Editor’s note: Here is another ‘Golden nugget’ from my former TI colleague from my days at TI. I recommend that you observe Keller’s insights---he is an expert in the RF arena.
High-speed data converters can now digitize or generate signals directly at radio frequencies (RFs), replacing traditional RF components such as mixers and local oscillators (LOs) with digital processing. In addition, the wide-bandwidth capabilities of RF-sampling data converters with multiple gigasample-per-second (GSPS) speeds enable radios to combine multiple bands for cellular infrastructure applications, resulting in smaller and lower-power systems that can ultimately reduce the number of remote radio head (RRH) boxes at each cell site. My previous article, “Signal Chain Basics #131: RF-sampling DACs for multiband transmitters” focused on RF-sampling transmitters; this article will cover ADCs and receivers.
Traditional radio architectures use either an intermediate frequency (IF) architecture (Figure 1a) or a zero-IF architecture (Figure 1b). In an IF architecture, the signal from the antenna is amplified and down-converted with a mixer to an IF, typically around 10% of the RF. A variable gain amplifier amplifies the IF signal, which passes through a band-pass filter before being digitized in the analog-to-digital converter (ADC). IF architectures are usually built from discrete components because it is difficult to integrate an IF filter.
In a zero-IF architecture, an analog quadrature demodulator amplifies and downconverts the RF signal directly to baseband. After filtering, a dual ADC converts the complex analog signal to a digital signal. Given the baseband ADC and low-pass baseband filters, a zero-IF receiver lends itself to integration.
Figure 1c illustrates an RF sampling architecture. The end-to-end functionality is the same, but the mixing and baseband stages are digital by directly sampling the RF from the antenna after amplification and filtering.
Adding a second band in an IF or zero-IF architecture typically requires a second signal chain with additional components because of bandwidth limitations. In comparison, RF-sampling ADCs run at multi-GSPS speeds and can digitize wide bandwidths at RF, covering multiple cellular bands with one ADC. In this case, adding two or more bands only requires additional digital downconverters to convert the additional bands to baseband signals.
Of course, RF sampling would not be attractive if the performance and power dissipation were worse than the IF or zero-IF architectures. As complementary metal-oxide semiconductor (CMOS) process technology has increased the speed and lowered the power of digital circuits, RF-sampling ADCs now have similar performance at lower power than traditional architectures.
Table 1 compares IF, zero-IF and RF-sampling architectures for a four-receiver (RX) system with 100MHz of bandwidth per band. The traditional architecture comprises the Texas Instruments TRF37B32, LMH6521, ADC16DX370 and LMX2581. The values taken from the TSW16DX370EVM evaluation module are described in the User’s Guide.
For zero-IF, I use the RX-mode specifications from a commercially available dual channel integrated transceiver, and for RF sampling the RX side of the AFE7686, a quad-channel transceiver with 9GSPS RF DACs and 3GSPS RF ADCs. I used several key performance parameters at 2.6GHz as a benchmark.
Architecture comparison for a quad receiver at 2.6GHz
Starting with size for a single band, one RF-sampling analog front end (AFE) and two dual-channel zero-IF transceivers are roughly the same size, while the discrete IF architecture is 10 times larger. Power for the zero-IF transceiver is 28% lower, but as discussed below with a lower performance in the presence of large interfering signals.
For a dual-band system, you can use the same RF-sampling AFE with a minimal increase in power. For IF and zero-IF architectures, the component count, size and power all double. The advantages are even larger when adding a third or fourth band.
Looking at the performance specifications, RF sampling has a somewhat higher noise figure than the IF or zero-IF examples. As a result, the low-noise amplifier needs more gain before the RF-sampling AFE to reduce the overall system noise figure. To receive small wanted signals in the presence of a large interfering signal, the RF-sampling AFE performance exceeds zero-IF receiver performance because of the latter’s worse in-band spurious free dynamic range (SFDR) and sideband image. However, the discrete IF architecture has several advantages – SFDR and phase noise – that could provide the best in-band dynamic range. Another benefit is that you can use a sharp IF filter, like a surface acoustic-wave filter, to significantly suppress out-of-band interfering signals.
One final difference: For multiband RF sampling, the integrated digital step attenuator at the input simultaneously controls the input level for all bands, although it is possible to use separate external variable gain amplifiers for each band. The presence of a large interfering signal in one band that requires an increase in attenuation will affect the system noise for the other band, reducing the sensitivity level. With two separate single-band receivers, you can independently control the gain for each band so that an interferer in one band does not affect the sensitivity in another band.
As demonstrated here and in my previous article, RF sampling has emerged as a new, competitive architecture for cellular infrastructure applications.
Stay tuned for the next Signal Chain Basics article, with advice on working with data converters, amplifiers, interface or other analog design challenges.
About the author
Robert Keller is a systems manager in TI’s Wireless Infrastructure group. He has 15 years of experience supporting high-speed products in wireless infrastructure communication, test and measurement, and military systems. He received a bachelor’s degree in physics and mathematics from Washington University in St. Louis, Missouri, and a Ph.D. in applied physics from Stanford University. He has 10 U.S. patents in networking and sensor applications. You can reach Robert at email@example.com.