Low-latency real-time FIR filtering is often realized using partitioned convolution algorithms, which split the filter impulse responses into a sequence of sub filters and process these sub filters efficiently using frequency-domain methods (e.g. FFT-based convolution). Methods which split both, the signal and the filter, into uniformly-sized sub filters define a fundamental class of algorithms, known as uniformly-partitioned convolution techniques. In these methods, both operands, signal and filter are usually partitioned with the same granularity. This contribution introduces uniformly-partitioned algorithms with independent partitions (block lengths) in both operands and regards viable transform sizes resulting from these. The relations of the algorithmic parameters are derived and the performance of the approach is evaluated.
展开▼
