Crafting software-defined radio for 3G

Before software-defined radios can be fielded for cellular systems, two simultaneous and often conflicting requirements must be met: wide instantaneous bandwidth and high dynamic range in both the digital and analog domains. Although progress has been made in the area of reconfigurable devices to support high-resolution, high-data-rate digital processing, only recently has technology evolved to meet the same requirements for the analog side of the processing chain.

For cellular communication systems, the software-defined radio (SDR) should support third-generation cellular services and also be backward-compatible with 2G and 2.5G. Thus, the transceiver should support wideband-CDMA, cdma2000, GSM, North American D-AMPS and other schemes. These services combine to generate certain requirements on bandwidth, dynamic range and other characteristics. As will be seen, GSM presents the greatest challenge and thus drives most of those requirements.

Because spectrum allocations throughout the world have evolved in an uncoordinated fashion, spectral allocations have been assigned in the bands from 400 MHz to more than 2.2 GHz. To eliminate multiple designs for different air interfaces, the SDR front end should be capable of covering that wide range with a high dynamic range. Covering the entire range will require a front end with switched or tunable filters.

Of far more importance than the tunable frequency range of the front end is the instantaneous bandwidth that may be covered. An SDR basestation must be capable of meeting the most stringent demands among all cellular services. In terms of bandwidth coverage, this means swallowing up to 75 MHz of simultaneous band coverage, the largest portion of the spectrum currently devoted to cellular communications. To allow enough room for filter transition bands, the bandwidth of operation should be restricted to not more than 30 to 35 percent of the sampling frequency. Choosing the smaller percentage simplifies filter design further or permits compatibility with future frequency bands. Thus, the sampling frequency should be on the order of 250 Msamples/second.

As the receiver bandwidth increases, the receiver becomes susceptible to more interferers. Thus, the analog-to-digital converter must have a high dynamic range. The most difficult basestation requirement to meet is the GSM900 blocking specification, which calls for a minimum sensitivity of -101 dBm in the presence of a -13-dBm blocking signal. Using reasonable assumptions for front-end noise figure and receiver LO and analog-to-digital clock phase noise levels so that the A/D is the limiting device, the system should have a signal-to-noise ratio dynamic range of approximately 75 dB full scale over the Nyquist bandwidth and a spurious-free dynamic range of 100 dBFS.

With those values, a GSM system based on a 250-Msample/s 14-bit A/D can maintain an Eb/No = 6 dB and a bit error rate below 10-5. Less-demanding services, such as UTRA FDD (Universal Terrestrial Radio Access frequency-division duplex ) and cdma2000, benefit from improved sensitivity and protection from real-world blocking conditions that exceed spec requirements. Devices are being introduced to meet those demands.

A basestation that captures a large bandwidth on receive should support the same spectrum on transmit. The desire for a wideband transmit path drives the need for multiple-carrier digital-to-analog converters and power amps. This, in turn, dictates the need for high-dynamic-range D/A converters and a method for performing amplifier linearization. The precision of the D/A is a function of the number of supported carriers, but for a typical four-carrier basestation, a 14-bit D/A should be sufficient. Digital predistortion technology is likely to be the primary method for performing linearization, and chips are available to take on that task.

In addition to dynamic-range design requirements, board-layout and spectral-planning issues suggest that the A/D and D/A intermediate frequencies should be separated to avoid crosstalk. Thus, the D/A should run faster than the A/D. No less than a factor of two in sampling rate should be used. Power amplifier linearization techniques may require higher rates and should be considered during system design.

Capturing a large bandwidth at high precision requires a mechanism for isolating individual signal bands out of the 75-MHz captured spectrum and reducing the data rate to something manageable by the baseband processors. A tuner chip provides that capability. For flexibility, it should cover bandwidths from 25 kHz for D-AMPS up to 5 MHz for W-CDMA. Since such a large section of RF spectrum is captured, efficiency concerns dictate that multiple channels be supported simultaneously.

For a truly flexible SDR, hardware should be included to support multimodes and multiservices simultaneously, as well as the capability to dynamically allocate capacity between 2G or 2.5G systems and 3G.

Large-scale integration with advanced processes such as IBM's Cu-11 permit the design of digital circuits that meet the demanding requirements described. On the analog side, with advanced technologies such as IBM's silicon germanium BiCMOS 5HP and 7HP available, front ends with high dynamic range and instantaneous bandwidth can be developed. By following SDR design concepts, an RF-to-bits-to-RF converter is realized. That converter then appears as a programmable extension to the digital hardware, thus realizing a universal software-defined-radio cellular basestation.

S. Davis Kent III is senior systems engineer at TelASIC Communications (El Segundo, Calif.).

See related chart