Introduction
In the evolving landscape of modern communication and radar systems, the demand for high-speed, high-frequency signal generation is paramount. Digital-to-Analog Converters (DACs) often serve as the critical interface in these systems, yet they frequently act as a performance bottleneck. Traditional DAC designs typically operate in Non-Return-to-Zero (NRZ) mode, which is characterized by a $sin(x)/x$ (SINC) frequency response. This inherent roll-off leads to significant power attenuation as the output frequency approaches the Nyquist limit—half the sampling rate. To generate signals in higher Nyquist zones (e.g., the 2nd or 3rd), designers often resort to complex digital equalization or external mixing, which increases system complexity and power consumption.
This paper presents a two-channel, 8-bit Radio Frequency (RF) DAC fabricated in 1 $\mu$m Gallium Arsenide (GaAs) technology. The design achieves a 1.2 Giga-samples per second (GSps) sampling rate and introduces a novel integrated circuit architecture that allows for seamless switching between standard NRZ and "Mixing" (Return-to-Zero, RZ) modes. This flexibility enables high-quality signal generation up to 1.8 GHz, effectively extending the device's utility into multiple Nyquist zones.
Architectural Design and Methodology
The researchers employed several strategic design choices to optimize speed, linearity, and versatility.
GaAs HBT Technology
The choice of a 1 $\mu$m GaAs Heterojunction Bipolar Transistor (HBT) process was central to the design. While CMOS is often preferred for low-power applications, GaAs offers superior breakdown voltages and a semi-insulating substrate. These characteristics are highly advantageous for high-performance mixed-signal RF circuits, providing better isolation and the ability to handle larger signal swings.
Segmented Current-Steering Architecture
To achieve an optimal balance between static linearity and high-speed operation, the DAC utilizes a "3+5" segmented architecture. The 3 Most Significant Bits (MSBs) are implemented using unary (thermometer) coding. This approach minimizes Differential Nonlinearity (DNL) and reduces glitch energy during transitions. The remaining 5 Least Significant Bits (LSBs) utilize a standard R-2R ladder network, which conserves die area and simplifies the layout without significantly impacting overall performance.
Integrated Mode Selection
The core innovation of this work is the integration of the mode-selection logic directly into the synchronous timing latches. In traditional designs, implementing a "Mixing Mode" (or RZ mode) often requires separate modules that introduce additional delays, power overhead, and potential signal artifacts. By adding only four transistors to the final synchronization latch, the authors enabled the DAC to switch between NRZ and mixing modes without the typical design penalties. This ensures that timing remains precise and "glitch" energy is minimized, regardless of the operating mode.
High-Speed Data Path
To support the 1.2 GSps throughput, the chip incorporates 2:1 multiplexers (MUXs) at the input stage. These MUXs combine two 600 Mbps data streams into a single 1200 Mbps stream for each channel, allowing the internal circuitry to process data at the full sampling rate.
Experimental Results
The performance of the fabricated chip, which occupies a die area of $5.5 \text{ mm} \times 3 \text{ mm}$, was rigorously evaluated through static and dynamic testing.
Static Performance
Static testing involved sweeping the digital inputs across the full 8-bit range (0 to 255). The DAC demonstrated exceptional linearity for its class:
- Differential Nonlinearity (DNL): Within $\pm 0.15$ LSB.
- Integral Nonlinearity (INL): Within $\pm 0.4$ LSB.
Dynamic Performance and Power
The dynamic range was measured using a spectrum analyzer to determine the Spurious Free Dynamic Range (SFDR) across various frequency bands.
- Sampling Rate: 1.2 GSps.
- SFDR (Normal Mode): Greater than 40 dB.
- SFDR (Mixing Mode): Greater than 30 dB for output signals reaching up to 1.8 GHz.
- Output Power: The device provides a substantial output swing of 1.1 Vpp.
- Power Consumption: The total power draw is 1.7 Watts for both channels (850 mW per channel) using a 5V supply.
The results confirm that in "mixing mode," the DAC effectively functions as both a converter and a mixer, allowing it to output high-quality signals in the third Nyquist band. This is a significant improvement over the 600 MHz limit imposed by standard NRZ operation at the same sampling rate.
Discussion and Conclusion
The 1.2 GSps 8-bit RF DAC successfully demonstrates a simplified yet effective approach to multi-Nyquist signal generation. By integrating mode-switching logic into the timing latches, the design avoids the complexity and performance degradation often associated with RZ-capable DACs.
However, the design is not without trade-offs. The power consumption of 850 mW per channel is considerably higher than modern CMOS alternatives, which may limit its application in power-constrained or high-density portable systems. Additionally, while 8-bit resolution is suitable for many radar and ultra-wideband (UWB) applications, higher-resolution (12-bit or 14-bit) designs are often required for advanced telecommunications.
Future work may involve porting the integrated latch technique to other semiconductor processes or scaling the architecture to higher resolutions. Nevertheless, this research reinforces the viability of GaAs technology for specialized RF applications where high breakdown voltage and substrate isolation are critical.
Summary updated