Part two of this two-part set examines the receive signal processing, transmit signal processing, and the D/A conversion sections of software radio designs.
By Brad Brannon, Senior Systems Application Engineer, Analog Devices
In part one of this article, we explored the applications for software-defined radios (see Overview of Software-Defined RadiosPart 1). We also started out discussion of the key building blocks by discussing A/D conversion. In this part, we will continue to explore the building blocks essential to software-defined radio design. Let's start with the receive signal processor (RSP).
The RSP is a numeric preprocessor for the DSP. The purpose of the RSP is to replace a local oscillator (LO), quadrature mixer, channel select filter, and data decimation filter (Figure 2). In a multi-carrier application, the RSP replaces the analog selectivity and tuning functions with digital equivalents. An RSP sets the receiver apart from traditional receivers for with this device, all channel characteristics are now programmable. This includes data rate, channel bandwidth, and channel shape.
In addition to the unlimited selection of channel characteristics, the digital filters will perform exactly alike across all boards, unlike analog solutions that always have their tolerances. To achieve this feat, a great deal of computational power is required.
An RSP is a special purpose DSP. The data from the analog-to-digital (A/D) converter may be as fast as 80 MSamples/s. Therefore, the input to the RSP will be running at this high rate.
In determining the number of concurrent operations, the RSP may be processing data at a sustained rate of 2200 millions of operations per second (MOPS), which includes the numerically controlled oscillator (NCO), quadrature mixer, decimation filters, and fast-input response (FIR) filters. If multiple channels are being processed, then the processing requirements are multiplied by the number of channels. This is clearly outside the capabilities of a general-purpose DSP.
Key RSP specs
There are several specifications that are important when selecting an RSP. First, the device must be capable of handling the data rates required by the analog-to-digital (A/D) converter interface. Since the sample rate of the A/D converter determines the bandwidth that can be processed via the Nyquist theory, the RSP must be capable of handling the same data rate in order to not be a limitation of performance.
The Nyquist theory states that the sample rate must be at least twice the bandwidth of signals being processed to recover the information contained therein. In practical systems, the sample rate is frequently run three times faster than the bandwidth of the signals being received to allow for anti-aliasing filter response. For US cellular operations on a 25 MHz total band, a minimum sample rate of 50 MSamples/s is required. Actual implements are typically 65 MSamples/s or higher.
The next specification of interest are the internal bus widths, which must be wide enough to preserve signal integrity. Although the A/D converter may only be 14 bits wide, oversampling followed by narrowband filtering improves the effective signal-to-noise ratio (SNR) [processing gain] of the A/D converter by up to 30 dB for some air interface standards. This is the equivalent of 5 more bits. Therefore, internal bus widths must have the equivalent of at least 19 bits to preserve signal integrity.
Since a large portion of any air interface is the channel bandwidth and shape, it is important that the RSP include flexible decimation and filtering configurations that allow for a wide variety of data rates and filter bandwidths. Fixed filter widths and shapes should be avoided since they limit channel bandwidths and usually preclude raised root cosine filtering.
Finally, one of the most distinct features of an RSP is the ability to select the desired analog frequency very precisely. Most RSPs have a 32-b NCO which provides a frequency resolution of about 1 in 4 billion. This is usually much more than adequate, but gives great flexibility. In addition to the flexibility, frequency hopping is greatly simplified. Since a phase-locked loop (PLL) is not used, changing frequencies is instantaneous. This can be a great benefit in time-division multiplexed (TDM) applications, such as GSM, where hopping must occur within the guardband of a few baud.
TSPs for software radios
The transmit signal processor (TSP) is a numeric post-processor for the DSP. The purpose of the TSP is to replace the first local oscillator (LO), quadrature modulator, channel filtering, and data interpolation (Figure 3).
As with the RSP, the TSP sets the transmitter apart from traditional designs because all channel characteristics are now programmable. This includes data rate, channel bandwidth, and channel shape. Since modulation, channel filtering, and other aspects of the modulation are done digitally, the filters will always perform exactly alike across all boards, unlike analog solutions that always have their tolerances.
