Software-defined radio (SDR) used to be rare and exotic. But today, most modern radios use SDR’s architecture and techniques. Each year with continuing advances in ICs and other technologies, SDR becomes more capable and widespread. In fact, new techniques like cognitive radio (CR) are making SDR more useful and beneficial to wireless communications.
SDR uses software to perform some of the signal processing in a receiver and transmitter. For example, a traditional receiver using the ubiquitous superheterodyne architecture performs all signal processing with basic electronic circuits (Fig. 1a). The superheterodyne downconverts the input signal to an intermediate frequency (IF) for demodulation and other processing.
Early SDR receivers (Fig. 1b) replaced the demodulator with an analog-to-digital converter (ADC) after the IF stage and performed the demodulation and some filtering in a digital signal processor (DSP). Today, because ADCs sample faster, DSPs can handle more functions.
To make DSPs work, the amplitude and phase of the signals both must be known. This has led to an architecture that divides the received signal into two paths, one producing an in-phase (I) signal and a 90° shifted quadrature (Q) signal. A basic carrier signal has the form:
V = Ac cos(2πfct + ?)
Where fc is the carrier frequency, ? is the phase, and Ac is the carrier amplitude. Any of these may be varied for modulation. For demodulation in the digital domain, a single signal is insufficient for existing algorithms. Therefore, the modulated signal is converted into the I and Q signals:
V = I(t) cos(2πfct) + Q(t) sin(2πfct)
With the quadrature signals, any variations in amplitude, frequency, or phase can be detected and used in a demodulation or other process.
Figure 2 shows a modern I/Q SDR receiver. A low-noise amplifier (LNA) usually boosts the input signal from the antenna before it is applied to the two mixers. The mixers develop the I and Q signals. Both receive a local oscillator (LO) signal from the phase-locked loop (PLL) frequency synthesizer. Note the 90° shift between the LO signals to the two mixers.
The LO frequency is set to the signal frequency so the difference signal from the mixers is zero without modulation. With modulation, the difference is the baseband or original modulating signal. This architecture is called direct conversion or zero IF.
After the baseband signals have been filtered in low pass filters to eliminate the sum components at the output of the mixers, the signals are digitized in a pair of ADCs. The digital baseband signals are then processed with digital downconverters (DDCs) to lower the sampling rate, making them more compatible with the digital signal processing circuits. The digital signal processing circuits then use both the I and Q signals to perform the demodulation, equalization, and additional filtering as the application demands.
In a modern SDR transmitter, the DSP modulator divides the data to be transmitted into I and Q signals and feeds them to digital upconverters (DUCs) that boost their sample rate (Fig. 3). The I and Q signals are next sent to digital-to-analog converters (DACs) that produce the final baseband signal. The signals are then low pass filtered and sent to the mixers that upconvert the signal to the final transmitted frequency. The signal is finally sent to a power amplifier before being applied to the antenna.
All modern SDR transceivers use some basic variation of the receiver and transmitter circuits shown here. Of course as ADCs and DACs get faster, the digital processing moves closer to the antenna. The ultimate receiver then becomes simply a filter at the antenna to limit the bandwidth and a LNA before a fast ADC (Fig. 4). Then, the DSP performs all other processing like demodulation and filtering. Commercially available amateur radio and shortwave receivers covering up to 30 MHz already use this advanced architecture.
Many functions are now performed digitally:
- Filtering (low pass, high pass, band pass, and band reject)
- Modulation (AM, FM, PM, FSK, BPSK, QPSK, QAM, OFDM, etc.)
- Spectrum analysis
New modulation methods and related procedures are generally known as waveforms. By changing waveform software, a radio for one application like FM voice could be reprogrammed for high-speed data on a different frequency with a different protocol.
The advantages of SDRs lie in the greater simplicity of the hardware. Standard RF circuits are reduced to a minimum, keeping the cost of ICs low. DSP software improves operation with functions (like filters) that provide better performance than equivalent analog circuits. Digital signal processing also can compensate for some failing of RF components.
Furthermore, the flexibility of reprogramming allows defects to be fixed, new features to be added, upgraded operations to be included, and performance to be improved. An SDR of flexible design can be quickly changed with software to accommodate new modulation methods, new protocols, and other major adjustments that would ordinarily require new hardware.
The downsides of SDR include software complexity, development costs and time, limited frequency range for some applications, and generally higher power consumption.
