizing multimode and multiband radio. It could present "on-the-fly" specification change and additional functions and/or services. It also could offer autonomous selection of air-interface standards according to environments (home, office, outdoor, vehicle) and user needs (voice, data, audio).
The driver for the development of software-radio base stations is likely to be the introduction of third-generation systems. But, huge investments are needed for first- and second-generation wireless systems. These make it unattractive—at least for the present time—for operators to consider a software-radio network deployment in the near future.
We can expect, then, that world- wide base-station and handset manufacturers see the software radio as a remote opportunity (a technological vision) and will focus on a commercially viable version of it. This pragmatic approach results in the digital programmable radio, which bridges the gap between second- and third- generation systems. We can also predict the acceleration of the development of software-radio architectures. The market discontinuity, represented by third-generation systems, will create a demand in the same way GSM did for real-time DSP in the late 1980s and early 1990s.
- Mitola, J., "The Software Radio," Proceedings of the IEEE National Telesystems Conference, Feb. 1998, Boston, Mass.
- Lackey and Upmall, "SPEAKeasy: The Military Software Radio," IEEE Communications Magazine, May 1995, pp. 56-61.
- Gratzek, T., "Software Radios for Cellular/PCS Base Stations: Fact or Fiction," Proceedings of the 1998 International Symposium on Advanced Radio Technologies, Sept . 1998, Boulder, Colo.
- Proceedings of First International Workshop on Software Radios, June 1998, Rhodes, Greece.
- Efstathiou, D., Zvonar, Z., "Transmitter and Receiver Design for Software Radio Base Stations: Enabling Technologies and Components," Proceedings of 3rd ACTS Mobile Communications Summit '98, June 1998, Rhodes, Greece.
- Brannon, B., "Wide-Dynamic Range A/D Converters Pave the Way for Wideband Digital Radio Receivers," EDN, Nov. 7, 1996, pp. 187-205.
- Brannon, B., "Overcoming Converter Non-Linearities with Dither," Analog Devices Application Note, number AN-410.
- Efstathiou, D., Aghvami, A. H., "Preamble-less Non-Decision-Aided (NDA) Feed-forward Synchronisation Techniques for 16-QAM TDMA Demodulators," IEEE Transactions on Vehicular Technology, Vol. 47, No. 2, May 1998, pp. 673-685.