Thermometer-Coded DFE Selection Element for High-Speed Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current silicon-based decision feedback equalizers (DFE) face challenges in achieving high-speed data rates due to performance limitations, particularly at symbol intervals of 100 picoseconds, and excessive complexity and power requirements in precomputation modules.
Innovation Solution
The implementation of a DFE selection element using thermometer-coded representations and bitwise multiplexers to reduce unrolling requirements, allowing for efficient generation and selection of symbol decisions, with a focus on optimizing the recursive selection unit to operate within the constraints of high-speed data transmission.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If standard textbook implementation of DFE is used, then ISI removal capability is improved, but device complexity and speed performance deteriorate due to cascaded circuit elements requiring completion within 100 ps symbol interval
Solution Approach 1:
The patent segments the DFE operation into precomputation phase and execution phase. The precomputation module generates multiple candidate feedback signals in advance, each corresponding to a possible preceding symbol value. During execution, the correct candidate is selected based on the actual preceding symbol, avoiding the need for complex real-time cascaded calculations while maintaining ISI removal capability.
Solution Approach 2:
The precomputation module performs preliminary generation of feedback signals for multiple possible preceding symbol values before the actual equalization operation. This allows the execution phase to simply select from pre-computed candidates rather than performing complex real-time calculations, reducing the circuit complexity required during the critical symbol interval.
2Device complexity
If precomputation modules are used to reduce DFE complexity, then device complexity is reduced, but power requirements increase excessively constituting dominant fraction of power consumption
Solution Approach 1:
The patent computes feedback signals for multiple possible preceding symbol values (excessive action) during the precomputation phase, but only the correct feedback signal corresponding to the actual preceding symbol is used in the final decision (partial action). This approach reduces the complexity of real-time processing while the power consumption is managed through efficient selection logic that activates only necessary computation paths.
Solution Approach 2:
The precomputation module creates copies of feedback signals for different possible preceding symbol values. These copies are generated in advance and stored or held ready for selection. During execution, only the appropriate copy corresponding to the actual preceding symbol is activated, reducing the active power consumption compared to continuously processing all possible feedback paths.
3Productivity
If higher-order signal constellations like PAM4 are used, then data rate is improved, but susceptibility to ISI and noise increases making symbol detection more difficult
Solution Approach 1:
The DFE uses feedback from previously detected symbols to generate feedback signals that are subtracted from the received signal. This feedback mechanism compensates for ISI effects introduced by higher-order constellations like PAM4, allowing accurate symbol detection even in the presence of dispersion and noise that would otherwise make detection difficult.
Solution Approach 2:
The patent replaces direct real-time calculation of feedback signals with a selection mechanism that chooses from precomputed candidates. This substitution of computation method reduces the processing burden during the critical symbol detection interval, maintaining detection accuracy for higher-order constellations without requiring excessive computational resources.
Data Source
AI summary
A disclosed DFE selection element reduces the degree of unrolling that might otherwise be required. In one illustrative embodiment of a method for converting a receive signal from a communication channel into a sequence of symbol decisions, the method includes, for each sampling interval: (a) generating a set of tentative symbol decisions each having a thermometer-coded representation with a least significant bit and a most significant bit; (b) providing each least significant bit as a thermometer-coded input to a first multiplexer; (c) providing each most significant bit as a thermometer-coded input to a second multiplexer; (d) applying a thermometer-coded representation of a preceding output symbol decision as selection inputs to the first and second multiplexers; and (e) capturing a current output symbol decision having a thermometer-coded representation that includes outputs of the first and second multiplexer.