SDR requires fast ADCs, DACs, and DSPs. Sampling rates of ADCs have been rising for years and now extend well into the gigahertz region. Many SDRs use a low IF architecture and an ADC boasting a range of 100 Msamples/s to several hundred Msamples/s. Even more is possible.
National Semiconductor’s (now Texas Instruments) recent ADC12Dxx00RF can sample at rates up to 3.6 Gsamples/s (see “ADCs Sample RF Directly” at www.electronicdesign.com). This dual-channel, 12-bit ADC can be used with a clock phase shift that allows interleaved or alternate channel sampling for even faster conversion. DAC sampling speed is closely following this trend.
While fast conversion is essential, the DSP must be fast enough to keep up. That has not been a problem as most processors have easily kept pace. SDR is software, of course, but you still need hardware that can be implemented physically in several forms.
For instance, you can write code to run on a general-purpose processor (GPP). This may not be an optimal approach as some of the algorithms call for math procedures that are awkward to handle on most GPPs. However, an Intel or AMD dual-core processor used in most PCs today does a great job in some applications. Some GPPs also have special instructions like the multiply and accumulate (function) that are so commonly used in DSP algorithms.
Then you can also use a DSP designed specifically to handle signal processing code. It has a special architecture (usually Harvard), memory, and arithmetic logic unit (ALU) instruction sets that make the DSP fast.
Texas Instruments’ popular line of DSPs has been used for years in SDR, such as the C5000 and C6000 series. Analog Devices and Freescale also have general-purpose DSPs. Like any processor, DSPs are fully programmable so they’re very flexible in applications where changes, additions, and updates may be required in the future. Clock speeds to 1 GHz are common in DSPs today.
More and more SDR designs are using FPGAs. The signal processing algorithms such as the fast Fourier transform (FFT) can be reduced to digital logic and quickly implemented in an FPGA. Since the cost of FPGAs has steadily declined, they have become a major alternative to DSPs. FPGAs are faster than some other processors with some functions but still have the flexibility of reprogramming. Altera and Xilinx support SDR on their FPGAs.
Finally, hard logic is also common today. When implementing fixed standards like cellular radio specifications, the flexibility or reprogramming is not necessary. Therefore, algorithms can be implemented in fixed on-chip logic. It is fast, uses less chip area, and can bring about a major decrease in power consumption. Such blocks of logic are generally called accelerators.
Many cell-phone basestation ICs like TI’s TMS320TC6614 system-on-a-chip (SoC) are examples that use accelerators. Figure 5 shows the 6614 with its ARM GPP and four 66x general-purpose floating-point DSP cores. Note the accelerator logic on the right. Most of the Layer 1 accelerators use DSP algorithms for the many SDR functions.
Real SDR Transceivers
Many SDRs have been developed for the military under the Joint Tactical Radio System (JTRS). This U.S. Department of Defense program aims to develop a complete line of SDR radios for voice, data, and video that can be used to form ad hoc networks on the battlefield. The program has been around since the late 1990s, with good progress over the years.
The whole basis of JTRS is the Software Communications Architecture (SCA). This open-architecture platform standard defines how the hardware and software work together. One of the primary objectives is to develop software that is fully transferrable between different hardware platforms, making all military radios multifunctional and interoperable.
The latest version, designated SCA 2.2.2, was recently made available to further improve the programmer’s ability to make the software more flexible and scalable. Called SCA Next, the software helps make programs smaller and require less testing.
SCA does not have specific provisions for cognitive features. But over the past few years, the U.S. Defense Advanced Research Projects Agency (DARPA) has been testing cognitive enhancements to SCA like Dynamic Spectrum Access that will hopefully be available in the coming next generation of JTRS radios.
The Thales Communications AN/PRC-148 JTRS Enhanced Multiband (JEM) Inter/Intra Team Radio, which is a JTRS radio, covers all the HF, VHF, and UHF military frequencies from 30 to 512 MHz (Fig. 6). Power output can be selected from 0.1 to 5 W. A wide range of modes and waveforms is available.
Cognitive radio (CR) expands the definition of SDR to include features that make a radio intelligent. The Wireless Innovation Forum defines CR as “Radio in which communication systems are aware of their environment and internal state and can make decisions about their radio operating behavior based on that information and predefined objectives. The environmental information may or may not include location information related to communicant systems.”