Eventually, the data from a TSP will be reconstructed with a digital-to-analog (D/A) converter and since as large a bandwidth as possible is required for tuning flexibility, data out of the TSP may be 100 MHz or higher. Therefore in terms of computational requirements, a TSP may process data at up to 2700 MOPS. This includes pulse shaping, interpolation and up conversion.
Selecting a TSP
There are several specifications that are important when selecting a TSP. First, the device must be capable of generating data at the rates required to preserve they Nyquist theory over the spectrum of interest. As with the analog-to-digital (A/D) converter (see part one), the sample rate of the D/A converter determines how much spectrum can be faithfully generated by the D/A converter. Therefore, the TSP must be capable of generating data at least twice as fast as the band of interest and preferably three times faster as reasoned earlier for anti-aliasing filter response.
Similar to the RSP, the TSP's bus widths are important, yet for different reasons. In the transmit direction, there are two different issues. If the TSP is used in a single channel mode, then the issue is simply quantization noise. It is usually not desirable to transmit excess in-band (or out of band) noise as this wastes valuable transmitter efficiency.
In a multicarrier application, the concern is a little different. In these applications, many channels would be digitally summed before reconstruction with a D/A converter. Therefore, each time the number of channels is doubled, an additional bit should be added so that dynamic range is not taken from one channel when another is added. As a general rule, the bit precision of the TSP+D/A converter should not be a major contributor to the noise figure of the transmitter. As with an A/D converter, the noise figure can be derived from the SNR of the converter.
Again, channel bandwidth and shape are important and require flexible interpolation and filtering options. As with the RSP, fixed filter widths and shapes should be avoided since they limit channel bandwidths and usually preclude raised root cosine filtering.
Finally, the ability to frequency hop is vital. Since a TSP implements frequency control with an NCO and mixer, frequency hopping can be very quick, allowing implementation of the most demanding hopping applications.
One final component necessary to develop a proper software-defined radio is the D/A converter. D/A converters are no different than A/D converters. Therefore, the first specification of interest when selecting these products is signal-to-noise ratio (SNR).
As with an A/D converter, SNR is primarily determined by quantization and thermal noise. If either is too large, then the noise figure of the D/A converter will begin to contribute to the overall signal chain noise.
While noise is not necessarily a concern spectrally, the issue does become important when the D/A converter is used to reconstruct multiple signals. In this case, the D/A converter output signal swing ("power") is shared among the carriers. Each time the number of channels is doubled, the SNR per channel is reduced by 6 dB because the noise level stays the same while the output signal power is split between the two channels. If low resolution or low quality D/A converters are used, then the noise spectral density of the transmitter could be too large.
The theoretical SNR of a D/A converter is determined by the same set of equations that govern an A/D converter (see equations one and two in part one). It is also important to use a low jitter clock source when driving the latch pin of a D/A converter as clock jitter on the encode pin of a D/A converter will cause wide-band noise on the D/A converter output. If the NF of a D/A converter is required, it may be calculated using the equations above as in the case for the A/D converter.
The final specification of interest for D/A converter is spurious response. For the case of a D/A converter, these spurious may at best cause in-band blocking of an adjacent cell site. In the worst case scenario, out-of-band spurious emissions may cause a Federal Communication Commission (FCC; Washington, DC) citation. Therefore, spurious products within the D/A converter must be kept to a minimum.
Most air interface standards require approximately 65 dB rejection of spurs at the antenna. In allowing for margin and degradation in the following stages, typical D/A converter requirements approach 75 dB spurious response. This will ensure -65 dBc rejection at the antenna.
By using high performance digitizers and fixed-function DSPs, it is possible to facilitate software-definable radios. These radio systems offer many distinct features that benefit both the manufacturer and the service provider. These features can also greatly increase the value of the investment into base station hardware.
About the author:
Brad Brannon, Senior Systems Application Engineer, Wireless Infrastructure Products, Analog Devices, 7910 Triad Center Dr., Greensboro, NC 27409. Phone: 336-605-4212; Fax: 336-605-4332