Blind Equalization in Digital Receivers With Parallel FIR Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional blind equalization techniques in digital receiver systems require a large number of iterations and long data samples to converge, leading to processing delays and inefficiencies, especially in applications with short packet data and dynamic channels, due to the need for sequential recalculation and recycling of channel output samples.
Innovation Solution
A digital receiver system utilizing a finite impulse response (FIR) filter with a channel estimator that calculates second and fourth order expectations from a batch of channel output signals to determine an error function, enabling steepest descent algorithms for faster convergence and parallel processing without sequential recalculation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If stochastic gradient descent with recycled data samples is used for blind equalization, then convergence can be achieved with limited data samples, but processing delays increase due to sequential recalculation and recycling
Solution Approach 1:
The patent segments the equalization process into independent parallel tasks by dividing the recycled data samples into multiple batches, each processed by separate processing units simultaneously. This segmentation eliminates the sequential dependency while maintaining the statistical accuracy benefits of data recycling.
Solution Approach 2:
The patent performs preliminary calculation of gradient components from recycled data samples before the main equalization update. By pre-computing these statistical quantities in parallel, the system avoids sequential recalculation during the critical update phase, reducing overall processing delay.
2Measurement precision
If a large number of iterations are used for steepest descent convergence, then equalization accuracy improves, but convergence time increases leading to long processing latency
Solution Approach 1:
The patent maintains continuous useful action by performing equalization updates on multiple data samples in parallel simultaneously rather than sequentially. This allows the system to process many more effective samples within the same time frame, achieving high accuracy without proportionally increasing convergence time.
Solution Approach 2:
The patent transitions from sequential iteration in time domain to parallel processing across multiple data samples in the batch dimension. By exploiting this additional dimensional space for parallel computation, the system achieves faster effective convergence while maintaining accuracy.
3Measurement precision
If sequential recalculation of equalizer output is performed for each updated parameter, then parameter optimization accuracy improves, but processing efficiency decreases due to idle time
Solution Approach 1:
The patent segments the parameter optimization process into independent gradient computation tasks that can be executed in parallel across multiple processing units. Each unit computes gradients for different data samples independently, eliminating sequential dependencies while preserving optimization accuracy.
Solution Approach 2:
The patent merges multiple gradient computations into a single parallel execution phase where all necessary gradient components are calculated simultaneously from recycled data samples. This consolidation eliminates idle time between sequential operations while maintaining the statistical accuracy of parameter updates.
Data Source
AI summary
A digital receiver system to recover signals from inter-symbol-interference includes a finite impulse response (FIR) filter using convolution to recover signals; and a channel estimator coupled to the FIR filter to estimate FIR coefficients, wherein the channel estimator uses a second order expectation and a fourth order expectation from a convolution to calculate error function.