Abstract
As satellite communications evolve, On-Board Processing (OBP) has become critical for enhancing efficiency and performance. However, the digital circuits within OBP transponders are highly susceptible to Single Event Upsets (SEUs) caused by high-energy particles in space. While Triple Modular Redundancy (TMR) is the standard mitigation technique, its high hardware overhead often exceeds the strict area and power constraints of satellite platforms. This paper proposes a cost-effective alternative: a Dual Modules plus Checking module based on Residue Code (DM-CRC) architecture. By implementing a Multi-Sample Checking (MSC) scheme, the authors achieve a fault missing rate as low as 0.05% and an area reduction of over 20% compared to traditional TMR.
The Challenge: Single Event Upsets in Space
Satellites operating in high-radiation environments face constant bombardment from high-energy particles. These particles can trigger Single Event Upsets (SEUs)—random bit flips in digital logic that can destabilize communication systems. For OBP platforms, Finite Impulse Response (FIR) filters are essential components that must be protected from such errors to ensure system stability.
Historically, Triple Modular Redundancy (TMR) has been the go-to solution. By running three identical modules and using a majority voter, TMR provides high reliability. However, it triples the hardware area and power consumption. In the resource-constrained environment of a satellite, this overhead is often prohibitive. The industry has sought "leaner" redundancy methods, such as Dual-With-Check (DWC) systems, to maintain reliability with significantly less hardware.
Proposed Methodology: DM-CRC and MSC
The authors propose a specialized architecture for FIR filters that leverages the mathematical properties of Residue Codes (modulo arithmetic). Because FIR filters are linear systems based on Multiply-Accumulate (MAC) operations, the residue of the final output should match the residue of the operations performed on the residues of the inputs.
The DM-CRC Architecture
The proposed DM-CRC (Dual Module + Checking) architecture replaces the third full module of TMR with a much smaller Checking Module (CRC). This CRC block operates on a small modulus (e.g., mod 7) to verify the outputs of two full-scale filters ($M_1$ and $M_2$).
Multi-Sample Checking (MSC)
A primary challenge with residue checking is the "fault missing" rate. In a Single-Sample Checking (SSC) setup, an error that is a multiple of the modulus (e.g., an error of +14 when using mod 7) will go undetected. To mitigate this, the authors introduced Multi-Sample Checking (MSC). By buffering and verifying $N$ consecutive samples, the system can statistically ensure that hardware faults—which rarely produce congruent errors across multiple different inputs—are detected. If a discrepancy is found between $M_1$ and $M_2$, the MSC logic uses the CRC module over several cycles to identify and output data from the healthy filter.
Experimental Results and Insights
The architecture was validated through FPGA-based fault injections on a Xilinx Virtex-4 XC4VLX100, using a 16-tap FIR filter as the test case.
- Optimal Modulus: The study found that using a modulus of $m = 2^n - 1$ (such as 7 or 15) is far more effective than $m = 2^n$ because it catches bit-flips across all bit positions, including higher-order bits.
- Hardware Efficiency: The MSC-DM-CRC design achieved an area reduction of over 20% in FPGA slices and LUTs compared to TMR when using a modulus of 7 ($n=3$).
- High Reliability: With a checking window of 4 samples ($N=4$) and a modulus of 7, the fault missing rate was reduced to 0.05%, providing near-TMR reliability at a fraction of the cost.
Conclusion and Future Work
The MSC-DM-CRC scheme provides a robust, low-cost alternative to TMR for satellite FIR filters. While highly effective, the current design assumes that SEUs occur infrequently enough to allow for a recovery cycle. Future research aims to extend this residue-checking logic to other digital signal processing structures like Fast Fourier Transforms (FFTs) and Digital Beamforming (DBF) modules, as well as hardening specific "VIP" registers that store critical intermediate samples.
Summary updated