CRs are sometimes called adaptive radios that automatically adjust their behavior or operations to achieve specific objectives. They can sense, learn, and adapt. They have an internal memory that stores instructions for various situations. Stored knowledge about their own capabilities makes it possible for these radios to make their own decisions.
A CR can also access external databases for additional decision-making intelligence. It senses by listening to a channel assessing the presence of other signals, their characteristics, and the noise background. The CR learns from its experience as well. With all of the knowledge it has or can access, the CR becomes a super-intelligent radio.
The transmitter (TX) and receiver (RX) are full frequency-agile SDRs with a mix of applicable waveforms and all the related SDR hardware and software (Fig. 7). A separate cognitive processor engine runs the cognitive aspects of the radio. It gets inputs (M) from the RX and TX to monitor their condition and parameters. It uses these inputs along with others to make decisions.
Other inputs can come from policy instructions stored in memory that define ways to operate under different conditions. External databases may also be accessed. Some CR units get location information by GPS. Decisions are then made, and controls (C) are issued to the radios to achieve the desired result.
CRs are also a good example of artificial intelligence (AI) in action. AI is a collection of software that is used to store and use knowledge to solve problems. It can use standard algorithms but it can also draw on several AI techniques such as expert systems, natural language processing, neural networks, fuzzy logic, and search techniques. AI/CR emulates the human user by assessing the situation and making decisions based on existing knowledge and taking actions to achieve the desired best result.
One important aspect of CR is dynamic spectrum access (DSA), which allows the CR to tune to a channel in the frequency spectrum for its operation after it decides that the channel is unused. A DSA radio uses unused spectrum, producing greater efficiency of limited spectrum space.
A cognitive transceiver essentially tells the SDR what to do in the way of frequency of operation, modulation, power level, protocol, and other factors and makes corresponding adjustments automatically. A CR is software that monitors the SDR and delivers commands and control instructions as needed.
CRs primarily seek to solve two major wireless problems: limited spectrum and interoperability between different radios or wireless systems. A CR can find open spectrum and use it. It also can change its waveforms or protocols to adapt to radios of a different nature, making communications possible or more reliable.
There are also several different classifications of CRs. For example, a policy-based radio is programmed with a predefined set of capabilities like waveforms and procedures. The radio is used by selecting one of several different preprogrammed fixed functions. The fixed functions are loaded during manufacturing, selected by the user, or downloaded over the air.
Another CR form is a fully reconfigurable radio. This fully generic transceiver can operate over a wide frequency and power range. This type of radio can be fully reconfigured on the fly for new applications or communications conditions.
The xG Technology xMax carrier-class CR system for mobile communications uses the unlicensed industrial, scientific, and medical (ISM) band spectrum in the 902- to 928-MHz range. In its first iteration, it was deployed as a trial garrison and battlefield cellular radio system for the U.S. Army.
The system was tested earlier this year at Fort Bliss and the White Sands Missile Range as part of a military Network Integration Evaluation (NIE) process used to validate and test systems for the field. The system did well based on written media accounts and quotes from Army personnel.
The prototype xMax system used a compact mobile basestation and its own TX70 handset. It also used a 900-MHz time division duplex (TDD) digital radio with cognitive technology in the handsets themselves. The system supported Voice over Internet Protocol (VoIP) calls and texting (SMS) as well. The handset included a full Wi-Fi radio.
The xMax system adheres to the Federal Communications Commission’s part 15 rules for the 902- to 928-MHz spectrum. Radios can radiate up to 4 W (EIRP), and that goes for the basestations too. The system divides the spectrum into 18 1.44-MHz channels and uses robust binary phase-shift keying (BPSK) modulation. Access is time division multiple access (TDMA), and each channel can handle up to 12 voice calls.
The cognitive feature of the radio listens to the band in use to determine where any interference is and then switches to a frequency with the lowest noise levels. The xG handsets scan the band 33 times per second looking for the interference and identifying the clear spots where a good link can be formed. It then notifies the basestation, causing the frequency to change as needed to keep a clean connection.
Now xG Technology is moving into its second-generation xMax design. This new xMax system drops the special handsets and uses standard smart phones. This is something that the military wants as soldiers can use standard off-the-shelf smart phones, laptops, or tablets to save money.
