Abstract
Traditional digital systems are constrained by the Nyquist-Shannon sampling theorem, requiring sampling rates at least twice the signal's highest frequency. As bandwidths enter the GHz range, high-speed Analog-to-Digital Converters (ADCs) become prohibitively expensive. This paper proposes Parallel Multi-rate Compressed Sampling (PMCS), a hardware-efficient architecture that utilizes multiple low-speed ADCs with varying sampling rates in parallel branches. By integrating over the entire signal period and employing diverse sub-Nyquist sampling rates, PMCS enhances sample diversity and reconstruction accuracy while mitigating common hardware errors like integrator reset jitter.
Introduction
The rapid expansion of signal bandwidth in modern wireless communications poses a significant challenge for conventional sampling techniques. While Compressed Sensing (CS) allows for the reconstruction of sparse signals from sub-Nyquist samples, practical implementation in Analog-to-Information Converters (AICs) remains difficult. Existing architectures, such as Parallel Segmented AICs, often suffer from timing jitters and errors due to frequent integrator resets. There is a critical need for a robust, parallelized architecture that can handle high-bandwidth signals using accessible, low-speed components.
Methods: The PMCS Architecture
The proposed Parallel Multi-rate Compressed Sampling (PMCS) system addresses these challenges through a parallelized multi-rate approach. The architecture consists of $P$ parallel groups, each processing the input analog signal $x(t)$.
As shown in the figure, each branch mixes the signal with a high-speed chipping sequence $c_p(t)$ and integrates the result over the full signal period $T$. This full-period integration avoids the "reset errors" typical of segmented designs. Following integration, the signal is fed into $I$ parallel ADCs within each group. These ADCs operate at different sub-Nyquist sampling rates ($T_1, T_2, \dots, T_I$). This multi-rate strategy ensures a diverse set of measurements, which are then mapped into a standard CS measurement matrix using a mathematical framework involving permutation matrices. Signal recovery is achieved through $l_1$ norm minimization.
Results and Discussion
The performance of PMCS was evaluated using simulated time-frequency signals with 10 random frequencies within a 100 MHz range (200 MHz Nyquist equivalent). Key findings include:
- Scalability: PMCS matches the performance of Parallel Segmented AICs when using a single ADC per group ($I=1$) but significantly outperforms them as the number of ADCs ($I$) increases.
- Rate Diversity: The selection of sampling rates is crucial. Simulations demonstrated that rates which are not simple multiples of each other provide higher "information gain" and better reconstruction probability. For instance, a configuration with 60 MHz and 70 MHz rates outperformed a 40 MHz and 60 MHz setup due to reduced data redundancy.
- Robustness: The system maintains high detection percentages across various Signal-to-Noise Ratios (SNRs), benefiting from the increased sample diversity provided by the multi-rate parallel structure.
Conclusion
PMCS offers a promising path toward practical sub-Nyquist sampling by leveraging parallelism and rate diversity. By shifting the burden from high-speed hardware to architectural diversity and backend processing, it enables the capture of high-bandwidth signals with lower-cost components. Future work should address the optimization of sampling rate selection and validate the architecture on physical hardware to account for real-world impairments like clock skew and mixer non-linearities.
Summary updated