To achieve the datasheet performance of ultra-high-speed data converters, now clocking to 64 GS/sec, the handoff to and from the microwave domain must be near perfect. To preserve the data converters’ spurious-free dynamic range, a new category of component has been developed that converts between the differential and single-ended signal domains while amplifying and filtering out-of-band signals.
By Seyed Tabatabaei, mmTron
As silicon geometries shrink, they increase the processing capability of each SoC generation. What is surprising: sampling frequencies for data converters are now running above 10 GS/sec and as high as 64 GS/Sec from companies such as Texas Instruments (TI), Analog Devices, Intel, and Teledyne e2v.
Among the many system benefits, this integration and speed combination is transforming the architecture of microwave front-ends. The historical receiver with down-conversion from RF to IF is bowing to direct RF sampling where the microwave signal is directly converted to digital. That’s the concept behind software-defined radio.
Although direct RF sampling systems running at the highest data rates can consume significant DC power — digital processors must keep up with the data rate — they simplify the RF block diagram with the attendant benefits of cost, size, and weight while adding flexibility and performance improvements. For systems with multiple parallel channels, such as fully active phased-array radar, direct RF sampling improves synchronization and phase coherence across the channels. These architectural tradeoffs favor direct RF sampling, judging by the defense, communications, and instrumentation systems that are adopting it.
Moving to mmWave
Designing a system that will use direct RF sampling, Nyquist sampling theory requires the sampling frequency, fs, to be more than twice the highest frequency being sampled to ensure the digital data accurately represents the original signal without aliasing. This means a 64 GS/Sec analog-to-digital converter (ADC) can theoretically convert signals with frequencies to 32 GHz. This frequency band is called the first Nyquist zone. To illustrate, Figure 1(a) shows a sampled baseband signal (the blue spectrum), with the images generated by sampling shown in red. Aliasing occurs when the band of the sampled signal is greater than fs/2, which causes the image and baseband spectrums to overlap.
Direct RF sampling extend to frequencies above the first Nyquist zone if the signal bandwidth being sampled does not exceed fs/2. This is termed under-sampling or harmonic sampling. Figure 1(b) illustrates this case, where the spectrum sampled lies in the second Nyquist zone. Because under-sampling will handle signal bandwidths to fs/2, it is useful for mmWave systems, where the bandwidth is typically large because of the spectrum available.
The data converter sets the performance
The performance of a direct RF sampling system is driven by the data converter. In the receiver, the RF signal presented to the ADC must be optimally matched to the ADC’s input requirements. In the transmitter, the interface at the output of the digital-to-analog converter (DAC) must maintain the fidelity of the digital signal.
Most ultra-high-speed ADCs and DACs use a differential signal flow, which provides significant advantages:
- reduced second-order distortion,
- suppressed common-mode interference,
- better grounding,
- immunity to substrate coupling,
- lower parasitic coupling, and
- improved power supply noise rejection.
RF systems are, however, typically single-ended with a 50 Ω characteristic impedance. While the conversion between differential and single-ended can be accomplished with a lossy passive balun, the sensitive interface between the differential data converters and the single-ended RF signal chains calls for active baluns specifically designed to maximize the performance of the expensive data converters.
Role of the Active Balun
In the receiver, the microwave signal driving the ADC must be filtered and amplified to the ADC’s maximum input level, adding minimal noise, harmonics, and intermodulation distortion or spurs. On the transmit side, the differential output of the DAC must be converted to single-ended and amplified sufficiently to drive the transmitter’s power amplifier chain, adding minimal noise, harmonics, and intermodulation. An active balun integrates the passive balun with filtering and low noise/high linearity amplification.
Figure 2 shows the block diagram of a direct RF sampling transceiver and where the active baluns fit in the signal flow. While the same circuit functions are integrated @mdash; a passive balun, anti-alias filtering, and quasi-differential low noise amplifiers @mdash; the transmit and receive paths have opposite signal flow, requiring separate MMICs for an ADC and DAC.
Active baluns optimize the interface with the data converter, providing wide bandwidth, high linearity, low noise figure, high common-mode rejection, and a high voltage swing compatible with the ADC or DAC. For example, the noise floor of the active balun should be much lower than the ADC’s noise floor. In addition to helping maximize system performance, the active balun will reduce the area on the printed circuit board (PCB), as maintaining the phase balance between differential signals adds to the circuit size and complexity of the PCB layout.
Powering the Active Balun
Wide bandwidth, low noise figure, high linearity, and high common-mode rejection ratio are key requirements for an active balun. The circuit must handle the full-scale voltage swing of the data converter, and its noise floor should be much lower than the noise of either the ADC or DAC.
You can meet these requirements with an active balun that uses a heterojunction bipolar transistor (HBT) process. GaAs HBTs can achieve good performance through 20 GHz, and InP has been used for low noise amplifiers to above 100 GHz. Active baluns fabricated in HBT are very small, enabling a compact surface-mount footprint on the PCB. They also have low phase noise, 10x to 20x lower than a design fabricated with a pseudomorphic high electron mobility transistor (pHEMT) process.
The main drawback of HBT devices is their relatively high noise figure. For example, a mmTron Ka-Band active balun designed for a 64 GS/Sec ADC has a minimum noise figure of 6 dB, compared to its pHEMT design with a noise figure below 3.5 dB. The area of the HBT-based design, however, is 3x smaller than the equivalent pHEMT MMIC.
Whichever process technology it uses, an active balun forces a tradeoff between achieving the lowest noise figure or the highest linearity. For example, the pHEMT devices used in mmTron’s TMC160 active balun achieve 1.5 dB average noise figure across a 3 GHz to 20 GHz operating bandwidth and an input IP3 of 15 dBm. Increasing the input IP3 to 21 dBm increases the noise figure to 3.5 dB. These and other trade-offs are part of the active balun’s design, which is often optimized to work with specific data converters.
Figure 3 shows the evaluation board for a TI multi-Nyquist, dual-channel DAC (DAC39RF10). With both channels processing, the maximum input data rate is 10.24 GS/SEC, which doubles to 20.48 GS/SEC if only a single channel is used. Each output of the DAC feeds an mmTron active balun, each integrated in a 7 mm x 7 mm air-cavity QFN package. The use of active baluns simplifies the design of the interface and doesn’t consume much PCB area. To support multi-channel data converters, several active balun channels can be integrated in a single QFN package.
The adoption of direct RF sampling in microwave and mmWave systems has created an opportunity for a new category of microwave component, the active balun. Its role is to interface between the high-speed data converters and traditional microwave and mmWave signal chains, preserving the spur-free dynamic range of the DAC and ADC. The active balun integrates the traditional passive balun with amplification and filtering in a small surface-mount package, simplifying system design and PCB layout.
Seyed Tabatabaei founded mmTron in 2020, inspired to develop disruptive MMIC products that deliver new levels of output power, efficiency, and linearity for mmWave applications.
Previously, Tabatabaei was CEO of Teramics, a design services company also focused on the mmWave market. His mmWave experience extends to Endwave Corporation, one of the pioneering companies in the industry, where he served as VP of semiconductor products.
Earlier roles include engineering manager of the microelectronics organization within Hewlett Packard (now Keysight Technologies), M/A-COM, and the Laboratory for Physical Sciences.
Dr. Tabatabaei has a Ph.D. from the University of Maryland, College Park, and a management certificate from the MIT Sloan School of Business.