In this new system, the smart phones talk to a bridging device called the xMod (Fig. 8). This device resembles the Novatel MiFi devices that let multiple laptops use Wi-Fi connections to talk to it and then backhauls these connections back to the Internet via a connection to the cellular network. The xMod works in a similar fashion supporting a Wi-Fi (or a direct USB wire) connection to commercial smart phones or computers and using the cognitive xMax network to connect back into the Internet or military network.
This new arrangement adds high-speed data connectivity to the system. Furthermore, it adds a managed VoIP capability via a smart-phone app. Normal 3G and 4G cellular smart phones use the cellular systems’ standard voice service via 2G or 3G cellular links, which currently isn’t of the VoIP kind. The xMax system puts a special app on the smart phone that permits it to identify and prioritize the voice packets. The xMax system then can provide landline quality voice, even though it is a 100% complete IP-based system.
The new system modifies the xMax waveform by leveraging orthogonal frequency division multiplexing (OFDM). Each of the previously defined 18 1.44-MHz channels is further subdivided into 128 subcarriers. Radio access is TDD. Another key addition, multiple input multiple output (MIMO), greatly improves the range, reliability, and data rate. The xMod devices use a 2x4 MIMO system with four receiver chains and two transmit chains.
The cognitive capabilities as well as the radio itself are primarily implemented in software, commonly called SDR. The system’s mix of CR techniques, MIMO, and advanced signal processing maximizes range, reliability, and throughput. The substantial processing complexity required leverages a new generation processor capable of supporting 50 GOPS in both the xMod and basestation. This kind of processing power is brand new and now available in a size and at a power consumption level suitable for battery powered devices like the xMod.
Rick Rotondo of xG explained that the cognitive capabilities and interference mitigation algorithms allow the xMax to operate reliably in “white spaces” as well as “gray spaces.” White space is defined as the unused TV channels that have been blessed by the FCC for unlicensed use. It’s used to mean a 6-MHz channel free of any interference from TV broadcasters.
However, white spaces can rapidly become gray spaces due to wireless microphones that operate in these channels as well as other white space devices operating nearby. In fact, the 900-MHz band that xG cut its teeth on is heavily used and “dark gray” with interference from cordless phones, wireless security systems, telemetry radios, and other devices.
So, the system had to operate reliability in the presence of high interference from day one. The result is that the xG cognitive system lets you maximize the capacity of gray spaces by reliably letting you operate where other radio systems may not be able to operate.
As for applications, xG Technology will soon offer an advanced system suitable for the military, rural broadband, and the enterprise. The new system will include the 902- to 928-MHz band and the 5.8-GHz ISM band. Future systems may also use the 700-MHz spectrum.
In addition, xG is evaluating its options in the TV white space arena both here in the U.S. as well as in the U.K., where the system would be a good fit. Imagine a powerful mobile voice and data cellular system that could leverage the free white space 6-MHz TV channels and offer the same services and better economics than commercial 3G and 4G cellular systems.
Speaking of white spaces, this is another excellent application for CR. White spaces comprise the unused 6-MHz TV channels that were abandoned in 2009 with the switch from analog to digital TV. TV stations still use channels 2 through 51 (54 to 698 MHz), though many of them are unused. The open channels vary widely by region but represent a huge waste of valuable spectrum.
The FCC has approved the use of these channels in a license-free service. The guidelines call for low power and knowledge of the available local channels. The FCC and a number of other organizations like Spectrum Bridge and Telcordia have developed comprehensive databases logging the TV stations and other wireless devices and services using these channels in most U.S. locations. To use the channels, a white space radio must access the database to see if it is being used. If it is, another channel is selected, preventing interference.
White space radios fall into two categories: basestations and customer premise equipment (CPE) terminals. The CPE terminals may be mobile. If they want to transmit, they send their location based on GPS coordinates to the basestation that accesses the database to see if the desired channel is open. If it is, the CPE is notified that it can transmit.
In some systems, the CPE actually listens to the desired channel to assess the presence of other signals. In any case, both the basestation and CPE radios use forms of cognitive radio to make intelligent decisions on what channel to use and when.
White space radios promise to make more efficient use of the unused TV spectrum, but CR methods make it practical to do so without interference. One key application for white space is wireless broadband in rural areas. There are still many places where good, high-speed Internet connections are not available. White space is well suited to this use.
The lower-frequency and non-line-of-sight (NLOS) characteristics of the white space channels make long-range connections of several miles not only possible but also reliable. In one potential business model, wireless Internet service providers (WISPs) would help fulfill the federal government’s National Broadband Initiative.
Remote monitoring and control is another potential use for white spaces. Machine-to-machine (M2M) applications like Smart Grid connections, video surveillance cameras, medical patient monitoring, and sensor networks all would benefit from access to white spaces.
Designed specifically for white space use, the RuralConnect IP Version II (RCIP VII) SDR/CR creates point-to-point and point-to-multipoint networks with priority routing for voice, data, and video (Fig. 9). The target application is rural broadband access but it can be used in IP video surveillance, well and pipeline monitoring, smart metering, and traffic signal communications.
The RCIP VII operates in the 470- to 786-MHz range. It uses TDD and can achieve data rates of 4, 6, 8, 12, or 16 Mbits/s in a 6-MHz channel using quadrature phase-shift keying (QPSK) or 16-phase quadrature amplitude modulation (16QAM). Its transmit power is +30 dBm, and its receive sensitivity is in the –86- to –89-dBm range. Security is by AES-128 with a shared secret key. The units are designed to use the Telcordia database and operate under the FCC’s Part 15 rules as well as the Ofcom regulations in the U.K.
The Neul NeulNET white space radio system is designed for M2M and rural broadband in the white spaces as well (see “First Commercial White Space Radios Target M2M And Broadband Applications” at www.electronicdesign.com).
An SDR transceiver is mainly a software project. Once you get a firm RF platform, the DSP processing choice is next. It could be a standard DSP, an FPGA, or some GPP. For first projects, a reference design with vendor development tools really speeds and simplifies things. Other than that there are few choices beyond going it alone.
One approach is to use GNU Radio, which is an open-source development platform for SDR. It includes a set of signal processing routines including modulation for Gaussian minimum shift keying (GMSK), phase-shift keying (PSK), QAM, OFDM, and a few others. The software also includes error correcting codes like Reed-Solomon, Viterbi, and turbo codes. There are routines for optimized filters, FFT, equalizers and timers. And, it allows coding in C++ or Python. The software runs under Windows, Linux, or MacOS.
The GNU Radio software relies on a basic RF platform called the Universal Software Radio Peripheral (USRP). It consists of several choices of RF boards covering frequency ranges up to 5.9 GHz. Also included is a complete data acquisition system of fast ADCs and DACs and related support circuitry. A USB port provides basic I/O.
Ettus Research, a good source for USRP, makes a line of RF boards covering different frequency ranges that use a basic I/Q architecture and are full duplex capable. Transmit/receive switching is provided. Transmitter power and receiver gain are controllable. Typical bandwidth capability is 30 MHz. You can use either a standard DSP like the TI OMAP3 that combines an ARM GPP and a TI C64xx DSP or a Xilinx Spartan 3A DSP1800 or Altera Cyclone FPGA for the software.
As for the ADC and DAC, some models of the Ettus USRP use 100-Msample/s, 14-bit ADCs and 400-Msample/s, 16-bit DACs. Other models use 64-Msample/s, 12-bit ADCs and 128-Msample/s, 14-bit DACs.
If you’re just beginning with SDR/CR or if you want to teach it, a great new choice is National Instruments’ USRP product (Fig. 10). National Instruments owns Ettus and uses its basic hardware in two products, the NI USRP 2120 with a frequency range of 50 MHz to 2.2 GHz and the NI USRP 2921 with a frequency range of 2.4 to 5.5 GHz.
Both transceivers use the standard direct conversion architecture and output their I/Q signals via a 1-Gigabit Ethernet port to a PC. Signals up to 25 Msamples/s baseband may be steamed this way. The bandwidth is 50 MHz. The units cost about $4000 each.
The digital signal processing and other procedures run on the PC. National Instruments’ LabVIEW software with the Modulation Tool Kit is used for development. Advanced software can be downloaded to NI’s Flex RIO PXI module with its Xilinx FPGA.
The NI USRP was developed primarily for university instruction and research, though it makes a good learning platform for anyone just beginning with SDR/CR. When you purchase a two-unit bundle, you get the full set of instructional materials developed at the University of Texas and Stanford University. The bundle costs about $6000. For initial exploration and prototyping, including projects with OFDM and MIMO, the NI USRP appears to be a good place to start